Nucl - Phys.B v.582

Nuclear Physics B 582 (2000) 318
www.elsevier.nl/locate/npe
Model independent bounds on the tau lepton

electromagnetic and weak magnetic moments
Gabriel A. Gonzlez-Sprinberg a , Arcadi Santamaria a , Jordi Vidal b
a Instituto de Fsica, Facultad de Ciencias, Universidad de la Repblica,
Igu 4225, 11400 Montevideo, Uruguay

b Departament de Fsica Terica, IFIC, CSIC-Universitat de Valncia,
E-46100 Burjassot, Valncia, Spain

Received 21 February 2000; revised 19 April 2000; accepted 1 May 2000
Abstract
Using LEP1, SLD and LEP2 data, for tau lepton production, and data from pp colliders and LEP2,
for W decays into tau leptons, we set model independent limits on non-standard electromagnetic and
weak magnetic moments of the tau lepton. The most general effective Lagrangian giving rise to
tau moments is used without further assumptions. Precise bounds (2 ) on the non-standard model
contributions to tau electromagnetic (0.007 < a < 0.005), tau Z-magnetic (0.0024 < aZ <
0.0025) and tau W -magnetic (0.003 < W < 0.004) dipole moments are set from the analysis.
2000 Elsevier Science B.V. All rights reserved.
PACS: 13.40.Em; 14.60.Fg; 13.10.+q
Keywords: Tau lepton; Electromagnetic and weak dipole moments; Effective Lagrangian; Bounds
1. Introduction
The present best bound on the tau lepton magnetic moment (a ) is indirect [1,2]. It
comes from the observation that in general extensions of the standard model (SM) it is
very difficult to generate a magnetic moment for a lepton without originating a coupling
of the Z boson to the lepton of the same order of magnitude. This anomalous Z coupling
(aZ ) is strongly bounded by LEP1, therefore, by assuming that the magnetic moment of
the lepton (a ) has the same size, one obtains a rather strong bound on it.
While this argument is plausible, the complete amount of data coming from tau-lepton
production at LEP1, SLD and LEP2, and data on W decays into tau leptons from LEP2
and pp colliders, allows for a more complete analysis of the magnetic moment couplings
of the tau to the photon, the Z and the W bosons.
Following Refs. [1,3], in order to analyze tau magnetic moments, we will use an effective
Lagrangian description [46]. Thus, in Section 2 we describe the effective Lagrangian we
0550-3213/00/$ see front matter 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 7 5 - 3
G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318
use and fix the notation. LEP1 and SLD are sensitive only to the Z magnetic moment of the
tau, however LEP2 is sensitive to both Z and photon tau magnetic moments. Far from the
resonance statistics decreases dramatically and the precision is not as good as the precision
obtained at LEP1. However, since magnetic moment couplings are non-renormalizable
couplings, their effects grow with energy and, therefore, LEP2 limits can still be relevant.
This is especially true for electromagnetic couplings for which LEP1 does not give much
information. That is also the reason why experiments performed at lower energies do not
provide, in general, stringent bounds (see for instance the bounds obtained from tau decays
[79]). To obtain relevant bounds at lower energies the suppression factor, (Elow /mZ )2 , has
to be compensated by higher precision in the experiment.
Tau magnetic moment contributions to tau production in LEP1-SLD and LEP2 are
studied in Section 3. In this respect we can classify the observables in three classes:
(i) universality test at LEP1, which are studied in Section 3.2, (ii) total production rates
at LEP2, considered in Section 3.3, and (iii) transverse polarization asymmetries, studied
in Section 3.4.
Tau magnetic moments flip chirality and in the standard model the only source of
chirality flips are fermion masses. This means that any contribution of magnetic moments
(weak or electromagnetic) to total rates are either suppressed by the fermion mass (relative
to the electroweak scale) or need two operator insertions and then, they come as the square
of the magnetic moments. In addition any new physics, not only that related with magnetic
moments, will appear in total rates. Hence, in order to study magnetic moments it is
interesting to look for observables that are exactly zero when chirality is conserved. Those
observables will only be sensitive to fermion masses and to magnetic moments. In addition
they will only depend linearly on magnetic moments. Some transverse tau polarization
asymmetries [1012] are observables of this type and they have been already measured
at LEP1 [13] and SLD [14]. We will show in Section 3.4 that they already give now the
best and cleanest bounds on tau magnetic moments. It would be interesting to study these
observables also at LEP2 to disentangle weak an electromagnetic magnetic moments.
Gauge invariance, after the spontaneous symmetry breaking, relates the magnetic
moment couplings of the tau to the Z, the photon, and the W gauge bosons. Therefore
one can also gain some insight on tau magnetic moments by studying W decays into tau
leptons. There are already rather good bounds on the universality of the leptonic decays
of the W coming from LEP2, UA1, UA2, CDF and D0, those are studied in Section 4.
In Section 5 we combine all bounds and obtain the best limits on the different magnetic
moment couplings. Finally in Section 6 we discuss the obtained results and compare with
other bounds found in the literature.
2. The effective Lagrangian for tau magnetic moments

The standard model gives a very good description of all physics at energies available at
present accelerators. Therefore, one expects that any deviation from the standard model,
at present energies, can be parametrized by an effective Lagrangian built with the standard
model particle spectrum, having as zero order term just the standard model Lagrangian,
and containing higher dimension gauge invariant operators suppressed by the scale of new
physics . The leading non-standard effects will come from the operators with the lowest
dimension. Those are dimension six operators. There are only two operators of this type
contributing to the tau magnetic moments:
OB =
g0
LL R B ,
22
(2.1)
OW =
g
E .
LL E R W
22
(2.2)
and
E are
Here LL = (L , L ) is the tau leptonic doublet, is the Higgs doublet, B and W
0
the U(1)Y and SU(2)L field strength tensors, and g and g are the gauge couplings.
In principle, one could also write dimension six operators like
g0
LL DL
/ L B ,
2
(2.3)
/ = D . However, these operators reduce to the operator Eq. (2.1) after using the
with D
standard model equations of motion. In doing so, the couplings will be proportional to the
tau-lepton Yukawa couplings. Note that operators in Eqs. (2.1) and (2.2) break chirality
while the operator (2.3) does not break it. Therefore, by using this last form we would
implicitly assume that the only source of chirality breaking are Yukawa couplings and that
any chirality breaking, including magnetic moments, should be proportional to fermion
masses. In order to be more general we will assume the forms in Eqs. (2.1) and (2.2),
having in mind that we are introducing in the standard model an additional source of
chirality breaking in addition to fermion masses. As we will see in the next sections this will
become very important in looking for the right observables sensitive to magnetic moments.
Thus, we write our effective Lagrangian as,
Leff = B OB + W OW + h.c.,
(2.4)
where, for simplicity, we will take the couplings B and W real. Note that complex
couplings will break CP conservation and lead to electric dipole moments.
If these operators come from a renormalizable theory, in the perturbative regime one
expects, in general, that they arise only at one loop and therefore their contribution will
be further suppressed. However, this does not need to be the case, therefore we leave the
couplings B and W as free parameters without any further assumption.
After spontaneous
symmetry
p breaking, the Higgs gets a vacuum expectation value
2 GF = 246 GeV, and the interactions (2.4) can be written

h 0 i = u/ 2 with u = 1/
in terms of the gauge boson mass eigenstates, A and Z , using that
B = sW Z + cW A ,
W3 = cW Z + sW A ,
1

W + = W1 iW2 ,
2
(2.5)
where, as usual, we define cW = cos W , sW = sin W , tan W = g 0 /g and e = gsW . Thus,

our Lagrangian, written in terms of the mass eigenstates, is 1
e
e
Leff =
F + Z
Z
2mZ
2mZ sW cW

e
+ W
L R W+ + h.c. ,
(2.6)
2mZ sW
where F is the electromagnetic field strength tensor, Z = Z Z and W+ =

W+ W+ are the corresponding strength tensors for the Z and W gauge bosons.
We have not written the non-abelian couplings involving more than one gauge boson
because they are not relevant to our purposes. We have normalized all couplings to mZ ,
the natural mass-scale at the energies we will consider, and we have defined the following
dimensionless couplings
umZ
,
(2.7)
= (B W )
22
umZ
2
2
+ B sW
,
(2.8)
Z = W cW
22

umZ
2
.
(2.9)
W = W 2 = 2 Z + sW
The Lagrangian (2.6) gives additional contributions to the electromagnetic moment of the
tau, which usually is expressed in terms of the parameter a . Similar parameters have been
introduced in the literature for the corresponding tau weak magnetic moment couplings to
the Z-boson, aZ [1012] and the W -boson, W [7]. They can be expressed in terms of our
s as follows,
2 m
,
(2.10)
mZ
2 m 1
Z ,
(2.11)
aZ =
mZ sW cW
2 m
W .
(2.12)
W = 2
mZ
Notice that, in the effective Lagrangian approach, exactly the same couplings that
contribute to processes at high energies also contribute to the magnetic moment form
factors, F new (q 2 ), at q 2 = 0. The difference F new (q 2 ) F new (0) only comes from higher
dimension operators whose effect is suppressed by powers of q 2 /2 , as long as q 2
as needed for the consistence of the effective Lagrangian approach.
In (2.6) we have all type of magnetic moment couplings except neutrinoneutrino
couplings. This is due to the fact that we have not included right-handed neutrinos, R , in
the particle spectrum. If we include them we can add two additional operators that will give
magnetic moment couplings of the neutrinos to the Z, and additional contributions to the
W + magnetic moment couplings. However they will also give rise to an electromagnetic
moment for the neutrinos which is extremely well bounded from a variety of sources,
a =
1 Similar results are found when using a non-linear realization of the gauge symmetry, as can be seen in Ref. [3].
energy loss in red giants, supernova cooling, etc., so we are not going to consider them in
our calculation.
Since the effect of the operators in (2.6) is suppressed at low energies the most interesting
bounds will come from the highest precision experiments at the highest available energies.
Presently this means LEP1 and SLD (Z decay rates and polarization asymmetries), LEP2
(cross sections and W decays rates), CDF and D0 (W decay rates). Consequently in the
following we will study all those observables.
3. e+ e + in presence of electromagnetic and weak magnetic moments
In this section we will consider e+ e + collision in a range of energies from
threshold to LEP2, therefore we will include both photon and Z-exchange with standard
model vector and axial couplings to fermions, plus additional magnetic moment couplings
given by Eq. (2.6). The amplitude for the process can be written as:
X

v(p
2 , s2 ) vek aek 5 u(p1 , s1 )Pk
M = i e2
k= ,Z,...

i
u(p
3 , s3 ) vk ak 5 g k
q v(p4 , s4 ),
2m
(3.1)
where the momenta p1 , p2 , p3 , p4 , correspond to e , e+ , , + respectively, q = p1 +

p2 = p3 + p4 , and Pk are the propagators of the different gauge bosons that contribute to
the process:
1
1
1
PZ =
.
(3.2)
P = 2 ,
2
2
2
q
(2sW cW ) q mZ + imZ Z
After squaring, summing and averaging over initial polarizations of electrons we obtain
X

M2 = 1 Re V V + A A q 2 g q q + pi pi
2

(3.3)
+ Im V A + A V i q pi ,
where pi = p2 p1 , pf = p4 p3 , and V, A carry all the coupling constants, final spin
and momentum dependences. After some algebra one finds:

X

1 k
k
k
k
3 , s3 )
ve
g pf Pk v(p4 , s4 ),
v g a 5
V = eu(p
2m
k= ,Z

X

1 k
3 , s3 )
aei ak g k vk 5
g pf Pk v(p4 , s4 ). (3.4)
A = eu(p
2m
k= ,Z
Here the couplings are:
ve = v = 1,
ae = a = 0,
1
2
,
veZ ve = vZ v = + 2sW
2
m
4m
,
gZ =
Z .
g = 2
mZ
mZ
(3.5)
1
aeZ ae = aZ a = ,
2
(3.6)
(3.7)
3.1. Total rates and cross sections

If we are not interested in final polarization we can sum over the polarizations of the
tau-leptons and obtain the angular distribution. This is usually written in the following
notation:

1
1
d 0
= 0 (s) 1 + cos2 + m (s) sin2 + FB (s) cos ,
d cos
2
2
(3.8)
where the coefficients i are given by:

0 (s) =

2 n
1 + 2v ve Re{} + ae2 + ve2 v2 + a2 2 ||2
s

+ 4rZ 1 + v ve Re{} + 4rZ2 2

8rZ ve Re{} + v ve2 + ae2 ||2 Z
(3.9)
o

+ (4rZ )2 ve2 + ae2 ||2 Z2 (4rZ )2 ve Re{} Z ,
2

4m
2
1 + 2v ve Re{} + v2 ve2 + ae2 ||2

(s) =
s
s

s
+ 4rZ 1 + v ve Re{} + 2 2
mZ
m

8rZ v ve2 + ae2 ||2 + ve Re{} Z

2 2
4s
4s 2
2
+ 2 ve + ae || Z 2 v Re{}Z ,
mZ
mZ
n
2

2
2a ae Re{} + 2v ve ||2
FB (s) =
s
o
+ 4rZ a ae Re{} 16rZ a ae ve ||2 Z ,
with
(3.10)
(3.11)
s
m
,
rZ =
mZ
= (s) =
= (s) =
4m2
,
s2
s
1
,
2
2
(2sW cW ) s mZ + iZ mZ
s = q 2 , and being the angle between the linear momentum of the e and the .
In the limit of massless fermions ( 1) all linear terms in the dipole moments
disappear and the coefficients i simplify to:
0 (s)|m 0 =

2
1 + 2v ve Re{} + ae2 + ve2 v2 + a2 ||2 ,
s
(3.12)
m (s)|m 0 =

2 2
2 s
4s 2
2
2
+
+
a

4v
Re{}

, (3.13)
v
||
e
e
Z
s
m2Z
m2Z
FB (s)|m 0 =

2
2a ae Re{} + 2v ve ||2 .
s
(3.14)
It is worth noticing that m (s), the term in sin2 , is proportional to the tau mass in the
standard model and, therefore, it vanishes for massless taus. However, in the presence of
magnetic moment couplings this term remains, even in the massless limit, and in fact it
carries all the information about the magnetic moments Z , . This contributionpeaks in
the angular region where the standard model contribution 12 (1 + cos2 ) 0 |m 0 reaches
its minimum.
By integrating out the angle in Eq. (3.8) we obtain the total cross section:
4
2
(3.15)
e+ e + = 0 (s) + m (s).
3
3
From Eqs. (3.9)(3.14) we find that there are basically three types of contributions: (i) the
standard model contribution 0 |m 0 , which is the dominant one, (ii) a contribution which
is proportional to the tau mass. This comes together with an insertion of the magnetic
moment operators (non-standard contribution to m and 0 ) or with an insertion of another
fermion mass (standard model contributions to m ) or both, two insertions of magnetic
moment operators and two mass insertions (non-standard contributions to 0 ), and (iii) a
contribution free of masses but with two insertions of the magnetic moment operators
(in m ).
This can be easily understood, since standard model couplings of gauge bosons to
fermions conserve chirality, while mass terms and magnetic moment couplings change
it, therefore interference of magnetic moment contributions with standard ones should
be proportional to the fermion masses and only the square of magnetic moments can be
independent of fermion masses. In the limit of zero tau mass only the contribution (iii) is
relevant, however there could be some range of the parameters in which contribution (ii) is
higher than (iii). In fact for any finite value of the tau lepton mass it is obvious that for
large enough (ii) will always dominate over (iii). In order to be as general as possible
we will include all three contributions.
3.2. Bounds from e+ e + at LEP1 and SLD
Bounds on the new couplings and Z can be obtained from LEP1-SLD universality
tests by assuming that only the tau lepton has anomalous magnetic moments (muon and
electron electromagnetic moments have been measured quite precisely [2]). In order to
compare with experimental data it is convenient to define the universality ratio:

(e+ e + )

R
=
=
RSM + R1 + R2 .
(3.16)
R =

+
(e e ) s=m2
R
Z
Here R had / and R = had / are the quantities measured directly [15], on the
other hand, RSM is the standard model contribution (including lepton-mass corrections),
10
and R1 and R2 are the linear and quadratic terms, respectively, in the tensor couplings. In
order to get bounds on the anomalous couplings, the theoretical expression for R can be
easily computed from Eqs. (3.8)(3.14),

q
2
2
2 v 2a
2
,
RSM = 1 4rZ 1 + 2rZ 2
v + a2
q
v
Z ,
(3.17)
R1 = 12 1 4rZ2 rZ 2
v + a2
q

1
2 ,
R2 = 2 1 4rZ2 1 + 8rZ2 2
v + a2 Z
with a ae = a = a , and v ve = v = v . Notice that in the ratio Eq. (3.16)
electroweak radiative corrections cancel to large extend and, therefore, we can use treelevel formulae. However, if needed, the expressions in Eq. (3.17) can be improved by
using effective couplings [16]. Combining the very precise experimental LEP1 and SLD
measurements [15],
R = 20.786 0.033,
R = 20.764 0.045,
we obtain
R = 1.0011 0.0027.
(3.18)
Comparing (3.18) with (3.16) and (3.17) one gets the condition for the Z coupling to
be
0.0007 6 7.967 Z2 + 0.037 Z 6 0.0061.
(3.19)
This equation leads to the following bounds on Z :

0.030 6 Z 6 0.012 or 0.007 6 Z 6 0.025,
(3.20)
so that, at 1 , the zero value for Z is excluded. This is so because the measured value of
R = 1.0011, given in Eq. (3.18), excludes the SM (at 1 ) due to the fact that the SM
mass correction (rZ ) to R = 1 is negative while the measured value is larger than one.
At 2 the effect disappears.
It must be noticed that, in Eq. (3.19), the linear term in Z is strongly suppressed by the
mass insertion rZ necessary to get a chirality even contribution to the observable and
also by the vector coupling v ( 41 effect), so that these bounds on Z come almost entirely
from the quadratic term in the coupling. Note also that, as expected from Eq. (3.17), no
bound is found on the -coupling , due to the Z dominance at the Z-peak.
3.3. Bounds from e+ e + at LEP2
The situation is quite different at LEP2, where contributions coming from the photonexchange are dominant over those coming from the Z-exchange. This is easily seen from
the expression of the e+ e + cross section given in Eq. (3.9) to Eq. (3.11).
Present limits from e+ e + cross section are much milder at LEP2 [1719].
A combination of the LEP2 data on this cross section, for a value of s 0 (the invariant mass
11
Table 1
Combined experimental data for the + cross section from ALEPH, DELPHI, L3 and OPAL at
LEP2 energies. s 0 /s is the cut in the invariant mass of the tau pair
s (GeV)
SM
(pb)
(pb)
182.7
188.6
3.45
3.21
3.43 0.18
3.135 0.102
s 0 /s >
0.85
0.85
Collaboration
LEP Electroweak
Working Group [15]
Table 2
Coefficients of the anomalous contributions to R , for the different center of mass measured
energies at LEP2
s (GeV)
130
136
161
172
183
189
F1
F2
F1Z
F2Z
FZ
0.079
0.083
0.092
0.094
0.096
0.096
0.682
0.784
1.221
1.427
1.642
1.765
0.028
0.026
0.022
0.021
0.020
0.019
5.258
5.152
5.272
5.497
5.789
5.971
0.286
0.304
0.384
0.424
0.467
0.491
p
of the pair of tau leptons) so that s 0 /s > 0.85, is listed in Table 1. This combination of
data has been only made for the 183 GeV and 189 GeV data-sets as they have the highest
luminosities and center-of-mass energies. For comparison we also present the standard
model prediction for the cross section. In both, experimental results and standard model
predictions, initial-final state radiation photon interference is subtracted.
For LEP2 let us define the ratio R as:
(e+ e + )
(e+ e + )SM
= 1 + F1 (s) + F2 (s) 2 + F1Z (s) Z + F2Z (s) Z2 + F Z (s) Z .
(3.21)
For the range of energies used by LEP2 experiments, the coefficients Fi (s), obtained from
Eqs. (3.8)(3.11), are given in Table 2. Direct comparison of Eq. (3.21) with experimental
data will provide bounds on anomalous couplings. We have checked that, even though
coefficients in Table 2 are obtained with no initial state radiation, its inclusion only changes
the coefficients by about a 10% and this does not affect significantly the obtained bounds.
In order to see how well the new couplings can be bound from LEP2 let us find the limits
on obtained by using only the data at 189 GeV (all data are used independently in the
global fit discussed in Section 5). The experimental value for the ratio R is:

(3.22)
R exp = 0.978 0.032,
which must be compared with our theoretical prediction (assuming that Z is well bounded
from LEP1-SLD, as it is)

(3.23)
R th = 1.00 + 1.765 2 + 0.096 .
12
From the two previous equations it is easy to find the following 1 bound:
0.10 < < 0.05,
(3.24)
which is comparable to the one obtained in the final global fit given in Section 5, where all
available data have been included.
3.4. Tau lepton transverse polarization asymmetry
Although magnetic moments change chirality, total cross sections are chirality even
observables. Thus, in the limit of massless taus, magnetic moment contributions to cross
sections come always squared. In addition all kind of new chirality even physics will also
contribute to total cross sections. Therefore, observables that vanish for massless taus are
superior because they depend linearly on magnetic moments and therefore they are more
sensitive to them. On the other hand they will not get contributions from physics conserving
chirality. In that sense they are truly magnetic moment observables.
At LEP1 [13], with the direction fully reconstructed in the semi-leptonic decays
e+ e + h+
1 X h2 , h1 h2 X (h1 , h2 = , ), it has been shown [1012] that
one can get relevant information about the anomalous weak magnetic moment just by
measuring the following azimuthal asymmetry of the -decay products:
A
cc =
(+) ()
cc
cc
,
cc
(+) + cc
()
is defined in the following angular regions

where cc

(+) = cos > 0, cos h > 0 + cos < 0, cos h < 0 ,

cc

() = cos > 0, cos h < 0 + cos < 0, cos h > 0 .

cc
(3.25)
(3.26)
(3.27)
In the 1 limit, the expression that one finds for the proposed asymmetry is:
A
cc = h
a
1
[vrZ + Z ],
2
2 v + a2
(3.28)
where h = (m2 2m2h )/(m2 + 2m2h ), is the polarization analyzer for each hadron channel
(h = , ), a ae = a , and v ve = v . The formula shows that A
cc , being sensitive to
the transverse polarization of the lepton, selects the leading contribution in the anomalous
weak magnetic coupling Z of the tau. In addition, the SM contribution to the observable
is doubly suppressed with respect to the non-standard one: by the fermion vector coupling
2 ) and by the r ( m /m ) factor.
v ( 12 + 2sW
Z
Z
Within 1 , the LEP1 [13] measurement of this asymmetry and the SLD [14] determination of the transverse tau polarization, translate in the following values for the Z
coupling

(0.0 1.7 2.4) 102 (LEP1),
(3.29)
Z =
(0.28 1.07 0.81) 102 (SLD).
Combining these results one gets the bound:
Z = 0.002 0.012.
(3.30)
13
Note that even though the transverse tau polarization has been measured at LEP1-SLD
with a precision one order of magnitude worse than the universality test R (24%
typically for the asymmetry, and 0.5% for the tau-muon cross section ratio), the obtained
bound Eq. (3.30) is as good as the one coming from universality Eq. (3.20). This is
so because the asymmetry depends linearly on the couplings. All the other observables
depend mainly quadratically on the s, therefore, if the new-physics contributions to
magnetic moments are ever found to be different from zero, the asymmetry will be the
only observable able to disentangle the sign of the couplings. In addition, as commented
before, the asymmetry A
cc is also qualitatively a better observable since it is independent
on any physics that does not break chirality.
At present we do not know any similar measurement at LEP2. However we think that
this measurement will be very interesting since it will allow us to disentangle cleanly the
components from the Z components of the magnetic moments.
4. Lepton universality in W
From Eq. (2.6) we see that the same couplings that give rise to tau electromagnetic and
Z-boson magnetic moments, also contribute to the couplings of the W gauge bosons to
tau leptons. The couplings appear in a different combination than that in the photon or
Z couplings, so their study gives us an additional independent information on magnetic
moment couplings. As was noticed in Ref. [7] the best place to look for effects of the W
coupling is in the W decay widths.
Using our effective Lagrangian we can easily compute the ratio of the decay width of
the W -gauge boson in tau-leptons (with magnetic moments) to the decay width of the W
to electrons (without magnetic moments).
(W )
(W e)

2
r2
2 3
2
W ,
= 1 rW
1 + W + 3 2 rW cW W + 1 + 2rW
2
RWe
(4.1)
where rW = m /mW , and W can be rewritten in terms of and Z as in Eq. (2.9). Similar
2 )
expression to Eq. (4.1) was given in Ref. [7] but we found and extra global (1 rW
missing, and a factor 3 instead of a (3/2) factor (in our notation) in the term linear in the
coupling. Note that RWe , like the cross sections studied in Section 3.1, is a chirality even
observable. Therefore, in the limit of massless taus, the only contribution from magnetic
moments comes squared.
The decay of the W into leptons has been measured to a rather good precision at LEP2,
UA1, UA2, CDF and D0. There, results are presented in the form of universality tests on
the couplings (for a review on tau lepton universality tests see Ref. [20])
g
= 0.987 0.025 (Colliders) [21,22],
(4.2)
ge
g
= 1.010 0.019 (LEP2) [15].
(4.3)
ge
14
We combine these results and rewrite them as a measurement on the ratio RWe defined
above
RWe = 1.002 0.030,
(4.4)
where we have assumed that the small effect (0.12%) of the tau lepton mass has been
subtracted to obtain the results shown in Eqs. (4.2) to (4.3).
Then, from Eqs. (4.4) and (4.1), we obtain the following limit on the W -boson magnetic
moments.
0.23 6 W 6 0.15.
(4.5)
5. Combined limits on electromagnetic and weak magnetic moments of the tau

lepton
We have performed a global fit, as a function of the two independent couplings and
Z , to the following studied observables:
(e+ e + )
Ratio of cross sections R = (e
+ e + ) (Eq. (3.16), Lepton Universality), at
LEP1 and SLD (Eq. (3.18));
+ + )
(Eq. (3.21)), for the two highest
Ratio of cross sections R (e(e+ ee
+ )
SM
energies measured at LEP2 (Table 1);
Transverse tau polarization and the tau polarization asymmetry
A
cc =
(+) ()
cc
cc
cc
(+) + cc
()
(Eq. (3.25)) measured at SLD and LEP1 (Eq. (3.30));

Ratios of decay widths of W -gauge bosons
RWe
(W )
(W e)
(Eq. (4.1)) measured at LEP2 and pp colliders (Eq. (4.4)).

In Fig. 1 we present, in the plane Z (or a aZ ) the allowed region of parameters at 1
and 2 . For comparison we also present (at 1 ) the relevant limits set independently by
the different observables, as discussed in the text. By projecting onto the axes one can read
off the 1 and 2 limits on the different couplings

0.12 < < 0.06,
(5.1)
(1 )
0.0072 < Z < 0.021,

0.18 < < 0.12,
(5.2)
(2 )
0.026 < Z < 0.027.
At 1 the allowed region is far from elliptic. This is because the dominant quadratic
dependence of the observables on the parameters and Z gives a tendency to provide
two symmetric zones around two different minima. This is especially true for the bounds
coming from universality tests at the Z-peak. In this situation the interpretation of contours
as 68% CL contours is not clear. Probably one can combine the two minima and obtain
15
Fig. 1. Global fit including all constraints discussed in the paper. 95% CL and 68% CL contours are
shown. The bands between straight lines show the allowed regions (1 ) coming from the different
experiments: solid (LEP2-189 GeV), dashed (LEP1-SLD cross section), dot-dashed (asymmetry).
2 (dotted line). This relationship appears when only the
We also have plotted the line Z = sW
operator OB contributes.
more stringent bounds. However in order to be conservative we just quote the maximum
allowed region. In any case, 2 contours do not show this effect and we think they can be
used reliably to obtain 95% CL limits.
Using the relationship among , Z , B and W at a given value of the scale of new
physics, one can easily obtain bounds on B and W . Alternatively, by assuming that
B /(4) or W /(4) are order unity one can obtain bounds on the scale of new physics
( > 9 TeV).
Finally the limits on and Z can be immediately translated into limits on the nonstandard contributions to the anomalous electromagnetic and weak magnetic moments a ,
aZ just by using Eq. (2.12). Thus we have:

0.005 < a < 0.002,
(5.3)
(1 )
0.0007 < aZ < 0.0019,

0.007 < a < 0.005,
(5.4)
(2 )
0.0024 < aZ < 0.0025.
For a these limits are only about one order of magnitude larger than the standard model
contribution, aSM 0.00117.
16
6. Discussion and conclusions

The above bounds are completely model independent and no assumption has been made
on the relative size of couplings B and W in the effective Lagrangian (2.4). For the
sake of comparison with published data [1] we present now the limits that can be found
by considering separately only operator OB or only operator OW in the Lagrangian (2.4).
Consider that only OB is present, as in Ref. [1], is equivalent to impose the relation Z =
2 . Thus, from Fig. 1, it is straightforward to obtain that the bounds on the anomalous
sW
magnetic moment (at 2 ) are reduced to 0.004 < a < 0.003, while little change is found
on the weak-magnetic moment 0.0019 < aZ < 0.0024. Universality tests in W decays do
not provide any interesting constraint on the couplings Z and . However, because of the
relationship (2.9), the LEP1-SLD and LEP2 constraints on Z and can be translated into
constraints on W (or W defined in Eq. (2.12)), the weak magnetic moment couplings of
the W -gauge-boson to taus and neutrinos. In Fig. 2 we present, in the plane W (or
a W ), the 1 and 2 regions obtained from our global fit to all data. Clearly LEP1-SLD
limits on Z coming from Z-decays and the asymmetry and LEP2 limits on constrain W
very strongly. For comparison we also plot, as straight lines, the 1 limits we have obtained
from the universality tests in W decays.
From Fig. 2, one immediately obtains the 95% CL limits
Fig. 2. The global fit of Fig. 1 is now plotted in the plane and W to show the combined bounds
on W . As in Fig. 1 95% CL and 68% CL contours are shown. For comparison we also draw as
straight lines the direct 1 bounds obtained from universality tests in W decays.
0.06 < W < 0.07,
or equivalently
0.003 < W < 0.004.
17
(6.1)
Bounds on the anomalous electromagnetic moment of the can also be obtained from the
radiative decay Z + at LEP1 [2325]. There, only the anomalous coupling a is
taken into account, while the contributions coming from the tau Z-magnetic coupling aZ
are neglected. Using this approach, with the inclusion of linear terms in a in the cross
section [24,25], and taking into account that the which emits the photon is off-shell, the
analysis of the L3 [26] and OPAL [27] collaborations lead to the limits (at 95% CL):
0.056 < a < 0.044 [L3],
(6.2)
0.068 < ha i < 0.065 [OPAL].
(6.3)
As can be seen from Eq. (5.4) our result, coming mainly from LEP2, is about one order
of magnitude better than the ones obtained from the radiative Z-decay. In both cases
some of the particles in the vertex are off-shell. The interpretation of off-shell form
factors is problematic since they can hardly be isolated from other contributions and gauge
invariance can be a problem. In the effective Lagrangian approach all those problems are
solved because form factors are directly related to couplings in the effective Lagrangian,
which is gauge invariant, and as discussed in Section 2, the difference F new (q 2 ) F new (0)
only comes from higher dimension operators whose effect is suppressed by q 2 /2 .
Concluding, we have shown that the use of all available data at the highest available
energies (LEP1, SLD, LEP2, D0, CDF) allowed us to strongly constrain all the magnetic
moments (weak and electromagnetic) of the tau lepton without making any assumption
about naturalness or fine tuning. The obtained bounds (Eqs. (5.4) and (6.1)), to our
knowledge, are the best bounds that one can find in published data.
Acknowledgements
This work has been supported by CICYT under the Grant AEN-99-0692, by DGESIC
under the Grant PB97-1261, by the DGEUI of the Generalitat Valenciana under the Grant
GV98-01-80, by Agencia Espaola de Cooperacin Internacional and by CSIC-Uruguay.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
R. Escribano, E. Mass, Phys. Lett. B 301 (1993) 419; Phys. Lett. B 395 (1997) 369.
C. Caso et al., Eur. Phys. J. C 3 (1998) 1.
R. Escribano, E. Mass, Nucl. Phys. B 429 (1994) 19.
W. Buchmuller, D. Wyler, Nucl. Phys. B 268 (1986) 621.
C.N. Leung, S.T. Love, S. Rao, Zeit. Physik C 31 (1986) 433.
M. Bilenky, A. Santamaria, Nucl. Phys. B 420 (1994) 47.
T.G. Rizzo, Phys. Rev. D 56 (1997) 3074.
M.T. Dova, J. Swain, L. Taylor, Phys. Rev. D 58 (1998) 015005.
M.T. Dova, P. Lacentre, J. Swain, L. Taylor, JHEP 9907 (1999) 026.
J. Bernabu, G. Gonzlez-Sprinberg, J. Vidal, Phys. Lett. B 326 (1994) 168.
J. Bernabu, G. Gonzlez-Sprinberg, M. Tung, J. Vidal, Nucl. Phys. B 436 (1995) 474.
18
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
J. Vidal, J. Bernabu, G. Gonzlez-Sprinberg, Nucl. Phys. B (Proc. Suppl.) 76 (1999) 221.

L3 Collaboration, M. Acciarri et al., Phys. Lett. B 426 (1998) 207.
SLD Collaboration, Kenji Abe et al., SLAC-PUB-8163.
LEP Collaborations ALEPH, DELPHI, L3, OPAL, LEP Electroweak Groups, SLD Heavy
Flavour and Electroweak Groups, CERN-EP/2000-016.
A.A. Akhundov, J. Bernabu, D. Gmez-Dumm, A. Santamaria, Nucl. Phys. B 563 (1999) 82.
OPAL Collaboration, G. Abbiendi et al., Eur. Phys. J. C 6 (1999) 1; CERN-EP/99-097; hepex/9908008.
ALEPH Collaboration, R. Barate et al., Eur. Phys. J. C 12 (2000) 183.
L3 Collaboration, M. Acciarri et al., CERN-EP/99-181; hep-ex/0002034.
A. Pich, hep-ph/9912294.
S. Protopopescu, in: Proceedings of the Fifth International Workshop on Tau Lepton Physics.
Santander, Spain, 1998; Nucl. Phys. B (Proc. Suppl.) 76 (1999) 91.
J. Ellison, in: Proc. EPS-HEP99 Conference, Tampere, 1999; hep-ex/9910037.
J.A. Grifols, A. Mndez, Phys. Lett. B 255 (1991) 611.
J. Biebel, T. Riemann, Zeit. Physik C 76 (1997) 53.
S.S. Gau et al., Nucl. Phys. B 523 (1998) 439.
L3 Collaboration, M. Acciarri et al., Phys. Lett. B 434 (1998) 169.
OPAL Collaboration, K. Ackerstaff et al., Phys. Lett. B 431 (1998) 188.

NLO corrections to the BFKL equation in QCD

and in supersymmetric gauge theories
A.V. Kotikov a,1 , L.N. Lipatov b,2
a II Institut fr Theoretische Physik, Universitt Hamburg, Luruper Chaussee 149, 22761 Hamburg, Germany
b St. Petersburg State University and Petersburg Nuclear Physics Institute, Gatchina, Orlova Roscha,
188300 St. Petersburg, Russia

Received 3 April 2000; accepted 23 May 2000
Abstract
We study next-to-leading corrections to the integral kernel of the BFKL equation for high-energy
cross-sections in QCD and in supersymmetric gauge theories. The eigenvalue of the BFKL kernel
is calculated in an analytic form as a function of the anomalous dimension of the local gaugeinvariant operators and their conformal spin n. For the case of an extended N = 4 SUSY, the kernel
is significantly simplified. In particular, the terms nonanalytic in n are canceled. We discuss the
relation between the DGLAP and BFKL equations in the N = 4 model. 2000 Elsevier Science
B.V. All rights reserved.
PACS: 12.38.Bx
1. Introduction
The BalitskyFadinKuraevLipatov (BFKL) equation [13] is used now together with
the DokshitzerGribovLipatovAltarelliParisi (DGLAP) equation [47] for a theoretical
description of structure functions of the deep-inelastic ep scattering at small values of the
Bjorken variable x. In this kinematical region the structure functions are measured by
the H1 and ZEUS collaborations [8,9] at DESY. For the DGLAP equation the radiative
corrections to the splitting kernels are well known [10,11]. Although the BFKL equation
in the leading logarithmic approximation (LLA) was constructed many years ago, the
calculation of the next-to-leading corrections to its kernel was started only in 1989 [12] and
completed recently [13,14]. The total cross-section for the colourless particle scattering is
q ,
q 0 ) of the BFKL equation for the t-channel
expressed in terms of the solution G (
1 Alexander von Humboldt fellow. On leave of absence from the Particle Physics Laboratory of the Joint
Institute for Nuclear Researches, 141980 Dubna, Russia.
2 Supported by the INTAS grant 97-31696 and by the Deutsche Forschungsgemeinschaft.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 2 9 - 1
20
A.V. Kotikov, L.N. Lipatov / Nuclear Physics B 582 (2000) 1943
partial wave fj (
q
q 0 ; t) (j = 1 + , t = 0) describing the scattering of the reggeized
gluons having their initial and final transverse momenta

q and
q 0 , correspondingly.
For unpolarized colourless particle collisions one can average the BFKL kernel over
the angle between the two-dimensional vectors

q and
q 0 . The eigenvalue of the averaged
kernel was calculated in [13]. The high energy behaviour of the total cross-sections s 1
is related with the pomeron intercept 1. It turns out, that in MS-renormalization scheme
the NLO corrections to 1 are negative and big. However, with the use of non-abelian
renormalization schemes and the BLM procedure for the optimal scale setting it was
shown [15,16] that the BFKL predictions for the total cross-section of the process
hadrons turn out to be in an agreement with the experimental data obtained by the L3
collaboration [17,18]. There are other resummation methods (see [1925] and references
therein) which lead to quite similar numerical results.
Because the initial virtual photons of the process hadrons can be polarized, one
should know also the NLO corrections to the spin correlators at high energies. It is one of
the reasons, why in this paper we calculate next-to-leading corrections to the intercept 1
of the BFKL pomeron with an arbitrary conformal spin n. The case n = 0 corresponds to
the cross-section 0 averaged over the angle between the photon spins (see [13]) and the
case n = 2 describes the high energy behaviour s 12 of the contribution 2 proportional to
cos2 12 .
On the other hand, the contribution of higher conformal spins is interesting from a pure
theoretical point of view, because it is related to remarkable mathematical properties of
the BFKL equation. To begin with, this equation in LLA is invariant under the Mbius
transformation of holomorphic impact parameters
k
ak + b
ck + d
(1)
for arbitrary complex values of a, b, c and d [26]. This invariance allows us to find
its solution for arbitrary momentum transfers q = t providing that we know the
eigenvalues 1(m, m
e) of the kernel for all conformal spins n. The conformal weights m
and m
e are related to the anomalous dimension = 12 + i of the local operators belonging
to a basic series of unitary representations of the Mbius group as follows:
n
n
m
e= .
(2)
m= + ,
2
2
Further, in LLA the homogeneous BFKL equation in the impact parameter representation
can be written as the Schrdinger equation
g2
min E12 ,
(3)
8 2
where 1 is the BFKL intercept. The pair Hamiltonian Hkl for the interaction of two
reggeized gluons with their holomorphic coordinates k , l and momenta pk = ik ,
pl = il can be presented as a sum of the holomorphic and antiholomorphic Hamiltonians
[27]:
H12 9 = E12 9,
Hkl = hkl + hkl ,
1=
where
(4)
hkl = ln(pk pl ) +
1
1
ln(kl ) pk + ln(kl ) pl + 2 ,
pk
pl
21
(5)
is the Euler constant and kl = k l .

The holomorphic separability of the Hamiltonian is very important for the solution of
the generalized BFKL equation [28,29] for a composite state of n reggeized gluons:
g2
E.
(6)
8 2
The Hamiltonian H in the multicolour QCD Nc can be presented as follows [27]:
H 9 = E9,
1n =
H = 12 (h + h ),
h=
n
X
hk,k+1 , [h, h ] = 0.
(7)
k=1
Due to the commutativity of h and h the function 9 has the property of the holomorphic
factorization [27]:
X
es ,
Crs r
(8)
9=
r<s
s are solutions of the Schrdinger equations:

where r and f
hr = r ,
es ,
es =
h
E = 12 ( + )
(9)
with the same values of energies and in the holomorphic and antiholomorphic
subspaces correspondingly. The coefficients Crs should be chosen from the requirement
of a single-valuedness of 9 in the two-dimensional impact parameter space.
The holomorphic Hamiltonian h has the integrals of motion qr (r = 1, 2, . . . , n) [27]:
X
i1 i2 i2 i3 . . . ir i1 pi1 pi2 . . . pir
(10)
qr =
i1 <i2 <<ir
with the properties:

[qr , qs ] = 0,
[qr , h] = 0.
(11)
Moreover, it coincides with the local Hamiltonian of an integrable Heisenberg spin

model [30,31].
The integrability of the model is related to a duality symmetry [32] of h and qr under
the canonical transformation
pr r,r+1 pr+1
(12)
combined with the transposition of these operators.

In particular, in the case of the compound state of three reggeized gluons, where we can
use the conformal ansatz

m m
e
23
23
12 30
m,e
,
(13)
9m,e
m=
m (x, x ), x =
20 30
20 30
10 32
the duality equation for m,e

m (x, x ) can be written in the pseudo-differential form:
Qm,e
m m,e
m = am am
e m,e
m = |m | 1m,1e
m,
am = x(1 x)(i)1+m .
(14)
22
Using this equation, new three-gluon states having the pomeron and odderon quantum
numbers and the intercept 1 = 0 were constructed [33,34]. Note, that the effective action
describing the interaction of reggeized gluons with quarks and gluons was derived in
Ref. [35].
To investigate a possibility of the holomorphic separability of the BFKL kernel at the
next-to-leading approximation in QCD and in extended supersymmetric gauge theories
is another reason for the calculation of the n-dependence of the pomeron intercept.
A presumable integrability of the BFKL dynamics in the N = 4 supersymmetric field
theory at Nc could be related to remarkable properties of the renormalization group
equations in this model. It is well known, that the -function here is zero in all orders of
the perturbation theory. Moreover, if we consider conformally invariant evolution equations
for matrix elements of the quasipartonic operators introduced in Ref. [36] it turns out [37]
that for N = 4 the pair anomalous dimensions for the supersymmetric twist-2 partonic
operators with the Lorentz spin j are proportional to (j 1) (1). It means that
for Nc the corresponding Hamiltonian coincides with the local Hamiltonian for an
integrable Heisenberg spin model similar to that obtained for the reggeized gluons (see [27,
30,31]). The integrability of the evolution equations at the N = 4 theory in the multicolour
limit is important for a verification of the Maldacena guess [38].
In this paper we extend the analysis of article [3] to the NLO level. Namely, we calculate
eigenvalues of the angle-dependent eigenfunctions of the NLO BFKL equation for the
t-channel partial waves in the framework of QCD and in gauge models with an extended
supersymmetry. In the case of QCD this analysis allows us to find main NLO corrections to
an angle-depended part of cross-sections. On the basis of the study of gauge models with an
extended supersymmetry one can attempt to find a possible source for the violation of the
Mbius symmetry, which is important for the calculation of the t-dependent BFKL-kernel
in QCD at the NLO level.
The article is organized as follows. In Section 2 we review the results of Ref. [13],
calculate the eigenvalues of the NLO BFKL equation in QCD and discuss the asymptotic
behaviour of cross-sections at large s. Section 3 contains the analysis of the gluino and
scalar particle contributions and our results for eigenvalues of the NLO BFKL equation in
supersymmetric theories. A summary is given in Conclusion. In Appendix A we discuss
the reasons for the appearance of the special function Ls3 (x) in our formulae and in
Appendix B we consider the Pomeranchuck singularity in QED and in its supersymmetric
extension.
2. NLO corrections to BFKL kernel in QCD

To begin with, we review shortly the results of Ref. [13], where the QCD radiative
corrections to the BFKL integral kernel at t = 0 were calculated. We discuss only the
formulae important for our analysis.
The total cross-section (s) for the high energy scattering of colourless particles A, B
written in terms of their impact factors 8i (qi ) and the t-channel partial wave G (q, q 0 )
23
for the gluongluon scattering is

Z
(s) =
d2 q d2 q 0
8A (q) 8B (q 0 )
(2)2 q 2 q 02
a+i
Z

d s
G (q, q 0 ),
2i s0
s0 = |q| |q 0|. (15)
ai
Here q and q 0 are transverse momenta 3 of virtual gluons and s = 2pA pB is the squared
invariant mass for the colliding particle momenta pA and pB .
Using the dimensional regularization in the MS-scheme to remove ultraviolet and
infrared divergences in intermediate expressions, the BFKL equation for G (q, q 0 ) can
be written in the following form:
Z
D2
(q q1 ) + dD2 q2 K(q, q2 ) G (q2 , q1 ), where
(16)
G (q, q1 ) =
K(q1 , q2 ) = 2 (q1 ) D2 (q1 q2 ) + Kr (q1 , q2 )
(17)
and the spacetime dimension D = 4 + 2 for 0. The gluon Regge trajectory

(q) and the integral kernel Kr (q1 , q2 ) related to the real particle production have been
calculated in [12,3947]. The final results (see [13]) can be written in the following form
correspondingly for the Regge trajectory:
2

q
2
+ 2 ln 2
(q) =
"

2
11 2 nf
1
2
2 q
ln

3
3 Nc
2
2
2

10 nf
q
1
404
67
2 (2)
+ 2 ln 2
+
9
9 Nc
27
#
56 nf
(18)
+ 2 (3) +
27 Nc
and for the kernel
Kr (q1 , q2 )
(
4 2
1
= 1+
0(1 ) (q1 q2 )2

2

(q1 q2 )2
11 2 nf 1
2
2 q1
1 +
1 (2)
1
ln
3
3 Nc
2
q22

(q1 q2 )2
+
2

10 nf
56 nf
67
404
2 (2)
+ 14 (3) +
+
9
9 Nc
27
27 Nc
3 To simplify equations hereafter (except Appendix B) we omit arrows in the notation of transverse momenta
q ,
q 0,
q1 ,
q2 , . . . , i.e. in our formulae the momenta
q ,
q 0,
q1 ,
q2 , . . . will be represented as q, q 0 , q1 , q2 , . . . ,
respectively. Note, however, that the momenta pA and pB are D-space momenta.
24
"

2
q
nf 3(q1 q2 )2 2q12q22 2
1
2
1
+ 1 + 3
+ 2+
2 ln 12
2
2
2
2
Nc
16q1 q2
q1
q2
q2
q1
q2

2
2
2
2
4
4
2
2q1 q2 3q1 3q2
(q1 + q2 )
nf
2
(q1 q2 )
1
3+ 1+ 3
Nc
8q12 q22
16q14q24

Z
1 + x
dx

ln
q12 + xq22 1 x
0
(q12 q22 )
(q1 q2 )2 (q1 + q2 )2
2 2 2

2
2
q
q q (q1 q2 )4
q1
q1
ln 12 ln 1 22
2Li
+
2Li
2
2
q2
(q1 + q22 )4
q22
q22
2 #)
Z1 Z !

xq1
(q12 q22 )
dx
ln
,(19)
1
(q1 q2 )2 (q1 + q2 )2
(q2 xq1 )2
q22
+
where Li2 (x) and (n) are the Euler dilogarithm and Riemann -functions:
Zx
Li2 (x) =
dt
ln(1 t),
t
(n) =
k n .
(20)
k=1
Here we introduced also

=
g2 Nc 0(1 )
(4)2+
(21)
for the colour group SU(Nc ) and g is the QCD coupling constant fixed at the
normalization point in the MS-scheme.
As it was shown in [3], a complete and orthogonal set of eigenfunctions of the
homogeneous BFKL equation in LLA is
2 1
q
ein .
(22)
Gn, (q/q 0 , ) =
q 02
The BFKL kernel in this representation is diagonalized up to the effects related with the
running coupling constant s (q 2 ):

s (q 2 )Nc
s (q 2 )Nc
(n, ) + (n, )
.
(23)
=
4
The calculation of the correction (n, ) for all conformal spins is one of main results of
our paper.
2.1. The technique of calculations
The integration over the angle of the transverse gluon momentum has the form
Z
d ein = 2
25
Z
d cos(n ) = 2
d Tn (cos ),
0
where Tn (cos ) are the Chebyshev polynomials (below n is a positive integer number).
In the framework of the dimensional regularization a natural extension of the Chebyshev
polynomial is the Gegenbauer polynomial Cn (cos ) (see [4855]) with the index related
to the transverse space dimension D 2 as = (D 2)/2 1 = D/2 2. When = 0
we return to the Chebyshev polynomials because
C0n (cos ) lim
2
1
C (cos ) = Tn (cos ).
n
n
To calculate the terms singular at 0 it is convenient to to use the formulae from

[49,50,55] for the traceless products
x (1 ,...,n ) =
[n/2]
X
n!(1)p 0(n p + )
p>0
22p p!(n 2p)!0(n + )
g 1 2 . . . g 2p1 2p x 2p x 2p+1 . . . x n .
The traceless products are related to the Gegenbauer polynomials Cn (cos ) as follows:
(1 ,...,n ) (1 ,...,n )
q2
,
n/2
2
2
(q1 q2 )
q
n1
C (q1 q2 ) = Sn(1) () 1
2 n
where the symbols q1 and q2 are the unit vectors parallel to the transverse momenta q1
and q2 :
q1 q2 =
(q1 q2 )
(q12 q22 )
1/2
and Sn(1) () is the following factor:

Sn(1) () =
2n1 0(n + )
.
0(n)0(1 + )
After the integration over the momentum q2 in Eq. (16) we obtain the product of the
same vectors [49,50,55]:
(1 ,...,n ) (1 ,...,n )
q1
2n
q1
q1
0()0(n + 2)
= Sn(2) ().
2n 0(2)0(n + )
Then in the front of the expressions, obtained from the contributions singular at 0,
the coefficient
Sn () = Sn(1) () Sn(2) () =
0(n + 2)
0(n)0(1 + 2)
appears. Because Sn () is factorized and Sn (0) = 1, it does not have any influence on our
results. Thus, the contributions singular at 0 can be calculated easily with the use of
the formulae from [49,50,55].
26
The -independent terms are obtained in a similar way, because one can use of the
well-known formulae (see [55]) to calculate the contributions of the Feynman diagrams
containing -functions

1, if y > 0,
(y) =
0, if y < 0,
in the integrands. These -functions come from an expansion of the expressions
(q1 + q2 )2 , (q1 xq2)2 and ln(q1 q2 )2 in the r.h.s. of Eq. (19) (see Eq. (24), for
example).
However, it is more convenient to work directly with the expansion of propagators over
the Chebyshev polynomials and to take initially integrals over the angle in the r.h.s. of
Eq. (19). The useful formulae for the expansion of propagators and the properties of the
Chebyshev polynomials are given below:
"
2 n/2 #
X

q
q12 q22
= 1+2
Tn (q1 q2 ) 12
q22 q12
2
(q1 q2 )
q2
n=1

(24)
[q1 q2 ] q12 q22 ,
"
ln(q1 q2 ) =
2
ln q22
X
1
n=1

Tn (q1 q2 )

+ [q1 q2 ] q12 q22 ,
2Tn (x)Tm (x) = Tn+m (x) + Tnm (x)
Tn (1) = 1,
T0 (x) = 1,
T2n (0) = (1) ,

n
q12
n/2 #
q22
q22 q12
(n > m),
Tn (x) = (1)n Tn (x),
(25)
T2n+1 (0) = 0.
2.2. The results of calculations

Applying formulae of the previous subsection to Eqs. (16)(19), we obtain the following
results for eigenvalues (23):

n
n
9 1 +
,
(26)
(n, ) = 29(1) 9 +
2

2

n
n
11 2 nf 1 2
(n, ) 9 0 +
+ 90 1 +
(n, ) =
3
3 Nc 2
2
2

n
10 f
67
2 (2)
(n, ) 6 (3)
9
9 Nc

nf
2 + 3 (1 )
2 cos( )
3+ 1+ 3
n0
+
Nc (3 2 )(1 + 2 )
sin2 ( )(1 2 )

nf
(1 )
2
1+ 3
Nc 2(3 2 )(1 + 2 ) n

n
n
00
00
9 1 +
+ 28(n, ) + 28(n, 1 ) , (27)
9 +
2
2
27
where nm is the Kronecker symbol, and 9(z), 9 0 (z) and 9 00 (z) are the Euler 9-function
and its derivatives. The function 8(n, ) is given below:
"

Z1
x 1+n/2 1
0 n+1
9
(2) + Li2 (x) + Li2 (x)
8(n, ) = dx
1+x
2
2
0
!
X
(x)k
+ ln(x) 9(n + 1) 9(1) + ln(1 + x) +
k+n
k=1
#
X

xk
1 (1)k
+
2
(k + n)
k=1

X
(1)k
9 0 (k + n + 1) 9 0 (k + 1)
=
k + + n/2
k=0

+ (1)k 0 (k + n + 1) + 0 (k + 1)

1
9(k + n + 1) 9(k + 1) ,
(28)
k + + n/2
and

1
0
0 z+1
0 z
9
.
(z) = 9
4
2
2
For the case n = 0 the results (27) and (28) coincide with ones obtained in [13].
Almost all terms in the right hand side of Eq. (27) except the contribution

n
n
11 2 nf 1 0
e
9 +
90 1 +
1(n, ) =
3
3 Nc 2
2
2
are symmetric under the transformation 1 . Moreover, as it was mentioned in [13],
e ) if one would redefine the function q 1 by including in it
it is possible to cancel 1(n,
the logarithmic factor

s (q 2 ) 1/2
.
s (2 )
Note also, that the term

10 nf
67
2 (2)
(n, ) = (n, )
(29)
9
9 Nc
is proportional to the LO contribution. It appears often in the MS-scheme for the processes
involving soft gluons and it is absent in the gluon-bremsstrahlung (GB) scheme [56,57]
being more natural for these processes. In the GB-scheme the term (29) can be incorporated
to a new coupling constant s (q 2 ) as

2
0 = 11
(30)
s q 2 = s q 2 eNc /0
3 Nc 3 nf
and Eq. (23) is replaced by

s (q 2 )Nc
s (q 2 )Nc
(n, ) + (n, )
,
=
where
(31)
28
)=
(n,

11 2 nf 1 2
n
n
(n, ) 9 0 +
+ 90 1 +
3
3 Nc 2
2
2

2
nf
2 + 3 (1 )
cos( )
3+ 1+ 3
n0
6 (3) + 2
Nc (3 2 )(1 + 2 )
sin ( )(1 2 )

nf
(1 )
n2
1+ 3
Nc 2(3 2 )(1 + 2 )

n
n
00
00
9 1 +
+ 28(n, ) + 28(n, 1 ) . (32)
9 +
2
2
Analogously to Ref. [13] we can obtain the eigenvalues (23) in the case of a nonsymmetric choice of the normalization s0 of the energy in Eq. (15). For example, for the scale
s0 = q 2 , which is natural for deep-inelastic scattering process, we have:
in MS-scheme:

s (q 2 )Nc
s (q 2 )Nc
0
(n, ) + (n, ) 2(n, ) (n, )
;
(33)
=
4
and in GB-scheme:

(q 2 )Nc
s (q 2 )Nc
) 2(n, ) 0 (n, ) s
(n, ) + (n,
,
4

d
n
n
+ 90 1 +
.
(n, ) = 9 0 +
0 (n, )
d
2
2
=
where
Considering the limit 0 we obtain for n = 0:

A
B
1
(0, ) 2(0, ) 0 (0, ) = 2 + + C + O 2 ,
(0, ) = + O 2 ,
e

) 2(0, ) 0 (0, ) = A + B + C + O 2 ,
(34)
(0,
2
where

nf
11 2 nf
+
,
B = 3 10Nc2 + 13 ,
A=
3
3 Nc3
9Nc

13 nf
67
e
2 (2) +
,
B =
9
9 Nc3

2nf
11
395
2 (3) (2) +
71 18 (2) .
(35)
C =
27
3
27Nc3
Following Ref. [13] and using Eqs. (34) and (35), one can obtain the expression for
anomalous dimensions of twist-2 operators at 0 (i.e. near j = 1):
in MS-scheme (coinciding with [13]):

2 (q 2 )Nc2 B
s (q 2 )Nc 1 A
+ + O() + s
+
O(1)
=
4 2

3
2
3

(q )Nc C
(36)
+ O 1 ;
+ s
4 3
2
29
and in GB-scheme:

e
s2 (q 2 )Nc2 B
s (q 2 )Nc 1 A
+ + O() +
+ O(1)
=
4 2

3 (q 2 )Nc3 C
1
+
O
+ s
.
4 3
2
(37)
Note, that the calculation of terms s3 for anomalous dimensions of the twist-2
operators at arbitrary values of will be performed in the near future (see [58,59] and
references therein).
For the case n > 2 from Eq. (33) one can find also the anomalous dimensions for the
local operators of the twist t > 2.
2.3. The asymptotics of cross-sections at s
As an example we consider the cross-sections for the inclusive production of two pairs
of charged particles with mass m in the collision (see [3]):

1
1 32
2
0 (s) + cos2
2 (s) ,
s2 2
(s) = em
2
m 81
where em is electromagnetic coupling constant and 2 (s) and 0 (s) are the coefficients
for angular dependent and independent contributions, respectively.
The asymptotic behaviour of the cross-sections k (s) (k = 0, 2) at s corresponds
(for the fixed coupling constant) to the unmoving singularities of f (t) of the type
( k )1/2 , where

1
1
1
s Nc
1
s Nc
s Nc
s Nc
k,
k,
c k,
k,
+
1
,
k =
2
4
2
2
4
2

9 5/2
s 0
1
+
O(
)
,
(38)
0 (s) =
s
1/2
32 7 (3) (ln( ss ))
0
2 (s) =

5/2
s 2
1 + O(s ) .
1/2
s
9 32 7 (3) 8 (ln( s ))
(39)
Here the symbol O(s ) denotes unknown s -corrections to the formfactors. Note, that the
running of the QCD coupling constant leads to the substitution of the unmoving cut by an
infinite set of the Regge poles [26].
Using our results (26) and (27), we obtain the following values for (k, 12 ) and c(k, 12 )
(k = 0, 2):

1
1
= 4 ln 2,
2,
= 4(ln 2 1),
(40)
0,
2
2

10 nf
11 2 nf
1
67
+ 2 (2) +
=2
c 0,
ln 2
2
3
3 Nc
9
9 Nc

nf
1
1
22 (3) 64Ls3
271 33 3 (2)
+
4 ln 2
2
16
Nc
30
nf
nf
= 25.8388 + 0.1869
+ 3.8442 3 ,
(41)
Nc
Nc

10 nf
11 2 nf
67
1
+ 2 (2) +
=2
(ln 2 1)
c 2,
2
3
3 Nc
9
9 Nc

nf
1
1
22 (3) + 64Ls3
+
893 3 3 (2) 64 ln 2
+
4(ln 2 1)
2
32
Nc
nf
nf
+ 0.3947 3 ,
(42)
= 3.2652 + 1.5203
Nc
Nc
where (see [60] and Appendix A)
Zx
Ls3 (x) =

y

ln2 2 sin
dy.
2
Note, that the function Ls3 (x) appears also in calculations of some massive diagrams (see,
for example, the recent articles [6164] and references therein).
The LO results (33) coincide with ones obtained in [3]. The value of c(0, 1/2) equals to
c(1/2) from [13] (see Appendix A).
As it was discussed in [13], the NLO correction c(0, 1/2) is large and leads to a quite
strong reduction of the value of the Pomeron intercept (see recent analyses [15,16,19
25] of various effective resummations of the large NLO terms). Contrary to c(0, 1/2), the
correction c(2, 1/2) is very small and does not change the small LO value [3] of the angledependent contribution.
In the GB-scheme we have analogously for k = 0, 2 (cf. (31)):

1
s Nc
1
s Nc
1
s Nc
1
s Nc
k,
+
k,
k,
1
c k,
,
k =
2
4
2
2
4
2
where

10 nf
1
67
1
2 (2)
.
= c k,
+
c k,
2
2
9
9 Nc
Note, that in the case of the nonsymmetric choice of the scale s0 = q 2 in r.h.s. of
Eq. (15), the values of the corrections c(0, 1/2) and c(2, 1/2) are not changed because
0 (k, 1/2) = 0. Thus, the asymptotics (41) and (42) can be applied directly to the deepinelastic scattering process at small values of Bjorken variable x, where x Q2 /s.
3. NLO corrections to BFKL kernel in supersymmetric field theories

3.1. Gluino and scalar particle pair production
As it was shown above, the eigenvalue of the BFKL kernel in the next-to-leading
approximation for QCD contains nonanalytic contributions proportional to n0 and n2 ,
which in particular leads to a violation of the holomorphic separability. It is natural to
investigate the BFKL equation in supersymmetric gauge theories where there could be
31
significant simplifications. In the case of the N = 1 supersymmetric YangMills theory we

should consider an additional contribution from a gluino loop. It can be obtained from the
QCD result for the next-to-leading kernel by a modification of the terms originated from
the quark loop contributions proportional to nf . Such terms in (n, ) contain the factors
nf /Nc or nf /Nc3 . The terms nf /Nc are related to the quark contribution in the running
coupling constant and will be considered below.
We discuss here only the terms in (n, ) proportional to nf /Nc3 . Their contribution to
depends on = 1/2 + i and can be written as follows:
nf s 2
QED ,
(43)
nf /Nc3 =
4Nc
where according to Eq. (B1) from Appendix B,

2
(1 ) |n|
(2 + 3 (1 ))n0
2 cos( )
QED = 2
2(3 2 )(1 + 2 )
sin ( )(1 2 ) (3 2 )(1 + 2 )
(44)
coincides with the position of the Pomeranchuck singularity of the t-channel partial wave
f (t) in QED (see [6567]). To give an physical interpretation of this result one should
take into account, that in the -plane the BFKL pomeron is the Mandelstam cut originated
from an exchange of two reggeized gluons. It means, that the amplitude for the transition
of two reggeized gluons into two others in the t-channel through the fermion loop should
contain the third spectral function (s, u). Therefore after the substitution s it should
coincide with the amplitude for the virtual photonphoton scattering in QED up to the
colour factor
a b
nf
1
tr t a t b t a t b ,
= nf 2
t , t = if abc t c ,
4Nc
Nc 1
for the corresponding nonplanar loop diagram.
Taking into account, that gluino is a Mayorana particle contrary to the charged electron
and that its field belongs to the adjoint representation (T a )bc = if abc of the gauge group,
we obtain for the gluino contribution in the N -extended supersymmetric gauge theory the
following result:
gluino =
N Nc2 s 2
QED .
2 2
(45)
Providing that N > 1, there is also a contribution of scalar superpartners of the gluon:
scalar =
ns (N) Nc2 s 2
SED ,
2
2
(46)
where ns (N) is the total number of scalar particles and SED is the position of the
Pomeranchuck singularity in the scalar electrodynamics (see Eq. (B2) from Appendix B
and Ref. [6567]):

2
(1 ) |n|
(1 + (1 ))n0
2 cos( )
+
.
(47)
SED = 2
sin ( )(1 2 ) 2(3 2 )(1 + 2 ) 4 (3 2 ) (1 + 2 )
32
For the number of scalar particles in the N -extended supersymmetric theories we have in
the case of N = 2 and N = 4:
ns (2) = 2,
ns (4) = 6,
(48)
taking into account, that for N = 4 the gluons and gluinos with both helicities = s
enter in one supermultiplet with the scalar particles. In particular, the total contribution to
from superpartners of the gluon for N = 4 is

2
3
2 s
QED + SED
1sp = Nc
2

2
2
2
(1 ) |n|
Nc s cos( ) (11 + 15 (1 ))n0
. (49)
= 2
8(3 2 )(1 + 2 )
sin ( )(1 2 ) 4(3 2 )(1 + 2 )
It is remarkable, that this term cancels the corresponding gluon contribution nonanalytic in
n (see (23) and (27)):
1g =
Nc2 s2 cos( )
4 sin2 ( )(1 2 )

(1 )
(2 + 3 (1 ))
0
2
3+
.
(3 2 )(1 + 2 ) n 2(3 2 )(1 + 2 ) |n|
3.2. Scalar particle contribution to the gluon Regge trajectory

Let us consider the contribution of scalar particles to the gluon Regge trajectory in the
two-loop approximation. We use the results of Ref. [39], where the corresponding quark
correction was calculated, which gives a possibility to find easily the gluino correction
using the substitution nf N Nc . The quark contribution can be presented in the form
Z

2nf 2 q 2
dD2 q
(q)
=
5q (q) ,
(50)
25q (q)
n(2)
2
2
f /Nc
3Nc
2
q (q q)
0 2 (1 + ) 0(1 )
2(1 + )
,
5q (q) =
(3 + 2)(1 + 2) 0(1 + 2) (q 2 )
where 5q (q)(g q 2 q q ) is one-loop quark self-energy correction to the polarization
tensor, and the is bare QCD constant:
=
g 2 Nc 0(1 )
,
(4)2+
with well-known relation between g2 and g 2 in MS scheme:

1 11 2 nf
g2 ,
g 2 = g2 2 1 +
3
3 Nc
(51)
where the last term in the r.h.s. of Eq. (51) is proportional to the leading-order term 0 (30)
of QCD -function.
The scalar particle correction can be presented in the similar form:
Z

2ns (N) a 2 q 2
dD2 q
(
q)
5
(q)
,
25
sc
sc
3Nc
2
q 2 (q q)2
0 2 (1 + ) 0(1 )
1
,
5sc (q) =
2(3 + 2)(1 + 2) 0(1 + 2) (q 2 )
(2)
sc
(q) =
33
(52)
where a 5sc (q)(g q 2 q q ) is one-loop self-energy correction to the polarization

tensor from scalar particles and a is the bare constant of N = 4 supersymmetric theory:
g 2 Nc
.
(4)2
As it is well known, the coupling constant of the N = 4 supersymmetric theory does not
run, i.e. its -function is zero, and, hence, a = a .
a=
3.3. The results of calculations

Putting all things together, in the case of N = 4 supersymmetric theory we have the
following representation for the Regge trajectory:
2

q
2
+ 2 ln 2
susy (q) = a

2

1
q
1
8
2 (2)
+ 2 ln 2
(53)
+ 2 (3) .
a2
3
9
Using (53), by analogy with the results of [13] for QCD we can easily find the
susy
corresponding kernel Kr (q1 , q2 ) for the N = 4 SUSY
susy
Kr
(q1 , q2 )
(
1
4a 2
= 1+
0(1 ) (q1 q2 )2

2
8
(q1 q2 )2
1
2 q1
2
(2)
+
+
14
(3)
ln
1 + a
3
9
2
q22
"
(q12 q22 )
+ a
(q1 q2 )2 (q1 + q2 )2
2 2 2

2
2
q
q q (q1 q2 )4
q1
q1
ln 12 ln 1 22
2Li
+
2Li
2
2
q2
(q1 + q22 )4
q22
q22
2 #)
Z1 Z!

xq1
(q12 q22 )
dx
ln
. (54)
1
(q1 q2 )2 (q1 + q2 )2
(q2 xq1 )2
q22
0
The set of eigenvalues for eigenfunctions (22) of the homogeneous BFKL equation in
the framework of N = 4 supersymmetric theory,

(55)
susy = 4 a (n, ) + susy(n, ) a ,
can be found analogously to the previous section in the following form:
34
h

susy (n, ) = 28(n, ) + 28(n, 1 ) 13 2 (2) (n, )

n
n i
9 00 1 +
,
6 (3) 9 00 +
2
2
where the function 8(n, ) was defined in Eq. (28). Note, that the term

1
3 2 (2) (n, )
(56)
appears as a result of the use of the MS-scheme in intermediate calculations (see (29)).
This term can be eliminated by a redefinition of the coupling constant

a a = a + 13 2 (2) a 2 ,
which is equivalent to the use of the BG-scheme (see (30)). For the new coupling constant
a above results can be written in the following form:

susy = 4 a (n, ) + susy(n, )a , where
h

n
susy (n, ) = 28(n, ) + 28(n, 1 ) 6 (3) 9 00 +
2

n i
.
(57)
9 00 1 +
2
Analogously to the previous section and Ref. [13] we can obtain the eigenvalues of the
kernel in the case of nonsymmetric choice of the energy-normalization s0 in Eq. (15). For
the scale s0 = q 2 we have:
in MS-scheme:

(58)
susy = 4 a (n, ) + susy (n, ) 2(n, ) 0 (n, ) a ;
and in GB-scheme:

susy = 4 a (n, ) + susy(n, ) 2(n, ) 0 (n, ) a .
The anomalous dimensions susy of the twist-2 operators at 0 (and n = 0) are:
in MS-scheme

susy

susy

1
B
C
1
+ O() + a
+ O(1) + a 2
+
O
; (59)
susy = 4 a
2
and in GB-scheme:

esusy

susy

B
1
susy
2 C
1
+ O() + a
+ O(1) + a
= 4 a
+O
,
(60)
where
B susy = 13 ,
esusy = 2 (2) and C susy = 2 (3).

B
(61)
For the case n > 2 one can calculate also the anomalous dimensions of the corresponding
local operators. Because in the case N = 4 SUSY the result is analytic in |n|, one can
continue these anomalous dimensions to the negative values of |n|. It gives a possibility
to find the singular contributions of the anomalous dimensions of the twist-2 operators not
only at j = 1 but also at other integer points j = 0, 1, 2, . . . . In particular, in the Born
35
approximation for the anomalous dimension of the operator corresponding to the sum of
distributions of gluons, gluino and scalar particles we obtain = 4a (9(1) 9(j 1))
which coincides with the result of the direct calculations (see [37]). Thus, in the case N = 4
the BFKL equation presumably contains the information sufficient for restoring the kernel
of the DGLAP equation.
4. Conclusion
In this paper we calculated next-to-leading corrections to eigenvalues of the BFKL
kernel in QCD for arbitrary conformal spins n (see (27)). Now with the use of Eqs. (39)
and (42) one can find the spin-dependent contribution to the total cross-section for the
annihilation into hadrons at high energies. We calculated also next-to-leading corrections
to the BFKL equation in the supersymmetric field theories and next-to-leading corrections
to anomalous dimensions of twist-2 operators (see (56) and (59)). In a particular case
of the extended N = 4 SUSY the result of our calculations is significantly simplified
(see Eqs. (56) and (57)). The absence of the coupling constant renormalization in this
model leads presumable to the Mbius invariance of the BFKL equation in higher orders
of the perturbation theory. This invariance can allow one to guess the next-to-leading
corrections to the BFKL kernel in QCD for a general case t 6= 0, which is very important
for the theory of the processes at small x. The remarkable cancellation of nonanalytic
contributions proportional to n0 and n2 at N = 4 SUSY is a possible manifestation of the
integrability of the reggeon dynamics in the Maldacena model [38] corresponding to the
limit Nc . Note, that in this model the eigenvalues of the LLA pair kernels in the
evolution equations for the matrix elements of the quasipartonic operators are proportional
to (j 1) (1) [37], which means that the corresponding Hamiltonian coincides with
the local Hamiltonian for the integrable Heisenberg spin model. The residues of these
eigenvalues in the points j = k can be obtained from the BFKL equation by the analytic
continuation of the anomalous dimensions to negative integer values of the conformal spin
|n|. Therefore the DGLAP equation is not independent from the BFKL equation in N = 4
SUSY and their integrability properties at Nc are related. We shall return to these
problems in our future publications.
Acknowledgements
A. Kotikov thanks the Alexander von Humboldt Foundation for its support. L. Lipatov
thanks the Hamburg University for its hospitality during the period of time when this
work was done. He is supported by the INTAS grant 97-31696 and by the Deutsche
Forschungsgemeinschaft.
We are indebted to J. Bartels, V. Fadin, V. Kim, R. Kirschner, R. Peschansky and
participants of the PNPI Winter School and of the Gribov-70 Workshop for helpful
discussions.
36
Appendix A
Here we demonstrate how the function Ls3 (x) appears in Eqs. (41) and (42). The direct
evaluation of the values of c(0, 1/2) and c(2, 1/2) leads to

11 2 nf
10 nf
67
1
=2
+ 2 (2) +
ln 2
c 0,
2
3
3 Nc
9
9 Nc
"
1

Z

1
1
dt
22 (3) + 16
arctan t ln
+
4 ln 2
t
1t
0
#

nf
1
81 + 33 3 (2) ,
+
(A.1)
16
Nc

11 2 nf
10 nf
67
1
=2
+ 2 (2) +
(ln 2 1)
c 2,
2
3
3 Nc
9
9 Nc
"
1

Z

1
1
dt
22 (3) 16
arctan t ln
+
4(ln 2 1)
t
1t
0
#

nf
1
189 3 3 (2) 64 ln 2 .
+
32
Nc
(A.2)
Note that (A.1) coincides exactly with Eq. (16) of [13].

To obtain Eqs. (41) and (42) we should evaluate the integral
Z1
0

1
dt
arctan t ln
.
t
1t
Using the substitutions

arctan(x + iy) =
t = p2 ,
i 1 + y ix
ln
+ 2k
2 1 y + ix

i = ei/2 ,
we obtain for it the following representation (for k = 0):

Z1
0

1
dt
arctan t ln
t
1t
Z1

dp
ln(1 p) + ln(1 + p) ln(1 ip) ln(1 + ip)
p
0

= i F1 (i) + F2 (i) F1 (i) F2 (i) ,
= i
where [68]
(A.3)
Z1
F1 (a)
37
dp
ln(1 p) ln(1 + ap) = Li3 (a) + S1,2 (a) and
p
Z1
dp
ln(1 + p) ln(1 + ap)
p
0

7
1a
a1
S1,2
+ S1,2 (a) + Li3 (a) (3)
= S1,2
1+a
1+a
8

1a
1
1a
ln a (2) Li2
.
+ ln2 a ln
2
1+a
1+a
F2 (a)
(A.4)
Here Sn,p (x) are the Nielsen generalized polylogarithms (see [6870]):
(1)n+p1
Sn,p (x) =
(n 1)!p!
Z1
dy n1
ln
y ln(1 xy)
y
(A.5)
and the polylogarithms Lin (x) = Sn1,1 (x).

With the use of (A.4) we have
F1 (i) = Li3 (i) + S1,2 (i),

i Li2 (i) + 12 (2) . (A.6)
2
Using the integral representation (A.5), we have for the Euler dilogarithm Li2 (i):
F2 (i) = 2S1,2 (i) S1,2 (i) + Li3 (i) 78 (3)
Zi
Li2 (1) Li2 (i) =
dp
ln(1 p).
p
If we replace s by s
= ei
corresponding to

i()/2
i
,
e
1 s = 1 e = 2 sin
2
we obtain
9
Li2 (1) Li2 (i) = (2) iG,
8
Z1
where G
dx
0
ln x
= Cl2 (/2).
1 + x2
Here, according to Ref. [60],

Zx
Cl2 (x) = Ls2 (x) and Lsn (x) =

y

lnn1 2 sin
dy.
2
Thus,
1
Li2 (i) = (2) iG.
8
Repeating above calculations, one can derive the relations:
(A.7)
38
3
3
(3) i(2),
32
16

i
7
29
+ (2) .
S1,2 (i) = (3) G Ls3
64
4
2
2
16
Li3 (i) =
(A.8)
(A.9)
The obtained results (A.7)(A.9) are in agreement with Ref. [60].

Putting (A.7)(A.9) to (A.6) and (A.3), we have
Z1
0

11

1
dt
arctan t ln
(2),
= 4Ls3
t
1t
2
8
and thus Eqs. (A.1) and (A.2) coincide with Eqs. (41) and (42), respectively.
Appendix B
Although the Pomeranchuck singularity in QED was investigated many years ago (see
[6567]), here we review the obtained results and generalize them to the case of the charged
scalar particle production.
To begin with, let us consider the electronelectron scattering in the Regge kinematics
2pA pB = s q 2 , where pA and pB are momenta of the colliding electrons. In the Born
approximation the corresponding amplitude equals
Aeeee = j (pA )

s
2/s
j (pB ) = j (pA ) pB 2 j (pB )pA

= 2e2 2 ,
2
q
q
q
where j (p) = e u(p0 ) u(p). The elastic cross-section is

Z
Z 2
1
d2 q |Aeeee |2
d q
2
=
4
.
el =
2
2
4 (2)
s
|q|4
This cross-section is divergent at small q due to an infinite radius of the Coulomb
interaction. We shall ignore this circumstance because in the case of the scattering of noncharged particles A and B the integrand is multiplied by a product of their impact factors
8A (q) and 8B (q) vanishing at q = 0
Z
d2 q
8A (q) 8B (q),
AB =
(2)2 |q|4
which leads to the cancellation of this divergency (see [6567]). Note, that for the electron
being a charged particle the impact factor is nonzero for q = 0:
8e (q) = 4.
The amplitude for the production of a massless electronpositron pair in the ee
scattering is
s
Aeeeee+ e = 2e4 2 2 A e+ e ,
q1 q2
39
where the factor A e+ e describes the transition of two virtual photons into the e+ e
pair:
A e+ e = ue (k1 ) Ae ve+ (k2 ),
q1 q2 = k1 + k2 .
The amplitude
!
c
c
c
q
+q
kc
k
2
1
1
2
1
p
c
p
c
c
p
c
Ae =
A
B +p
B
A ,
s
|k1 q1 |2 + |k1|2 21
|k2 q1 |2 + |k2 |2 12
where we introduced the Sudakov variables for momenta k1 and k2 of the produced electron
and positron:
2
sr r = kr = |kr |2
kr = r pB + r pA + kr ,
and used for the momenta of the virtual photons the following relations:
q1 = (1 + 2 ) pA + q1 ,
q2 = (1 + 2 ) pB + q2
valid at high energies when 2k1 k2 s (corresponding to small r and r ). Note, that with
the use of the Dirac equation for the produced particles we can rewrite Ae in the another
form:
!
c
c
c
c
s 1 qc
s 2 qc
c
2
B
B q2 q1
1 + q1 q2 p
1 +p
Ae
s(1 + 2 ) |k1 q1 |2 + |k1 |2 2
|k1 + q2 |2 + |k1 q1 + q2 |2 12
1
and verify its vanishing at q1 = 0 or q2 = 0.
Analogously for the production of a pair of charged scalar particles s we have
s
Aeeees + s = 2e4 2 2 As , where
q1 q2
As = 2 +
2 |k1|2
|k1 q1 |2 + |k1 |2 21
2 |k2|2
|k2 q1 |2 + |k2 |2 12
It can be written in the form

2
1 (
q1 , 2 k1
q1 )
q1 , 2 k2
q1 )
2 (
+
As
2
1 + 2 |k1 q1 |2 + |k1 |2 2
|k2 q1 | + |k2 |2 12
1
to make obvious its vanishing at q1 = 0 or at q2 = 0. The total cross-section for the pair
production in the collision of the particles A and B after calculating the integrals over i
(i = 1, 2) with the use of the -functions (si i |ki |2 ) can be presented as follows:
Z 2

s
d q1 8A (q1 ) d2 q1 8B (q2 )
K
q1 ,
q2 .
pair = log 2
2
4
2
4
m
(2) |q1 |
(2) |q2 |
The kernel K( q , q ) for electrons e and scalar particles s is, correspondingly:

1
Ke,s
q2 = 16 2
q1 ,
d2 k1
(2)2
Z1
0
dx fe,s k1 ; x ,
x=
1
,
1 + 2
40
and
tr(kb1 Ae kb2 A+
e)
,
fe k1 ; x =
16 x(1 x)
fs k1 ; x =
|As |2
.
16 x(1 x)
The factor log(s/m2 ) was obtained as a result of the integration over = 1 + 2 in

the region m2 /s 1, where the mass scale m equals to an essential value for
transverse momenta of charged particles inside the colliding particles. We obtain for the
e+ e production:

fe k1 ; x =
1
2
2
2 |k1 | |k1 q1 |
(|k1 xq1|2 + x(1 x)|q1|2 )2
1
2
2
2 |k2 | |k1 + q2 |
(|k1 + q2 xq1 |2 + x(1 x)|q2|2 )2
1
(|k1 xq1 |2 + x(1 x)|q1|2 )(|k1 + q2 xq1 |2 + x(1 x)|q1 |2 )

q1 ,
q2 + k1 k2 , k1 k1
q1 , k2 k1 +
q2 , k1
k1

q2 , k2 k1
q1 , k1
k1 +
and for charged scalar pair production:

1
1

2

4 x(1 x) ( q1 , 2 k1 q1 )
4 x(1 x)( q1 , 2 k2 q1 )
+
fs k1 ; x =
(|k1 xq1|2 + x(1 x)|q1|2 )2 (|k1 + q2 xq1 |2 + x(1 x)|q2|2 )2

1

2 x(1 x) ( q1 , 2 k1 q1 )( q1 , 2 k2 q1 )
.
+
(|k1 xq1 |2 + x(1 x)|q1|2 )(|k1 + q2 xq1 |2 + x(1 x)|q1 |2 )
By introducing the Feynman parameter y to unify two factors in the denominators of the
last contributions to fe ( k1 ; x) and fs ( k1 ; x) and integrating the kernel over k1 , one can
present it in the form (cf. [6567])
Ke,s
8 2
q2 =
q1 ,
Z1
dx
0
where for the electrons

q2 ; x, y =
q1 ,
e
Z1
dy e,s
q2 ; x, y ,
q1 ,
x(1 x)|q1
|2
1
+ y(1 y)|q2 |2

x(1 x) + x(1 y) 4x(1 x)y(1 y) |q1 |2 |q2 |2
2x(1 x)y(1 y)(

q1 ,
q 2 )2
and for scalar particles
q2 ; x, y =
q1 ,
s
1
4(x(1 x)|q1 |2 + y(1 y)|q2 |2 )

x(1 x) + x(1 y) 8x(1 x)y(1 y) |q1 |2 |q2 |2
+ 4x(1 x)y(1 y)(

q ,
q )2 .
1
Note, that for the supersymmetric QED, where there are scalar charged particles, the total
contribution to the kernel is
41
e (
q1 ,
q2 ; x, y) + 2s (
q1 ,
q2 ; x, y) 3x(1 x) + 3x(1 y) 16x(1 x)y(1 y)
=
2
2
|q1 | |q2 |
2(x(1 x)|q1|2 + y(1 y)|q2|2 )
q 2 )2 .
and does not contain the term proportional to (
q1 ,
The total cross-section in LLA can be written in the form
a+i
Z
tot =
d s
2 m2
ai

d2 q1 8A (q1 ) d2 q1 8B (q2 )
f
q2 ,
q1 ,
2
2
2
2
(2) |q1 |
(2) |q2 |
q1 ,
q2 ) for the -scattering satisfies the Bethe
where the t-channel partial wave f (
Salpeter equation
Z

d2 q10
2 2
0
K
q1 ,
q 10 f
q2 .
q 10 ,
f q1 , q2 = (2) (q q ) +
(2)2 |q1 |2 |q10 |2
The solution of this equation can be obtained in the form of the Fourier expansion:
Z
X
q2 =
q1 ,
f
de
i n (1 2 )
n=

|q1 | 2i
2
1
,
|q1 ||q2 | |q2 |
(n, )
qk written in the complex form as

where k is the phase of the two-dimensional vector
i
k
|qk |e , and (n, ) is the eigenvalue of the homogeneous BetheSalpeter equation:
0 2i3
Z
d2 q 0
0 i n ( 0 ) |q |
K q, q e
.
(n, ) =
|q|
(2)2 |q|4
q,
q 0 ), it is easy to calculate (n, ) for the
Using the above expressions for Ke, s (
spinor and scalar electrodynamics (cf. [6567]):

2 sinh()
1 + 4 2 2
1 1
2 0
11
+
12
(B.1)
QED (n, ) =
n
|n| ,
32 (cosh())2 1 + 2
2

2 sinh()
1 + 4 2 2
1 1
2 0
5 + 4 n +
|n| .
(B.2)
SED (n, ) =
64 (cosh())2 1 + 2
2
The high energy behaviour of the total cross-section is
tot

cn s 1(n) cos n( 0 ) ,
n=0
where 1 is the Pomeron intercept:

1QED (n) = 2
11
32
5
1SED (n) = 2 64
n0
n0
1 2
64 n ,

1 2
128 n .
(B.3)
(B.4)
Note, that for the supersymmetric QED the results are especially simple:
SUSY (n, ) =
2 sinh() 0
,
2 (cosh())2 n
1SUSY (n) =
2 0
.
2 n
42
References
[1] L.N. Lipatov, Sov. J. Nucl. Phys. 23 (1976) 642.
[2] E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Phys. Lett. B 60 (1975) 50; Sov. Phys. JETP 44 (1976)
45; Sov. Phys. JETP 45 (1977) 199.
[3] Ya.Ya. Balitsky, L.N. Lipatov, Sov. J. Nucl. Phys. 28 (1978) 822; Sov. Phys. JETP Lett. 30
(1979) 355.
[4] V.N. Gribov, L.N. Lipatov, Sov. J. Nucl. Phys. 18 (1972) 438, 675.
[5] L.N. Lipatov, Sov. J. Nucl. Phys. 20 (1975) 93.
[6] G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298.
[7] Yu.L. Dokshitzer, Sov. Phys. JETP 46 (1977) 641.
[8] H1 collaboration, S. Aid et al., Nucl. Phys. B 470 (1996) 3.
[9] ZEUS collaboration, M. Derrick et al., Z. Phys. C 69 (1996) 607.
[10] G. Gurci, W. Furmanski, R. Petronzio, Nucl. Phys. B 175 (1980) 27.
[11] W. Furmanski, R. Petronzio, Phys. Lett. B 97 (1980) 437.
[12] L.N. Lipatov, V.S. Fadin, Sov. J. Nucl. Phys. 50 (1989) 712.
[13] V.S. Fadin, L.N. Lipatov, Phys. Lett. B 429 (1998) 127.
[14] G. Camici, M. Ciafaloni, Phys. Lett. B 430 (1998) 349.
[15] S.J. Brodsky, V.S. Fadin, V.T. Kim, L.N. Lipatov, G.B. Pivovarov, JETP. Lett. 70 (1999) 155.
[16] V.T. Kim, L.N. Lipatov, G.B. Pivovarov, IITAP-99-013, hep-ph/9911228; IITAP-99-014, hepph/9911242.
[17] L3 collaboration, M. Acciarri et al., Phys. Lett. B 453 (1999) 94.
[18] Talk presented by M. Kienzle-Focacci at 8th Blois Workshop on Elastic and Diffractive
Scattering (EDS99), Protvino, Russia, June 27July 2, 1999, to appear.
[19] B. Andersson, G. Gustavson, J. Samuelson, Nucl. Phys. B 467 (1996) 443.
[20] B. Andersson, G. Gustavson, H. Kharrazina, Phys. Rev. D 57 (1998) 5543.
[21] G. Salam, JHEP 9807 (1998) 019.
[22] M. Ciafaloni, D. Colferai, G.P. Salam, JHEP 9910 (1999) 017; Phys. Rev. D 60 (1999) 114036.
[23] M. Ciafaloni, D. Colferai, Phys. Lett. B 452 (1999) 372.
[24] R.S. Thorne, Phys. Rev. D 60 (1999) 054031.
[25] G. Altarelli, R.D. Ball, S. Forte, CERN-TH-99-317, hep-ph/9911273; CERN-TH-2000-010,
hep-ph/0001157.
[26] L.N. Lipatov, Sov. Phys. JETP 63 (1986) 904.
[27] L.N. Lipatov, Phys. Lett. B 309 (1993) 394, Preprint DFPD/93/TH/70, hep-th/9311037.
[28] J. Bartels, Nucl. Phys. B 175 (1980) 365.
[29] J. Kwiecinski, M. Prascalowich, Phys. Lett. B 94 (1980) 413.
[30] L.N. Lipatov, Sov. Phys. JETP Lett. 59 (1994) 571.
[31] L.D. Faddeev, G.P. Korchemsky, Phys. Lett. B 342 (1995) 311.
[32] L.N. Lipatov, Nucl. Phys. B 548 (1999) 328.
[33] L.N. Lipatov, in: Proc. Int. Workshop DIS99, Zeuthen, April 1999, pp. 207209.
[34] J. Bartels, G.P. Vacca, L.N. Lipatov, Phys. Lett. B 477 (2000) 178.
[35] L.N. Lipatov, Nucl. Phys. B 452 (1995) 369; Phys. Rep. 286 (1997) 132.
[36] A.P. Bukhvostov, G.V. Frolov, E.A. Kuraev, L.N. Lipatov, Nucl. Phys. B 258 (1985) 601.
[37] L.N. Lipatov, in: Perspectives in Hadronic Physics, Proc. of the ICTP Conf., World Scientific,
Singapore, 1997.
[38] J. Maldacena, Adv. Theor. Phys. 2 (1998) 231; Int. J. Theor. Phys. 38 (1998) 1113.
[39] V.S. Fadin, R. Fiore, M.I. Kotsky, Phys. Lett. B 359 (1995) 181; Phys. Lett. B 387 (1996) 593.
[40] V.S. Fadin, L.N. Lipatov, Nucl. Phys. B 406 (1993) 259.
[41] V.S. Fadin, R. Fiore, A. Quartarolo, Phys. Rev. D 50 (1994) 5893.
[42] V.S. Fadin, R. Fiore, M.I. Kotsky, Phys. Lett. B 389 (1996) 737.
[43] V.S. Fadin, L.N. Lipatov, Nucl. Phys. B 477 (1996) 767.
43
[44] V.S. Fadin, M.I. Kotsky, L.N. Lipatov, Phys. Lett. B 389 (1996) 737.
[45] S. Catani, M. Ciafaloni, F. Hautman, Phys. Lett. B 242 (1990) 97; Nucl. Phys. B 366 (1991)
135.
[46] G. Camici, M. Ciafaloni, Phys. Lett. B 386 (1996) 341; Nucl. Phys. B 496 (1997) 305.
[47] V.S. Fadin, R. Fiore, A. Flashi, M.I. Kotsky, Phys. Lett. B 422 (1998) 287.
[48] L. Durand, P.M. Fishbane, L.M. Simmons, J. Math. Phys. 17 (1976) 1973.
[49] K.G. Chetyrkin, A.L. Kataev, F.V. Tkachov, Nucl. Phys. B 174 (1980) 447.
[50] D.I. Kazakov, A.V. Kotikov, Theor. Math. Phys. 73 (1987) 126; Nucl. Phys. B 307 (1988) 791.
[51] M.J. Levine, E. Remiddi, R. Roskies, Phys. Rev. D 20 (1979) 2068.
[52] W. Celmaster, R.J. Gonzalves, Phys. Rev. D 21 (1980) 3112.
[53] A. Terrano, Phys. Lett. B 93 (1980) 349.
[54] B. Lampe, G. Kramer, Phys. Scripta 28 (1986) 585.
[55] A.V. Kotikov, Phys. Lett. B 375 (1996) 240.
[56] S. Catani, G. Marchesini, B.R. Webber, Nucl. Phys. B 366 (1991) 135.
[57] Yu.L. Dokshitzer, V.A. Khoze, S.I. Troyan, Phys. Rev. D 53 (1996) 89.
[58] J.A.M. Vermaseren, Acta Phys. Polon. B 29 (1998) 2599; Int. J. Mod. Phys. A 14 (1999) 2037.
[59] S. Moch, J.A.M. Vermaseren, NIKHEF-99-030, hep-ph/9912355; NIKHEF-99-022, hepph/9909269.
[60] L. Lewin, Polylogarithms and Associated Functions, North Holland, Amsterdam, 1981.
[61] A.I. Davydychev, J.B. Tausk, Phys. Rev. D 53 (1996) 7381.
[62] J. Fleischer, M.Yu. Kalmykov, A.V. Kotikov, Phys. Lett. B 462 (1999) 169.
[63] J. Fleischer, M.Yu. Kalmykov, Phys. Lett. B 470 (1999) 168.
[64] A.I. Davydychev, Phys. Rev. D 61 (2000) 087701.
[65] V.N. Gribov, L.N. Lipatov, G.V. Frolov, Phys. Lett. B 31 (1970) 34; Sov. J. Nucl. Phys. 12
(1971) 543.
[66] H. Cheng, T.T. Wu, Phys. Rev. D 1 (1970) 2775.
[67] H. Cheng, T.T. Wu, Expanding Protons: Scattering at High Energies, MIT press, Cambridge,
MA, 1987.
[68] A. Devoto, D.W. Duke, Riv. Nuovo Cimento 7 (1984) 1.
[69] N. Nilsen, Nova Acta 90 (1909) 125.
[70] K.S. Kolbig, J.A. Mignaco, E. Remiddi, BIT 10 (1970) 38; Nuovo Cimento A 11 (1972) 824.

Asymmetric orbifolds, noncommutative geometry

and type I string vacua
Ralph Blumenhagen, Lars Grlich, Boris Krs, Dieter Lst
Humboldt-Universitt zu Berlin, Institut fr Physik, Invalidenstrasse 110, 10115 Berlin, Germany
Received 16 March 2000; accepted 2 May 2000
Abstract
We investigate the D-brane contents of asymmetric orbifolds. Using T-duality we find that the
consistent description of open strings in asymmetric orbifolds requires to turn on background gauge
fields on the D-branes. We derive the corresponding noncommutative geometry arising on such Dbranes with mixed NeumannDirichlet boundary conditions directly by applying an asymmetric
rotation to open strings with pure Dirichlet or Neumann boundary conditions. As a concrete
application of our results we construct asymmetric type I vacua requiring open strings with mixed
boundary conditions for tadpole cancellation. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w; 11.25.Mj
Keywords: Asymmetric orbifolds; Asymmetric orientifolds; Type I vacua; Noncommutative geometry; D-branes
1. Introduction
As is known since the work of Connes, Douglas and Schwarz [1,2], matrix theory
compactifications on tori with background three-form flux lead to noncommutative
geometry. Starting with the early work [3,4] one has subsequently realized that open
strings moving in backgrounds with non-zero two-form flux or non-zero gauge fields have
mixed boundary conditions leading to a noncommutative geometry on the boundary of the
string world-sheet [517]. As pointed out in [12], also the effective theory on the D-branes
becomes a noncommutative YangMills theory.
We know from the discovery of D-branes, that Dirichlet branes made their first
appearance by studying the realization of T-duality on a circle in the open string sector [18].
For instance, starting with a D9 brane, the application of T-duality leads to a D8-brane
where the ninth direction changes from a Neumann boundary condition to a Dirichlet
E-mail addresses: blumenha@physik.hu-berlin.de (R. Blumenhagen), goerlich@physik.hu-berlin.de
(L. Grlich), koers@physik.hu-berlin.de (B. Krs), luest@physik.hu-berlin.de (D. Lst).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 7 0 - 4
R. Blumenhagen et al. / Nuclear Physics B 582 (2000) 4464
45
boundary condition. Thus, one may pose the question how D-branes with mixed Neumann
Dirichlet boundary conditions fit into this picture. Does there exist a transformation relating
pure Dirichlet or Neumann boundary conditions to mixed NeumannDirichlet boundary
conditions?
At first sight unrelated, there exists the so far unresolved problem of what the Dbrane content of asymmetric orbifolds is. The simplest asymmetric orbifold is defined
by modding out by T-duality itself, which is indeed a symmetry as long as one chooses
the circle at the self-dual radius. Thus, as was argued in [19] and applied to type I
compactifications in [20], in this special case D9- and D5-branes are identified under the
asymmetric orbifold action. However, the general T-duality group for compactifications
on higher-dimensional tori contains more general asymmetric operations. For instance,
the root lattice of SU(3) allows an asymmetric b
Z3 action. (A leftright asymmetric ZN
symmetry is denoted by b
ZN .) The closed string sector can very well live with such
non-geometric symmetries [21] but what about the open string sector? Since all type II
string theories contain open strings in the non-perturbative D-brane sector, in order for
asymmetric orbifolds to be non-perturbatively consistent, one has to find a realization of
such non-geometric symmetries in the open string sector, as well. Thus, the question arises
what the image of a D9-brane under an asymmetric b
ZN action is.
The third motivation for the investigation performed in this paper is due to recently
introduced orientifolds with D-branes at angles [2227]. We investigated orientifold
models for which the world-sheet parity transformation, , is combined with a complex
conjugation, R, of the compact coordinates. After dividing by a further leftright
symmetric ZN spacetime symmetry the cancellation of tadpoles required the introduction
of so-called twisted open string sectors. These sectors were realized by open strings
stretching between D-branes intersecting at non-trivial angles. As was pointed out in [23,
24], these models are related to ordinary orientifolds by T-duality. However, under
this T-duality the former leftright symmetric ZN action is turned into an asymmetric
b
ZN action in the dual model. Thus, we are led to the problem of describing asymmetric
orientifolds in a D-brane language. Note, that using pure conformal field theory methods
asymmetric orientifolds were discussed recently in [28].
In this paper, we study the three conceptually important problems mentioned above, for
simplicity, in the case of compactifications on direct products of two-dimensional tori.
It turns out that all three problems are deeply related. The upshot is that asymmetric
rotations turn Neumann boundary conditions into mixed NeumannDirichlet boundary
conditions. This statement is the solution to the first problem and allows us to rederive
the noncommutative geometry arising on D-branes with background gauge fields simply
by applying asymmetric rotations to ordinary D-branes. The solution to the second
problem is that asymmetric orbifolds necessarily contain open strings with mixed boundary
conditions. In other words: D-branes manage to incorporate asymmetric symmetries
by turning on background gauge fluxes, which renders their world-volume geometry
noncommutative. Gauging the asymmetric symmetry can then lead to an identification
of commutative and noncommutative geometries. In this sense asymmetric type II
orbifolds are deeply related to noncommutative geometry. Apparently, the same holds
46
for asymmetric orientifolds, orbifolds of type I. Via T-duality the whole plethora of R
orientifold models of [2326] is translated into a set of asymmetric orientifolds with Dbranes of different commutative and noncommutative types in the background. We will
further present a D-brane interpretation of some of the non-geometric models studied in
[28] and some generalizations thereof.
In Section 2 we describe a special class of asymmetric orbifolds on T 2 . Employing
T-duality we first determine the tori allowing an asymmetric b
ZN action, where we discuss
b
the Z3 example in some detail. Afterwards we study D-branes in such models and also
determine the zero-mode spectrum for some special values of the background gauge
flux. In Section 3 we apply asymmetric rotations to give an alternative derivation of the
propagator on the disc with mixed NeumannDirichlet boundary conditions. Moreover,
we compute the commutator of the coordinate fields confirming the well known results in
the literature. In the final section we apply all our techniques to the explicit construction of
Z3 orientifold containing D-branes with mixed boundary conditions.
a Z3 b
2. D-branes in asymmetric orbifolds

In this section we investigate in which way open strings manage to implement
asymmetric symmetries. Naively, one might think that asymmetric symmetries are an issue
only in the closed string sector, as open strings can be obtained by projecting onto the left
right symmetric part of the spacetime. However, historically just requiring the asymmetric
symmetry under T-duality on a circle led to the discovery of D-branes. This T-duality acts
on the spacetime coordinates as
(XL , XR ) (XL , XR ).
(2.1)
Thus, the open string sector deals with T-duality by giving rise to a new kind of
boundary condition leading in this case to the well known Dirichlet boundary condition.
Compactifying on a higher-dimensional torus T d , in general with non-zero B-fields, the
T-duality group gets enlarged, so that one may ask what the image of Neumann boundary
conditions under these actions actually is.
In the course of this paper we restrict ourselves to the two-dimensional torus T 2 and
direct products thereof. For concreteness consider type IIB compactified on a T 2 with
complex coordinate Z = X1 + iX2 allowing a discrete ZN symmetry acting as

: (ZL , ZR ) ei ZL , ei ZR
(2.2)
with = 2/N . The essential observation is that performing a usual T-duality operation
in the x1 -direction

(2.3)
T : (ZL , ZR ) Z L , ZR
yields an asymmetric action on the T-dual torus Tb2

b = T T 1 : (ZL , ZR ) ei ZL , ei ZR .
(2.4)
47
Diagram 1. Duality relation
The aim of this paper is to investigate the properties of asymmetric orbifolds defined by
actions like (2.4).
The strategy we will follow is depicted in the commuting Diagram 1. In order to obtain
the features of the asymmetric orbifold, concerning some questions it is appropriate to
b For other questions it turns that it is better to
directly apply the asymmetric rotation .
first apply a T-duality and then perform the symmetric rotation and translate the result
back via a second T-duality. If not explicitly present in the equations, we have set 0 = 1
both for the closed and the open string.
2.1. Definition of the T-dual torus
The first step is to define the T-dual torus Tb 2 allowing indeed an asymmetric action
(2.4). Let the torus T 2 be defined by the following two vectors
e1 = R1 ,
e2 = R2 ei ,
(2.5)
so that the complex and Khler structures are given by

e2 R2 i
=
e ,
e1 R1
T = b + iR1 R2 sin .
U=
(2.6)
The left and right moving zero-modes, i.e., KaluzaKlein and winding modes, can be
written in the following form

1
U m1 m2 T (n1 + U n2 ) ,
pL =
i U2 T 2

1
U m1 m2 T (n1 + U n2 ) .
pR =
i U2 T 2
(2.7)
Applying T-duality in the x1 -direction exchanges the complex and the Khler modulus
b 2 defined by the vectors
yielding the torus T
e1 =
1
,
R1
e2 =
b
+ iR2 sin
R1
(2.8)
and the two-form flux

R2
cos .
b =
R1
For the KaluzaKlein and winding modes we get
(2.9)
48

1 b
b n2 ,
pL = p
T m1 + U
U n1 + m2 b
b2 Tb2
i U

1 b
b m1 + U
b n2 ,
U n1 + m2 T
pR = p
b2 Tb2
i U
(2.10)
from which we deduce the relation of the KaluzaKlein and winding quantum numbers
m
b1 = n1 ,
m
b2 = m2 ,
b
n1 = m1 ,
b
n2 = n2 .
(2.11)
If the original lattice of T 2 allows a crystallographic action of a ZN symmetry, then the

T-dual Narain-lattice of Tb 2 does allow a crystallographic action of the corresponding
asymmetric b
ZN symmetry. In view of the orientifold model studied in Section 4, we present
the Z3 case as an easy example.
2.2. The b
Z3 torus
One starts with the Z3 lattice defined by the basis vectors

1
3
A
A
+i
e2 = R
e1 = R,
2
2
(2.12)
and arbitrary b-field. The complex and Khler moduli are
1
3
A
A
2 3
,
T = b + iR
.
(2.13)
U = +i
2
2
2
This lattice has the additional property that it allows a crystallographic action of the
reflection at the x2 -axis, R. This was important for the study of R orientifolds in [23,24].
We call this lattice of type A. Recall from [23,24], that under R all three Z3 fixed points
are left invariant. For zero b-field one obtains for instance for the T-dual A lattice
1
3
A
A
e2 = iR
,
(2.14)
e1 = ,
R
2
Z3 symmetry and that
and b A = 1/2. That this rectangular lattice features an asymmetric b
all three fixed points of the b
Z3 are left invariant under is not obvious at all. This shows
already how T-duality can give rise to fairly non-trivial results.
As we have already shown in [23,24] there exists a second Z3 lattice, called type B,
allowing a crystallographic action of the reflection R. The basis vectors are given by
e1B = R,
e2B =
R
R
+i ,
2
2 3
(2.15)
with arbitrary b-field leading to the complex and Khler moduli

UB =
1
1
+i ,
2
2 3
R2
TB =b+i .
2 3
(2.16)
For the B lattice only one Z3 fixed point is invariant under R, the remaining two are
interchanged. For b = 0 the T-dual lattice is defined by
e1B =
1
,
R
R
e2B = i
2 3
49
(2.17)
with b B = 1/2. It is a non-trivial consequence of T-duality that only one of the three b
Z3
fixed points is left invariant under .
If one requires the lattices to allow simultaneously a symmetric Z3 and an asymmetric
b
Z3 action one is stuck at the self-dual point U = T yielding R = 1 and b = 1/2. Note,
that this is precisely the root lattice of the SU(3) Lie algebra. Since now we are equipped
with lattices indeed allowing a crystallographic action of asymmetric b
ZN operations, we
can move forward to discuss their D-brane contents.
2.3. Asymmetric rotations of D-branes
In order to divide a string theory by some discrete group we first have to make sure that
the theory is indeed invariant. For the open string sector this means that the D-branes also
have to be arranged in such a way that they reflect the discrete symmetry. Thus, for instance
we would like to know what the image of a D0-brane under an asymmetric rotation is. In
the compact case we can ask this question for the discrete b
ZN rotations defined in the
last subsection, but we can also pose it quite generally in the non-compact case using a
continuous asymmetric rotation

0
cos
sin
X1,L
X1,L
=
,
0
X2,L
X2,L
sin cos

0
cos sin
X1,R
X1,R
=
.
(2.18)
0
X2,R
X2,R
sin
cos
As outlined already in the beginning of Section 2 (see Diagram 1), instead of acting with the
asymmetric rotation on the Dirichlet boundary conditions of the D0-brane, it is equivalent
to go to the T-dual picture, apply first a symmetric rotation on the branes and then perform
a T-duality transformation in the x1 -direction. In the T-dual picture the D0-brane becomes
a D1-brane filling only the x1 -direction. Thus, the open strings are of Neumann type in the
x1 -direction and of Dirichlet type in the x2 -direction. The asymmetric rotation becomes a
symmetric rotation, which simply rotates the D1-brane by an angle in the x1 x2 plane.
Thus, after the rotation the D1 boundary conditions in these two directions read
X1 + tan X2 = 0,
X2 tan X1 = 0.
(2.19)
If we are on the torus T 2 there is a distinction between values of , for which the rotated
D1-brane intersects a lattice point, and values of , for which the D1-brane densely covers
the entire T 2 . In the first case, one still obtains quantized KaluzaKlein and winding modes
as computed in [7].
If the D1-brane runs n-times around the e1 circle and m times around the e2 circle until
it intersects a lattice point, the relation
n
(2.20)
cot = cot +
mU2
50
Fig. 1. A rotated D1-brane with n = 2 and m = 1.
holds. As an example we show in Fig. 1 a rotated D1-brane with n = 2 and m = 1.

In the following we will mostly consider D-branes of the first kind, which we will
call rational D-branes. Finally, T-duality in the x1 -direction has the effect of exchanging
X1 X1 , leading to the boundary conditions [29]
X1 + cot X2 = 0,
X2 cot X1 = 0.
(2.21)
As emphasized already, one could also perform the asymmetric rotation directly on the
Dirichlet boundary conditions for the D0-brane and derive the same result. Thus, we
conclude that an asymmetric rotation turns a D0-brane into a D2-brane with mixed
boundary conditions. The last statement is the main result of this paper. As has been
discussed intensively in the last year, mixed boundary conditions arise from open strings
travelling in a background with non-trivial two-form flux, B, or non-trivial gauge flux, F ,
X1 + (B + F ) X2 = 0,
X2 (B + F ) X1 = 0.
(2.22)
Thus, we can generally identify

cot = F = B + F ,
(2.23)
which in the rational case becomes (note that cot is not necessarily rational)
cot = cot +
n
= B + F.
mU2
(2.24)
Since the B field is related to the shape of the torus T 2 and the F field to the D-branes,
from (2.24) we extract the following identifications
B = cot ,
F=
n
.
mU2
(2.25)
In Section 3 we will further elaborate the relation between asymmetric rotations and
D-branes with mixed boundary conditions and will present an alternative derivation of
some of the noncommutativity properties known for such boundary conditions. In the
remainder of this section we will focus our attention on the zero mode spectrum for open
strings stretched between D-branes with mixed boundary conditions. In particular, we will
demonstrate that in the compact case open strings stretched between identical rational
51
D-branes do have a non-trivial zero mode spectrum. This is in sharp contrast to some
statements in the literature [30] saying that Neumann boundary conditions allow Kaluza
Klein momentum, Dirichlet boundary conditions allow non-trivial winding but general
mixed D-branes do have neither of them.
2.4. KaluzaKlein and winding modes
Since we can not easily visualize a D-brane with mixed boundary conditions, we first
determine the zero-mode spectrum in the closed string tree channel and then transform the
result into the open string loop channel. Thus, we are looking for boundary states (see also
[31]) in the closed string theory satisfying the following boundary state conditions
[ X1,cl + cot X2,cl ]|Bi = 0,
[ X2,cl cot X1,cl ]|Bi = 0.
(2.26)
Rewriting (2.26) in terms of the complex coordinate the boundary condition reads
[ Zcl i cot Zcl ]|Bi = 0.
(2.27)
Using the mode expansion

z0 1
1
i X n in( + ) e
n
+ (pL + pR ) + (pL pR ) +
e
+ ein( ) ,
2
2
2
n
2 n6=0 n

1
i X n in( + ) e
z0 1
n in( )
+ ( pL + pR ) + ( pL pR )
e
e
Zcl =
+
2
2
2
n
n
2
Zcl =
n6=0
(2.28)
one obtains

(pL + pR ) i cot (pL pR ) |Bi = 0,

n |Bi = 0
n + e2i e
(2.29)
with similar conditions for the fermionic modes. Inserting (2.10) and (2.11) into the first
equation of (2.29) one can solve for the KaluzaKlein and winding modes
n
n
n2 ,
n1
m
b2 = b
(2.30)
m
b1 = b
m
m
giving rise to the following zero-mode spectrum

r + s U
b2 n + m T
b2
2
(2.31)
Mcl =
b2
U
Tb2
with r, s Z. We observe that this agrees with the spectrum derived in [7] by employing
T-duality. Note, that the formula (2.31) is explicitly SL(2, Z) SL(2, Z) invariant. Thus,
the bosonic part of a boundary state satisfying (2.29) is given by
!
X
X1
2i
e ne
exp
n |r, si(n,m) .
(2.32)
|Bi(n,m) =
n
r,sZ
nZ
52
Using this boundary state we compute the tree channel annulus partition function.
Transforming the result via a modular transformation into the loop channel, we can extract
the zero-mode contribution and conclude that open strings stretching between identical
rational D-branes carry non-vanishing zero modes giving rise to masses

r + s U
b2
Tb2
2
(2.33)
Mo =
.

b2
n + m T
b2
U
It would be interesting to derive this quantization condition directly in the open string
sector. Again making use of T-duality in the x1 -direction allows use to extract also the
individual KaluzaKlein and winding contributions
s

mTb2
Tb2
1
b1 ,
b2 + i r + s U
sU
P = (pL + pR ) =

2
b2 n + m T
2
b
U
s

Tb2 n + mTb1
1
b1 is U
b2 .
(2.34)
L = (pL pR ) =
r +sU
b2 n + m T
2
b2
U
Note, that the zero modes indeed satisfy the boundary condition (2.21). Summarizing, we
now have the means to compute annulus amplitudes for open strings stretched between
different kinds of D-branes with rational mixed boundary conditions. As an example, we
discuss the b
Z3 case in some more detail.
2.5. D-branes in the asymmetric b
Z3 orbifold
Consider the b
Z3 lattice of type A and start with a D1 -brane with pure Dirichlet boundary
conditions ( = 0)
X1 = 0,
X2 = 0.
(2.35)
Successively applying the asymmetric b

Z3 this D-brane is mapped to a mixed D2 -brane
with boundary conditions ( = 2/3)
1
1
X2 + X1 = 0,
X1 X2 = 0,
3
3
and a mixed D3 -brane with boundary conditions ( = 2/3)
(2.36)
1
1
X2 X1 = 0.
(2.37)
X1 + X2 = 0,
3
3
In the orbifold theory these three kinds of D-branes are identified. This reflects that their
background fields are being identified according to
1
F F +
3
or equivalently
+
2
.
3
(2.38)
(2.39)
53
The two coordinates X1 and X2 yield the following contribution to the annulus partition
function for open strings stretched between identical D-branes

!
!
X 2t r 2
X 2t 3s 2

2
2
R
4R
e
e
(2.40)
Aii = 3
rZ
sZ
independent of i {1, 2, 3}. Open strings stretched between different kinds of D-branes
give rise to shifted moding and yield the partition function
h1 i
+
3
(2.41)
Ai,i+1 = ni,i+1 h 1 i
+
31
2
which looks like a twisted open string sector. As we know from [2326] here we have to
take into account extra multiplicities, ni,i+1 , which have a natural geometric interpretation
as multiple intersection points of D-branes at angles in the T-dual picture. By this reasoning
we find that for the A type lattice the extra factor is one. However, for the three D-branes
generated by b
Z3 when one starts with a D-brane with pure Neumann boundary conditions,
{/2, /6, /6}, T-duality tells us that there must appear an extra factor of three
in front of the corresponding annulus amplitude (2.41). In the orientifold construction
presented in Section 4 these multiplicities are important to give consistent models.
3. Asymmetric rotations and noncommutative geometry

In Section 2 we have pointed out that on T 2 or R2 D-branes with mixed boundary
conditions can be generated by simply applying an asymmetric rotation to an ordinary Dbrane with pure Neumann or Dirichlet boundary conditions. Thus, it should be possible to
rederive earlier results for the two-point function on the disc

(3.1)
Xi (z) Xj (z0 ) ,
for the operator product expansion (OPE) between vertex operators on the boundary
eipX ( ) eiqX ( 0 )
or for the commutator of the coordinate fields

Xi (, ), Xj (, 0 )
(3.2)
(3.3)
by applying an asymmetric rotation on the corresponding quantities for open strings ending
on D0-branes in flat spacetime.
3.1. Two-point function on the disc
The two-point function on the disc for both X1 and X2 of Dirichlet type reads

Xi (z) Xj (z0 ) = 0 ij ln |z z0 | ln |z z 0 |

1
= 0 ij ln(z z0 ) + ln( z z 0 ) ln(z z 0 ) ln( z z0 )
2
(3.4)
54
from which, formally using

Xi (z) = Xi,L (z) + Xi,R ( z ),
(3.5)
we can directly read off the individual contributions from the left- and right-movers.
Performing the asymmetric rotation
XL A XL ,
XR AT XR ,
(3.6)
where A denotes an element of SO(2), leads to the following expression for the propagator
in the rotated coordinates

Xi (z) Xj (z0 ) = 0 ij ln |z z0 | 0 ij sin2 cos2 ln |z z 0 |

z z0
.
(3.7)
0 ij sin cos ln
z z0
This expression agrees precisely with the propagator derived in [3,4] with the identification

0
cot
F=
.
(3.8)
cot
0
Thus, by applying an asymmetric rotation we have found an elegant and short way of
deriving this propagator without explicit reference to the boundary conditions or the
background fields. Moreover, since the commutative D0-brane is related in this smooth
way to a noncommutative D2-brane, it is suggesting that also both effective theories arising
on such branes are related by some smooth transformation. Such an explicit map between
the commuting and the noncommuting effective gauge theories has been determined in
[12].
3.2. The OPE of vertex operators
In this subsection we apply an asymmetric rotation also to the operator product
expansion of tachyon vertex operators O(z) = eipX (z) on the boundary. Of course,
this OPE is a direct consequence of the correlator (3.7) restricted to the boundary, but
nevertheless we would like to see whether we can generate the noncommutative -product
directly via an asymmetric rotation. Taking care of the left- and right-moving contributions
in the OPE between vertex operators living on a pure Dirichlet boundary we can write for
|z| > |z0 |
0
ipX
(z) e
iqX
(z ) =
(z z0 ) 2 pL qL ( z z 0 ) 2 pR qR
(z z 0 )
0
2 pL q R
( z z0 )
0
2 pR q L
ei(p+q)X (z0 ) + .
(3.9)
Now we apply an asymmetric rotation (3.6) together with

pL A pL ,
pR AT pR ,
qL A qL ,
qR AT qR ,
and, after all, identifying pL = pR , qL = qR we obtain
(3.10)

0
ipX
(z) e
iqX
(z ) =
[(z z0 )( z z 0 )] 2 pq
[(z z 0 )( z z0 )]
0
2
cos(2) pq
z z0
z z0
55
0 ij pi qj sin(2)
2
ei(p+q)X (z0 ) + .
(3.11)
Restricting (3.11) to the boundary and choosing the same branch cut as in [12] we finally
arrive at
0
eipX ( ) eiqX ( 0 ) = ( 0 ) pq(1+sin
2 cos2 )
exp(i 0 sin cos ij pi qj ) ei(p+q)X ( 0 ) + .
(3.12)
This is precisely the OPE derived in [10,12]. It shows that it is indeed possible to derive
the -product eipX ( )eiqX ( 0 ) eipX eiqX ( 0 ) directly via an asymmetric rotation, where
the noncommutative algebra A of functions f and g is defined as
f g = fg i 0 sin cos ij i f j g + .
(3.13)
3.3. The commutator of the coordinates

While the two-point function derived above already implies that the commutator of the
coordinate fields is non-vanishing, i.e., the geometry on the D-brane noncommutative, we
would like to rederive this result directly via studying D-branes with mixed boundary
conditions, as well. This is done by the quantization of the bosonic coordinate fields of
the open string. We start with the T-dual situation with two D-branes intersecting at an
arbitrary angle 2 1 (see Fig. 2).
Fig. 2. D-branes at angles.
The open string boundary condition at = 0 are

X1 + tan 1 X2 = 0,
X2 tan 1 X1 = 0,
(3.14)
X2 tan 2 X1 = 0.
(3.15)
and at = we require
X1 + tan 2 X2 , = 0,
The mode expansion satisfying these two boundary conditions looks like
56
X n+ i(n+)

e
X1 = x1 + i 0
cos (n + ) + 1
n+
nZ
X m i(m)

e
cos (m ) 1 ,
+ i 0
m
mZ
X n+ i(n+)

0
e
sin (n + ) + 1
X2 = x2 + i
n+
nZ
X m i(m)

e
sin (m ) 1 ,
i 0
m
(3.16)
mZ
with = (2 1 )/ . Using the usual commutation relation

[n+ , m ] = (n + ) m+n,0
(3.17)
and the vanishing of the commutator of the center of mass coordinates x1 and x2 one can
easily show that for D-branes at angles the general equal time commutator vanishes

(3.18)
Xi (, ), Xj (, 0 ) = 0.
Therefore, the geometry of D-branes at angles, but without background gauge fields, is
always commutative.
Performing T-duality in the x1 direction one gets the two mixed boundary conditions for
the open strings
X1 + cot 1 X2 = 0,
X2 cot 1 X1 = 0
(3.19)
X2 cot 2 X1 = 0
(3.20)
at = 0 and
X1 + cot 2 X2 = 0,
at = . The mode expansion satisfying these boundary conditions is

X n+ i(n+)

e
sin (n + ) + 1
X1 = x1 0
n+
nZ
X m i(m)

e
sin (m ) 1 ,
0
m
mZ
X n+ i(n+)

0
e
sin (n + ) + 1
X2 = x2 + i
n+
nZ
X m i(m)

e
sin (m ) 1 ,
i 0
m
(3.21)
mZ
which one can also derive performing an asymmetric rotation on the mode expansion for
pure Dirichlet type branes. Now one can compute the commutator
X 2 0

sin (n + ) + 1 sin (n + ) 0 + 1 .
X1 (, ), X2 (, 0 ) = [x1 , x2 ] + i
n+
nZ
(3.22)
For and 0 not both equal to zero or the second term in (3.22) is constant, so that the
whole expression can be set to zero by choosing
[x1 , x2 ] =
2i 0
.
F2 F1
57
(3.23)
Note, that this value also can be obtained directly from the canonical quantization [3,4].
However, as has been shown in [14,17], for = 0 = 0 the evaluation of the sum in (3.22)
yields

2i 0 F1
X1 (, 0), X2 (, 0) =
1 + F12
(3.24)
and for = 0 = one analogously obtains

2i 0 F2
.
X1 (, ), X2 (, ) =
1 + F22
(3.25)
Thus, the coordinates only noncommute at the boundary of the word-sheet, where the
commutator can be expressed entirely in terms of the gauge field on the local D-brane. If
both ends of the open string end on the same D-brane with 1 = 2 , both terms in (3.22)
become singular, but the sum of them give rise to the same expression (3.24) and (3.25)
for the commutator at the ends of the open string. Thus, only for 1 = 2 {0, /2} the
X1 and X2 coordinates commute on the D-brane, in all other cases the end points see
a noncommuting spacetime. Moreover, in the compact case for rational D-branes the
noncommutative theory on the D-branes is mapped via T-duality to a commutative theory
on D-branes at angles. This is only a special example of the more general rule pointed out
in [12] that for rational points the noncommutative torus is T-dual to a commutative one.
At the end of this section let us briefly comment on the algebraic structure of the
noncommutative torus we have obtained by the asymmetric rotation on the D-branes.
As shown in the previous section, the tachyon vertex operator O = eipX ( ) leads to a
noncommutative algebra A, defined in Eq. (3.13). As explained in [12], the algebra A of
tachyon vertex operators can be taken at either end of the open string. Therefore the open
string states form a bimodule A A0 , where A is acting on the boundary = 0 and A0 on
the boundary = of the open string. Specifically, for an open string whose first boundary
= 0 is related to a D-brane with parameter 1 and whose second boundary = is
attached to a D-brane with parameter 2 , the algebra A of functions on the noncommutative
torus is generated by

2 2 0 F1
,
U1 = exp iy1
1 + F12 y2

2 2 0 F1
,
(3.26)
U2 = exp iy2 +
1 + F12 y1
which obey

2 0 F1
U2 U1 .
U1 U2 = exp 2i
1 + F12
On the other hand, the algebra A0 is generated by
(3.27)
58

2 0
e1 = exp iy1 + 2 F2 ,
U
1 + F22 y2

2 0
e2 = exp iy2 2 F2 ,
U
1 + F22 y1
(3.28)
obeying

0
e2 U
e2 = exp 2i 2 F2 U
e1 .
e1 U
U
1 + F22
(3.29)
4. Asymmetric orientifolds
Another motivation for studying such asymmetric orbifolds arises in the construction of
type I vacua. In [2326] we have considered so-called supersymmetric orientifolds with
D-branes at angles in six and four spacetime dimensions which in the six-dimensional
case were defined as
Type IIB on T 4
{R, }
(4.1)
with R : zi zi , the zi being the complex coordinates of the T 4 . Upon T-dualities in

the directions of their real parts one obtains an ordinary orientifold where, however, the
spacetime symmetry becomes asymmetric
Type IIB on Tb 4

.
b
,
(4.2)
In the entire derivation in Section 2 we have identified the two constructions explicitly via
T-duality, relating branes with background fields to branes at angles. While in the R
b now maps branes
orientifolds identified branes at different locations on the tori,
with different values of their background gauge flux upon each other. As the background
fields determine the parameter which rules the noncommutative geometry, branes with
different geometries are identified according to (2.38). In this manner asymmetric orbifolds
and orientifolds provide extremely exotic string backgrounds. In the particular example
of the b
Z3 orientifolds, as obtained from the Z3 R orientifold in [23,24] via T-duality,
all background fields are equivalent to a vanishing field, all geometries equivalent to a
commutative one. But in any of the models, where the orbifold group contains an element
of order 2, i.e., with even N , the background fields can only be gauged away on one half
of the D-branes, the other half stays noncommutative.
From the above mentioned identification it is now clear that the N = (0, 1) supersymmetric asymmetric b
ZN orientifolds (4.2) have the same one loop partition functions as
the corresponding symmetric ZN orientifolds (4.1). The only difference is that instead of
D7-branes at angles, we introduce D9-branes with appropriate background fields. Thus, a
whole class of asymmetric orientifolds has already been studied in the T-dual picture involving D-branes at angles. One could repeat the whole computation for the asymmetric
orientifolds (4.2), getting of course identical results. Note, the model (4.2) is really a type
59
I vacuum, as itself is gauged. Thus, in principle there exist the possibility that heterotic
dual models exist. Of course, in six dimensions most models have more than one tensormultiplet so that no perturbative heterotic dual model can exist. It would be interesting to
look for heterotic duals for the four-dimensional models discussed in [26].
Z3
In the following we will construct the even more general six-dimensional Z3 b
orientifold
Type IIB on Tb 4

(4.3)
b
, ,
which is T-dual to
Type IIB on T 4

,
b
R, ,
(4.4)
where in fact, as shown in Section 2.2, the two tori are identical T 4 = Tb 4 = SU(3)2 . The
freedom to choose their complex structures gives rise to a variety of three distinct models,
which are denoted by AA, AB, BB as in [23,24]. Note, that the same orbifold group is
b 1 , and a pure right-moving Z3R , R =
generated by a pure left-moving Z3L , L =
b
. As was also shown in [28] this model actually has N = (1, 1) supersymmetry, but
one can get N = (0, 1) supersymmetry by turning on non-trivial discrete torsion.
4.1. Tadpole cancellation
The computation of the various one-loop amplitudes is straightforward. For the loop
channel Klein bottle amplitude we obtain
Z h
8c
dt
(ab)
(ab)
(ab)
=
12
00 a b + 01 + 02 + n(ab)
K
2 11 + n,
b 10 + n,
b
b
,
12
t4
0
i
(ab)
(ab)
(ab)
+ nb2 , 20 + nb2 , 2 21 + nb2 , 22 ,
(4.5)
where c V6 /(8 2 0 )3 and is a phase factor defining the discrete torsion. Further we
have introduced a similar notation as in [28]
4
X

2+2+4
(1)
,
00 =
12
1
,=0, 2

+h
(1)2+2+4
2 sin(hi ) h 1 i i , h 6= 0,
0h =
i=1
12
,=0, 21
2 +hi
2 2 +gi
X
Y +hi
(1)2+2+4
gh =
h 1 +g i , g, h 6= 0,
6
i
i=1 12
,=0, 12
X
2
2
Y
(4.6)
2 +hi
with g, h {(1/3, 1/3), (2/3, 2/3)} for which we use the shorter notation g, h {1, 2}.
The index (ab) denotes the three possible choices of lattices, AA, AB and BB, and a are
the zero mode contributions (2.33) to the partition function
60
A =
et
m21 + 43
2
m1
2 m2
B =
m1 ,m2
et
m21 +12
2
m1
2 m2
(4.7)
m1 ,m2
(ab)
Finally, n61 ,62 denotes the trace of the action of 62 on the fixed points in the 61 twisted
Z3 ,
sector. Taking into account that the origin is the only common fixed point of Z3 and b
they can be determined to be
9 for (AA),
(ab)
=
3 for (AB),
(4.8)
n,
b
1 for (BB),
and
(ab)
n,
b
(ab)
= nb2 , 2
for (AA),
3
= i 3 for (AB),
1
for (BB).
(4.9)
The remaining numbers are given by complex conjugation of (4.9). Applying a modular
transformation to (4.5) yields the tree channel Klein bottle amplitude
e(ab)
32c
=
3
Z
1
1 (ab)
dl na0 nb0 00 a b + n(ab)
02
b2 , 01 + n,
3
3 b
0
(ab)
+ 310 n(ab)
2 12
b2 , 2 11 n,
b

(ab)
(ab)
+ 320 nb2 , 21 n,
22 ,
b
(4.10)
3 and nB
0 = 1/ 3. The lattice contributions are
X 3l m2 + 4 m1 m 2
2
2
1 3
e
,
A =
with nA
0 =
m1 ,m2
B =
1
2
m1
2
3 m1 +4 2 m2
(4.11)
m1 ,m2
In order to cancel these tadpoles we now introduce D-branes with mixed boundary
conditions. For both the A and the B lattice we choose three kinds of D-branes with
{/2, /6, /6}. The asymmetric b
Z3 cyclically permutes these three branes, whereas the
symmetric Z3 leaves every brane invariant and acts with a ,i matrix on the ChanPaton
factors on each brane. Since b
Z3 permutes the branes, all three ,i actions must be the
same. The computation of the annulus amplitude gives
A
(ab)
c
=
12
dt h 2
M 00 a b + (Tr )2 01 + (Tr 2 )2 02
t4
0
2 (ab)
2 (ab)
+ M 2 n(ab)
b 10 + (Tr ) n,
b 11 + (Tr 2 ) n,
b 2 12
,1
i
(ab)
(ab)
(ab)
+ M 2 nb2 ,1 20 + (Tr )2 nb2 , 21 + (Tr 2 )2 nb2 , 2 22 ,
(4.12)
61
where the twisted sector is given by open strings stretched between D-branes with i and
(ab)
the number
i+1 . Thus, n(ab)
b denotes the intersection number of two such branes and n,
b
,1
of intersection points invariant under . The actual numbers turn out to be the same as the
multiplicities of the closed string twisted sectors in (4.8) and (4.9). For the tree channel
amplitude we obtain
c
A (ab) =
6

Z
1 (ab)
1 (ab)
2
a b
a b
dl M n0 n0 00 + nb2 ,1 01 + n,1
02
3
3 b
0

(ab)
(ab)

+ (Tr )2 310 nb2 , 11 n,
12
b

(ab)

n

+ (Tr 2 )2 320 n(ab)
21
22 .
b2 , 2
b 2
(4.13)
Finally, one has to compute the Mbius amplitude
(ab)
c
=
12
dt h
1
1
T
T
01 + Tr
M 00 a b + Tr
2 2 02
4
t
1 (ab)
T
+ M n(ab)
12
b 11 + Tr n,
b
,

(ab)
(ab)
1
T
+ Tr
2 2 n,
2 10 + M n
b
b2 , 22
i

(ab)
1
1 (ab)
T
T
+ Tr

,
nb2 , 20 + Tr
21
2 2 n
b2 , 2
(4.14)
with argument q = exp(2t). Transformation into tree channel leads to the expression
e (ab)
8c
=
3

Z
1 (ab)
1 (ab)
n
dl M na0 nb0 00 a b + n,
01
02
3 b
3 b2 ,
0

(ab)
(ab)
1
T
311 n,
+ Tr
2 2
2 12 n
b
b2 , 2 10

(ab)
(ab)
1
T
20 nb2 , 21 . (4.15)
+ Tr 322 n,
b
The three tree channel amplitudes give rise to two independent tadpole cancellation
conditions
M 2 16 M + 64 = 0,
1
T
(Tr )2 16 Tr
2 2 + 64 = 0.
(4.16)
Thus, we have M = 8 D9-branes of each kind and the action of Z3 on the ChanPaton
labels has to satisfy Tr = 8 implying that we have the simple solution that is the
identity matrix.
62
Table 1
Closed string spectra

(ab)
Spectrum
(1, 1) Sugra + 4 V1,1
e2i/3
AA
AB
BB
(0, 1) Sugra + 6 T + 15 H
(0, 1) Sugra + 9 T + 12 H
(0, 1) Sugra + 10 T + 11 H
Table 2
Open string spectra

(ab)
Spectrum
V1,1 in SO(8)
e2i/3
AA
AB
BB
V in SO(8) + 4 H in 28
V in SO(8) + 1 H in 28
V in SO(8)
4.2. The massless spectrum

Having solved the tadpole cancellation conditions we can move forward and compute the
massless spectrum of the effective commutative field theory in the non-compact space
time. In computing the massless spectra we have to take into account the actions of the
operations on the various fixed points. In the closed string sector we find the spectra
shown in Table 1. The computation of the massless spectra in the open string sector is
also straightforward and yields the result in Table 2.
All the spectra shown in Tables 1 and 2 satisfy the cancellation of the non-factorizable
anomaly. Note, that the configurations AB and BB were not analyzed in [28]. Thus, we
have successfully applied the techniques derived in Sections 2 and 3 to the construction of
asymmetric orientifolds.
5. Conclusions
In this article we have pointed out a relationship between the realization of asymmetric
operations in the open string sector and noncommutative geometry arising at the boundary
of the world-sheet of open strings. More concretely, we have shown that a leftright
asymmetric rotation transforms an ordinary Neumann or Dirichlet boundary condition into
a mixed NeumannDirichlet boundary condition. We have employed this observation to
rederive the noncommutativity relations for the open string. Moreover, we have solved the
problem of how the open string sector manages to incorporate asymmetric symmetries.
It simply turns on background gauge fluxes. In asymmetric orbifolds different values
63
of background gauge fields on the D-branes get identified and correspondingly different
geometries, commutative or noncommutative, as well. Finally, we have considered a
concrete asymmetric type I vacuum, where D-branes with mixed boundary conditions were
introduced to cancel all tadpoles.
We have restricted ourselves to the case of products of two-dimensional tori. It would be
interesting to generalize these ideas to more general asymmetric elements of the T-duality
group and to discuss the dual heterotic description. Furthermore, it would be interesting
to see whether via the asymmetric rotation one can gain further insight into the relation
between the effective noncommutative and commutative gauge theories on the branes.
Acknowledgements
We would like to thank Volker Braun and Andreas Recknagel for encouraging
discussions. The work is partially supported by the EU-TMR project ERBFMRX-CT960045 and B.K. likes to thank the Studienstiftung des deutschen Volkes, as well.
References
[1] A. Connes, M.R. Douglas, A. Schwarz, Noncommutative geometry and matrix theory:
compactification on tori, JHEP 9802 (1998) 003; hep-th/9711162.
[2] M.R. Douglas, C. Hull, D-branes and the noncommutative torus, JHEP 9802 (1998) 008; hepth/9711165.
[3] C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, String loop corrections to beta functions, Nucl.
Phys. B 288 (1987) 525.
[4] A. Abouelsaood, C.G. Callan, C.R. Nappi, S.A. Yost, Open strings in background gauge fields,
Nucl. Phys. B 280 (1987) 599.
[5] Y.-K.E. Cheung, M. Krogh, Noncommutative geometry from D0-branes in a background Bfield, Nucl. Phys. B 528 (1998) 185; hep-th/9803031.
[6] J. Frhlich, O. Grandjean, A. Recknagel, Supersymmetric quantum theory and noncommutative
geometry, Commun. Math. Phys. 203 (1999) 119; math-ph/9807006.
[7] F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, Noncommutative geometry from strings and
branes, JHEP 9902 (1999) 016; hep-th/9810072.
[8] C.-S. Chu, P.-M. Ho, Noncommutative open string and D-brane, Nucl. Phys. B 550 (1999) 151;
hep-th/9812219.
[9] C.-S. Chu, P.-M. Ho, Constrained quantization of open string in background B field and
noncommutative D-brane, hep-th/9906192.
[10] V. Schomerus, D-branes and deformation quantization, JHEP 9906 (1999) 030; hep-th/9903205.
[11] F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, Dirac quantization of open strings and noncommutativity in branes, hep-th/9906161.
[12] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032;
hep-th/9908142.
[13] H. Arfaei, D. Kamani, Branes with background fields in boundary state formalism, Phys. Lett.
B 452 (1999) 54; hep-th/9909167.
[14] B. Chen, H. Itoyama, T. Matsuo, K. Murakami, pp0 system with B-field, branes at angles and
noncommutative geometry, hep-th/9910263.
[15] T. Lee, Canonical quantization of open string and noncommutative geometry, hep-th/9911140.
64
[16] J. Frhlich, O. Grandjean, A. Recknagel, V. Schomerus, Fundamental strings in DpDq

systems, hep-th/9912079.
[17] C.-S. Chu, Noncommutative open string: neutral and charged, hep-th/0001144.
[18] J. Dai, R.G. Leigh, J. Polchinski, New connections between string theories, Mod. Phys. Lett.
A 4 (1989) 2073.
[19] M. Dine, E. Silverstein, New M-theory backgrounds with frozen moduli, hep-th/9712166.
[20] R. Blumenhagen, L. Grlich, Orientifolds of non-supersymmetric asymmetric orbifolds, Nucl.
Phys. B 551 (1999) 601; hep-th/9812158.
[21] K.S. Narain, M.H. Sarmadi, C. Vafa, Asymmetric orbifolds, Nucl. Phys. B 288 (1987) 551.
[22] M. Berkooz, M.R. Douglas, R.G. Leigh, Branes intersecting at angles, Nucl. Phys. B 480 (1996)
265; hep-th/9606139.
[23] R. Blumenhagen, L. Grlich, B. Krs, Supersymmetric orientifolds in 6D with D-branes at
angles, hep-th/9908130.
[24] R. Blumenhagen, L. Grlich, B. Krs, A new class of supersymmetric orientifolds with Dbranes at angles, hep-th/0002146.
[25] R. Blumenhagen, C. Angelantonj, Discrete deformations in type I vacua, Phys. Lett. 473 (2000)
86; hep-th/9911190.
[26] R. Blumenhagen, L. Grlich, B. Krs, Supersymmetric 4D orientifolds of type IIA with D6branes at angles, JHEP 0001 (2000) 040; hep-th/9912204.
[27] G. Pradisi, Type I vacua from diagonal Z3 -orbifolds, hep-th/9912218.
[28] M. Bianchi, J.F. Morales, G. Pradisi, Discrete torsion in non-geometric orbifolds and their openstring descendants, hep-th/9910228.
[29] M.M. Sheikh-Jabbari, More on mixed boundary conditions and D-branes bound states, Phys.
Lett. B 425 (1998) 48; hep-th/9712199.
[30] Z. Kakushadze, G. Shiu, S.-H.H. Tye, Type IIB orientifolds, F-theory, type I strings on orbifolds
and type I Heterotic Duality, Nucl. Phys. B 533 (1998) 25; hep-th/9804092.
[31] I. Brunner, A. Rajaraman, M. Rozali, D-branes on asymmetric orbifolds, Nucl. Phys. B 558
(1999) 205; hep-th/9905024.
[32] A. Sagnotti, M. Bianchi, On the systematics of open string theories, Phys. Lett. B 247 (1990)
517.
[33] H. Arfaei, M.M. Sheikh Jabbari, Different D-brane interactions, Phys. Lett. B 394 (1997) 288;
hep-th/9608167.
[34] J.C. Breckenridge, G. Michaud, R.C. Myers, New angles on D-branes, Phys. Rev. D 56 (1997)
5172; hep-th/9703041.
[35] M.M. Sheikh Jabbari, Classification of different branes at angles, Phys. Lett. B 420 (1998) 279;
hep-th/9710121.

String theory and noncommutative field theories

at one loop
Adel Bilal, Chong-Sun Chu, Rodolfo Russo
Institute of Physics, University of Neuchtel, CH-2000 Neuchtel, Switzerland
Abstract
By exploiting the boundary state formalism we obtain the string correlator between two internal
points on the one loop open string world-sheet in the presence of a constant background B-field.
From this derivation it is clear that there is an ambiguity when one tries to restrict the Green function
to the boundary of the surface. We fix this ambiguity by showing that there is a unique form for the
correlator between two points on the boundary which reproduces the one loop field theory results of
different noncommutative field theories. In particular, we present the derivation of one loop diagrams
for 63 and 44 scalar interactions and for YangMills theory. From the 2-point function we are able
to derive the one loop -function for noncommutative gauge theory. 2000 Elsevier Science B.V.
All rights reserved.
PACS: 11.25.w; 11.25.Db; 11.15.Tk
Keywords: String theory in backgrounds; Noncommutative geometry; Perturbative string theory;
Renormalization
1. Introduction
Recently, noncommutative field theory has shown up as an effective description of string
theory in a certain background [16]. The non-standard commutation relation among space
coordinates takes the form
[x , x ] = i
with , 6= 0,
(1.1)
is an antisymmetric real constant matrix of dimension length squared. In the

where
dual language, the algebra of functions is described by the Moyal product
(f g)(x) = e
i2

f (x + )g(x + ) = =0 ,
(1.2)
E-mail addresses: adel.bilal@iph.unine.ch (A. Bilal), chong-sun.chu@iph.unine.ch (C.-S. Chu), rodolfo.

russo@iph.unine.ch (R. Russo).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 1 7 - 5
66
A. Bilal et al. / Nuclear Physics B 582 (2000) 6594
which is associative and noncommutative. The noncommutative nature of string theory in

the presence of a non-vanishing F -field was emphasized in [3,7]. There it was shown that
when quantizing an open string with the boundary condition
X + X F = 0,
(1.3)
where F = B dA is the gauge invariant BornInfeld field strength, the noncommutativity

is an unavoidable feature of open string dynamics. In fact one finds that the coordinates x
of the open string endpoints have to satisfy the new relation (1.1) and also the commutation
relations for the modes of the string expansion are modified.
Recent interest in noncommutative quantum field theory was boosted by the paper
of Seiberg and Witten [6] where it is systematically shown how tree-level open string
theory, in the presence of a non-zero F -field and of Dp-branes, leads to a noncommutative
quantum field theory. One of the main observations in [6] was that, in the presence of
a F -field, the open string world-sheet Greens function on the boundary of the disk is
modified [8]. In particular, Seiberg and Witten showed that in the limit
0 ,
gij 2 ,
F 1/,
(1.4)
(1.5)
the tree level amplitudes just consist of a phase factor, which corresponds to the vertices of
a noncommutative field theory. The limit gij 2 is necessary if one wants to kill the string
propagators to get an irreducible field theory vertex from an M-point string amplitude. In
fact, the basic building block in open string theory is the 3-string vertex: thus in order
to get higher point interactions (e.g., n , n > 4), one may sew together a couple of 3string vertices and contract some of the propagators in between. This is accomplished by
(1.5). Due to some existing confusions in the literature, we feel that it is worthwhile to
stress that the contraction of propagator (1.5) should be imposed only when necessary
and should not always be understood. Indeed, for our purposes of obtaining more general
field theory amplitudes, it would be wrong to always insist on this limit, as this would
kill all the propagators, including the loop propagators. However, the other scaling limit
(1.4) is necessary to obtain a noncommutative field theory and we will refer to it as the
noncommutative limit. In this paper, we will show that by taking the limit (1.4) (and (1.5)
only when necessary), one can reproduce from string theory different noncommutative
field theories, at the tree and one loop levels.
The noncommutative scalar [914] and gauge theories [1525] have been much
studied in their own right. Intriguing phenomena occur, in particular there are important
distinctions between planar and nonplanar Feynman diagrams even in theories of a single
scalar. Moreover, nonplanar diagrams are automatically regulated by an effective UVcutoff 2eff 1/(p )2 , where p is some combination of external momenta. This implies
a non-analyticity in , and an IR singularity 1/(p )2 is generated from integrating out
the high momentum modes. This UV/IR-mixing has attracted quite some attention. Within
quantum field theory it appears as a puzzling feature, but if we think in terms of string
theory there is no natural distinction between UV and IR since high energy open string
loop excitations are mapped via a modular transformation to low energy closed string ones.
67
It is thus quite natural to suspect that the SeibergWitten limit of string theory with
a non-vanishing F -field can be extended beyond tree-level. In fact, by now the low energy
behavior ( 0 0) of string amplitudes is well understood also at one-loop level and
has provided a reliable and flexible tool for analyzing various aspects of very different
field theories. For instance, string amplitudes or string-inspired techniques were used to
evaluate one-loop QCD scattering amplitudes [2628] (see also references therein) and
renormalization constants [2931]; graviton scattering amplitudes were computed and their
relation to gauge amplitudes explored [32,33]; progress was made towards the extension
of the method to more than one loop [3441], and to off-shell amplitudes [4244]. String
techniques also served to stimulate the development of new techniques in field theory, that
preserve some of the nice features of the string formalism [4549]. Basically, the flexibility
of these techniques has its root in the fact that string theory has a two-dimensional structure,
describing the world-sheet dynamics beyond the usual space-time structure. If one can find
a corner of string moduli space which at low energies yields the field theory under study,
it is possible to use the string description to perform the calculations and thus exploit all
the conformal theory features one has already studied for other problems. Thus, it is not
really a surprise that field theory computations, which are largely independent of each
other, appear to be related if viewed from the string theory point of view.
As already discussed above, Seiberg and Witten have pointed out a regime of string
theory which at low energies is described by a non-commutative theory and thus in the
spirit of the above papers it is natural to exploit string computations to derive field theory
results. Specifically, we will show how the non-commutative parameter arises in the field
theory limit of one-loop amplitudes of the simple open bosonic string. We would like to
insist that string theory not only conceptually leads to the noncommutative quantum field
theories, but it also represents a simplifying technique for the computation of perturbative
amplitudes. As we will explicitly see, once we have computed the objects entering in
the string master formula also in the presence of a non-vanishing F -field, the one-loop
diagrams of different noncommutative quantum field theories can be obtained by following
the calculations performed in the commutative case; in particular, we refer to [31,35,40,41].
A nice feature discussed in detail in those references is the existence of a one-to-one
correspondence between Feynman diagrams and different corners of the integration region
over the string parameters. It is worth to stress that this mapping is preserved in the noncommutative case and is identical to the one found for = 0.
In order to incorporate a non-vanishing F -field in one-loop string computations, the
first non-trivial task is to obtain the conformal field theory propagator with F 6= 0 on
a world-sheet with topology of the annulus, rather than simply a disk. This will be
done in Section 2. Starting from the boundary state formalism, we discuss the possible
ambiguity for the open string Green function that exists in the literature and provide an
unambiguous computation to fix its form. Once this Green function is known, one can
apply the techniques developed in [31,35,40,41] in order to extract quantum field theory
Feynman diagrams from string loop amplitudes. By means of this formalism, we compute
various one-loop amplitudes in noncommutative scalar 3 and 4 field theories. In all cases
we show that string theory exactly reproduces the previously known results obtained from
68
quantizing the noncommutative field theory action. We then study the 2-gluon amplitude
in noncommutative gauge theories and determine the leading and subleading singularity
in . By exploiting the fact that in the field theory limit, string theory gives results in the
background field method [31], we easily obtain the -function for the noncommutative
U (1) gauge theory.
Note added
After completing this work and while the present paper was typed, two related
papers [63,64] appeared where also one-loop noncommutative field theory amplitudes
are obtained from one-loop string theory amplitudes. However, the Green function of
[63] differs from ours since the above mentioned ambiguity was differently resolved. In
Section 2 we will argue that to obtain the correct gluon two-point function imposes our
choice for resolving the ambiguity.
2. One-loop open string Green function in the presence of an F -field

In this section we focus on the one-loop Green function of bosonic string theory and, in
particular, we want to generalize the usual calculation to the case where a constant field
F = B dA is present. In fact, once the explicit form of the Green function is known,
it is possible to write in a compact form a generic string amplitude with an arbitrary
number of legs. The situation is thus very different from the one in field theory, where
each diagram represents an independent calculation and one has always to start from the
very first building blocks, i.e., the Feynman rules. As we said, this simplification is possible
because string calculations rely on the world-sheet structure, which is described by a twodimensional theory, more than on the space-time structure. In fact, the h-loop bosonic
string amplitude among M-tachyons with inflowing momenta p1 , . . . , pM can be written as
"
#2 0 pi pj
Z
Y exp Gr(h)
i ,rj (i , j )
(h)
M
M
q
.
(2.1)
[dm]h
AM (p1 , . . . , pM ) = Ch N
Vi0 (0) Vj0 (0)
i<j
Here Ch and N are the normalization factors depending on the world-sheet topology and
the string vertices respectively; their explicit form in terms of the dimensionless string
coupling constant will be given later. The other building blocks of (2.1) have a clear
(h)
geometrical interpretation: Gri ,rj is the correlator of two world-sheet bosons located at
i on the boundary labeled ri , and at j on the boundary rj ; [dm]M
h is the measure
of integration over the moduli space for an open Riemann surface with h loops and M
punctures; Vi0 () are M projective transformations which define local coordinate systems
around each puncture i . Here we do not give the explicit expressions of these quantities
in general (see, for instance, [50]), but we want to stress that their definition depends
only on the geometrical properties of the string world-sheet and in general on the twodimensional conformal theory living on it. From this point of view it is natural that different
computations are much more related to each other than in the usual field theory approach.
Here we want to exploit the great flexibility of this technique in order to derive the
one-loop Feynman diagrams of noncommutative field theories; from the string point of
69
view noncommutativity is easily implemented: one changes the commutation relations

of the open string modes [3] or, equivalently, the boundary conditions of the open string
coordinates X (, ). At the tree and one-loop level this modification basically only shows
up in the Green function. This means that the definition of the field theory limit of the string
master formula is not modified by the presence of F ; in particular, the mapping between
the corners of integration over the moduli and Feynman diagrams can be read from the
calculations of the usual commutative case [31]. Since in our approach all the differences
between commutative and noncommutative field theory are resumed in the string Green
function, we want to derive here this key ingredient from first principle.
2.1. Boundary state formalism
In [8], among other things, the tree level Green function in the presence of a constant
F -field was derived by solving the defining differential equation G(, 0 ) =
2 0 (, 0 ) with the following boundary condition 1

G (, 0 ) iF k G (, 0 )= = 0,
(2.2)
where k ( ) is the derivative parallel (normal) to the world-sheet border. There is,
however, an ambiguity in this approach: one can always add to a given solution a constant
piece (i.e., independent of the punctures , 0 ) with arbitrary dependence on F and on
the annulus width and obtain another Green function which gives different results in the
0 0 limit. As we will see, these terms play a crucial role in the field theory limit of
noncommutative amplitudes, so it is important to understand the actual form of the Green
function appearing in the string master formula (2.1). In order to clarify this point, we
derive the Green function in the boundary state formalism using a simple trick that reduces
the actual calculation to the one encountered in the usual case F = 0.
2.1.1. Tree level
Let us consider the correlation function of two closed string tachyons on a disk, from
which one may extract the tree-level Green function G (0) by simply looking at the term
with explicit dependence on both punctures
2 2 0

(0)
ip2 X( 0 , 0 )
0 d d
e
A2 0eip1 X(,)
= ep1 G
(0) (, 0 )p
2 +p1 C
(0) ()p
1 +p2 C
(0) ( 0 )p
d 2 d 2 0.
(2.3)
This same amplitude can be calculated in the boundary state formalism. In this approach
one starts from a world-sheet with the topology of the sphere and thus the string coordinates
X depend on two independent sets of oscillators n and n . Then one introduces in

the amplitudes a coherent state |Bi (see, [53,54] and references therein) which basically
identifies the left and the right sector of the closed strings with the appropriate boundary
conditions and thus inserts a boundary on the string world-sheet
1 Notice that there is a factor of i different from (1.3) due to a Wick rotation on the world-sheet.
70

ip1 X(z,z) ip2 X(z0 ,z0 )

A(0)
e
B(P ) d 2 z d 2 z0 ,
2 0, 0 T e
(2.4)
where T is the radial ordering. The P in the boundary state |B(P )i is put there to emphasize
that in order to sew to a boundary state to a given Riemann surface, it has to contain
a closed string propagator 1/(L0 + L 0 2) [51]. At tree-level its effect in the amplitudes
is just to shift the positions of the external legs, so it does not modify the form of the
Green functions. Notice that in (2.3) and (2.4) we have used two different ways to label the
world-sheet coordinates. This is because the two approaches naturally give rise to different
parameterizations of the string world-sheet. In (2.3) the poles of open string exchanges
between the two vertex operators are manifest and the fundamental region is the upper half
of the complex plane. On the other hand, the boundary state calculation is written in the
closed string channel and the world-sheet is mapped inside the disk of unit radius.
The advantage of the boundary state formulation is that it is very simple to introduce
a constant F -field; in fact, one only needs to slightly modify the identification brought by
the boundary state [52,55] (here we use the convention of [55] in the particular case where
there are no Dirichlet directions)

n B F = 0, n > 0.
(2.5)
(1 + F )
n + (1 F )
From this identification it is easy to see that the part of the tachyon vertex depending on
the non-zero modes n satisfies the following relation

p
p 0
i 1+F X 0 1z
(2.6)
: |BiF ,
: ei 2 X (z) : |BiF =: e 2 1F
where X0 is the oscillator part of the string coordinate without the zero modes. In fact,
since we only consider noncompact and Neumann directions in |Bi, both p and p vanish
on the boundary; thus the zero mode contribution can be calculated separately and modifies
only the F -independent part of G. Using this identification repeatedly, one can reduce
(2.4) to the the usual computation of the expectation value of four open string-like vertex
operators and, in general, a N -point function of closed string on a disk is equivalent to
a 2N -point amplitude among open strings. The only difference is that the former X part of
the vertices is evaluated in the unusual image point 1/z and its Lorentz index is contracted
with the external momentum through a non-trivial matrix depending on F . After these
considerations, it is easy to find the expression for the Green function G which depends on
two different points z and z0

1 1F
1 (0)
0
0
G
(z,
z
)
=
ln
|z
z
|
+
ln(1 zz0 )
0
2 1 + F

1 1+F
ln(1 z0 z ).
(2.7)
+
2 1 F
Here the antisymmetric nature of the field F has been used to rewrite the final result in the
standard form where the index always precedes . It is clear that G is symmetric under
the simultaneous exchange z z0 and .
Note that the above result is written in the z-coordinates, the parameterization chosen
by the boundary state calculation. In order to do the comparison with the one of [8], it
71
is necessary to perform a conformal transformation z = ( i)/( + i) and rewrite the

Green functions in terms of the -coordinates. A small subtlety in this mapping comes from
the fact that the Green function in (2.7) is not a scalar under conformal transformations.
(0)
The simplest thing to do is to transform the scalar combination (A2 d 2 z d 2 z0 ) to the
(0)
0
parameterization and read off G (, ) from there. One finds indeed the same result
(i.e., Eq. (2.15) of [8]).
A couple of remarks are in order. Notice that the tree-level Green function in the zcoordinates cannot satisfy a boundary condition similar to the one in (2.2). This peculiarity
has already been stressed in [8] where it was pointed out that the condition (2.2) may be in
contradiction with the equation of motion because Gauss theorem requires that the sum of
all boundary integrals is equal to 2 0 . This is indeed the case in the z coordinate and the
boundary condition is modified as

(2.8)
G (z, z0 ) iF k G (z, z0 )|z|=1 = 0 .
As already anticipated in the amplitudes among closed string vertices on a disk, one also
gets a contribution C (0) depending on a single point: this comes from the contraction of
the former left and right moving part of the same vertex, since the oscillators of these two
parts are now identified by the presence of a boundary (2.5)

1 1+F

1 1F
1 (0)
2
C (z) =
ln 1 |z| +
ln 1 |z|2 .
(2.9)
0
4 1 + F
4 1 F
2.1.2. One loop
The same approach easily generalizes to one-loop, where now two boundary states must
be inserted. Since the noncommutativity we are interested in at present is related to the
global U (1), we choose the boundary states to each enforce the same identification on the
oscillators sets of closed strings,

ik1 X(z,z) ik2 X(z0 ,z0 )

e
(2.10)
B(P ) F d 2 z d 2 z0 .
A(1)
2 F B(P ) T e
Again, one can use (2.6) in order to eliminate the dependence of the two vertices on the
right moving oscillators, so that the scalar product hB|Bi in this sector transforms into
a trace over the remaining left moving n s. The final evaluation of the trace is most
easily performed by using coherent states and canonical forms [50]; let us here report and
comment on the three basic pieces of the result. First the measure. The two propagators P
in (2.10) combine and give rise to the usual q 2L0 factor present in one-loop amplitudes. It
is clear from this point of view that the contribution to (2.10) not appearing in the exponent
depends trivially on F only through the normalization of the boundary state [55]. This part
is absorbed in the definition of the measure (the relation between [d] and [dm] is given
below) which thus becomes
"
#

dq 2 2 0 Y
M=2
2n 2d
.
(2.11)
1q
[d]h=1 = det(1 + F ) 3 d z d z
q
n=1
72
Notice here that there is no (ln q)d/2 in [d] since we did not have to perform any Gaussian
integral in the closed string channel. Next, by looking at the exponent of the result (2.10),
one can extract the one-loop Green function
1 (1)
G (z, z0 )
0
"
0
= ln |z z | + ln

#
0

Y
1 q 2n zz 1 q 2n zz0
n=1
(1 q 2n )2

"
Y
(1 q 2n zz0 ) 1
1 1F
ln(1 zz0 ) + ln
+
2 1 + F
(1 q 2n )2
n=1

"
Y
(1 q 2n z0 z ) 1
1 1+F
0
ln(1 z z ) + ln
+
2 1 F
(1 q 2n )2
q 2n
zz0
q 2n
z0 z
#
.
(2.12)
n=1
Finally as in the tree-level case, the full amplitude A(1)

2 contains also contractions between
the left and the right part of a single vertex which are encoded in the following C (1)
2n #

"
(1 q 2n |z|2 ) 1 q
Y

2
1 1F
1 (1)
|z|
C (z) =
ln 1 |z|2 + ln
0
4 1 + F
(1 q 2n )2
n=1
2n #

"
(1 q 2n |z|2 ) 1 q
Y

1 1+F
|z|2
2
ln 1 |z| + ln
. (2.13)
+
4 1 F
(1 q 2n )2
n=1
We note that (2.12) is exactly the Green function obtained in [8] (Eq. (3.6) there), which
satisfies a certain particular form of boundary condition. We remark that the result (2.12)
are written in the closed string channel and the natural modular parameter ln q = ic is
related to the length of the surface viewed as a cylinder. In this case a fundamental region
for the string world-sheet is the annulus with inner radius q and outer radius 1. The result
of [8] is obtained by setting q = a/b and rescaling the coordinate z z/b.
In order to extract from the string amplitude the contribution of the open string
channel, where the world-sheet degenerates into a circle, one has to perform a modular
transformation on both the Green function and the measure. In particular, at one-loop level
the relation between z and -coordinates is [56]
2 2
.
(2.14)
ln k
Notice that this identification fixes the cut of the log function in the complex plane. In
fact we want that the segment (1, k) of the negative real axis is mapped by (2.14) on
the inner border of the z-parameterization; thus we take 2
ln
z = e2i ln k ,
ln q =
ln = ln || + i arg ,
6 arg < .
(2.15)
2 Of course, insisting on 6 arg < makes the logarithm a single valued function at the expense of
continuity.
73
However, before explicitly performing the modular transformation (2.14) on the various
building blocks of the string amplitude, we want to make two important remarks about the
Green function (2.12).
First as one can see from (2.1), the string amplitude does not contain simply G, but its
combination with the derivatives of the local coordinates around each punctures V 0 (0) [50],
and this combination has conformal weight zero. Usually the V 0 (0) dependence drops out
on-shell, since the factor coming from the exponent cancels against the one present in the
definition of the the measure [35]
M
[dm]M
1 = [d]1
m
Y
1/Vi0 (0).
(2.16)
i=1
However, in order to exploit the off-shell continuation of the string results which is possible
in the field theory limit [31,35], it is more useful not to perform this simplification. Thus
we use in the amplitudes the measure [dm] instead of the one of (2.11), and a shifted Green
function

0
0
(1)
0
ln Vz0 (0) + ln Vz00 (0)
G(1)
(z, z ) = G (z, z )
2
0
(1)
= G
(z, z0 ) ln |zz0 |.
(2.17)
2
Here, as in [31], we have related the derivative of the local coordinate to the one-loop
Abelian differential since this is the only well defined object on the annulus having
conformal dimension 1. In particular we have Vi0 (0) = zi .
The second remark is related to an ambiguity in the determination of G from the
boundary state approach. In fact, as we have already seen, the computation of the
amplitude (2.10) always gives a combination of G and C. Thus, by exploiting momentum
conservation, it is possible to shift terms of a particular form between G and C. In particular
one can extract from the closed string interaction on the annulus a different definition for
G and C which is still compatible with the final result of (2.10)
0
(1)
0
0
G(1)
(z, z ) = G (z, z ) + M (z, z ),
(1)
(1)
(z) = C
(z) N (z).
C
(2.18)
(2.19)
The general form of this ambiguity is the addition to G of a term M (z, z0 ) with M
satisfying the Laplace equation, as well as M (z, z0 ) = M (z0 , z) in order to preserve
the exchange symmetry discussed after (2.7). Moreover, in order to ensure that a shift of
the form (2.18), (2.19) does not change the string amplitude for all mass levels, one also
needs the following properties M (z, z0 ) = M (z, z0 ), z z0 M = 0. Thus M and N
are related by M (z, z0 ) = N (z) N (z0 ). This freedom in the definition of the Green
function also appears in the calculation of [8] where G is derived by solving the Laplace
equation on the world-sheet. In fact, as in the tree level case, also here it is not possible to
strictly impose on the Green function the same boundary condition imposed on the string
coordinates (1.3); and hence there is a certain degree of freedom in the choice of what
constraint is satisfied by G. Indeed, the shift in (2.18) corresponds simply to a redefinition
74
of the boundary
condition of the Green function which leaves unmodified the value of the
H
integral G ds = 2 0 that is fixed by Gauss theorem. We stress that G and C give
the same results as those obtained with G and C when they are used in the contraction of
closed string fields X(z) (where thus z must be a point not located on the boundary of the
surface). However, when one wants to restrict the Green function to the boundary in order
to calculate open string amplitudes, the two Green functions may give different results.
For the case of scalar amplitudes, we note that the kind of shift in (2.18) does not modify
the amplitude at all, as it is antisymmetric in and and the Green function is contracted
only with external momenta. Therefore the new contributions due to the additional term
sum up to zero using momentum conservation. Thus, the two Green functions G(1) and
G (1) actually give the same result for tachyon amplitudes. However this is not the case for
gluon amplitudes as the string master formula will involve derivatives of the punctures, see
(3.26) below. Using this in the next subsection, we will find that the correct Green function
is given by (2.18).
2.2. One-loop open string Green function with F 6= 0
We begin by evaluating the Green function G (z, z0 ) of Eq. (2.17), as obtained from
the boundary state approach, when the arguments take values on any of the two boundaries.
Then we will argue that we need to exploit the above-mentioned ambiguity and shift
(1)
0
0
G(1)
(z, z ) to G (z, z ) as in (2.18) in order to correctly reproduce the gluon two-point
amplitude in the noncommutative field theory limit.
(1)
First, we rewrite G (z, z0 ) in the following form, splitting it into its symmetric and
antisymmetric part (in , ):

1 + F2
F
1 (1)
0
G (z, z ) = I +
J
K S + A,
(2.20)
0
1 F 2
1 F 2
(1)
where

0

Y
p

p
1 q 2n zz 1 q 2n zz0
0
0

I = ln z/z z /z + ln
(1 q 2n )2
n=1

2n
Y
|1 q 2n zz0 |1 qzz0
0
J = ln |1 zz | + ln
(1 q 2n )2
n=1
2n

Y
(1 q 2n zz0 ) 1 qzz0
1 zz0
.
K = ln
+ ln
2n
1 z0 z
(1 q 2n z0 z ) 1 q 0
n=1
(2.21)
z z
It is easy to see that G(1)(z, z0 ) is invariant under z q/z, z0 q/z0 (equivalently

k/, 0 k/ 0 ) which maps the outer boundary of the annulus to the inner one and
vice versa. Note also that G(1) (z, z0 ) is single valued on the annulus.
As we have already said, the field theory limit of string amplitudes is more easily
performed in the Schoktty representation of the annulus, since there the open string
contributions are manifest. To go to the coordinate, the conformal transformation (2.14)
implies the following transformation
ln
z/z0
z0 /z
+ ln
Y
1 q 2n zz
n=1
1 q 2n zz0
(1 q 2n )2

+ ln
75
ln q

0
1 kn 1 kn
Y
p
p

0
ln2 (/ 0 )
+ ln / 0 0 / + ln
.
=
2 ln k
(1 k n )2
(2.22)
n=1
Let us discuss the symmetric part of G(1) first. The non-planar open string Green
function has one argument on each boundary, i.e., |z| = 1 and |z0 | = q corresponding to
> 0 and 0 < 0, or vice versa. For the planar case, a priori, one has to distinguish the
case where both arguments are on the outer boundary (|z| = |z0 | = 1 corresponding to
, 0 > 0) from the case where both arguments are on the inner boundary (|z| = |z0 | = q
corresponding to , 0 < 0). It is easy to see that for both planar cases (/ 0 > 0)

,
(2.23)
I = I0 ln
ln q
where we have separated the ln(ln q) term, which will eventually combine with the modular
transformation of the measure, from the usual Green function I0
I0 (, 0 ) =
p
p

ln2 (/ 0 )
+ ln / 0 0 /
2 ln k

0
1 kn 1 kn
Y
0

+ ln

.

(1 k n )2
(2.24)
n=1
For the nonplanar case we have / 0 < 0 and taking into account (2.15),
ln(/ 0 ) = ln |/ 0 | i,
(2.25)
so that
ln2 / 0
e 2 ln k = e 2 ln1 k (ln2 |/ 0 | 2 )
and hence
I =

2
+ I0 ln
2 ln k
ln q
(2.26)
with I0 given by
I0 (, 0 ) =
p
p

ln2 |/ 0 |
+ ln |/ 0 | + | 0 /|
2 ln k

Y
(1 + k n |/ 0 |)(1 + k n | 0 /|)

.
+ ln

(1 k n )2
(2.27)
n=1
Next one has to transform the other piece J to the coordinate, and then restrict the
values of and 0 to the appropriate boundaries. While this is straightforward, it is simpler
to remark, that when first restricting z and z0 to the boundaries one easily sees that
(
0,
planar (outer or inner),
(2.28)
J =I + 1
2 ln q, nonplanar.
76
As a result, we obtain for the symmetric part

( 0,

(1 F 2 )
S = I0 ln
+
2
2
ln q
, nonplanar
8 02 ln k
(2.29)
where we have identified

= 2 0 F .
(2.30)
As mentioned before, the noncommutative field limit is defined by 0 0 with 0 F fixed

and hence is a fixed quantity in field theory.
Now we turn to the antisymmetric part A. We will transformthis to the coordinates
and then evaluate it on the boundaries. Since zz0 = exp 2i

ln k ln 0 , one has to be careful to
correctly take into account the cut of the logarithm. For the planar case with both z, z0 on the
outer boundary (, 0 > 0) there is no subtlety. With the choice of the cut (2.15), the nonplanar cases also present no difficulties as we can always take 6 arg(/ 0 ) < 0, and
the value of the Green function on the boundary is just obtained by analytically continuing
the Green function from the interior of the -domain. However, to get the planar Green
function with both z and z0 on the inner boundary, we cannot analytically continue the
resulting expression without encountering the cut, since now the argument of / 0 would
have to go from 0 to 2 . The simplest and safest way to get the Green function for this
case is to observe that G (, 0 ) must always be periodic under k, 0 k 0 , as
well as under k/, 0 k/ 0 . The second transformation exchanges the outer
and inner boundary, so that the planar Green function on the inner boundary equals the one
on the outer boundary. Thus, rather than insisting on analytic continuation which physically
is irrelevant here, since only the Green functions on the boundaries are relevant, we insist
on the physical equivalence of the two boundaries.
As already pointed out, A has an ambiguity that cannot be fixed in the boundary state
computation. This ambiguity does not affect the scalar amplitudes as it gives a vanishing
contribution when substituted into the string master formula (2.1). However, there is an
important difference in the gluon amplitude. Indeed, the gluon master formula (3.26)
contains terms that depend on the derivatives with respect to the punctures and there is
only one form of the Green function that can give the correct field theory result. The
ambiguity can be most easily fixed by looking at the gluon 2-point function. We recall
from field theory that the planar gluon 2-point function is independent of . It is easy to
see that if |z| = |z0 | = 1, KP = ln(z/z0 ) and using this in the master formula would lead
to a -dependence of the planar 2-gluon amplitude. Indeed on the boundary we have
z
(2.31)
KNP = 0,
KP = ln 0 ,
z
where + () refers to the outer (inner) boundary. However, the field theory result can be
reproduced if K is shifted as follows
2i
ln | 0 |,
(2.32)
ln k
which amounts to choose the N (z) introduced in the previous section as 0 F /(1
F 2 ) ln(z). Thus, the final result is
KP = i( 0 ),
KNP =
(1 F 2 )
A=
2
i
0
4
0 ( ),
2i
0 ln k
ln | 0 |,
nonplanar,
77
(2.33)
where () is the step function that is 1 or 1 for positive or negative . Note that this
open string Green function is to be taken only on the boundaries (real , 0 ). Note also that
(1)
G (, 0 ) is symmetric under the exchange of the two particles:
0
(1) 0
G(1)
(, ) = G ( , ).
(2.34)
We see that the only modification in the planar case is a step function and it gives rise to
the usual phase factor of [57] 3 .
Before substituting (2.29) and (2.33) into the string master formula for the open string
amplitude, we note that one has to scale them by a factor of (1 F 2 )/2 first. The factor of
2 is simply because a different normalization for the Green function was adopted for the
boundary state formalism and the open string amplitude (for instance, because of (2.12),
the exponent of (2.3) is proportional to 0 p2 , while in (2.1) a factor of 2 0 is present). As
for the scaling (1 F 2 ), it is needed when one passes from the closed to the open string
amplitudes. The reason is simple. Lets consider the case of tachyon states whose vertex
operator is eipX when = 0. As usual one may read the mass of the ground state described
by the above vertex by simply looking at the Virasoro constraint L0 1 = 0. When the
non-commutative parameter is turned on, the commutation relations for the modes become
[3,7],

(2.35)
an , x0 = an , p0 = p0 , p0 = 0,

(2.36)
am , an = 2 0 mM 1 m+n x0 , p0 = i 2 0 M 1 ,

,
(2.37)
x0 , x0 = i 2 0 M 1 F
where M = 1 F 2 and
X ein
ian cos n an F sin n ,

X = x0 + p0 p0 F +
n
(2.38)
n6=0
is the mode expansion for the open string coordinates. The change in (2.35) and (2.36) is
gentle since it is simply an F dependent rescaling [8,58],
x0 = x0 (1 F ) ,
p0 = p0 (1 F ) ,
a n = an (1 F ) .
(2.39)
In terms of these operators, the commutation relations (2.35), (2.36) take the standard form,
with the F -dependence concentrated on

(2.40)
x0 , x0 = 2i 0 F .
However, since x0 does not show up in L0 , the computation of the mass parallels the
calculation for = 0 and is found to be F dependent. On the other hand, in the field theories
3 Note that putting the labels r in increasing order in clockwise or anticlockwise direction is a matter of
convention, changing to and hence in individual Feynman diagrams. However, the total amplitude is always
an even function of . We will take the convention of clockwise ordering in this paper.
78
we want to reproduce, the presence of the noncommutative parameter has no effect on the
quadratic part of the Lagrangian, so the mass of the fields do not depend on . In order
to reproduce this feature in the string amplitude, we rescale quantum numbers like the
external momenta and polarizations by an appropriate factor p (p (1 F )) . Notice,
as a check, that this rescaling exactly absorbs the overall F -dependent normalization of the
measure found in the boundary state calculation (2.11). Equivalently one can introduce the
hatted variables as in the previous subsection. With this, the mass of the tachyon takes the
usual value 1/ 0 . We remark that this step of rescaling the modes by 1F is equivalent to
b = X(1 F ).
using a vertex operator V = eipX with the rescaled open string coordinate X
As we have mentioned before, we note that there is no ln q factor in the closed string
amplitude just like the = 0 case, but there is a power of (ln k)d/2 (ln q)d/2 in the open
string amplitude. There are three sources that these ln q factors can arise when passing from
the closed string to the open string amplitude: from the modular transformation (2.22);
from the measure of integration over the moduli; and from the partition function. All these
factors are independent of F and they combine to give the desired ln k dependence of the
open string amplitude.
Summarizing, the open string Green function with constant F -field in the Schottky
representation of the annulus is given by
GP (, 0 ) = I0
i
( 0 ),
4 0
(2.41)
with
I0 (, 0 ) =
p
p

ln2 (/ 0 )
+ ln / 0 0 /
2 ln k

0
1 kn 1 kn
Y

0

+ ln

(1 k n )2
(2.42)
n=1
in the planar case; and the nonplanar Green function is

GNP (, 0 ) = I0 +
i ln | 0 |
( 2 ) 1
2 0 ln k
8 02 ln k
(2.43)
with
I0 (, 0 ) =
p
p

ln2 |/ 0 |
+ ln |/ 0 | + | 0 /|
2 ln k

Y
(1 + k n |/ 0 |)(1 + k n | 0 /|)

.
+ ln

(1 k n )2
(2.44)
n=1
The sign +() in (2.43) refers to the outer (inner) borders. Note that G is still symmetric
with respect to the exchange of particles
G (, 0 ) = G ( 0 , ).
(2.45)
Exactly these open string Green functions are obtained from the open string operator
formalism [65].
79
3. String amplitudes in the presence of a constant F -field and their field theory
limits
After this detailed discussion of the Green functions and the measure in the previous
section, we now turn to the actual evaluation of the one-loop open string amplitudes. In
Section 3.1 we start our analysis by focusing on scalar interactions; this is the simplest
example since the scalar amplitudes involve only the ground state of open bosonic string
theory (that is the tachyon). In Section 3.2, we study the YangMills case.
As a general remark, we want to stress that string amplitudes yield the correct
overall normalization of the various Feynman diagrams without having to calculate the
combinatorial factors typical of field theory and this agreement holds also in the nonplanar case [41]. It is natural then to expect that also the coefficient of noncommutative
amplitudes are reproduced by the string master formula. We find that this is indeed the
case without having to change the definition of Ch and N . We have [31,35,41,51,52]
Ch =
1
1
gop 2h2
,
dh
(2)
(2 0 )d/2
N=
d2
2r gop (2 0 ) 4 ,
(3.1)
where gop is the open string coupling constant and r is related to the normalization
chosen for the ChanPaton factor Tr(a b ) = 1r ab . Note, in particular, that the vertex
normalization N is independent of the particular string state chosen [31] and will be used
to derive both scalar and photon interactions.
3.1. One-loop amplitudes in scalar theories
We are now in the position to derive from the master formula (2.1) a compact expression
that will generate the noncommutative Feynman diagrams for scalar theories 4
(1)
AM (p1 , . . . , pM ) =

Z1
( 2 gop )M
dk 0 m2 ln k
ln k d/2
0 (Md2M2d)/4
e
(2 )
(4 )d/2
k
2
0
X

Z
Z
d2
dM
0 (E)
exp
2 pi pj G (i , j ) .
(3.2)
2
M
i<j
The string projective invariance has been used to choose the fixed points of the single
Schottky generator as = 0 and , and to fix 1 = 1. Also, following [35], we
have to neglect all O(k) terms in the measure of the integration. In fact they will not be
relevant to the field theory limit where we want to single out the contributions where only
scalar particles (tachyon) run in the loop. Notice also that the original string measure
is quadratically divergent when k 0, which just signals the presence of a tachyon
instability in the bosonic string. In the field theory limit, however, we want to deal with
0 2
scalars of positive m2 , so we replace the tachyon mass 1/ 0 m2 and k 1 e m ln k .
4 Notice that the overall normalization, in terms of g
op is different from the one of [41] because there the
convention r = 2 has been used. Here we focus on U (1) interaction and thus is more natural to fix r = 1.
80
This may look strange at first sight, but it has a natural interpretation if one goes back to the
old dual model [35] and is sufficient to reproduce correctly all scalar field theory diagrams.
In the same spirit one does not have to use the full Green function derived in the previous
sections, but can use the effective G(E) where all positive powers of k, corresponding to
higher string modes, have been dropped
r

r
i
j
i
1
(E)
2 i

+
ln
(i j ),
G P (i , j ) = ln

j
i
2 ln k
j
4 0
s
!
#
" s
|j |
1
|i |
(E)
2 |i |
+
+
ln
G NP (i , j ) = ln
|j |
|i |
2 ln k
|j |
i ln |i j | ( 2 ) 1
+
,
(3.3)
2 0
ln k
8( 0 )2 ln k
for the planar and non-planar case, respectively.
As it is clear from the above expression, for = 0, the differences between the planar
and nonplanar Green functions are just some absolute values and signs and it is easy to
see that they lead to the same field theory integrand. This is no longer true when a nonzero F -field is turned on. In this case, the non-planar Green function is modified by some
non-trivial terms which depend both on and on the moduli and k. As we will see, this
modification is enough to account for all the differences in noncommutative field theory
between planar and non-planar diagrams.
3.1.1. 3 in 6 dimensions
The first noncommutative field theory we want to reproduce has only 3 interactions:
Z
1
1
1
m2 2 + g3 ( ).
(3.4)
S3 =
2
2
3!
By comparing the result for the simplest tree-level amplitude between string and field
theory, one can fix the relation between gop and g3
g3 = 25/2 gop (2 0 )
d6
4
(3.5)
Let us now turn to loop amplitudes. As a warm-up exercise we will first compute the oneloop two-point function in scalar 3 -theory from string theory. This example will show
most of the necessary ingredients.
By writing (3.2) for the special case M = 2, one gets
AP2 (p1 , p1 ) =
0 2d/2 g32
(4)d/2 4
Z1
Z1
dk 0 m2 ln k
d2 [2 0 p p GP (1,2 )]
d/2
1 1
e
( ln k)
e
,
k
2
0
(3.6)
since 1 = 1 is fixed. The field theory limit ( 0 0) of the above expression has to
be performed, as usual, by keeping fixed all quantities that have a meaning in the field
theory. In particular the logarithmic divergences in the Green function are related to the
81
dimensionfull Schwinger parameters via a factor of 0 . For instance, one always associates
ln k to the total length of the loop by taking
T
,
(3.7)
0
thus T has to be kept finite as 0 0. In this limit k goes to zero exponentially which
means shrinking the annulus to a one-loop Feynman graph. After this replacement the 0
dependence of (3.6) simplifies and the whole amplitude is just proportional to a single
power of 0 . This means that, in order to have a finite answer for our field theory limit, it
is necessary to introduce one more Schwinger parameter: since we want to reproduce the
irreducible diagram, we perform the 0 0 limit by keeping fixed also t2
ln k =
ln 2 =
t2
0
(3.8)
and get
AP2 (p1 , p1 ) =
g32
1
4 (4)d/2
dT m2 T
e
T d/2
ZT

t2
,
dt2 exp p12 t2 1
T
(3.9)
where we have taken the 0 limit, since no overall 0 factor remained. In this simple
case, the -term does not contribute since p1 p1 = 0. Eq. (3.9) matches the standard
field theory result written in the Schwinger parameterization, numerical factor included.
Notice that the overall 1/4 is half of the usual result determined by the calculation of the
symmetry factor, because we have restricted ourself to the planar diagram.
Now we look at the nonplanar contribution to the 2-point one-loop open string
amplitude. Then 1 = 1 and 2 [1, k], so (3.8) is replaced by
t2
.
0
Then, using the nonplanar Green function (2.43) yields
ln |2 | =
ANP
2 (p1 , p1 ) =
g32
1
4 (4)d/2
Z
0
dT m2 T
e
T d/2
ZT
(3.10)

2
t2
dt2 exp p12 t2 1

+ p1 p1
.
T
4T
(3.11)
Again, we are left with the standard Schwinger proper time integral, but with the additional
factor
1
e 4T p1 p1 = e 4T p1
with p = p .
(3.12)
As noted in [10,11], for p 2 6= 0 this serves as an effective UV cutoff and is at the origin of
the UV/IR mixing. The amplitude (3.11) exactly coincides with the one obtained from a
direct one-loop calculation in noncommutative 3 field theory.
Let us now consider the irreducible part of the 3-point amplitude. In the noncommutative
3
field theory, there are 2 planar and 6 nonplanar diagrams. The 2 planar ones correspond
to the two different cyclic orderings of p1 , p2 and p3 and are each equal to one half
82

i
of the standard commutative diagram times an overall factor e 2 p1 p2 respectively,

i
e+ 2 p1 p2 . In string theory we have one planar diagram with k < 2 , 3 < 1 1 which
can be splitted into 2 < 3 and 2 > 3 , leading to the 2 different planar noncommutative
field theory diagrams as 0 0. We also have a nonplanar string amplitude where we
have three choices of which is alone on one boundary, and for each such choice there
are two possible orderings of the two s which are not fixed. Again we will show how
the noncommutative field theory one-loop amplitudes are obtained from the open string
one-loop amplitude by studying a particular example. The other diagrams are obtained in
a completely analogous manner. We will start with the nonplanar string loop having 1 = 1
on one boundary and 2 , 3 [1, k] on the other boundary. From the general formula
(3.2) we get
ANP
3 (p1 ; p2 , p3 )
g 3 (2 0 )3d/2
= 3
8 (4)d/2
Z1
dk
( ln k)d/2
k2
Zk
d2
2
Zk
2
d3
3

0 NP
0 P
exp 2 0 p1 p2 GNP
(1, 2 ) + 2 p1 p3 G (1, 3 ) + 2 p2 p3 G (2 , 3 ) .
(3.13)
Now let (1 < 2 < 3 < k)
ln k =
T
,
0
ln |2 | =
t2
,
0
ln |3 | =
t2 + t3
,
0
(3.14)
so that as 0 0, k, 2 , 3 all go to zero exponentially. Again all 0 dependence disappears

from the measure and after using momentum conservation the resulting amplitude as
0 0 can be written as
g33 e 2 p2 p3
8 (4)d/2
i
ANP
3 (p1 ; p2 , p3 ) =
Z
0
dT m2 T
e
T d/2
ZT
dt2
0

t2 + t3
t2
+ p1 p3 (t2 + t3 ) 1
dt3 exp p1 p2 t2 1
T
T
0

t3
1
t3
p1 2 p1 ip1 p3
+
,
(3.15)
+ p2 p3 t3 1
T
4T
T

TZt2
where we denote p1 p3 p1 p3 etc. This result exactly equals the field theory
amplitude for the nonplanar Feynman diagram with p2 and p3 being non-planar, as
shown in the Fig. 1. The corresponding planar amplitude with the same cyclic ordering,
is easily obtained from (3.15) by dropping all -dependent terms from the exponent inside
i
the integral, but keeping the factor e 2 p2 p3 in front of the integral. All the other six
amplitudes are obtained in a completely analogous way and we conclude that the oneloop string amplitudes correctly reproduce all the eight 3-point one-loop diagrams of the
noncommutative 3 field theory.
83
Fig. 1. Field theory limit of a nonplanar amplitude.
We could now go on and derive higher-point one-loop amplitudes of the noncommutative 3 theory from the string amplitudes but it is pretty clear how this will work and that
it gives the correct result.
3.1.2. 4 in 4 dimensions
It is now interesting to show that the same string master formula (3.2) used for 3 also
correctly reproduces the one-loop amplitudes of noncommutative 4 theory in d = 4. We
follow [40,41,59] where 4-point vertices are obtained by a different scaling of the string
moduli. Again we start by fixing the new relation between the string coupling constant and
the one appearing in the field theory Lagrangian
1
1
1
L4 = m2 2 + g4 ( ).
2
2
4!
Comparing the 4-point vertex a tree level one can read
2
,
g4 = 4! (2 0 )(d4)/2gop
(3.16)
(3.17)
where one can immediately check that the critical dimension of 4 is correctly reproduced.
Since now gop will bring in (3.2) less powers of 0 , in order to have a finite one-loop
amplitude in the field theory limit, one has to consider a limit in which the distance between
two cubic vertices is vanishing as 0 0.
A first simple example is to check how the non-planar 4 2-point function follows from
(3.2). In the field theory diagram we have only one Schwinger parameter which is related
in the usual way to the length of the string annulus through (3.7). No more rescaling of
string moduli are needed, since the overall factor is now independent of 0 and thus we
shrink to zero the second propagator which would have been present in the 3 2-point:
1 2 0, but with 0 ln |2 | 0. As expected, this corner of the region of integration
gives the correctly normalized non-planar diagram
ANP
2 (p1 , p1 ) =
g4
1
6 (4)d/2

2
dT m2 T

e
exp
p
p
.
1 1
T d/2
4T
(3.18)
The planar result is similar to the above equation; a first difference, of course, is the absence
of the dependent factor; however in this case also the overall coefficient is different and
84
the factor of 1/6 is replaced by 1/3. As already noticed in the commutative case also,
this relative factor of 1/2, between the planar and non-planar contribution, is reproduced
by string theory [40,41]. In fact in the planar amplitude, on can consider the region of 2
(2 1) which gives in the 3 case the tadpole diagram. Again it is possible to shrink the
reducible propagator to zero and produce the same 4 diagram. Notice that this possibility
is allowed only for planar diagram and thus in this case the final result is doubled.
In the next examples we obtain the four point one-loop diagrams where p1 and p2 enter
the loop at one vertex and p3 and p4 at the other, with p1 , p2 , p3 , p4 ordered clockwise.
In this case one has to scale the moduli 2 , 3 , 4 in such a way that no Schwinger
proper time is associated to 2 /1 2 or to 3 /4 . Thus, for the planar diagram, we
start with the planar open-string four-point one-loop amplitude AP4 and do the following
scalings
4
and 2 1 but finite,
(3.19)
0 ln 3 = t3 ,
3
implying also
0 ln 4 = t3 + O( 0 ),
0 ln 2 0.
(3.20)
From (3.2) we then get for the planar 4-point amplitude

AP4 (p1 , . . . , p4 ) =
g42
1
36 (4)d/2
dT m2 T
e
T d/2
Z1
eT /
d2
2
X

0 P
exp
2 pi pj G (i , j ) .
et
Z 3 /
ZT
dt3
0 ln 2
eT /
d4
4
(3.21)
i<j
Note again that all factors of 0 have canceled from the measure. Since 4 4 /3 is finite
it is more convenient to replace
et
Z 3 /
eT /
d4

4
Z1
e(T t3 )/
d 4
so that as 0 0 we get
AP4 (p1 , . . . , p4 ) =
Z
Z1
ZT
Z1
g42
1
dT m2 T
d2
d 4
e
dt3
d/2
d/2
36 (4)
T
2
4
0
0
0
0

i
t3
2
(p1 p2 + p3 p4 ) .
exp (p1 + p2 ) t3 1
T
2
(3.22)
The integrals on 2 and 4 are divergent in the 0 0 limit and are handled much as
1
2 0
we did with divergence in k [41], that is we write 1/2 = e 0 ln 2 em ln 2 0

as 0 0 and similarly for 1/2 . Hence the integrals simply give 1 and we finally
get
85
AP4 (p1 , . . . , p4 )
g2
i
1
e 2 (p1 p2 +p3 p4 )
= 4
d/2
36 (4)
Z
0
dT m2 T
e
T d/2
ZT
dt3 e[(p1 +p2 )
2t
t3
3 (1 T )]
. (3.23)
Just like the two-point amplitude, one can also get contributions from other regions
of the string moduli in the present case. It is easy to see that the reducible 3
diagrams may degenerate giving the desired 4 diagram: this can be done in three
different ways according to where are the reducible propagators in the original cubic
diagram. One can also reproduce the integrand of (3.23) from regions different from
the one considered in (3.19). For instance, it is possible to exchange the relative order
of the punctures 3 and 4 and obtain the same result written above with p3 and
p4 exchanged. Notice however that the two contributions come from the common
boundary of two contiguous regions (3 < 4 and 4 < 3 ), so that it is natural to
weight each of them with a factor of one half [41]. A similar observation holds also
for the punctures 1 and 2 once one remembers that the points 1 and k of the
Schottky parameterization are identified and mapped to a single point on the annulus.
Since we have pinched together two pairs of external legs, this brings a factor of
1/4 which exactly cancels the degeneracy coming from the contributions related to
reducible 3 diagrams and (3.23) really represents the final result for the diagram under
study.
Next we look at the same 4-point function but with p3 and p4 nonplanar, i.e., putting
the vertex on the inner border of the loop rather than on the outer one. Then one has to
start with the non-planar string amplitude with 1 = 1 and 2 on one boundary and 3 , 4
on the other, i.e., 3 , 4 [1, k]. Inserting absolute values and signs at the appropriate
places, the scalings (3.19), (3.20) remain the same. The final result equals (3.23) with the
i
i
phase factor e 2 (p1 p2 +p3 p4 ) replaced by e 2 (p1 p2 p3 p4 ) and with an additional factor
1
2
e 4T (p3 +p4 ) (p3 +p4 ) inserted in the integrand, again in perfect agreement with the field
theory result. Following the above arguments, one can also easily work out the global
coefficient by adding again the three contribution coming from the regions related to the
reducible 3 diagrams.
Our last scalar example is this same 4-point diagram but now with only p3 being
nonplanar. Going through the same steps as before, one is led to
Z
ZT
g42
1
dT m2 T
2i (p1 p2 p3 p4 )
e
e
dt3
36 (4)d/2
T d/2
0
0

1
t
t
3
3
+
p3 2 p3 ip3 p4
.
exp (p1 + p2 )2 t3 1
T
4T
T
ANP
4 (p1 , p2 ; p3 ; p4 ) =
(3.24)
From the string point of view, it is also quite easy to fix the global normalization by
counting how many contributions one can get from the reducible 3 diagrams. As is
clear in t Hoofts double line notation, two legs can be pulled away and form a reducible
propagator only if they are on the same border. This is also accounted for in string theory
86
where reducible diagrams are related to the singularity ln |zi zj | [31] present only in the
planar case. Thus for this diagram one can get only an additional contribution beside the
one already computed, so that the final result is twice the one reported in (3.24).
A small remark is due at this point. Contrary to what one expects, the results obtained
above for scalar interaction do not seem to be real because of the presence of phase
factors. However, one has to remember that the physically meaningful quantity is the whole
amplitude which contains also the contribution coming from the non-cyclical permutation
of the diagrams calculated here. It is easy to see that in the sum the linear terms in
disappear and the final result is real. However, this observation does not apply to the
last diagram we computed; in fact, in this case, the punctures 3 and 4 stay on different
boundaries and thus can not be exchanged without changing the topology of the diagram.
i
This means that the final result has to be real by itself, except for the factor e 2 p1 p2 .
One can check this by doing the change of variables t3 T t3 in (3.24) whose effect is
exactly to switch the signs of the p3 p4 terms. Of course one or the other form is obtained
through a direct field theory computation.
3.2. Noncommutative photon
Next we study the noncommutative gauge theory with the action
Z
1
F F ,
S =
4
(3.25)
where F = A A + ig[A , A ] . The master formula for reproducing the

gluon amplitude can be easily derived from (2.1) by the usual trick of shifting the external
momenta by pi pi + Vi0 (0) i i and then isolating from the result the part linear in
all i . We obtain
(1)
AM (p1 , . . . , pM )
M

Z1
2 gop
dk
ln k d/2 Y
0 (Md2M2d)/4
(2
)
(1 k n )2d
=
(4)d/2
k2
2
n=1
0
X

Z
Z
d2 dM exp
2 0 pi pj G (i , j )
i<j

X
1
2 0 i i G (i , j )pj + i i j G (i , j )j
,(3.26)
exp
2
m.l.
i6=j
where the subscript m.l. stands for multilinear, meaning that only terms linear in each
polarization should be kept. As usual, we have used the worldsheet projective invariance
to fix = 0, and 1 = 1. Note that the measure for the punctures differs from the
one of scalar amplitude in the expected way.
The identification of the string coupling with the gluon coupling can be determined by
comparing the tree level field theory 3-point function with the tree level string 3-point
amplitude and the result is
g=
2gop (2 0 )(d4)/4.
87
(3.27)
Before we start doing computations using this master formula, we would like to
comment on the U (1) (non)decoupling in a (non)commutative U (N) YangMills theory
from the point of view of string theory. In the commutative case, the U (1) factor is free and
decouples. As it should be, this can be accounted for in string theory. For example, there
is no 3-point tree level amplitude in field theory. The way string theory produces this fact
is quite simple, the disc diagram with 3 vertex operators inserted in one order cancels the
diagram with the vertex operators inserted in the reversed order. This is a consequence of
the fact that vector states of the open bosonic theory carry a quantum number 1 under the
world-sheet parity operator . Thus a 3-point interaction among photons is forbidden
by the conservation of this quantum number. Note that this symmetry argument not only
implies that the 3-point tree level amplitude is zero, but also forbids the presence of internal
3-photon vertices in more complicated diagrams. At the loop level, it has been shown
in detail [31] that string theory again reproduces the one-loop correction to the 3 and
4-point vertex of YangMills. It is instructive to briefly review what is the mechanism
that implements the U (1) decoupling at the one loop. To understand this, we consider
the simple case of a 2-point amplitude where only one cyclical order of the external legs
is possible. A vanishing result is now assured by the cancellation among diagrams with
a different number of legs on the inner and outer border of the annulus. The reason is
quite clear: due to the form of the vector string state, each vertex brings in the amplitude
a derivative of the Green function with respect of the insertion point of the interaction. Now
since the field theory limit of the Green function is an even function (G () = G ()),
diagrams with legs on different borders will differ simply by a change of sign. In the 2point case this implies that the planar and the non-planar cancel each other to ensure the
U (1) decoupling, even if they are not separately zero. In the noncommutative case, the
U (1) factor is not free any more and does not decouple. This non-decoupling can again be
accounted for in string theory: the various diagrams, which used to cancel each other in
the = 0 case, are now dressed up with different dependent phases and do not cancel
anymore. Thus it is clear that the presence of a noncommutative parameter has the same
effect as the ChanPaton trace in the SU(N) case, and it is no surprise to see that a
dependent factor here plays the same role as the SU(N) structure constant.
After this comment on the interaction of U (1), we are ready to do the computation.
We will analyze the gluon 2-point amplitude as an example. Restricting (3.26) to the case
M = 2, the gluon 2-point amplitude is given by
2

Z1
Z

2gop
dk
ln k d/2 Y
0 1
n 2d
(2
)
k
d 0
A2 (p1 , p2 ) =
2
(4)d/2
k2
0
n=1

0
ij
0

e2 pi G (, )pj 1i
Gij 2j 2 0 1i Gij pj pi 0 Gij 2j

0
,
=1
(3.28)
88
where Gij is the Green function (2.41) or (2.43) depending on whether we are considering
the planar diagram or the nonplanar diagram; the integration region for 0 is (k, 1) in the
first case, while in the latter 0 (1, k). In order to perform the field theory limit, as
before, one has to introduce the Schwinger parameters
0
k = eT / ,
0 = et / ,
(3.29)
with t, T fixed. The modification in the planar case is simple, the additional term in the
Green function is a step function (2.41). In general one should drop the contribution coming
from the derivatives of the step function, as it corresponds to put two vertex operators at
the same point, which is not allowed in string interaction. Therefore the only modification
to the planar M gluon amplitude is the usual phase factor.
The nonplanar case is more interesting. Using (2.43) and (2.27), we arrive at
2
Z1
Zk
2
2gop
0 1
0 2 0 p 2 I0 (p)
4T J,
(2
)
[dk]
d
e
(3.30)
A2 (p, p) =
(4)d/2
0
where I0 is given by (2.27), [dk] is the measure

ln k d/2 Y
dk
(1 k n )2d ,
[dk] = 2
k
2
(3.31)
n=1
and J = J0 + J1 + J2 are the polarization dependent pieces in (3.28)

J0 = 1 2
I0 I0
2 I0
2 0 (1 p)(2 p)
,
0

i I0
1 I0
(2 p)(1 p) ,
(1 p)(2 p) + 0
(3.32)
J1 =
ln k 0

1
1
(1 p)(2 p),
J2 = 0
2
2 (ln k) 0
where the subscript of J denotes the powers of .
The 1-loop amplitude, as computed in (3.28), contains the propagation of all the string
states within the loop. This is partially reflected in the fact that the measure contains all
powers of k. As explained in [31], the contribution from the gluon (which is at level 1 of
the open string spectrum) can be identified with the contribution coming from the pieces
linear in k in the integrand (3.28). It is straightforward to do the expansion in k in order to
isolate the relevant factors

Z1
Z
dT 1
0 d/21
+ (d 2) + ,
(3.33)
[dk] = 2 (2 )
T d/2 k
0

1
1 t
k
I0
1 I0
0
0
2
.
(3.34)
=
+
k
+
O
k
= 0
0
0
0
T
2

If we now pass from the string moduli to the Schwinger parameters by means of (3.29), we
get
2
g2
A2 (p, p) = 0
(4)d/2
89
ZT

2
dT 1
(2 0 )p 2 I0 (p)
4T ( 0 J ).
+
(d
2)
+
dt
e
T d/2 k
(3.35)
Let us focus now on the contribution which is linear in

1
i 1 t
k
0
0
+ 0 k + ,
J1
ln k 0 T
2
and since 0 +
(3.36)
k 0 0, so the contribution of J1 in the two point function is
k
0

g2
p)(
p)
(
p)(
p)
(
1
2
2
1
(4)d/2

Z
ZT
1 p2 (t t 2 ) (p)2
t
dT
T
4T .
e
dt
T
2
T d/2+1
A2 (J1 ) = 2 i (d 2)
(3.37)
It is easy to see that the integral of t is is a total derivative and integrates to zero. Therefore
the linear term J1 is vanishing:
A2 (J1 ) = 0.
As for J0 , one can do an integration by part on 0 and it is easy to see that one reproduces
the usual tensor structure for the 2-point amplitude. Thus the explicit form of the J0
contribution is

g2
(1 2 )p2 (1 p)(2 p)
A2 (J0 ) = 4
(4)d/2

Z
ZT
(p)2
t2
1 2
t
dT
2
dt
(d
2)
2
ep (t T ) 4T .
(3.38)
d/2
T
2
T
0
For d = 4, the integral is

1
2
dT p 2
e 4T
T
Z1
dx eTp
2 x(1x)
(1 2x)2 4
Z1
=
p

dx (1 2x)2 4 K0 x(1 x)|p||p|
.
Now the leading divergence for the Bessel function is

2
K0 (x) ln ,
x
So we obtain
x 0.

g2
22
2
2
2
ln
|p|
|
p|

)p
(
p)(
p)
.
+
(
1
2
1
2
(4)d/2 3
Notice that there is a ln -type singularity.
A2 (J0 ) =
(3.39)
(3.40)
90
Finally we look at J2 , which is more interesting. It is

0 1
(1 p)(2 p),
2T 2 0
and hence it is contribution in A2 is
J2 =
g2
A2 (J2 ) = (d 2)
(4)d/2
(3.41)
ZT
dT
T d/2+2
dt ep
2 (t t 2 ) (p)
T
4T
(1 p)(2 p).
(3.42)
For d = 4, the integral is equal to

Z
dT (p)2
e 4T
T3
Z1
dx eTp
2 x(1x)
16 2
= 4 2
p a
=
32 2
p 4 a 2
dT a(p) 2
e 4T
T
Z1
dx K0
Z1
dx e
Tp 2 x(1x)

x(1 x)|p||p|
a
a=1
a=1
(3.43)
By expanding the above formula for small momenta, we can extract the leading IR behavior
of J2
A2 (J2 ) =
2g 2 (1 p)(
2 p)
+ .
2
4
(3.44)
This form of 1/ 2 singularity in the two gluon amplitude was first obtained in [18,19] and
is another example of UV/IR mixing in noncommutative field theory which was stressed
in [10,11]. Also in the 4 dimensional scalar 4 theory the leading singularity is of the form
1/(p)2 as one can easily see from the result of the nonplanar 2-point function (3.18).
Since the commutative gauge theory is less divergent in the UV than the commutative
scalar theory, one could have expected that the leading singularity in the noncommutative
gauge theory is of the form ln(p)2 . However, as was shown in [1820], this expectation is
not correct and one still gets the 1/(p)2 behavior. As pointed out in [20], the appearance of
the 1/(p)2 piece does not spoil gauge invariance. In fact it does not correspond to a mass
shift because of its unusual tensor structure. Here we have rederived this result using string
theory. In addition, we have been able to determine also the subleading singularity from
string theory which is of the form ln(p)2 from string theory.
Summarizing, we have

g2
22
2 p)
(1 p)(
2
2
2
ln
|p|
+
|
p|

)p
(
p)(
p)
+ ,
(
32
ANP =
1
2
1
2
3
16 2
p 4
(3.45)
where dots stay for terms which are finite when 0. It is interesting to note that the
whole result comes from a single corner of the moduli space. This is because we performed
91
an integration by parts which has the effect of including the tadpole contribution in the
region here considered. We note that the vacuum polarization of the photon was also
obtained by Hayakawa [18,19] in the Feynman gauge. The coefficient of the ln term is
clearly related to the wave-function renormalization constant of the commutative case and
thus is gauge dependent. The result found in [18,19] is thus different from ours, because
his calculation was performed in the Feynman gauge. On the contrary, as explained in [31]
for the commutative case, string theory amplitudes naturally give rise to field theory results
in the background field method [60]. Recall that a distinct property of Feynman diagrams
in the background field gauge is to leave the gauge invariance of the external particles
unbroken. For the noncommutative case, the situation is the same and we conclude that
string theory again gives field theory results computed in the background field method.
One of the advantages of our result is that the function can be determined directly from
the two point function without the need of computing the vertex corrections. By exploiting
this and the fact that the planar diagram contribution is exactly the same (apart from a
trivial overall group theory factor of N ) as those in a usual commutative SU(N) gauge
theory, we easily obtain the 1-loop function
=
g 2 22
.
16 2 3
(3.46)
The -function for the noncommutative QED has also been computed in [1519] in the
Feynman gauge, by combining vertex corrections and wave function renormalization. The
YangMills -function can be elegantly derived also from the string generating functional
by following the approach 5 of Mastaev and Tseytlin [61]. Again, the noncommutative
parameter can be easily implemented in the string calculation by switching on a constant
magnetic field. It is easy to see that the derivation of [61] is not substantially modified and
one finds the distinctive factor 26 D: this is, from this point of view, the source of the
YangMills -function 22/3.
4. Discussion
In this paper, we used the boundary state approach to compute the one-loop open string
Green function by calculating the string amplitude with two boundaries inserted. We note
that generally there is an ambiguity in the resulting Green function that cannot be fixed
from the boundary state approach. We pointed out that the scalar amplitudes do not care
about this ambiguity. However, this ambiguity can be fixed by simply comparing with
the planar gluon 2-point function and one obtains the correct open string Green function
which can then be used in the string master formula in a uniform manner for all mass levels.
A more satisfactory way [62] to obtain the desired Green function would be to calculate
directly the open string scattering amplitudes using the vertex operators constructed from
the open string modes (2.35)(2.37) and extract from there the Green function.
5 We would like to thank A. Tseytlin for pointing out this possibility to us.
92
An interesting behavior of noncommutative quantum field theory is that although

classically the noncommutative description is a smooth deformation of the commutative
description, quantum mechanically they are different and generally one cannot turn off
to smoothly reduce back to the commutative description. The UV/IR mixing [10,11]
and the induced ChernSimons action in odd dimensions [25] are examples. The fact that
a commutative field theory differ intrinsically from a noncommutative field theory is clear
from the present approach. It is important to recall that in deriving the noncommutative
field theory limit from string theory, one has to take the double scaling limit (1.4) with
0 0, F and 0 F fixed in order to have a nonzero . The commutative case = 0,
however, is obtained from a different limit with F fixed instead and it is thus clear that the
commutative and the noncommutative field theories thus obtained differ in character.
We now discuss some possible further directions of work. In this paper, we have shown
that by taking the limit (1.4), (1.5) appropriately, one can obtain field theory results from
a single string master formula in a uniform manner. We presented explicit calculations for
both the case of noncommutative scalar theories and noncommutative gauge theory with
gauge group U (1). However, it is straightforward to add ChanPaton factors and obtain
results for noncommutative field theory with gauge group U (N). One difference is that
in the U (N) case, the action is no longer even in and one will get field theory result
with odd power dependence in also. This can definitely be accounted for by the Green
function we obtained. Another possible generalization is to consider superstring amplitudes
in the presence of F -field and to extract the field theory limit. The field theory is of course
a noncommutative supersymmetric one. It may be interesting to understand some of the
field theory issues (like non-renormalization theorem and duality for noncommutative field
theory) from the string point of view. Also, it is possible to generalize all this to higher
loops [62] and we hope to report on some of these topics in the future.
Acknowledgements
R.R. wants to thank P. Di Vecchia, M. Frau, A. Lerda and R. Marotta for useful
discussions and A. Frizzo and L. Magnea for e-mail exchange. We are also grateful to
H. Dorn for helpful correspondence after the submission of the first version of the present
paper. This work was partially supported by the Swiss National Science Foundation, by the
European Union under TMR contract ERBFMRX-CT96-0045 and by the Swiss Office for
Education and Science.
References
[1]
[2]
[3]
[4]
[5]
[6]
A. Connes, M.R. Douglas, A. Schwarz, JHEP 02 (1998) 003, hep-th/9711162.

M.R. Douglas, C. Hull, JHEP 02 (1998) 008, hep-th/9711165.
C.-S. Chu, P.-M. Ho, Nucl. Phys. B 550 (1999) 151, hep-th/9812219.
V. Schomerus, JHEP 06 (1999) 030, hep-th/9903205.
C.-S. Chu, P.-M. Ho, M. Li, hep-th/9911153.
N. Seiberg, E. Witten, JHEP 09 (1999) 032, hep-th/9908142.
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
93
C.-S. Chu, hep-th/0001144.

A. Abouelsaood, C.G. Callan, C.R. Nappi, S.A. Yost, Nucl. Phys. B 280 (1987) 599.
I. Chepelev, R. Roiban, hep-th/9911098.
S. Minwalla, M.V. Raamsdonk, N. Seiberg, hep-th/9912072.
M.V. Raamsdonk, N. Seiberg, hep-th/0002186.
I.Ya. Arefeva, D.M. Belov, A.S. Koshelev, hep-th/9912075.
I.Ya. Arefeva, D.M. Belov, A.S. Koshelev, hep-th/0001215.
W. Fischler, E. Gorbatov, S. Paban, A. Kashani-Poor, P. Pouliot, J. Gomis, hep-th/0002067.
C.P. Martin, D. Sanchez-Ruiz, Phys. Rev. Lett. 83 (1999) 476479, hep-th/9903077.
M.M. Sheikh-Jabbari, JHEP 06 (1999) 015, hep-th/9903107.
T. Krajewski, R. Wulkenhaar, hep-th/9903187.
M. Hayakawa, hep-th/9912094.
M. Hayakawa, hep-th/9912167.
A. Matusis, L. Susskind, N. Toumbas, hep-th/0002075.
J. Madore, S. Schraml, P. Schupp, J. Wess, hep-th/0001203.
F. Ardalan, N. Sadooghi, hep-th/0002143.
J.M. Gracia-Bondia, C.P. Martin, hep-th/0002171.
L. Bonora, M. Schnabl, A. Tomasiello, hep-th/0002210.
C.-S. Chu, hep-th/0003007.
Z. Bern, D.A. Kosower, Phys. Rev. Lett. 66 (1991) 16691672.
Z. Bern, D.A. Kosower, Nucl. Phys. B 379 (1992) 451.
Z. Bern, L. Dixon, D.A. Kosower, Ann. Rev. Nucl. Part. Sci. 46 (1996) 109, hep-ph/9602280.
Z. Bern, D.A. Kosower, Phys. Rev. D 38 (1988) 1888.
P. Di Vecchia, A. Lerda, L. Magnea, R. Marotta, Phys. Lett. B 351 (1995) 445, hep-th/9502156.
P. Di Vecchia, A. Lerda, L. Magnea, R. Marotta, R. Russo, Nucl. Phys. B 469 (1996) 235,
hep-th/9601143.
Z. Bern, D.C. Dunbar, T. Shimada, Phys. Lett. B 312 (1993) 277284, hep-th/9307001.
Z. Bern, L. Dixon, D.C. Dunbar, M. Perelstein, J.S. Rozowsky, Nucl. Phys. B 530 (1998) 401,
hep-th/9802162.
K. Roland, Phys. Lett. B 289 (1992) 148.
P. Di Vecchia, A. Lerda, L. Magnea, R. Marotta, R. Russo, Phys. Lett. B 388 (1996) 65, hepth/9607141.
K. Roland, H. Sato, Nucl. Phys. B 485 (1996) 241, hep-th/9604152.
K. Roland, H. Sato, Nucl. Phys. B 515 (1997) 488, hep-th/9709019.
L. Magnea, R. Russo, Two loop gluon diagrams from string theory, in: Proceedings of the
Workshop Deep Inelastic Scattering 97, Chicago, 1997.
Z. Bern, J.S. Rozowsky, B. Yan, Phys. Lett. B 401 (1997) 273, hep-ph/9702424.
R. Marotta, F. Pezzella, hep-th/9912158.
A. Frizzo, L. Magnea, R. Russo, hep-th/9912183.
L. Cappiello, R. Marotta, R. Pettorino, F. Pezzella, Mod. Phys. Lett. A 13 (1998) 2433, hepth/9804032.
L. Cappiello, R. Marotta, R. Pettorino, F. Pezzella, Mod. Phys. Lett. A 13 (1998) 2845, hepth/9808164.
A. Liccardo, R. Marotta, F. Pezzella, Mod. Phys. Lett. A 14 (1999) 799, hep-th/9903027.
M.J. Strassler, Nucl. Phys. B 385 (1992) 145, hep-ph/9205205.
M.G. Schmidt, C. Schubert, Phys. Lett. B 331 (1994) 6976, hep-th/9403158.
C. Schubert, Acta Phys. Polon. B 27 (1996) 3965, hep-th/9610108.
H.-T. Sato, M.G. Schmidt, Nucl. Phys. B 524 (1998) 742, hep-th/9802127.
H.-T. Sato, M.G. Schmidt, Nucl. Phys. B 560 (1999) 551, hep-th/9812229.
P. Di Vecchia, F. Pezzella, M. Frau, K. Hornfeck, A. Lerda, S. Sciuto, Nucl. Phys. B 322 (1989)
317.
94
[51] M. Frau, I. Pesando, S. Sciuto, A. Lerda, R. Russo, Phys. Lett. B 400 (1997) 52, hep-th/9702037.
[52] P. Di Vecchia, M. Frau, I. Pesando, S. Sciuto, A. Lerda, R. Russo, Nucl. Phys. B 507 (1997)
259, hep-th/9707068.
[53] P. Di Vecchia, A. Liccardo, hep-th/9912161.
[54] P. Di Vecchia, A. Liccardo, hep-th/9912275.
[55] P. Di Vecchia, M. Frau, A. Lerda, A. Liccardo, hep-th/9906214.
[56] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, Cambridge University Press, 1987.
[57] T. Filk, Phys. Lett. B 376 (1996) 53.
[58] C.-S. Chu, F. Zamora, JHEP 02 (2000) 022, hep-th/9912153.
[59] R. Marotta, F. Pezzella, hep-th/0003044.
[60] L.F. Abbott, Nucl. Phys. B 185 (1981) 189.
[61] R.R. Metsaev, A.A. Tseytlin, Nucl. Phys. B 298 (1988) 109.
[62] Work in progress.
[63] O. Andreev, H. Dorn, hep-th/0003113.
[64] Y. Kiem, S. Lee, hep-th/0003145.
[65] C.-S. Chu, R. Russo, S. Sciuto, hep-th/0004188.

Superconformal hypermultiplets in superspace

Sergei V. Ketov a,b,1
a Niels Bohr Institute, University of Copenhagen, 2100 Copenhagen , Denmark
b Institut fr Theoretische Physik, Universitt Hannover Appelstrae 2, Hannover 30167, Germany
Received 9 February 2000; revised 17 April 2000; accepted 26 April 2000
Abstract
We use the manifestly N = 2 supersymmetric, off-shell, harmonic (or twistor) superspace
approach to solve the constraints implied by four-dimensional N = 2 superconformal symmetry
on the N = 2 non-linear sigma-model target space, known as the special hyper-Khler geometry.
Our general solution is formulated in terms of a homogeneous (of degree two) function of
unconstrained (analytic) FayetSohnius hypermultiplet superfields. We also derive the improved
(N = 2 superconformal) actions for the off-shell (constrained) N = 2 projective hypermultiplets,
and relate them (via non-conformal deformations) to the asymptotically locally-flat (ALF) Ak and
Dk series of the gravitational instantons. The same metrics describe KaluzaKlein monopoles in
M-theory, while they also arise in the quantum moduli spaces of N = 4 supersymmetric gauge field
theories with SU(2) gauge group and matter hypermultiplets in three spacetime dimensions. We
comment on rotational isometries versus translational isometries in the context of N = 2 NLSM in
terms of projective hypermultiplets. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.30.Pb; 11.10.Lm; 11.15.Tk; 12.60.Jv
1. Introduction
Better understanding of non-conformal supersymmetric field theories is often facilitated
by the study of superconformal field theories. One may, therefore, expect, that studying
the N = 2 (rigid) superconformal Non-Linear Sigma-Models (NLSM) in 3 + 1 spacetime
dimensions may shed light on the structure of the hypermultiplet low-energy effective
actions in quantized N = 2 supersymmetric gauge field theories, as well as provide
more insights into the moduli spaces of hypermultiplets in the type-II superstring
compactifications on CalabiYau threefolds in the limit where the supergravity decouples,
since they are all governed by hyper-Khler metrics. The N = 2 superconformal
hypermultiplets also appear in describing the D3-brane world-volume field theories
1 E-mail: ketov@itp.uni-hannover.de; also at High Current Electronics Institute of the Russian Academy of
Sciences, Siberian Branch, Akademichesky 4, Tomsk 634055, Russia.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 5 2 - 2
96
S.V. Ketov / Nuclear Physics B 582 (2000) 95118
[13], and in relation to the AdS/CFT correspondence [4]. The universal formulation of
hypermultiplets is indispensable for those purposes.
The most natural formulation of supersymmetry is provided by superspace. However,
as regards N = 2 supersymmetry in 1 + 3 dimensions, the standard N = 2 superspace
is not suitable to describe off-shell hypermultiplets. Since N = 2 supersymmetry in the
four-dimensional NLSM amounts to hyper-Khler geometry in the NLSM target space [5],
there should be a good reason for this failure inside the hyper-Khler geometry. As is well
known, the hyper-Khler geometry is characterized by the existence of three, covariantly
constant, complex structures, I (a) , a = 1, 2, 3, satisfying a quaternionic algebra. In the
conventional superspace approach one picks up a single complex structure to be manifest.
In fact, any linear combination, c1 I (1) + c2 I (2) + c3 I (3) , of the complex structures with
some real coefficients ca is again a covariantly constant complex structure provided
that c12 + c22 + c32 = 1. There is, therefore, the whole sphere S 2 of the hidden complex
structures on the top of the manifest one. To treat all the complex structures on equal
footing, one should add S 2 to the hyper-Khler manifold, which essentially amounts to its
twistor extension. In the context of N = 2 supersymmetry, the extension of the standard
N = 2 superspace by the two-sphere S 2 gives rise to the off-shell (model-independent)
formulation of a hypermultiplet with manifest (linearly realized) N = 2 supersymmetry.
The twistor extension of N = 2 superspace comes in two versions known as Projective
Superspace (PSS) and Harmonic Superspace (HSS). The PSS appears after adding a
single projective holomorphic coordinate of CP1 S 2 . The PSS construction naturally
leads to holomorphic potentials for hyper-Khler metrics via the so-called generalized
Legendre transform in terms of projective O(2p) hypermultiplets [68]. In the HSS
construction [9] one uses another isomorphism S 2 SU(2)/U (1), one adds harmonics
belonging to the group SU(2), and one considers only equivariant (Grassmann-)analytic
functions with respect to the U (1) charge (cf. the notion of a flag manifold). The HSS thus
naturally leads to analytic potentials for hyper-Khler metrics. An off-shell, manifestly
N = 2 supersymmetric formulation of the most basic FayetSohnius (FS) hypermultiplet
is only possible with the infinite number of the auxiliary fields. This problem is elegantly
solved in HSS that provides the universal formulation of a hypermultiplet in terms of
a single unconstrained N = 2 superfield in the analytic subspace of HSS [9]. At the
same time, the infinite number of the auxiliary fields leads to a quite obscure connection
(bridge) between the HSS superfields and their physical components, which highly
complicates a derivation of hyper-Khler metrics out of the N = 2 NLSM in terms of
the FS hypermultiplet superfields. The projective hypermultiplets with the finite numbers
of the auxiliary fields, are more suitable for describing the N = 2 NLSM metrics with
isometries, either translational or rotational.
The combined constraints implied by hyper-Khler geometry and conformal invariance
on the target space geometry of the (1 + 3)-dimensional N = 2 supersymmetric NLSM
were recently investigated by de Wit, Kleijn and Vandoren [10] who called the N =
2 superconformal NLSM geometry special hyper-Khler. They also found remarkable
relations between the special hyper-Khler, quaternionic and 3-Sasakian metrics, by using
the component approach with on-shell N = 2 supersymmetry for hypermultiplets. The
97
approach adopted in Ref. [10] is invariant under reparametrizations, but it formulates the
special hyper-Khler geometry in terms of complicated geometrical constraints that are
unavoidable in any component approach. We use the manifestly N = 2 supersymmetric
HSS approoach to solve those constraints in terms of an unconstrained homogeneous (of
degree two) function of hypermultiplets, similarly to the standard N = 2 superconformally
invariant action of N = 2 vector multiplets.
Though our general HSS construction uses the infinite number of auxiliary fields,
we also derive the non-trivial N = 2 superconformal NLSM in terms of the projective
hypermultiplets with the finite numbers of the auxiliary fields. As an application, we
easily reproduce Ak and Dk series of four-dimensional gravitational instanton metrics
by combining the (improved) special hyper-Khler potentials with different moduli and
adding simple non-conformal deformations to them. Those metrics naturally arise in
modern gauge and string theories that supply more motivation for the alternative and more
transparent derivation of the metrics from HSS.
Distant physical problems in field theory often share common hyper-Khler moduli
space. For example, the vacuum structure of the quantized SU(n)-based N = 4 supersymmetric (pure) gauge field theory in 2 + 1 dimensions (in the Coulomb branch) is known
to be equivalent to the moduli space of n BPS monopoles in the classical SU(2)-based
(N = 0) YangMillsHiggs system in 3 + 1 dimensions [11,12]. The four-dimensional
hyper-Khler spaces are necessarily self-dual, while the latter naturally arise as gravitational instantons in quantum gravity, or as the moduli spaces of solutions to the (integrable)
system of Nahm equations with appropriate boundary conditions [1316].
M-theory provides the unifying framework for a study of those remarkable correspondences, while it also offers their explanation via dualities. For example, the M-theory compactification on the Ak1 Asymptotically Locally Flat (ALF) self-dual space is equivalent
to the background of k parallel D6-branes in the type-IIA string picture [17]. 2 Probing
this background with a parallel D2-brane gives rise to the (2 + 1)-dimensional N = 4 supersymmetric effective gauge field theory with k matter hypermultiplets in the D2-brane
world-volume, whose moduli space is given by the ALF space. In the (dual) type-IIB string
picture one gets a BPS configuration of intersecting D5- and D3-branes, which gives rise
(in the infra-red limit) to the same effective field theory in the D3-brane world-volume,
and whose moduli space is given by the charge-two (centered) BPS monopole moduli
space with k singularities [18].
Though the brane technology appears to be very efficient in establishing the equivalence
between the apparently different moduli spaces, it does not offer any means for a
calculation of the exact moduli space metrics. The standard hyper-Khler quotient
construction of the multi-monopole moduli space metrics from the Nahm data is very
complicated in practice, so that it does not allow one to establish a simple and natural
way of describing the metrics. The alternative approach is provided by twistor methods
whose essential ingredients are given by holomorphic vector bundles over the twistor
space [19]. The ALF metrics can be obtained from the (moduli-dependent) holomorphic
2 In the case of D ALF space, one gets k D6-branes parallel to an orientifold O6 [17].
k
98
potentials in the twistor space by the generalized Legendre transform that was originally
deduced from PSS [6,7]. However, even the generalized Legendre transform techniques are
usually limited to the hyper-Khler metrics having isometries, whereas a generic monopole
moduli space does not have any isometries. It is, therefore, very desirable to develop a more
universal approach to this problem.
The ALF spaces asymptotically approach R3 S 1 / , where is a discrete subgroup
of SU(2). After sending the radius of S 1 to infinity, one gets the Asymptotically Locally
Euclidean (ALE) metrics that asymptotically approach R 4 / . Kronheimer [20] found their
ADE classification into two infinite (Ak and Dk ) and three exceptional E6,7,8 cases,
according to the intersection matrix of their two-cycles. Since the ALE spaces are naturally
related to the enhanced symmetries, most notably, conformal invariance, it is natural to
approach an ALF moduli space metric from its ALE limit, being guided by hyper-Khler
geometry and conformal invariance on the top of it. In the context of N = 2 NLSM in
superspace, the ALE potentials can be constructed by switching on vacuum expectation
values of the hypermultiplet scalars, whereas the associated ALF potentials are obtained
by non-conformal deformations of them. To solve the hyper-Khler constraints, we use
an unconstrained hyper-Khler potential in HSS. The N = 2 superconformal symmetry
implies extra constraints on the special hyper-Khler potentials, which can be easily solved
in HSS too.
Our paper is organized as follows. In Section 2 we review some basic properties of
the special hyper-Khler geometry and its relation to N = 2 superconformal symmetry
[10]. In Section 3 we introduce rigid N = 2 superconformal transformations in superspace
[21,22]. The N = 2 superconformal rules for various types of the N = 2 hypermultiplet
superfields are discussed in Section 4. In Section 5 a simple general solution to the special
hyper-Khler geometry in N = 2 HSS is given. In Section 6 we turn to a construction of
the improved (N = 2 superconformal) actions of the O(2p) projective multiplets in HSS,
and then relate them to the (ALE and ALF) Ak and Dk series of gravitational instantons.
Section 7 is devoted to our conclusion. All efforts were made to make our presentation as
simple as possible.
2. Special hyper-Khler NLSM geometry

Let (P , M ; D, K ) be the generators of the standard conformal extension of the
Poincar algebra in 3 + 1 spacetime dimensions, , , . . . = 0, 1, 2, 3, where P stand for
translations, M for Lorentz rotations, D for dilatations and K for special conformal
transformations. The commutation relations of the conformal algebra are given by a
contracted so(4, 2) algebra. We use middle Greek letters to denote spacetime vector
indices, whereas early Greek letters are reserved for the 2-component spinor indices,
, , . . . = 1, 2. A vector index () is equivalent to a bi-spinor index ( ).

The N = 2 superconformal algebra extends the conformal algebra to a contracted
su(2, 2|2) superalgebra. The new generators are given by bosonic charges of the
99
SU(2)conf. U (1)ch. internal symmetry, eight fermionic supersymmetry charges, Qi and
Qi , and eight fermionic special supersymmetry charges, Si and S i , where i = 1, 2.

In the context of N = 2 supersymmetric NLSM in 3 + 1 dimensions, N = 2
supersymmetry amounts to the hyper-Khler NLSM target space M of real dimension 4k,
k = 1, 2, . . . , whose holonomy group is in Sp(k) [5]. Given the full N = 2 superconformal
invariance, its internal su(2)conf. part implies an su(2) isometry of the hyper-Khler NLSM
m
obeying Killing equations,
metric gmn , i.e., the existence of three Killing vectors K(A)
(m;n)
K(A)
=0,
m = 1, 2, . . . , 4k,
A = 1, 2, 3,
(2.1)
and forming an su(2) algebra. This non-abelian isometry is necessarily rotational, i.e.,
it rotates complex structures in the hyper-Khler NLSM target space M [10]. The
dilatational invariance of a Riemannian manifold M is equivalent to the existence of
another (Eulerian) vector Xm satisfying an equation [10]
Xm;n = g mn .
(2.2)
Its geometrical significance was clarified in Ref. [23], where Eq. (2.2) was shown to be
equivalent to the following form of the metric:
gmn dx m dx n = dr 2 + r 2 hab dx a dx b ,
(2.3)
where
x n = (r, x a ),
a, b, c = 1, 2, . . . , 4k 1,
and hab = hab (x c ).
(2.4)
Eq. (2.3) means that M can be considered as a cone C(B) over a base manifold B of
dimension 4k 1 [23]. The vector Xm in the coordinates (2.4) reads
X=r
,
r
(2.5)
so that it is associated with the dilatations (r, x a ) (r, x a ) indeed.

Eq. (2.2) is obviously equivalent to the conformal Killing equation,
LX gmn = Xm;n + Xn;m = 2gmn ,
(2.6)
together with a condition X[m;n] = 0, or, equivalently,

Xm = m f.
(2.7)
Eqs. (2.6) and (2.7) imply that the metric gmn admits a potential f (x n ),
gmn = m n f,
or gmn Xm Xn = 2f.
(2.8)
In the context of hyper-Khler geometry, the potential f also generates the complex
structures by differentiation, so that the function f is sometimes called the hyper-Khler
potential [10,24]. It is worth noticing here that the potential f is a constrained function in
any geometry (e.g., m n p f = 0), whereas we are going to introduce the hyper-Khler
(pre-)potential as an unconstrained function (Section 5).
100
Though the Euler vector X is not a Killing vector, it is easy to verify that it implies the
existence of a Killing vector Y in the presence of a covariantly constant complex structure I
on M [10,23],
Y n = I n m Xm ,
LX I = 0,
[X, Y ] = 0.
(2.9)
The second equation means that the vector X is holomorphic, i.e., it preserves the complex
structure. The corresponding base manifold B then carries the so-called Sasakian structure
to be obtained by projection of the (I, X, Y ) structure of M on B [25]. Since in our case
M is a (special) hyper-Khler manifold, the base manifold B should admit a 3-Sasakian
structure because of the existence of three independent and covariantly constant complex
structures on M,
n
n
m
= I(A)
Y(A)
mX ,
[X, YA ] = 0.
(2.10)
One easily finds that the vector X is tri-holomorphic, and [23]

LYA IB = 2ABC IC ,
[YA , YB ] = 2ABC YC .
(2.11)
This means that the Killing vectors YA are rotational and form an su(2) algebra indeed
[10]. The complex structures are invariant under dilatations.
The base manifold B associated with a special hyper-Khler manifold M is called 3Sasakian (see Ref. [26] for a recent mathematical account of the 3-Sasakian manifolds). A
3-Sasakian manifold of real dimension (4k 1) is an Einstein space,
Rab = (4k 1)hab ,
(2.12)
while it takes the form of an Sp(1) fibration over a quaternionic space [26]. In Ref. [10] the
special hyper-Khler manifolds (of real dimension 4k) were described as the local products
of flat 4-dimensional space with a (4k 4)-dimensional quaternionic manifold, i.e., as the
manifolds of Sp(k 1) holonomy. The special hyper-Khler manifolds can be equally
defined as cones over 3-Sasakian manifolds [23]. Some applications of the 3-Sasakian
manifolds in M-theory were discussed in Refs. [24,27].
Our purpose is to solve the constraints implied by the special hyper-Khler geometry,
in terms of an unconstrained special hyper-Khler (pre-)potential. By using the established
relation between the special hyper-Khler geometry and the N = 2 superconformal symmetry, the solution amounts to a formulation of the most general N = 2 superconformally
invariant NLSM. To do this in superspace, we need a HSS realization of N = 2 superconformal transformations.
3. N = 2 superconformal transformations in HSS

The N = 2 superconformal transformations in the ordinary N -extended superspaces are
known for a long time [21]. It is, therefore, straightforward to rewrite them to the case of
N = 2 supersymmetry in HSS [22]. We follow Ref. [22] in this section.
101
In the HSS approach the standard N = 2 superspace coordinates Z = (x , i , i ) are
extended by bosonic harmonics (or twistors) ui , i = 1, 2, belonging to the group SU(2)

and satisfying the unimodularity condition
u+i u
i = 1,
ui+ = u
i .
(3.1)
The original motivation for an introduction of harmonics [9] was the desire to make
manifest the hidden analyticity structure of the N = 2 superspace constraints defining
both N = 2 vector multiplets and FS hypermultiplets, and find their manifestly N = 2
supersymmetric solutions in terms of unconstrained N = 2 superfields.
Instead of using an explicit parametrization of the sphere S 2 , in HSS one deals with the
equivariant functions of harmonics, which have definite U (1) charges defined by U (u
i )=
1. The simple harmonic integration rules,
Z
Z
(3.2)
du = 1 and
du u+(i1 u+im uj1 ujn ) = 0 otherwise,
are similar to the (Berezin) integration rules in superspace. In particular, any harmonic
integral over a U (1)-charged quantity vanishes. The harmonic covariant derivatives,
preserving the defining equations (3.1) in the original (central) basis, are given by
++ ,
D = ui +i ,
i
u
u
0
+i
i
u
.
(3.3)
D =u
u+i
ui
They satisfy an su(2) algebra and commute with the standard (flat) N = 2 superspace
D ++ = u+i
covariant derivatives Di and D i . The operator D 0 measures U (1) charges.

The key feature of HSS is the existence of an analytic subspace parametrized by
= x 4i i j u+
( ; u) = xanalytic
(i uj ) ,

+ i +
+ = i u+
i , = ui ; ui ,
(3.4)
which is invariant under N = 2 supersymmetry [9]:
+
= 4i i + + + i u
+ + ,
xanalytic
i 4i
+
+ = i u+
i ,
+
+
i u+
=
i ,
u
i = 0,
(3.5)
where only + are present, not .

,
The usual complex conjugation does not preserve analyticity. However, it does, after
being combined with another (star) conjugation that only acts on the U (1) indices as
i
i
(u+
i ) = ui and (ui ) = ui . One has u = ui and ui = u .
(q)
Analytic off-shell N = 2 superfields ((Z, u), u) of any positive (integral) U (1)
charge q in HSS are defined by (cf. N = 1 chiral superfields)
D+ (q) = D + (q) = 0,
where D+ = Di u+
i
and D + = D i u+
i .
(3.6)
102
The analytic measure reads d (4) du d4 xanalytic d2 + d2 + du, and it has the U (1)
charge (4). The harmonic derivative D ++ in the analytic subspace (3.4) takes the form
++
= ++ 4i + +
Danalytic
(3.7)
,
x
it preserves analyticity, and it allows one to integrate by parts. Similarly, one easily finds
that
0
= u+i +i ui i + + + + + .
(3.8)
Danalytic
u
u
+
In what follows we omit the explicit references to the analytic subspace, in order to simplify
our notation.
The use of harmonics allows one to make manifest (i.e., linearly realized) the SU(2)R
symmetry of N = 2 supersymmetry algebra, in addition to manifest N = 2 supersymmetry
(see Ref. [28] for more details). The relation to PSS, where the SU(2)R rotations take the
form of projective transformations, becomes clear in a particular parametrization

1
1
i
=
(1,
),
u
=
,
(3.9)
u+
i
2
|
|
1+
where is the projective (complex) CP1 coordinate.
The translational and Lorentz transformation properties of the HSS coordinates are
obvious, and we do not write them down. The transformation rules with respect to
dilatations with the infinitesimal parameter are also rather evident, being dictated by
conformal weights w,
1
w[u] = 0.
(3.10)
w[ ] = w[ ] = ,
2
The non-trivial part of N = 2 superconformal transformations is given by the U (2)conf. rotations with the parameters l ij , special conformal transformations with the parameters k ,
w[x] = 1,
and N = 2 special supersymmetry with the parameters i and i .

The N = 2 superconformal extension of the spacetime conformal transformations,
x = x + k
x x ,
(3.11)
is dictated by the requirement of preserving the unimodularity and analyticity conditions

in Eqs. (3.1) and (3.6), respectively. As regards the non-trivial part of the N = 2
superconformal transformation laws, one finds [22]

+ +
+ k x x + 4i x + x + ,
x = 4iij u
i uj

+
+ k x + 2i + + + x + ,
+ = ij u+
i uj
+ = ( + ),
h

i
kj + +
+ +
+ 4i + + + + + u
u+
i = uk uj + 4ik
i ,
u
i = 0.
103
(3.12)
Since the building blocks of any invariant action in HSS are given by the measure,
analytic superfields and HSS covariant derivatives, only their transformation properties
under rigid N = 2 superconformal transformations are going to be relevant for our
purposes. It follows from Eq. (3.12) that [22]
Ber
( 0 , u0 )
= 1 2,
(, u)
or

d (4) du = 2 d (4) du ,
where the HSS superfield parameter

j
= + k x + ij + 4i i j + 4i i u+
i uj
(3.13)
(3.14)
has been introduced. Similarly, one easily finds that

(D ++ )0 = D ++ (D ++ )D 0
and (D 0 )0 = D 0 .
(3.15)
The truly N = 2 superconformal (infinitesimal) component parameters can thus be

nicely encoded into the single scalar superfield subject to the HSS constraint [22]
(D ++ )2 = 0,
(3.16)
and the reality condition
(++ )= ++ ,
where ++ D ++ .
(3.17)
The transformations rules of the harmonics in Eq. (3.12),

++
ui ,
u+
i =
u
i = 0,
(3.18)
together with Eqs. (3.13), (3.15), (3.16) and (3.17) represent the very simple and convenient
description of rigid N = 2 conformal supersymmetry in HSS.
4. Superconformal hypermultiplet superfields

The O(2p) projective (or generalized tensor) multiplets in the standard N = 2
superspace are described by the N = 2 superfields Li1 i2p that are totally symmetric with
respect to their SU(2)R indices, being subject to the constraints [8]
D(k Li1 i2p ) = D (k Li1 i2p ) = 0,
(4.1)

Li1 i2p (Li1 i2p ) = i1 j1 i2p j2p Lj1 j2p .
(4.2)
The N = 2 projective multiplets are all irreducible off-shell representations of N = 2

supersymmetry with superspin Y = 0 and superisospin I = p 1 as long as p > 1. The
list of their off-shell, 8(2p 1) bosonic and 8(2p 1) fermionic, field components is most
conveniently represented in terms of the SU(2) Young tableaux [8]:
104
(4.3)
where the boxes with dots and stars denote the N = 2 superspace covariant derivatives, Di
and D , respectively. It follows from matching the numbers of the bosonic and fermionic
i
degrees of freedom in Eq. (4.3) that the vector V in the N = 2 tensor multiplet (p = 1) is
i i
conserved, V = 0. The vector V 3 2p of any projective N = 2 multiplet with p > 1
is an unconstrained (general) vector field.

The PSS naturally comes out of the efforts to construct an N = 2 supersymmetric
self-interaction of the projective multiplets in N = 2 superspace [68]. Lets introduce
a function G(LA ; , ), whose arguments are given by some number (A = 1, 2, . . . , k) of
the O(2p) projective superfields (of any type) and two extra complex coordinates, and .
Lets also impose four linear differential equations on this function,

1
2
1 G D + D G = 0.
(4.4)
G D1 + D2 G = 0,
It follows from the defining constraints (4.1) that a general solution to Eq. (4.4) reads

G = G QA ( ); , = ,
Q(2p) ( ) i1 i2p Li1 i2p ,
i (1, ),
(4.5)
in terms of an arbitrary function G(QA ; ). Since the function G does not depend upon
a half of the Grassmann coordinates of N = 2 superspace by construction, its integration
over the rest of the coordinates is invariant under N = 2 supersymmetry. This leads to the
following universal N = 2 supersymmetric action for the projective multiplets in PSS [6,
8]:
Z
I
4 2 2
1
e 1
e G(QA , ) + h.c.,
d 1 + 2
(4.6)
S[LA ] = d4 x
2i
C
105
where the new derivatives,

e D1 D2 ,
e D 1 D 2 ,
1
(4.7)
in the directions orthogonal to the vanishing directions of Eq. (4.4) have been introduced.
The integration contour C in the complex -plane is supposed to make the action (4.6)
non-trivial. The factor (1 + 2 )4 in the action (4.6) was introduced to simplify the
transformation properties of the integrand under SU(2)R [28]. The projective variable
CP1 has the rational transformation law,
a
b
,
(4.8)
0 =
a + b
whose complex parameters (a, b) are constrained by the condition |a|2 + |b|2 = 1. In
general, the action (4.6) is neither conformally nor SU(2)R invariant.
After being expanded in components, the action (4.6) depends upon the bosonic
Lagrange multipliers given by the vector fields (V ) and the scalars (C), that can be
removed via their algebraic equations of motion or by a duality transformation. This
procedure is known as the generalized Legendre transform [6] that leads to a hyper-Khler
metric in the bosonic NLSM part of the theory (4.6).
The constraints (4.1) and (4.2) take the simple form in HSS,
D ++ L++ = 0,
L++ = L++ ,
(4.9)
where (cf. Eq. (4.5))

+ ii i2p
.
L++ = u+
i1 ui2p L
(4.10)
Requiring the invariance of the constraints (4.9) under the N = 2 superconformal

transformations (3.15) and (3.18) gives rise to the covariant transformation laws for the
projective superfields,
L++ = wL++ ,
with
w = 2p.
(4.11)
The choice of 2p = 1 in Eq. (4.1) defines the most basic FS hypermultiplet (with
vanishing central charge), whose physical components comprise only scalars and spinors. It
is not difficult to verify that the constraints (4.1) in this case imply free equations of motion,
Li = 0. An off-shell FS hypermultiplet is naturally described in HSS by an unconstrained
complex analytic superfield q + of U (1) charge (+1) [9]. Its free HSS action reads [9]
Z
(4.12)
S[q]free = d (4) du q + D ++ q + .
This action is invariant under the N = 2 superconformal transformations provided that
D ++ q + transforms covariantly like q + itself, which implies that q + is of conformal weight
one [22],
q + = q + .
(4.13)
The free HSS equations of motion, D ++ q + = 0, imply q + = Li (Z)u+

i together with
the on-shell FS hypermultiplet superspace constraints, D (i Lj ) = D (i Lj ) = 0.
106
It is worth noticing that the SU(2)conf. transformations, which are the part of the N = 2
superconformal symmetry, are different from the SU(2)R transformations, with the latter
being defined by their natural action on the Latin indices, i, j = 1, 2, as SU(2)R ui =
i j uj , etc. For example, as regards the free off-shell theory (4.12), one finds [22]
++ +
q ,
SU(2)R q + = SU(2)conf. q + + ij u
i uj D
(4.14)
so that the SU(2)conf. and SU(2)R transformations coincide only if D ++ q + = 0, i.e., onshell.
5. General N = 2 superconformal NLSM actions
We are now in a position to discuss the N = 2 superconformal hypermultiplet actions in
HSS. We use the pseudo-real Sp(1) notation for a single FS hypermultiplet superfield,

(5.1)
qa+ = q + , q + , a = 1, 2, q a+ = ab qb+ ,
and further generalize it to the case of several FS hypermultiplets, q a+ q A+ and qA+ =
AB q B+ , with a constant (antisymmetric) Sp(k)-invariant metric AB , A, B = 1, . . . , 2k.
First, we recall that the most general N = 2 supersymmetric NLSM can be most
naturally formulated in HSS, in terms of the FS hypermultiplet superfields, as

Z

1
1
,
(5.2)
SNLSM [q] = 2 (4) u qA+ D ++ q A+ + K(+4) q A+ , u
i
2
where the real analytic function K(+4) = K(+4) of U (1) charge (+4) is called a hyperKhler (pre-)potential [29]. 3 By manifest N = 2 supersymmetry of the NLSM action (5.2), the NLSM metric must be hyper-Khler for any choice of K(+4) . Unfortunately,
an explicit general relation between a hyper-Khler potential and the corresponding hyperKhler metric is not available (see, however, Refs. [2932] for the explicit hyper-Khler potentials of the (ALE) multi-EguchiHanson, (ALF) multi-TaubNUT and AtiyahHitchin
metrics, and their derivation from the NLSM (5.2) in terms of FS hypermultiplet superfields, in HSS).
Eq. (5.2) formally solves the hyper-Khler constraints on the NLSM metric in terms
of an arbitrary function K(+4) . It is, therefore, quite natural to impose extra N = 2
superconformal invariance on this function, in order to determine a general solution to the
special hyper-Khler geometry, since the free part (4.12) of the action (5.2) is automatically
N = 2 superconformally invariant. In general, the invariance of the HSS action (5.1)
merely implies the invariance of the HSS Lagrangian up to a total derivative, because of
the identity
Z
(5.3)
d (4) du D ++ X++ = 0.
3 We now choose our HSS superfields to be dimensionless, by the use of the dimensionful coupling constant
in front of their actions.
107
However, in the case of unconstrained FS analytic superfields q + , the hyper-Khler

potential should be invariant too. Eqs. (3.13), (3.18) and (4.13) now imply

(+4)
(+4)
K
A+
(+4)
++ K
q
2K
u
(5.4)
+
i = 0.
q A+
u+
i
Equation (5.4) is equivalent to two constraints,
K(+4) A+
q = 2K(+4)
q A+
and
K(+4)
= 0.
u+
i
(5.5)
This means that the special hyper-Khler potential of the N = 2 superconformally invariant
NLSM, in terms of the analytic FS superfields q A+ in HSS, is given by a homogeneous (of
A+ . There is no restriction on the dependence of
degree two) function K(+4) (q A+ , u
i ) of q
(+4)
upon ui , while it should be independent upon u+
K
i . This represents one of our main
new results in this paper.
Our simple description of the N = 2 superconformal hypermultiplet actions in terms
of the off-shell N = 2 superfields is to be compared to the well-known description of the
(abelian) N = 2 vector multiplet actions in the standard N = 2 (chiral) superspace [34,35],
Z
(5.6)
S[W ] = d4 x d4 F (WA ) + h.c.,
in terms of the N = 2 restricted chiral superfields WA representing the N = 2 abelian
gauge field strengths. The N = 2 superconformal invariance of the action (5.6) implies
that F (WA ) is a homogeneous (of degree two) function of WA [36,37]. The special Khler
geometry, associated with the scalar NLSM part of the action (5.6), and the special hyperKhler geometry (Section 2) are, however, very different at the level of components, as
well as in the geometrical terms.
A non-trivial special hyper-Khler potential exists even in the case of a single FS
hypermultiplet, e.g.,
"
#2

q +q +
(+4)
+ +
q , q , ui = C
,
(5.7)
K
+
+
q u2 q u1
where C is a real constant. Eq. (5.7) is not invariant with respect to SU(2)R because of its
explicit dependence upon harmonics, whereas it is invariant under the SU(2)conf. part of the
N = 2 superconformal symmetry by construction. The corresponding special hyper-Khler
metric interpolates between the EguchiHanson (ALE) metric described by a hyper-Khler
potential
"
#2
++
(+4)
,
(5.8)
KEH =
+ u
q + u
q
2
1
in the limit (q + q + ) ++ = ij ui+ uj + with constant ij , and the TaubNUT (ALF)
+
metric described by a hyper-Khler potential (in the limit q + u
2 q u1 = const)
2
(+4)
KTaubNUT = const q + q + .
(5.9)
108
6. N = 2 superconformal projective multiplets

We now turn to a HSS construction of the N = 2 superconformal (improved) actions for
the O(2p) projective multiplets introduced in Section 4. Unlike the FS hypermultiplets
described by unconstrained (analytic) superfields in HSS, the projective multiplets
are described by the constrained (off-shell) analytic superfields that give rise to the
finite numbers of the auxiliary fields. It is, therefore, straightforward to deduce the
component hyper-Khler metrics out of their HSS actions. The N = 2 supersymmetric
selfinteraction (4.6) of the projective multiplets (p < ) in PSS is known to be merely a
subclass of the most general N = 2 NLSM described by Eq. (5.2) in HSS [38], while the
projective superfields are of higher conformal weight than FS superfields see Eqs. (4.11)
and (4.13). We should, therefore, expect severe constraints on the N = 2 superconformal
actions in terms of the projective multiplets. This is known to be the case for the standard
N = 2 tensor multiplet (p = 1) indeed, whose N = 2 improved action was constructed
many years ago, first in components [39], then in N = 1 superspace [40] and N = 2 PSS
[6,7], and finally in HSS [41]. We begin with a simple derivation of the improved N = 2
tensor multiplet action in HSS, and then discuss its N = 2 superconformal generalizations
and non-conformal deformations.
On dimensional reasons, the most general N = 2 supersymmetric action of a single N =
2 tensor multiplet superfield L++ , subject to the constraints (4.9), reads in HSS as [38]
Z

1
(6.1)
S[L] = 2 d (4) du L(+4) L++ ; u
i .
A free bilinear action in L++ is obviously not N = 2 superconformally invariant, so that

it has to be improved in some non-trivial way. A power series in terms of L++ , as the
naive ansatz for the HSS Lagrangian L(+4) , also does not work here, because of the need
to balance the conformal weights defined by Eqs. (3.13) and (4.11). A resolution of this
problem was suggested in Ref. [39], where it was noticed that the improved Lagrangian
must be topologically non-trivial, i.e., it should contain a Dirac-like string of singularities
parametrized by an arbitrary SU(2) triplet of constants cij ,
cij = cj i ,
(cij ) = ik j l ckl ,
c2 = 12 cij cij .
(6.2)
This essentially amounts to extracting a fake vacuum expectation value out of the N = 2
tensor superfield Lij ,
Lij = cij + l ij ,
or, equivalently, L++ = c++ + l ++ .
(6.3)
The N = 2 superconformal invariance of the action makes it to be independent upon the

constants cij because of the SU(2)conf. symmetry. Since the normalization of c2 can always
be changed by dilatations, we temporarily set c2 = 1 for simplicity of our calculations in
what follows. The definitions
+
c++ = cij u+
i uj ,
c+ = cij u+
i uj ,
c = cij u
i uj ,
(6.4)
imply the identities [41]

D ++ c = 2c+ ,
D ++ c+ = c++ ,
D ++ c++ = 0,
(6.5)
109
and
c++ c (c+ )2 = c2 = 1,
(6.6)
with the latter being the corollary of the completeness relation for harmonics,
+ i
i
ui+ u
j uj u = j .
(6.7)
Equations (6.2), (6.3) and (6.5) imply that l ++ also satisfies the initial off-shell
constraints (4.9),
D ++ l ++ = 0,
l ++ = l ++ .
(6.8)
The natural ansatz for the improved N = 2 tensor multiplet action in HSS is given
by [41]
Z
1
(6.9)
S[L]impr. = 2 d (4) du (l ++ )2 f (y), y l ++ c ,
where the function f (y) is at our disposal. Since the action in question is supposed to
improve the naive (quadratic) one, the function f should obey the boundary condition
f (0) = 1.
(6.10)
A more general HSS ansatz for the HSS Lagrangian may include a term (c++ )2 g(y, c) in
Eq. (6.9), with yet another function g(y, c) to be discussed below.
The identities (6.5) and (6.6) together with the constraint (6.8) further imply that
2y(D ++ )2 y (D ++ y)2 = 4(l ++ )2
and (D ++ )3 y = 0.
(6.11)
The N = 2 superconformal transformation laws of the new variables l ++ and y follow

from their definitions in Eqs. (6.3) and (6.9), by the use of Eqs. (3.18) and (4.11) with
p = 1 and w = 2. We find

(6.12)
l ++ = 2l ++ + 2D ++ c+ ++ c ,
and

y = 2y + 2 c++ ++ c+ c

= 2y + 2D ++ c c+ ++ (c )2 + 4c+ ++ c c+ . (6.13)
Varying the action (6.9) by the use of Eqs. (3.13), (6.12) and (6.13), integrating by parts
via Eq. (5.3), and using the identities (6.11) yield
Z
d (4) du (l ++ )2 f (y)

Z
1
3
= d (4) du (D ++ y)2 y(y + 1)f 00 + (7y + 6)f 0 + f
2
2
Z

(6.14)
+ d (4) du ++ (D ++ y) y 2 f 00 + y(6 y)f 0 yf = 0.
110
Note that the second line of Eq. (6.14) is also a total derivative in the case of a single N = 2
tensor multiplet. The N = 2 superconformal invariance of the action (6.9) thus amounts to
the second-order ordinary differential equation on the function f (z),
1
3
z(1 z)fzz + (6 7z)fz f = 0,
2
2
z = y.
(6.15)
This is the very particular hyper-geometric equation, whose well-known general form
depending upon three parameters (, , ) is given by

(6.16)
z(1 z)Fzz + ( + + 1)z Fz F = 0.
Hence, a regular solution to our problem (6.15) with the boundary condition (6.10) is given
by the hyper-geometric function F (, , ; z) with = 1, = 3/2 and = 3. It appears
to be an elementary function

1 + 1 z 2
3
,
(6.17)
f (z) = F 1, 2 , 3; z =
2
in full agreement with Ref. [41], where the recursive methods were used. As was
demonstrated in Ref. [41], integration over the Grassmann and harmonic coordinates of
HSS in the action defined by Eqs. (6.9) and (6.17),

2
Z
4
L++ c++
p
,
(6.18)
S[L]impr. = 2 d (4) du
1 + 1 + (L++ c++ )c /c2

results in the improved component action of Ref. [39]. The equivalent PSS action (4.6) has
a holomorphic potential [6,7]

G Q( ), = Q( ) c( ) ln Q( ) c( ) 1 ,
(6.19)
where Q = i j Lij and c( ) = i j cij with i = (1, ), while the contour C in the
complex -plane encircles the roots of a quadratic equation (Q c)( ) = 0. Like the HSS
action (6.9), the equivalent PSS action (4.6) with the potential (6.19) does not depend upon
the constants cij .
It is not difficult to verify that another ansatz for the N = 2 tensor multiplet HSS
Lagrangian of the form (c++ )2 g(y, c+ ) gives rise to another N = 2 superconformal
invariant,
Z
(6.20)
d (4) du (c++ )2 L++ c .
However, it vanishes after integration over harmonics and the anticommuting N = 2
superspace coordinates, because of the conservation law V = 0 for the vector

component of the N = 2 tensor multiplet in Eq. (4.3).
The N = 2 superconformally invariant action (6.18) leads to a flat NLSM metric (in
disguise) after the generalized Legendre transform [39]. The improved N = 2 tensor
multiplet action can, nevertheless, serve as the key building block for a construction of
non-trivial four-dimensional hyper-Khler metrics. For example, the Ak series of the ALE
111
gravitational instanton metrics arise when one sums the improved N = 2 tensor multiplet
ij
Lagrangians with different moduli ca ,
2
Z
k+1
X
4
L++ ca++
q
.
(6.21)
S[L]ALEAk = 2 d (4) du
++
a=1 1 + 1 + (L++ ca )ca /ca2
The associated PSS potential reads
G Q( ), =
k+1
X

Q( ) ca ( ) ln Q( ) ca ( ) 1 ,
(6.22)
a=1
while its generalized Legendre transform is known to lead to the Ak ALE metrics
indeed [42]. Another simple non-conformal deformation is given by the naive (bilinear)
action of the N = 2 tensor multiplet,
Z
(6.23)
S[L]naive = m d (4) du (L++ )2 .
After being added to the action (6.21), it leads to the Ak series of the ALF (multi-Taub
NUT) metrics with a TaubNUT mass parameter m (see, e.g., Ref. [28]). As is clear from
Eq. (6.22), the moduli of the Ak metrics are naturally described by fixed real sections
ca ( ) of the O(2) holomorphic bundle, while one of them can be arbitrarily chosen. In
the context of M-theory/type-IIA superstring compactification (Section 1), the Ak moduli
describe the D6-brane positions in the transverse space.
Having established the improved action of a single N = 2 tensor multiplet (p = 1),
it is natural to look for N = 2 superconformal actions in terms of several N = 2 tensor
multiplets or the higher O(2p) projective multiplets as well (a geometrical motivation of
the latter is discussed at the end of this section).
A generalization of the ansatz (6.9) to the case of several N = 2 tensor multiplets
(p = 1) is given by
Z
q
X
1
la++ lb++ fab ({y}), a = 1, 2, . . . , q,
(6.24)
S[La ] = 2 d (4) du
a,b=1
where fab ({y}) is the symmetric matrix of q(q + 1)/2 functions depending upon q
variables, {y} = (l1++ c , . . . , lq++ c ), with the same c . Requiring the action (6.24)
be invariant under the N = 2 superconformal transformations gives rise to a system of
q(q + 1)/2 second-order ordinary differential equations 4

q
q
X
X
yb ya + 2(1 + ya ) fa(c,d)
q+
a=1
b=1

q
X
3
3
+
1 ya fcd,a fcd = 0,
2
2
a=1
4 We denote differentiations by commas, like in general relativity.
(6.25)
112
and extra consistency condition on the vector

Va
q
X
yb (2 yc )fab,c
b,c=1
q
X
fab yb
(6.26)
b=1
to be a total derivative, i.e., a Vb b Va = 0. This gives rise to the additional equations

q
X
ya (2 yb )fa[c,d]b 2
a,b=1
q
X
fa[c,d] ya = 0,
(6.27)
a=1
that make the full set of q 2 equations to be overdetermined. We are unaware of any nontrivial solutions to these equations, except of the one given by a non-interacting sum of the
improved actions for each N = 2 tensor multiplet.
We now turn to a single O(4) projective multiplet L++++ L(+4) satisfying the
constraints (4.9), and construct its improved action in HSS. Instead of writing down a
new ansatz, it is much simpler to use the already established improved action (6.18) for the
O(2) projective (tensor) multiplet, and then take into account the known transformation
property (4.11) of L(+4) with p = 2 and w = 4. The latter implies that L(+4) transforms as
(L++ )2 under the N = 2 superconformal transformations. By using the obvious identities,
p
(6.28)
L++ c = (L++ )2 (c )2
and
p
(L++ c++ )2 = (L++ )2 4(D ++ )2 (L++ )2 (c )2 + (c++ )2 ,
(L++ )2
(6.29)
by
in Eq. (6.18). It yields the N = 2 superconforwe can simply substitute
mally invariant (improved) action of an O(4) projective multiplet in the form
"
#2
r
Z
++ c
p
(+4)
4
1
c
= 2 d (4) du 1 + 1 + 2 L(+4) (c )2
S L
impr.
c
c2

8c++ c
L(+4) 1 p
+ (c++ )2 .
(6.30)
(+4)
2
L
(c )
This action is one of our main new results in this paper.
p The associated PSS potential is
obtained from Eq. (6.19) after a substitution Q(2) Q(4) , i.e.,
h p

i
p
Q(4) ( ) c( ) ln Q(4) ( ) c( ) 1 ,
(6.31)
G Q(4) ( ), =
L(+4)
while the integration

p contour C2 in the complex -plane is now given by two circles around
the branch cuts of Q(4) [42,43].
p
It is worth noticing that the HSS superfield L(+4) (c )2 transforms covariantly under
the N = 2 superconformal transformations with conformal weight w = 2,
p
p
(6.32)
L(+4) (c )2 = 2 L(+4) (c )2 .
This implies the existence of another non-trivial N = 2 superconformal invariant that
originates from Eq. (6.20) and has the form
Z
p

(6.33)
S L(+4) , c ext. = d (4) du (c++ )2 L(+4) (c )2 .
113
The holomorphic PSS potential associated with the HSS Lagrangian (6.33) is obvious,
p

(6.34)
G Q(4) ( ) ext. = Q(4) ( ).
We are now in a position to make use of the new improved action (6.30) of a single
O(4) projective multiplet. The generalized Legendre transform in application to Eq. (6.31)
is supposed to yield a free NLSM metric (in disguise), similarly to the improved O(2)
projective multiplet action in Eqs. (6.18) and (6.19). However, a sum of the improved
ij
actions (6.30) and (6.33) with different moduli ca ,

S L(+4) ALED
k
k

X

=
S L(+4) , ca impr. + S L(+4) , ca impr. + S L(+4) , c0 ext. ,
(6.35)
a=1
gives rise (after the generalized Legendre transform) to the N = 2 NLSM whose non-trivial
metric can be identified with the ALE Dk metric [18,43]. The Z2 symmetric combination
of the improved terms in Eq. (6.35) is necessary to produce the dihedral group Dk out of
the cyclic Ck group. Similarly, the ALF series of Dk metrics are obtained after adding to
Eq. (6.35) a non-conformal deformation (cf. Eq. (6.23)),
Z
(6.36)
m d (4) du L(+4) .
ij
In the context of M-theory/type-IIA compactification (Section 1), the parameters {ca } in

the action (6.35) describe the positions of D6-branes and an orientifold in the transverse
directions. In the context of the related three-dimensional N = 4 supersymmetric gauge
field theory with k matter hypermultiplets in the probe D2-brane world-volume, the moduli
ij
{ca } parametrize the quantum moduli space (= ALF Dk with k singularities) of the lowenergy effective field theory, being related to the positions of monopoles in the type-IIB
picture (Section 1).
In particular, the N = 2 NLSM with the famous (regular and complete) AtiyahHitchin
(AH) metric [19] is obtained by the non-conformal deformation (6.36) of the N = 2
superconformal action (6.33). The corresponding PSS data,
I
I
I
p

m
Q(4) ( )
1
+
Q(4) ( ),
(6.37)
G Q(4) ( ), =
2i
2i
C0
C2
with the contour C0 encircling the origin in the clock-wise direction, just describes the
AH metric [42]. The AH metric also appears in the quantum moduli space of the threedimensional N = 4 supersymmetric pure gauge SU(2)-based quantum field theory [44],
and in the hypermultiplet low-energy effective action (NLSM) of a (magnetically charged)
hypermultiplet in the Higgs branch [45]. From the AH prospective, the ALF Dk metrics,
associated with the HSS potential given by a sum of Eqs. (6.35) and (6.36), can be equally
interpreted as the deformations of the AH metric. Some of those metrics were discovered
by Dancer [46], by using the hyper-Khler quotient construction. Their derivation via the
generalized Legendre transform is due to Chalmers [47] who also noticed the significance
114
of the same metrics for the monopole moduli spaces in the completely broken SU(3) gauge
theory investigated by Houghton [48].
To the end of this section, we would like to comment on the geometrical significance of
an O(4) projective multiplet versus an O(2) projective (tensor) multiplet, in the context of
N = 2 supersymmetric NLSM with four-dimensional (self-dual or hyper-Khler) target
spaces. The PSS construction of N = 2 NLSM takes the universal form (4.6) for all
projective multiplets, it generically breaks the SU(2)R automorphism symmetry of N = 2
supersymmetry algebra, but it leaves a U (1) symmetry. The latter implies an SO(2)
isometry in the target space of the associated N = 2 NLSM. The nature of this isometry
is, however, dependent upon whether one uses an O(2) or O(4) projective multiplet in the
PSS action (4.6). Any such action in terms of a single O(2) tensor multiplet necessarily
leads to a translational (tri-holomorphic) isometry that arises after trading the conserved
vector component of the O(2) projective multiplet for a scalar by the generalized Legendre
transform.
Quite generally, the existence of an isometry amounts to the existence of a Killing vector
m
K , m = 1, 2, 3, 4, obeying Eq. (2.1) or, equivalently, the existence of the coordinate
system (x a , ), a = 1, 2, 3, where the metric components are independent upon one of
the coordinates ( ),
ds 2 = H 1 (d + Ca dx a ) + H ab dx a dx b .
(6.38)
A translational isometry implies extra condition on the Killing vector [4951],

K [m;n] = K [m;n] ,
(6.39)
where the star denotes the four-dimensional dual tensor. Equation (6.39) gives rise to
the existence of the coordinate system (6.38) where, in addition, we have the monopole
equation [6,7,4951]
E =
E C,
E
H
and ab (x) = ab .
(6.40)
Self-duality then amounts to a linear Laplace equation on the potential H (x a ),

1H = 0,
(6.41)
whose localized solutions,

H (x) = +
k
X
s=1
m
,
|E
x xEs |
(6.42)
just describe the Ak series of self-dual metrics (ALE multi-EguchiHanson metrics in the
case of = 0 and ALF multi-TaubNUT metrics in the case of = 1). It is now not very
surprising that those metrics arise from the N = 2 superspace NLSM in terms of an O(2)
tensor multiplet only.
However, if one wants to construct the hyper-Khler metrics possessing merely a
rotational isometry, the conventional way towards their derivation (in components) is
much more involved. Equation (6.38) still holds, but no Eqs. (6.39), (6.40) and (6.41)
115
are available. Nevertheless, a single real scalar potential for those metrics also exists in the
so-called Toda frame defined by the conditions [4951]
H = 3 ,
C1 = 2 ,
C2 = 1 ,
C3 = 0,
(6.43a)
and
11 = 22 = e ,
33 = 1.
(6.43b)
In the Toda frame self-duality amounts to the non-linear 3d Toda equation [4951] 5

(6.44)
12 + 22 + 32 e = 0,
which is very hard to solve. By the use of an O(4) projective multiplet having no conserved
vector components, in the N = 2 superspace construction of 4d hyper-Khler metrics,
we just deal with the self-dual metrics having merely a rotational isometry. The basic
geometrical difference between the N = 2 superspace NLSM actions in terms of O(2)
or O(4) projective multiplets thus amounts to the nature of their abelian isometry: it is
tri-holomorphic in the O(2) case, whereas it is not triholomorphic in the O(4) case. The
N = 2 NLSM in terms of higher O(2p) projective multiplets are similar to that with p = 2.
It is worth mentioning in this context that a self-dual metric with rotational isometry gives
rise to a solution to the 3d Toda equation (6.44).
7. Conclusion
By the use of harmonic superspace describing hypermultiplets with manifest N = 2
supersymmetry, we arrived at a general solution to the N = 2 non-linear sigma-models with
special hyper-Khler geometry. Our solution is parametrized by a single (of degree two)
homogeneous function, which is quite similar to the well-known N = 2 superconformal
description of N = 2 vector multiplets in the standard N = 2 superspace.
We also constructed the improved (N = 2 superconformal) actions of the O(2) and O(4)
projective multiplets, which lead to flat four-dimensional metrics in disguise. However,
after being added together with different moduli, those N = 2 superconformal actions
naturally lead to the Ak and Dk series of the highly non-trivial self-dual metrics in the
target space of the associated N = 2 NLSM. It gives us the very natural way of derivation
and classification of those self-dual metrics. In particular, the ALE metric potentials in
superspace can be interpreted as the interpolating potentials between different improved
(flat) potentials in terms of the O(2) projective multiplet in the Ak case and in terms of the
O(4) projective multiplet in the Dk case, whereas the ALF potentials can be understood as
non-conformal deformations of the ALE ones. It would be interesting to find the self-dual
metrics, associated with the exceptional (simply-laced) E6,7,8 Lie groups, in the superspace
approach.
The (improved) N = 2 superconformally invariant actions (6.18) and (6.30) of the O(2)
and O(4) projective multiplets, respectively, are not easily generalizable to the case of
5 The 3d Toda equation arises from the standard (2d) SU(N )-based Toda system in the large-N limit [5052].
116
higher O(2p) projective multiplets with p > 2. An O(2p) projective multiplet is described
in HSS by the L(2p+) superfield satisfying the off-shell constraints (4.9). It can be used to
introduce the HSS superfield
(2p+) p 1/p
(c )
(7.1)
L
that covariantly transforms under the rigid N = 2 superconformal transformations with
conformal weight w = 2. It is, however, unclear to us how to define a covariant HSS
superfield of U (1) charge (+4) and conformal weight w = 4, in terms of the L(2p+)
superfield, in order to substitute (L++ )2 in Eq. (6.18). This obstruction may be related
to the existence of only two (Ak and Dk ) regular series of gravitational instantons. It is,
however, possible to define N = 2 superconformal generalizations of Eqs. (6.20) and (6.33)
to the case of higher projective multiplets,
Z

1/p
(2p+)
, c ext. = d (4) du (c++ )2 L(2p+) (c )p
.
(7.2)
S L
It would be interesting to study non-conformal deformations of Eq. (7.2).
Our final remark is devoted to local N = 2 superconformal symmetry that implies
coupling N = 2 (rigidly) superconformal NLSM to N = 2 (conformal) supergravity.
It leads to the deformation of a given special hyper-Khler NLSM metric gmn to a
quaternionic metric Gmn [53]. This deformation in four dimensions preserves self-duality
of the Weyl tensor of the metric gmn , but it also turns it into an Einstein metric with negative
scalar curvature [54]. The associated quaternionic metric reads [10]

1
1 1
Xm Xn + 2YmA YnA
(7.3)
Gmn = gmn 2
f
f
2
in terms of the Euler vector Xm , the potential f defined by Eq. (2.7), and the Killing
(A)
vectors Ym defined by Eq. (2.9). Therefore, an explicit derivation of the quaternionic
metric, associated with a given special hyper-Khler metric, seems to be possible without
going into details of the N = 2 NLSM coupling to N = 2 supergravity.
Acknowledgements
It is my pleasure to thank Jan Ambjorn and the High Energy Theory Group at Niels Bohr
Institute for nice hospitality extended to me in Copenhagen during the completion of this
work. I also thank Luis Alvarez-Gaum, Francois Delduc, Olaf Lechtenfeld, Werner Nahm
and Galliano Valent for useful discussions.
References
[1]
[2]
[3]
[4]
A. Sen, Nucl. Phys. B 475 (1996) 562.

T. Banks, M. Douglas, N. Seiberg, Phys. Lett. B 387 (1996) 278.
M. Douglas, D. Lowe, J.H. Schwarz, Phys. Lett. B 394 (1997) 297.
O. Aharony, S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory
and gravity, hep-th/9905111.
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
117
L. Alvarez-Gaum, D.Z. Freedman, Commun. Math. Phys. 80 (1981) 443.

A. Karlhede, U. Lindstrm, M. Rocek, Phys. Lett. B 147 (1984) 297.
N. Hitchin, A. Karlhede, U. Lindstrm, M. Rocek, Commun. Math. Phys. 108 (1987) 535.
S.V. Ketov, Selfinteraction of N = 2 multiplets in 4d, and UV finiteness of 2d, N = 4 NLSM,
in: Group-Theoretical Methods in Physics, Nauka, Moscow, 1985, p. 87.
A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, S. Kalitzin, E. Sokatchev, Class. Quant. Grav. 1
(1984) 469.
B. de Wit, B. Kleijn, S. Vandoren, Rigid superconformal hypermultiplets, in: J. Wess,
E.A. Ivanov (Eds.), Supersymmetries and Quantum Symmetries, Springer, 1999, p. 37; hepth/9808160; Nucl. Phys. B 568 (2000) 475.
G. Chalmers, A. Hanany, Nucl. Phys. 489 (1997) 223.
A. Hanany, E. Witten, Nucl. Phys. 492 (1997) 152.
W. Nahm, The construction of all self-dual multi-monopoles by the ADHM method, in:
N.S. Craigie et al. (Eds.), Monopoles in Quantum Field Theory, World Scientific, Singapore,
1982.
S.K. Donaldson, Commun. Math. Phys. 96 (1985) 387.
J. Hurtubise, Commun. Math. Phys. 120 (1989) 613.
D. Diaconescu, Nucl. Phys. 503 (1997) 220.
A. Sen, JHEP 09 (1997) 1; Adv. Theor. Math. Phys. 1 (1998) 115.
S.A. Cherkis, A. Kapustin, Adv. Theor. Math. Phys. 2 (1998) 1287; Commun. Math. Phys. 203
(1999) 713.
M. Atiyah, N. Hitchin, The Geometry and Dynamics of Magnetic Monopoles, Princeton
University Press, 1988.
P.B. Kronheimer, J. Diff. Geometry 29 (1989) 665; J. Diff. Geometry 29 (1989) 685.
A. Ferber, Nucl. Phys. 132 (1978) 55.
A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E. Sokatchev, Conformal invariance in harmonic
superspace, in: C.J. Isham, G. Vilkovisky (Eds.), Quantum Field Theory and Quantum Statistics,
Vol. 2, Adam Hilger, 1987, p. 233.
G.W. Gibbons, P. Rychenkova, Phys. Lett. B 443 (1998) 138.
I. Klebanov, E. Witten, Nucl. Phys. 356 (1998) 199.
D. Blair, Compact Manifolds in Riemannian Geometry, Springer, 1976.
C.P. Boyer, K. Galicki, 3-Sasakian manifolds, hep-th/9810250, to appear in: M. Wang et al.
(Eds.), Essays on Einstein Manifolds.
B. Acharya, J. Figueroa-OFarrill, C. Hull, B. Spence, Adv. Theor. Math. Phys. 2 (1999) 1249.
S.V. Ketov, Exact low-energy effective actions for hypermultiplets in four dimensions, hepth/9909177; to appear in Int. J. Mod. Phys. A.
A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E. Sokatchev, Ann. Phys. (N.Y.) 185 (1988) 1;
Ann. Phys. (N.Y.) 185 (1988) 22.
A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E. Sokatchev, Commun. Math. Phys. 103 (1986)
515.
A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, P.K. Townsend, Class. Quant. Grav. 3 (1986) 625.
G.W. Gibbons, D. Olivier, P.J. Ruback, G. Valent, Nucl. Phys. 296 (1988) 679.
S.V. Ketov, Ch. Unkmeir, Phys. Lett. B 422 (1998) 179.
S.J. Gates Jr., Nucl. Phys. 238 (1984) 349.
S.V. Ketov, I.V. Tyutin, JETP Lett. 39 (1984) 703.
B. de Wit, J.W. van Holten, A. van Proeyen, Nucl. Phys. 167 (1980) 186.
B. de Wit, P. Lauwers, A. van Proeyen, Nucl. Phys. 255 (1985) 569.
A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, Nucl. Phys. 282 (1987) 74.
B. de Wit, R. Philippe, A. van Proeyen, Nucl. Phys. 219 (1983) 143.
U. Lindstrm, M. Rocek, Nucl. Phys. 222 (1983) 285.
A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, Sov. J. Nucl. Phys. 45 (1987) 157.
118
[42] I.T. Ivanov, M. Rocek, Commun. Math. Phys. 182 (1996) 291.
[43] G. Chalmers, M. Rocek, S. Wiles, JHEP 01 (1999) 009.
[44] N. Seiberg, E. Witten, Gauge dynamics and compactification to three dimensions, hepth/9607163; in: Saclay96. The Mathematical Beauty of Physics.
[45] S.V. Ketov, Phys. Lett. B 469 (1999) 136.
[46] A.S. Dancer, Quart. J. Math. Oxford 45 (1994) 463.
[47] G. Chalmers, Phys. Rev. D 58 (1998) 125011.
[48] C.J. Houghton, Phys. Rev. D 56 (1997) 1220.
[49] C. Boyer, J. Finley, J. Math. Phys. 23 (1982) 1126.
[50] J. Gegenberg, A. Das, Gen. Rel. Grav. 16 (1984) 817.
[51] I. Bakas, K. Sfetsos, Int. J. Mod. Phys. A 12 (1997) 2585.
[52] M. Saveliev, Commun. Math. Phys. 121 (1989) 283; Theor. Math. Phys. 92 (1993) 1024.
[53] J. Bagger, E. Witten, Nucl. Phys. 222 (1983) 1.
[54] A. Swan, Math. Ann. 289 (1991) 421.
[55] K. Galicki, Class. Quant. Grav. 9 (1992) 27.

Anomalous N = 2 superconformal Ward identities

Sergei V. Ketov a,b,1
a Niels Bohr Institute, University of Copenhagen, 2100 Copenhagen , Denmark
b Institut fr Theoretische Physik, Universitt Hannover Appelstrae 2, D-30167, Hannover, Germany
Received 7 April 2000; revised 19 April 2000; accepted 1 May 2000
Abstract
The N = 2 superconformal Ward identities and their anomalies are discussed in N = 2 superspace
(including N = 2 harmonic superspace), at the level of the low-energy effective action (LEEA)
in four-dimensional N = 2 supersymmetric field theories. The (first) chiral N = 2 supergravity
compensator is related to the known N = 2 anomalous Ward identity in the N = 2 (abelian) vector
mulitplet sector. As regards the hypermultiplet LEEA given by the N = 2 non-linear sigma-model
(NLSM), a new anomalous N = 2 superconformal Ward identity is found, whose existence is related
to the (second) analytic compensator in N = 2 supergravity. The celebrated solution of Seiberg and
Witten is known to obey the (first) anomalous Ward identity in the Coulomb branch. We find a few
solutions to the new anomalous Ward identity, after making certain assumptions about unbroken
internal symmetries. Amongst the N = 2 NLSM target space metrics governing the hypermultiplet
LEEA are the SU(2)-YangMillsHiggs monopole moduli-space metrics that can be encoded in
terms of the spectral curves (Riemann surfaces), similarly to the SeibergWitten-type solutions. After
a dimensional reduction to three spacetime dimensions (3d), our results support the mirror symmetry
between the Coulomb and Higgs branches in 3d, N = 4 gauge theories. 2000 Elsevier Science
B.V. All rights reserved.
PACS: 11.30.Pb
Keywords: Supersymmetry; Anomaly; Supergravity
1. Introduction
Field theories with N = 2 extended supersymmetry are known to possess remarkable
properties that sometimes allow one to obtain their exact (non-perturbative) low-energy
solutions, pioneered by Seiberg and Witten [1]. The natural way to produce such exact
results is to relate the field theory problem to an integrable system (or Whitham dynamics
in the SeibergWitten case) [2]. The origin of an elliptic curve behind the exact solution
1 E-mail: ketov@itp.uni-hannover.de; also at HCEI, Academy of Sciences, Akademichesky 4, Tomsk 634055,
Russia.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 6 6 - 2
120
[1] also becomes apparent in this approach. Another (related) interpretation of the
SeibergWitten result is possible from the viewpoint of anomalous breaking of N = 2
superconformal symmetry and uniformization theory on Riemann surfaces [3]. It is not
very surprising since N = 2 superconformal symmetry is well-known to be instrumental in
deriving the classical structure of all (non-conformal) N = 2 supersymmetric field theories,
including N = 2 supergravity, in four spacetime dimensions. It is natural to examine first
the constraints imposed by N = 2 superconformal invariance and then study compensation
of unwanted N = 2 superconformal symmetries. Or one can start from a classical field
theory that is N = 2 superconformally invariant, and then investigate its superconformal
anomalies that can be developed in quantum field theory.
Much work in the recent past was devoted to investigating the constraints imposed by
N = 4 superconformal symmetry on the correlation functions of the N = 4 super-Yang
Mills theory in the context of AdS/CFT correspondence [46]. More recently, similar
constraints on the correlation functions were studied in the context of four-dimensional
N = 2 conformal supersymmetry [7,8]. We recall that all N = 2 supersymmetric field
theories can be brought into the manifestly N = 2 supersymmetric form, by using offshell (unconstrained) N = 2 superfields in Harmonic Superspace (HSS) [9]. It makes the
effective field theory methods to be very efficient in the N = 2 case contrary to the N = 4
case where only on-shell N = 4 supersymmetry is possible. The next obvious step is to
make manifest N = 2 superconformal invariance of the N = 2 supersymmetric quantum
effective action, and then study its superconformal anomalies. Truly non-perturbative
results are expected to be obtained along these lines.
As regards quantum field theories with rigid N = 2 supersymmetry, their building blocks
are given by N = 2 vector multiplets and hypermultiplets. At the level of the Low-Energy
Effective Action (LEEA), on the (abelian) N = 2 vector multiplet side we have to deal with
the SeibergWitten-type action specified by a holomorphic potential and the associated
special Khler geometry. This topic is well-known, and we are going to use it as our
basic pattern to follow. Hypermultiplets (e.g., if they are magnetically charged) may also
develop the non-trivial LEEA that takes the form of a hyper-Khler Non-Linear SigmaModel (NLSM) by N = 2 supersymmetry. After being formulated in HSS, the N = 2
NLSM also possess an (analytic) potential. Our purpose in this paper is to impose and
make manifest N = 2 superconformal invariance in the N = 2 NLSM, and then formulate
an anomalous N = 2 superconformal Ward identity on the hyper-Khler potential.
The paper is organized as follows. In Section 2 we briefly review the N = 2
supercurrents, the anomalous N = 2 superconformal Ward identities in the N = 2 gauge
sector, and their relation to the solution of Seiberg and Witten [1]. In Section 3 we construct
in HSS the most general N = 2 superconformal NLSM that gives a general solution to the
special hyper-Khler geometry. The anomalous N = 2 superconformal Ward identity on
the hypermultiplet LEEA (N = 2 NLSM) is found for the first time in Section 4, by using
the N = 2 supergravity compensators in HSS. In Section 5 we give our conclusions. In
Appendix A we briefly review a construction of N = 2 supergravity in HSS.
121
2. N = 2 supercurrent and Ward identities

An N = 2 supercurrent is the irreducible representation of N = 2 supersymmetry in
four spacetime dimensions, having superspin one. The independent field components of
the N = 2 supercurrent (of some N = 2 matter system) include the energymomentum
tensor, the N = 2 supersymmetry current, the central charge current, the axial current, the
SU(2) current of R-symmetry and some auxiliary field components of lower dimension.
The relevant field components of the N = 2 supercurrent were first identified in
Ref. [10] by analyzing the free field theory of a massive (FayetSohnius) hypermultiplet.
The systematic way of derivation of the N = 2 supercurrent superfield is provided
by a construction of the irreducible N = 2 superfields in the conventional (flat) N =
2 superspace {Z} = (x m , i , ), where m = 0, 1, 2, 3, = 1, 2 and i, j = 1, 2. All

j
irreducible N = 2 superprojectors were found in Ref. [11], whereas all irreducible (scalar)
N = 2 superfields were explicitly derived in Refs. [12,13]. Amongst the irreducible N = 2
superfields, comprising a general N = 2 real scalar superfield, one finds almost all off-shell
N = 2 supermultiplets that usually appear in any discussion of N = 2 supersymmetry (with
a finite number of the auxiliary fields). In particular, an N = 2 (restricted) chiral superfield
is defined by the N = 2 superspace (off-shell) constraints
(x, , )
D i = 0,
D 4 = ,
(2.1)
j
where (Di , D ) are the usual (flat) N = 2 superspace covariant derivatives. 2 As a
consequence of the constraints (2.1), the N = 2 restricted chiral superfield possess

a two-form F = Fmn dx m dx n satisfying the Bianchi identity dF = 0. A solution
to the Bianchi identity, F = dA, in terms of the one-form A subject to the gauge
transformations A = d, allows one to represent the N = 2 superfield strength W of an
abelian N = 2 vector multiplet by a restricted chiral N = 2 superfield too.
The N = 2 restricted chiral superfield is dual to an N = 2 linear superfield Lij . The
latter is symmetric with respect to its SU(2) indices and satisfies the off-shell constraints
D (i Lj k) = D (i Lj k) = 0,
(Lij ) = ik j l Lkl .
(2.2)
The duality relation in N = 2 superspace is just given by

Lij = D ij .
(2.3)
Both superfields and Lij represent the irreducible N = 2 multiplets of superspin zero
and superisospin zero.
Similarly, an irreducible N = 2 scalar superfield R of superspin zero and superisospin
one is defined by the constraints [12,13]
h
i
(2.4)
D R = D R = i D j , D j R = 0.
1
i
2 We use the notation D
ij
i j
4
ij
= Di D , D = D D and D = 12 Dij D , and similarly for the conjugated
quantities.
122
The irreducible N = 2 scalar superfield R is dual to an N = 2 projective superfield T ij kl ,

T ij kl = D (ij D kl) R,
(2.5)
T ij kl
is totally symmetric with respect to its SU(2) indices and obeys the (off-shell)
where
constraints
D (i T j klm) = D T j klm) = 0 ,
(i
(T i1 i2 i3 i4 ) = i1 j1 i2 j2 i3 j3 i4 j4 T j1 j2 j3 j4 .
(2.6)
The N = 2 supercurrent J is also in the list of the irreducible N = 2 scalar superfields,

being defined by the constraints [12,13]
Dij J = D ij J = 0.
(2.7)
The N = 2 superspace constraints (2.7) imply that the energy-monentum tensor is

symmetric, conserved and traceless, whereas all the vector currents are conserved. In other
words, J is a multiplet of N = 2 superconformal currents.
An N = 2 superconformal anomaly amounts to breaking the N = 2 supercurrent
conservation relations (2.7) by an N = 2 anomaly multiplet of lower superspin, i.e., of
superspin zero. This is equivalent to activating an irreducible superspin-zero superfield in
the N = 2 scalar superfield J . As is clear from the above discussion, there are potentially
two ways of assigning the N = 2 anomaly multiplet with an N = 2 irreducible superspinzero superfield: either (or, equivalently, Lij ) or R (or, equivalently, Lij kl ). The main
difference between the two choices is the fact that Lij still contains a conserved vector
current (associated with unbroken central charge transformations), whereas the vector
current in Lij kl is not conserved. The first choice yields the N = 2 superconformal anomaly
relation in the standard form [14,15]
i
4 Dij J = Lij .
(2.8)
For example, the N = 2 supercurrent conservation law in the quantum N = 2 SYM theory
takes the form [1416]
c
i

(2.9)
S = tr W 2 ,
4 Dij J = Dij S ,
2
where the (Lie algebra-valued) N = 2 SYM superfield strength W has been introduced.
The constant c is proportional to the one-loop renormalization group beta-function. Though
tr W 2 is merely a chiral (not a restricted chiral) N = 2 superfield, Eq. (2.9) can be easily
brought into the form (2.8) by a local shift of the supercurrent, J J 4iS.
The most general N = 2 supersymmetric ansatz for the LEEA of some number (r)
of abelian N = 2 vector multiplets is governed by a holomorphic potential F (W ) of the
N = 2 (restricted chiral) superfield strengths Wp [1719], 3
Z
I [W ] = d4 x d4 F (Wp ) + h.c.
(2.10)
The superfield W has conformal weight +1, in accordance with its canonical dimension,
whereas the N = 2 chiral superspace measure in Eq. (2.10) has conformal weight (2).
3 Our normalization differs from that used in Ref. [1] by a factor i/(16 ).
123
It is, therefore, clear that the N = 2 superconformal Ward identity for the LEEA (2.10) is
given by
r
X
Wp
p=1
F
2F = 0.
Wp
(2.11)
The N = 2 superconformal solution to the holomorphic potential F (W ) is thus given by

a homogeneous (of degree two) function. The (rigid) N = 2 superconformal invariance
of the action (2.10) is the necessary pre-requisite for its coupling to N = 2 conformal
supergravity [17].
Given a non-trivial renormalization flow (like in the N = 2 SYM theory), the N = 2
superconformal Ward identity (2.11) is going to be broken by the anomaly S,
r
X
p=1
Wp
F
2F = 4S.
Wp
(2.12)
This equation is just the anomalous N = 2 superconformal Ward identity for (abelian) N =
2 vector multiplets, which was found in Ref. [20] by averaging the anomaly relation (2.9)
with respect to the quantum effective action (2.10). A simple derivation of Eq. (2.12) by
using the N = 2 supergravity compensators is discussed in Section 4.
Eq. (2.12) can be applied to a derivation of the SeibergWitten solution [1] provided that
one knows the anomaly S as the function of W , in the context of the N = 2 SYM theory
based on the gauge group SU(2) spontaneously broken to U (1). The original derivation [1]
made use of the electric-magnetic duality and renormalization flow. Since the anomaly S
is N = 2 chiral, gauge-invariant and of dimension two, its vacuum expectation value has to
be proportional to the order parameter u = 12 htr W 2 i, with the coefficient being dictated by
the one-loop beta-function 1 see Eq. (2.9). A comparision with Ref. [1] yields [20,21]
c = 2i1 .
(2.13)
To close Eq. (2.12), as the equation on F , one needs a relation between a = hW i and u.
It was obtained in Ref. [3] by using the modular invariance of u = u(a), in the form of a
non-linear differential equation,
1
(2.14)
(1 u2 )u00 + au03 = 0.
4
A connection to integrable systems arises after identifying the moduli space of the
Coulomb branch in the SeibergWitten model with the moduli space of complex structures
on an elliptic curve. The SeibergWitten solution then appears to be a classical solution to
the equations of motion of a particular spin chain system [2], while the origin of the elliptic
curve underlying the dynamics becomes apparent in this approach. 4 Generalizations to the
larger gauge groups and the presence of N = 2 matter are straightforward, in principle.
Our main goal, however, is to explore what can happen on the hypermultiplet side, at the
level of the LEEA. Unlike the N = 2 vector multiplets, the universal and most symmetric
4 The origin of the elliptic curve in the SeibergWitten exact solution is also explained by brane technology in
the context of M-theory [2224].
124
off-shell formulation of hypermultiplets is only possible in HSS, with the infinite number
of the auxiliary fields.
3. Rigid N = 2 superconformal symmetry in HSS

In the HSS approach [9] the standard N = 2 superspace coordinates {Z} = x m , i , j
are extended by bosonic harmonics (or twistors) ui , i = 1, 2, belonging to the group
SU(2) and satisfying the unimodularity condition
u+i u
i =1,
ui+ = u
i .
(3.1)
The hidden analyticity structure of the N = 2 superspace constraints defining both

N = 2 vector multiplets and FS hypermultiplets, as well as their solutions in terms of
unconstrained N = 2 superfields, can be made manifest in HSS [25].
Instead of an explicit parametrization of the twistor sphere S 2 = SU(2)/U (1), the
SU(2)-covariant HSS approach deals with the equivariant functions of harmonics, having
the definite U (1) charges defined by U (u
i ) = 1. The simple harmonic integration rules,
Z
Z
(3.2)
du = 1, and
du u+(i1 u+im uj1 ujn ) = 0 otherwise,
are similar to the (Berezin) integration rules in superspace. In particular, any harmonic
integral over a U (1)-charged quantity vanishes. The harmonic covariant derivatives,
preserving the defining equations (3.1) in the original (central) basis, are given by
, = ui +i , 0 = u+i +i ui i .
(3.3)
i
u
u
u
u
They satisfy an su(2) algebra and commute with the standard (flat) N = 2 superspace
j
covariant derivatives Di and D . The operator 0 measures U (1) charges.
The key feature of HSS is the existence of the analytic subspace parametrized by
o
n
+
i + +
i +
= x 4i i j u+
u
,
u
,
u
;
u
M ; u = xanalytic
i
i
i , (3.4)
(i j )
++ = u+i
which is invariant under N = 2 (rigid) supersymmetry [9]:

= 4i i + + + i u 4i + + + ,
xanalytic
i
+
+ = i u+
i ,
+
+
i u+
=
i ,
u
i = 0.
(3.5)
but not
, where = (, ).
The analytic dependence includes +
The usual complex conjugation does not preserve analyticity. However, it does, after
being combined with another (star) conjugation that only acts on the U (1) indices as
i
i
(u+
i ) = ui and (ui ) = ui . One has u = ui and ui = u .
(q)
Analytic superfields ( (Z, u), u) of any positive (integral) U (1) charge q in HSS
are defined by the off-shell constraints (cf. the definition of N = 1 chiral superfields)
(q)
= 0,
D + (q) = D +

where D+ = Di u+
i
i +
and D +
= D ui .
(3.6)
125
m
The analytic measure reads d (4) du d4 xanalytic
d2 + d2 + du, and it is of U (1) charge
(4). The covariant derivatives in the analytic basis (3.4) receive certain connection terms.
For example, the harmonic derivative ++ in the analytic subspace is replaced by
D ++ = ++ 4i + +
(3.7)
.
x
This derivative preserves analyticity and permits integration by parts. Similarly, one easily
finds the U (1) charge operator in the analytic subspace reads
D 0 = u+i
i
+
+ .
u
+
u+i
ui
+
+
(3.8)
In what follows we always use the analytic basis and the associated HSS covariant
derivatives denoted by capital D, without making explicit references.
The use of harmonics also gives us control over the (linearly realised) SU(2)R symmetry
(or its absence), in the context of manifest N = 2 supersymmetry (see Ref. [26] for more
details). Since the translational and Lorentz symmetries, as well as N = 2 supersymmetry,
are manifestly realized in HSS, the latter provides us with the natural arena for a study of
truly N = 2 superconformal symmetries on the top of N = 2 non-conformal (rigid or
Poincar) supersymmetry.
The superfield transformation rules with respect to dilatations (with the infinitesimal
parameter ) are dictated by conformal weights w of the superfields, together with the
weights of the N = 2 superspace coordinates,
1
(3.9)
w[ ] = w = , w[u] = 0.
2
The non-trivial part of the N = 2 superconformal transformations is given by SU(2)conf.
internal rotations with the parameters l ij , special conformal transformations with the
w[x] = 1,
parameters k , and N = 2 special supersymmetry with the parameters i and i .

The N = 2 superconformal extension of the spacetime conformal transformations,
x = x + k
x x ,
(3.10)
is dictated by the requirement of preserving the unimodularity and analyticity conditions

in Eqs. (3.1) and (3.6), respectively. As regards the non-trivial part of the N = 2
superconformal transformation laws, one finds [27]

+ +

+
+
x
+
k
x
+
4i
x
x = 4iij u
i j
,

+
+
+ +
+
x
u
+
k
2i
+
x
,
+ = ij u+
i j
+ = ( + ),
h

i
+ +
kj + +
+ +
+ +
u
u
+
4ik
+
4i
u+
u
i
k j
i ,
u
i = 0.
(3.11)
126
Since the building blocks of invariant actions in HSS are given by the measure, analytic
superfields and HSS covariant derivatives, only their transformation properties under the
rigid truly N = 2 superconformal transformations are needed. It follows from Eq. (3.11)
that [27]

( 0 , u0 )
= 1 2, or d (4) du = 2 d (4) du ,
(3.12)
Ber
(, u)
where the HSS superfield parameter

j
(3.13)
= + k x + ij + 4i i j + 4i i u+
i uj
has been introduced. Similarly, one easily finds that

0
0
D ++ = D ++ D ++ D 0 and D 0 = D 0 .
(3.14)
The truly (rigid) N = 2 superconformal infinitesimal parameters can, therefore, be

encoded into the single scalar harmonic superfield that is subject to the constraint [27]
2
(3.15)
D ++ = 0,
(++ )= ++ ,
where ++ D ++ .
(3.16)
The transformations rules of the harmonics,

++
ui ,
u+
i =
u
i = 0,
(3.17)
together with Eqs. (3.12), (3.14), (3.15) and (3.16) yield the very simple and convenient
description of rigid N = 2 conformal supersymmetry (on the top of N = 2 Poincar
supersymmetry) in N = 2 HSS.
The special hyper-Khler geometry of the N = 2 (rigidly) superconformal NLSM in
components was investigated in Ref. [28]. A general solution to the special hyper-Khler
geometry in HSS was described in our recent paper [29]. We use the pseudo-real Sp(1)
notation for a FayetSohnius (FS) hypermultiplet superfield,

(3.18)
qa+ = q + , q + , a = 1, 2, q a+ = ab qb+ ,
which can be easily generalized to the case of several FS hypermultiplets, q a+ q A+
and qA+ = AB q B+ , with a constant (antisymmetric) Sp(k)-invariant metric AB , A, B =
1, . . . , 2k.
First, we recall that the most general (rigidly) N = 2 supersymmetric NLSM can be
formulated in terms of the FS hypermultiplet superfields,
Z
h
i
1
)
,
(3.19)
INLSM [q] = 2 d (4) du 12 qA+ D ++ q A+ + K(+4) (q A+ , u
i
where the real analytic function K(+4) = K(+4) of U (1) charge (+4) is known as a hyperKhler (pre-)potential [30]. 5 By manifest N = 2 supersymmetry of the NLSM action
5 The HSS superfields q are dimensionless. The dimensionality of the measure in the action (3.19) is
compensated by the coupling constant of dimension of length.
127
(3.19), the NLSM metric must be hyper-Khler for any choice of K(+4) . Unfortunately,
an explicit general relation between a hyper-Khler potential and the corresponding hyperKhler metric is not available (see, however, Refs. [3032] for the explicit hyper-Khler
potentials of (ALF) multi-Taub-NUT and AtiyahHitchin metrics, and their derivation
from the NLSM (3.19) in HSS, and Ref. [33] for a review or a general introduction into the
supersymmetric NLSM).
Eq. (3.19) formally solves the hyper-Khler constraints on the NLSM metric in terms
of an arbitrary function K(+4) that may be considered as the (analytic) hypermultilet
counterpart to the (holomorphic) potential F of abelian N = 2 vector superfields in
Eq. (2.10). It is, therefore, natural to impose extra N = 2 superconformal invariance on the
action (3.19), in order to determine a general solution to the special hyper-Khler geometry,
since the free part of the action (3.19) is N = 2 superconformally invariant [27]. The FS
superfields q + have conformal weight one, q + = q + . Together with Eqs. (3.12), (3.17)
and (3.19) it implies two constraints [29]:
K(+4) A+
q = 2K(+4)
q A+
and
K(+4)
= 0.
u+
i
(3.20)
This means that the special hyper-Khler potentials are given by homogeneous (of degree
A+ . There is no restriction on the dependence of K(+4)
two) functions K(+4) (q A+ , u
i ) of q
upon ui , though there should be no dependence upon u+

i .
The HSS description of the N = 2 superconformal hypermultiplet actions depending
upon the FS superfields in terms of a homogeneous (degree 2) potential is thus formally
the same as that of the N = 2 superconformal (abelian) vector multiplet actions in the
standard N = 2 (chiral) superspace (Section 2). However, the special Khler geometry in
the target space of the NLSM arising in the scalar sector of the N = 2 superconformal
action (2.10) is very different from the special hyper-Khler geometry arising from the
N = 2 superconformal NLSM action (3.19) in components.
4. Ward identities and supergravity compensators

The best way of derivation of the superconformal anomaly relations is based on the use
of the supergravity (SG) compensators [34,35]. To compensate the unwanted (local) N = 2
superconformal symmetries in N = 2 conformal SG, one needs two compensators [3638].
This also implies the existence of two anomaly relations in the N = 2 case (cf. Section 2).
In the case of N = 2 superfield SG, its most universal formulation is provided by HSS
[27,39,40] see Appendix A for a short introduction and our notation. Having an N = 2
matter action I coupled to N = 2 SG, one can naturally define an N = 2 supercurrent J
by a variation of the action I with respect to the N = 2 conformal SG potential G,
I
,
(4.1)
G
in the flat limit where all N = 2 conformal SG fields vanish. The N = 2 conformal
SG potential G is defined by Eq. (A.4), where it is introduced as the general N = 2
J=
128
harmonic real superfield, subject to the pre-gauge transformations (A.5) and the gauge
transformations (A.7) whose linearized form reads [40]
G = D ++ l .
(4.2)
The N = 2 superconformal anomalies are also naturally defined in HSS by variational

derivatives of the N = 2 matter action with respect to the N = 2 SG compensators
v5++ (, u) and (, u) [40],
L++ =
I
,
v5++
(4.3)
I
,
(4.4)
and
T ++++ =
where v5++ is the real analytic gauge superfield, associated with the central charge and
transforming under the (linearized) gauge transformations (A.13) as
v5++ = D ++ 5 ,
(4.5)
whereas the analytic density transforms according to Eq. (A.14) that implies (in the
linearized approximation)
= D ++ = D (D + )4 l ,
(4.6)
where we have used the last Eq. (A.6). The HSS superfields, L++ and T (+4) T ++++ ,
representing the N = 2 superconformal anomalies, are analytic by their definition.
Given the N = 2 matter action I that is entirely formulated in the ordinary N = 2
superspace without harmonics (it is certainly the case in the N = 2 gauge sector, without FS
hypermultiplets), one can make a connection to the standard N = 2 anomaly relation (2.8).
The invariance of the action I with respect to the gauge transformations (4.2) obviously
yields
D ++ J = 0,
(4.7)
whereas the invariance of the same action with respect to the gauge transformations (4.5)
implies
D ++ L++ = 0.
(4.8)
Eq. (4.7) means that J is independent upon harmonics too, whereas Eq. (4.8) implies that
+ ij
ij
L++ = u+
i uj L (x, , ), where L satisfies the N = 2 linear multiplet constraints (2.2)
++
due to the analyticity of L . Finally, the invariance of the action I with respect to the
pre-gauge transformations (A.5) and (A.11) yields [40]
i + 2
J = L++ ,
D
4
which is equivalent to Eq. (2.8), as expected.
(4.9)
129
The HSS results about N = 2 SG potentials and compensators imply the natural
definitions of the latter in the ordinary N = 2 superspace by linear relations,
Z

(4.10)
G = du G , u,
and
Z
=
2
du D v5++ (, u).
(4.11)
gives the N = 2 conformal SG potential (cf.

The N = 2 real superfield G(x, , )
Ref. [41]), whereas the abelian N = 2 (restricted chiral) superfield strength squared, 2 ,
can serve as an N = 2 (unrestricted) chiral density, i.e., as the N = 2 chiral compensator
(cf. Ref. [20,21]). The anomaly relation (2.8) is then the direct consequence of the
definitions
I
I
and S =
.
(4.12)
J=
G
2
Having identified the compensator with one of the (abelian) N = 2 gauge superfield
strengths, = W1 , taking the N = 2 gauge LEEA (2.10) to represent the N = 2 matter
action I above results in the anomalous N = 2 superconformal Ward identity (2.12). The
second compensator decouples from the effective N = 2 gauge matter action, so that it
does not have any impact on its anomaly structure. The situation is just the opposite one
in the case where the effective N = 2 matter action represents a selfinteraction of FS
hypermultiplets. The analytic compensator can be considered as a part (density) of the
hypermultiplet matter, while the invariance of the most general N = 2 matter action in HSS
with respect to the gauge transformations (4.2) and (4.6) gives rise to the second anomaly
relation in the form [40]
D ++ J = D T (+4) .
(4.13)
Being applied to the hypermultiplet LEEA I in the form of the N = 2 NLSM (3.19) in HSS,
Eq. (4.13) gives rise to the following anomalous N = 2 superconformal Ward identity:
X
q A+
K(+4)
2K(+4) = T (+4) .
q A+
(4.14)
This is the key equation in our paper. A generic anomaly T (+4) (q, u) is analytic, while
it has to be invariant under the unbroken gauge symmetries, internal symmetries, and
modular transformations, if any. The anomalous Ward identity (4.14) then gives us the
equation on the hyper-Khler potential K(+4) of the effective N = 2 NLSM (LEEA).
5. Examples
To close the anomalous N = 2 superconformal Ward identity (4.14), as the equation
on the effective hyper-Khler potential K(q, u), one has to know the anomaly T as a
function of (q, u) explicitly. In the absence of a general solution for the anomaly (at
130
least, we are unaware ot it), it is worthy to discuss some examples. The non-anomalous
symmetries play the major rle in determining the form of the anomaly T (q, u).
The crucial simplification arises when the SU(2)R automorphisms of N = 2 supersymmetry algebra are not broken (together with the N = 2 supersymmetry that we always
assume). Since the SU(2)R transformations are linearly realised in HSS, the R-invariance
of the hypermultilet LEEA amounts to the independence of the anomaly T (and, hence,
of the hyper-Khler potential K) upon harmonics. Since both have U (1) charge (+4), the
most general (analytic) invariant ansatz is given by a real quartic polynomial of the analytic FS superfields q A+ [26],
T (+4) (q) K(+4) (q) = ABCD q A+ q B+ q C+ q D+ ,
(5.1)
whose coefficients (ABCD) are totally symmetric, being subject to the reality condition,
K (+4) = K(+4) . Not all of the coefficients in Eq. (5.1) are really significant since the
FS kinetic terms in Eq. (3.19) have the manifest global Sp(n) symmetry. It may be not
accidental that this Sp(n) symmetry coincides with the maximal Sp(n) holonomy group of
the hyper-Khler manifolds in 4n real dimensions.
In the case of a single FS hypermultiplet, Eq. (5.1) is simplified to

+ 2 + 2
(+4)
(+4)
4
3 +
+
+
K
=
q
(q ) + (q ) + (q ) q + h.c.
(5.2)
T
2
with one real () and two complex (, ) parameters. The Sp(1) transformations of qa+
leave the form of Eq. (5.2) invariant, but not its coefficients, which can be used to reduce
the number of coupling constants in the family of the hyper-Khler metrics described by
the hyper-Khler potential (5.2) from five to two. In addition, Eq. (5.2) implies the (onshell) conservation laws
D ++ K(+4) = D ++ T (+4) = 0,
(5.3)
which are valid on the equations of motion of the hypermultiplet (FS) superfield,
D ++ q + =
K(+4)
q +
and D ++ q + =
K(+4)
(5.4)
To understand the physical significance of Eq. (5.2), it is instructive to consider first a

simpler case, by assuming the additional (translational) U (1)T symmetry that acts on the

complex superfields q + , q + by phase rotations (with a constant parameter ),
q + ei + ,
q + ei q + ,
(5.5)
but does not move the hyper-Khler structure in the target space of the N = 2 NLSM (3.19).
It happens, e.g., in the N = 2 supersymmetric QED with a single charged hypermultiplet,
or in the Coulomb branch of the SeibergWitten model [42]. In geometrical terms, the
U (1)T symmetry amounts to the existence of a tri-holomorphic (translational) isometry
in the N = 2 NLSM target space. It is worth mentioning that the SU(2)R isometries are
not tri-holomorphic but rotational: they rotate three independent complex structures in the
131
Fig. 1. The one-loop harmonic supergraph contributing to the N = 2 superconformal anomaly.
N = 2 NLSM hyper-Khler target space. Given the SU(2)R U (1)T isometry of the N = 2
NLSM target space, it must be the symmetry of the NLSM hyper-Khler potential, which
implies further restrictions in Eq. (5.2). The unique, SU(2)R U (1)T invariant, hyperKhler potential is obviously given by
+ + 2
(+4)
=
,
(5.6)
KTaub
- NUT 2 q q
while it is known as the hyper-Khler potential of the Taub-NUT metric with the mass
parameter M = 12 1/2 [31].
The induced coupling constant in the one-loop approximation is determined by
the HSS graph depictured in Fig. 1. As was shown in Ref. [43,44], this graph does
lead to the non-vanishing anomalous contribution having the form (5.6), provided that
the matter hypermultiplet has a non-vanishing central charge. The wave lines in Fig. 1
denote the analytic propagators of the N = 2 (abelian) vector superfields V ++ (in N = 2
supersymmetric Feynman gauge) [45],

4
1
(5.7)
D1+ 12 (Z1 Z2 ) (2,2)(u1 , u2 ),
i V ++ (1)V ++ (2) =
21
where (2,2)(u1 , u2 ) stands for the harmonic delta-function [45]. The hypermultiplet
analytic propagators (the solid lines in Fig. 1) with non-vanishing central charges are given
by [43,44]
4
4

D1+ D2+
1
e3 [v(2)v(1)] 12 (Z1 Z2 ),
(5.8)
i q + (1)q + (2) =
+ + 3
21 + m2
u u
1
++ v
e = 0, and m2 = |Z|2 stands for the
where the bridge v satisfies an equation DZ
hypermultiplet (BPS) mass. We find [43]
132

g4 1
m2
1
ln
1
+
,
2 m2
2
2 + m2
(5.9)
where the gauge coupling constant g and the IR-cutoff have been introduced. It is not
difficult to check that 6= 0 only if Z 6= 0. The naive non-renormalization theorem
usually forbids the quantum corrections given by the integrals over a subspace of the full
N = 2 superspace, like the one in Eq. (3.19). However, this theorem does not apply
here, because of the non-vanishing central charges that give rise to the explicit dependence
of the superpropagators (5.8) upon the N = 2 superspace Grassmann (anticommuting)
coordinates via the bridges v(, ) that are responsible for the N = 2 superconformal
anomaly.
The more general SU(2)R -invariant anomaly (5.2) cannot be generated in N =
2 perturbation theory, but it can be generated non-perturbatively, due to instanton
contributions [46]. Of course, in an abelian N = 2 supersymmetric field theory no
instantons exist. This means that the N = 2 perturbative anomaly described by the TaubNUT metric is exact in the abelian case. If, however, the underlying N = 2 supersymmetric
quantum field theory has a non-abelian gauge group of rank larger than one (say, SU(3)),
then one may expect the nonperturbative contributions to the hypermultiplet LEEA (in the
Higgs branch) from instantons and anti-instantons that break the U (1)T symmetry. The onshell relations (5.3) imply that in this case both the anomaly T (+4) and the hyper-Khler
potential K(+4) can be expressed in terms of a real analytic superfield T ++++ satisfying
the constraints
D ++ T ++++ = 0
and
++++
= T ++++ ,
(5.10)
which are the same as the off-shell constraints (2.6) defining an O(4) projective superfield
+ + + ij kl
Unlike
(x, , ).
T ij kl , in the ordinary N = 2 superspace, T ++++ (, u) = u+
i uj uk ul T
the O(2) tensor multiplet defined by Eq. (2.2), the O(4) multiplet does not have a
conserved vector amongst its field components. Hence, the U (1) isometry, if any, in the
N = 2 NLSM to be constructed in terms of T ++++ , is no longer tri-holomorphic (or
translational). The Taub-NUT NLSM (5.6) arises in the limit T ++++ (L++ )2 .
The two-parametric family of the hyper-Khler potentials (5.2) describes the (hyperKhler and SU(2)R -invariant) deformations of the AtiyahHitchin (AH) metric [46]. The
N = 2 (projective) superspace description of the AH metric in terms of an O(4) projective
supermultiplet (2.6) was found in Ref. [47]. The AH metric is known to be the only regular
metric in the family [48]. The difference between the AH and Taub-NUT metrics, being
considered as the metrics in the quantum moduli space of an N = 2 gauge theory (in the
region where quantum perturbation theory applies), can be interpreted as the (exponentially
small) instanton contributions [49,50]. Similar remarks are valid in the more general case
(5.1) [26].
Another simple example of the N = 2 superconformal anomaly, which is still under
control, is possible in the case of the unbroken U (1)T symmetry (5.5). It implies that the

function T (+4) q + , q + ; u be a function of the invariant product q + q + and harmonics
u
i only. The most general ansatz reads
(+4)
q q ;u =
(2l) (q
l=0
+ q + )l+2
l+2
133
(5.11)
whose harmonic-dependent coefficients (2l) are given by
(2l) = (i1 i2l ) u

i1 ui2l ,
l = 1, 2, . . . ,
(5.12)
and obey the reality condition
(2l)
= (1)l (2l) .
(5.13)
The associated solution to Eq. (4.14) takes the similar form (5.11), while it appears to be
the hyper-Khler potential describing the multi-Taub-NUT metrics [32]. The multi-TaubNUT metrics are known to describe static configurations of several (BPS) monopoles [48].
This is not surprising from the viewpoint of the brane engineering of the effective N = 2
supersymmetric gauge field theories in M-theory [2224], where the N = 2 field theory
hypermultiplets are associated with (parallel) D6-branes whose configurations in M-theory
are just described by the multi-Taub-NUT metrics [51,52].
The brane technology/M-theory also suggest a possibility of yet another generalization,
by adding a (parallel) orientifold O6 to the D6-branes [51,52]. In geometrical terms,
this means replacing the U (1)T translational isometry by a rotational U (1)R isometry
in the N = 2 NLSM target space. Though the associated N = 2 NLSM in HSS were
recently constructed in terms of an O(4) projective superfield [29], we are unaware about
their explicit reformulation in terms of the FS superfields. We conjecture that this should
give rise to the Dk series of the asymptotically locally flat (self-dual) metrics in the fourdimensional target space of N = 2 NLSM.
6. Conclusion
In the preceeding section we gave a few non-trivial examples of the N = 2 superconformal anomaly T (+4) that saturates the hyper-Khler potential of the effective hypermultilet
LEEA given by an N = 2 NLSM. To get more (non-perturbative) solutions to our main
Eq. (4.14), it may be better to find first the organizing principle behind all those solutions.
In the SeibergWitten case (2.12), it was the underlying Riemann surface or an integrable
system. It is natural to expect a similar hidden curve behind the hypermultiplet LEEA too
[26]. The SU(2)R -invariant effective NLSM metrics in our examples (Section 5) coincide
with the standard metrics in the (BPS) monopole moduli space of the classical SU(2)YangMillsHiggs system, with magnetic charge n = 1 (Taub-NUT) or n = 2 (Atiyah
Hitchin). In the N = 2 NLSM, whose target space metric is given by the monopole moduli
space metric of higher magnetic charge n > 2, the SU(2)R symmetry is necessarily broken
[26]. As was shown in Ref. [53], a BPS monopole of magnetic charge n > 1 can always
be described by the spectral curve (Riemann surface) of genus (n 1)2 . It is therefore,
conceivable that more general solutions to the anomalous N = 2 superconformal Ward
identity (4.14) are also encoded in terms of the spectral curve, like the SeibergWitten-type
134
solutions to Eq. (2.12). After a dimensional reduction to three spacetime dimensions, our
results support the conjectured mirror symmetry between the Coulomb and Higgs branches
[5456].
Quantum breaking of conformal symmetry in a (classically) superconformal field
theory is related to the appearance of dynamically generated scales. The superconformal
compensators can be naturally interpreted as the superfield extensions of the scale
parameters. From this physical point of view, the (first) N = 2 chiral compensator is
apparently related to the N = 2 superfield extension of the renormalization scale squared,
whereas the second N = 2 compensator appears to be the N = 2 superfield extension of
the induced N = 2 NLSM coupling constant that is (roughly) proportional to the inverse
central charge squared (Section 5). It would be interesting to understand better the physical
significance of the second N = 2 compensator from the viewpoint of the underlying (nonabelian) N = 2 gauge theory and brane technology.
Acknowledgements
I am indebted to L. Alvarez-Gaum, J. Ambjorn, S. Cherkis, S. Gates Jr., A. Gorsky and
E. Ivanov for useful discussions. I am also grateful to Niels Bohr Institute in Copenhagen
for hospitality extended to me during the completion of this work.
Appendix A. N = 2 supergravity in HSS
All (rigid) N = 2 supersymmetric field theories can be naturally defined in HSS, in
terms of unconstrained analytic superfields, with manifest N = 2 supersymmetry [9,25].
N = 2 matter (FS) hypermultiplets are described by complex analytic superfields q + of
U (1) charge +1, whereas their coupling to N = 2 super-YangMills fields is described
via the (Lie algebra-valued) extension of the FS hypermultiplet kinetic operator D ++ by
a gauge connection V ++ , so that the new gauge-covariant operator D++ = D ++ + V ++
preserves analyticity. The rigid N = 2 superconformal transformations also preserve the
analytic subspace of flat N = 2 HSS (Section 3), so it is natural to define N = 2 conformal
supergravity (SG) in HSS along the similar lines [27,39]: by preserving analyticity (of
N = 2 matter & gauge fields), the analytic conjugation (= the product of usual complex
conjugation and Weyl reflection of the sphere S 2 ), and unimodularity (of harmonics). In
this appendix we briefly review the HSS formulation of N = 2 SG along the lines of
Ref. [40] see Refs. [25,27,39,40] for more details.
} be the coordinates of the full N = 2 HSS in the analytic basis,
=
Let { M , u,
(, ). The conformal N = 2 SG transformations are naturally defined in HSS as the

analyticity-preserving diffeomorphisms [27,39],
M = M (, u),
ui+ = ++ (, u)ui ,
(, u, ).
ui = 0,
(A.1)
135
Accordingly, the covariant derivative D++ in N = 2 conformal SG can be put into the form
D+
,
D++ = D ++ + H M++ DM + H (+4)D + H +
(A.2)
) have been introduced in front

where the SG vielbeine (H M++ , H (+4) H ++++ , H +

and D+
. The N = 2
,D
of the flat N = 2 HSS covariant derivatives DM = , D
conformal SG parameters (A.1) can be similarly organized into the one-forms
= M DM + ++ D
and =
D+
.
(A.3)
D++ (+p) = 0, of
Since the SG derivative D++ is supposed to preserve analyticity, D+
+ (+p)
(+p)
of a positive U (1) charge p, D
= 0, the vielbeine of
the analytic superfield
Eq. (A.2) have to obey certain linear constraints, whose solution is given by [40]
H ++ = iD + D + G,
4
H (+4) = D + G,
2
1
H +++ = D + D + G,
8
(A.4)
1
(D + )2 (D + )2 , and the real unconstrained (general HSS superfield) N =
where (D + )4 = 16
2 SG pre-potential G(, u, ) is subject to the pre-gauge transformations [40]
G =
1 + 2 1 + 2
D
+ D
4
4
(A.5)
with the complex unconstrained HSS parameter (, u, ). Accordingly, the gauge

HSS superfield parameters (M , ++ ) are to be expressed in terms of a single real
unconstrained HSS superfield l (, u, ) [40],
= iD + D + l ,
4
++ = D + l .
2
1
+ = D + D + l ,
8
(A.6)
is pure gauge, so that we can simply ignore it, H +

= 0. Then the
It is easy to see that H +
transformation law of G under the remaining gauge transformations with the parameters
(, l ) is given by a simple formula,
G = G + D++ l .
(A.7)
In the absence of superconformal anomalies, there exist a (WZ-type) gauge, where the HSS
superfield G is independent upon harmonics, being subject to the constraints

(A.8)
Dij G x, , = D ij G x, , = 0.
This equation coincides with Eq. (2.7) defining the N = 2 conformal supercurrent, so that
the independent field components of G (in the WZ gauge) are in one-to-one correspondence
with the field content of the off-shell N = 2 conformal SG, as it should.
To construct N = 2 matter and gauge couplings in N = 2 Poincar (non-conformal or
Einstein) SG, one has to compensate truly N = 2 superconformal gauge transformations, namely, dilatations, special conformal transformations, U (1) chiral and SU(2)conf.
rotations, and N = 2 special supersymmetry [3638]. Some of them are compensated by a
136
real analytic gauge superfield V5++ (, u) as the first compensator, subject to abelian gauge
transformations in HSS,
V5++ = D++ 5 ,
(A.9)
with the real analytic parameter 5 (, u). In the context of gauging the N = 2
supersymmetry algebra, the V5++ superfield is associated with the central charge generator
b Though the N = 2 conformal SG itself has the vanishing central charge, N = 2 matter
Z.
hypermultiplets may have non-vanishing central charges. It is natural to incorporate the
++
=
central charge generator into the HSS covariant derivatives by redefining D++ to DZ
++ b
++
++
D +V5 Z, etc. At the component level the N = 2 supermultiplet V5 (in a WZ gauge)
adds extra 8B + 8F off-shell (field) degrees of freedom to the N = 2 Weyl multiplet, thus
forming together the so-called minimal off-shell N = 2 SG multiplet with 32B + 32F field
components [3638].
++
also has to preserve analyticity, which
The extended HSS covariant derivative DZ
++
implies a linear constraint on V5 . A solution to the constraint reads [40]
V5++ =
2
i + 2
i
G D + G + v5++ ,
D
4
4
(A.10)
where v5++ is real analytic. The pre-gauge transformations (A.5) are to be appended by
[40]

4
(A.11)
v5++ = i D + .
The related restrictions on the HSS parameter 5 in Eq. (A.9) are [40]
i + 2 i + 2
l D
l + 5 ,
D
4
4
where 5 is real analytic. One easily finds
5 =
(A.12)
v5++ = v5++ + D++ 5 .
(A.13)
To compensate the remaining unwanted gauge symmetries (e.g., SU(2)conf. ), one needs
a second compensator that may have either a finite or the infinite number of the auxiliary
field components. The three minimal formulations of N = 2 Poincar SG, each having
40B + 40F off-shell (field) degrees of freedom, were described in Ref. [3638]. Unfortunately, all of them impose some restrictions on allowed N = 2 matter couplings, which makes
them of limited use in the context of generic N = 2 NLSM. The most universal choice is
given by a real analytic compensator that compensates the N = 2 local supersymmetry
transformations of the analytic measure,

( 0 , u0 )
(, u).
(A.14)
0 ( 0 , u0 ) = Ber1
(, u)
It allows one to accommodate any N = 2 NLSM in N = 2 SG via the covariantization
of the (rigid) N = 2 NLSM action in HSS, by using the invariant analytic measure
d (4) du . It is the absence of an analytic density that is responsible for the restricted
N = 2 matter couplings in the more conventional formulations of N = 2 SG [3638].
137
Coupling to N = 2 SG deformes hyper-Khler geometry of the N = 2 NLSM target space

into quaternionic geometry [57,58]. The actions of quaternionic NLSM coupled to N = 2
SG in HSS were recently investigated in Ref. [59], where the density was constructed in
a+ 2
terms of a FS hypermultiplet superfield q + as (u
a q ) . When going in the opposite
direction, a rigidly N = 2 superconformal (special hyper-Khler) NLSM arises from the
quaternionic one after putting all the N = 2 conformal SG fields to zero, G = 0, together
with the vanishing Maxwell multiplet, v5++ = 0, and = 1.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
N. Seiberg, E. Witten, Nucl. Phys. B 426 (1994) 19; Nucl. Phys. B 431 (1994) 484.
A. Gorsky, I. Krichever, A. Marshakov, A. Morozov, Phys. Lett. B 355 (1995) 466.
M. Matone, Phys. Lett. B 357 (1995) 342.
J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231.
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105.
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 235.
P.S. Howe, P.C. West, Phys. Lett. B 389 (1996) 273; Phys. Lett. B 400 (1997) 307.
B. Eden, P.S. Howe, A. Pickering, E. Sokatchev, P.C. West, Four-point functions in N = 2
superconformal field theories, hep-th/0001138.
A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, S. Kalitzin, E. Sokatchev, Class. Quantum
Grav. 1 (1984) 469.
M.F. Sohnius, Phys. Lett. B 81 (1979) 8.
S.J. Gates, W. Siegel, Nucl. Phys. B 189 (1981) 295.
S.V. Ketov, I.V. Tyutin, Theor. Math. Phys. 61 (1984) 1126.
S.V Ketov, Fortschr. Phys. 36 (1988) 361.
P.C. West, Introduction to Supersymmetry and Supergravity, World Scientific, 1990.
P.C. West, Introduction to rigid supersymmetric theories, hep-th/9805055.
S. Marculescu, Phys. Lett. B 188 (1987) 203.
B. de Wit, P.G. Lauwers, R. Philippe, S.-Q. Su, A. van Proeyen, Phys. Lett. B 134 (1984) 37.
S.J. Gates, Nucl. Phys. B 238 (1984) 349.
S.V. Ketov, I.V. Tyutin, JETP Lett. 39 (1984) 703.
P.S. Howe, P.C. West, Nucl. Phys. B 486 (1997) 425.
M. Margo, L. ORaifeartaigh, I. Sachs, Nucl. Phys. B 508 (1997) 433.
A. Klemm, W. Lerche, P. Mayr, C. Vafa, N. Warner, Nucl. Phys. B 477 (1996) 746.
A. Hanany, E. Witten, Nucl. Phys. B 492 (1997) 152.
E. Witten, Nucl. Phys. B 500 (1997) 3.
A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky and E. Sokatchev, Harmonic Superspace, Cambridge University Press, to appear.
S.V. Ketov, Exact low-energy effective actions for hypermultiplets in four dimensions, hepth/9909177; to appear in Int. J. Mod. Phys. A.
A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E. Sokatchev, Conformal invariance in harmonic
superspace, in: C.J. Isham, G. Vilkovisky (Eds.), Quantum Field Theory and Quantum
Statistics, Vol. 2, Adam Hilger, 1987, p. 233.
B. de Wit, B. Kleijn, S. Vandoren, Rigid superconformal hypermultiplets, in: E.A. Ivanov,
J. Wess (Eds.), Supersymmetries and Quantum Symmetries, Springer-Verlag, 1999, p. 37; hepth/9808160; Nucl. Phys. B 568 (2000) 475.
S.V. Ketov, Superconformal hypermultiplets in superspace, Nucl. Phys. B 582 (2000) 95,
preceding article in this issue.
138
[30] A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E. Sokatchev, Ann. Phys. (N.Y.) 185 (1988) 1;
Ann. Phys. (N.Y.) 185 (1988) 22.
[31] A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E. Sokatchev, Commun. Math. Phys. 103 (1986)
515.
[32] G.W. Gibbons, D. Olivier, P.J. Ruback, G. Valent, Nucl. Phys. B 296 (1988) 697.
[33] S.V. Ketov, Quantum Non-Linear Sigma-Models, Springer-Verlag, to appear.
[34] S.J. Gates , M.T. Grisaru, W. Siegel, M. Rocek, Superspace, Benjamin/Cummings, 1983.
[35] I.L. Buchbinder, S.M. Kuzenko, Ideas and Methods of Supersymmetry and Supergravity, IOP
Publishing, 1998.
[36] B. de Wit, J.W. van Holten, A. van Proeyen, Nucl. Phys. B 167 (1980) 186.
[37] B. de Wit, R. Philippe, A. van Proeyen, Nucl. Phys. B 219 (1983) 143.
[38] B. de Wit, P. Lauwers, A. van Proeyen, Nucl. Phys. B 255 (1985) 569.
[39] A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E. Sokatchev, Class. Quantum Grav. 4 (1987)
1255.
[40] S.M. Kuzenko, S. Theisen, Correlation functions of conserved currents in N = 2 superconformal theory, hep-th/9907107.
[41] P.S. Howe, K.S. Stelle, P.K. Townsend, Nucl. Phys. B 192 (1981) 332.
[42] S.V. Ketov, Fortschr. Phys. 47 (1999) 643.
[44] E.A. Ivanov, S.V. Ketov, B.M. Zupnik, Nucl. Phys. B 509 (1998) 53.
[45] A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E. Sokatchev, Class. Quantum Grav. 2 (1985) 601;
Class. Quantum Grav. 2 (1985) 617.
[47] I.T. Ivanov, M. Rocek, Commun. Math. Phys. 182 (1996) 291.
[48] M. Atiyah, N. Hitchin, The Geometry and Dynamics of Magnetic Monopoles, Princeton Univ.
Press, 1988.
[49] N. Seiberg, E. Witten, Gauge dynamics and compactification to three dimensions, hepth/9607163; Saclay96. The Mathematical Beauty of Physics.
[50] N. Dorey, V. Khoze, M. Mattis, D. Tong, S. Vandoren, Nucl. Phys. B 502 (1998) 59.
[51] P.K. Townsend, Phys. Lett. B 350 (1995) 184.
[52] A. Sen, JHEP 09 (1997) 1; Adv. Theor. Math. Phys. 1 (1998) 115.
[53] N.J. Hitchin, Commun. Math. Phys. 83 (1982) 579; Commun. Math. Phys. 89 (1983) 145.
[54] R. Brooks, S.J. Gates, Nucl. Phys. B 432 (1994) 205.
[55] K. Intriligator, N. Seiberg, Phys. Lett. B 387 (1996) 513.
[56] A. Kapustin, M.J. Strassler, JHEP 9904 (1999) 21.
[57] J.A. Bagger, E. Witten, Nucl. Phys. B 222 (1983) 1.
[58] J.A. Bagger, A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, Nucl. Phys. B 303 (1988) 522.
[59] E. Ivanov, G. Valent, Quaternionic metrics from harmonic superspace: Lagrangian approach
and quotient construction, hep-th/0001165.

Target space pseudoduality

between dual symmetric spaces
Orlando Alvarez 1
Department of Physics, University of Miami, P.O. Box 248046, Coral Gables, FL 33124, USA
Received 4 May 2000; accepted 19 May 2000
Abstract
A set of on shell duality equations is proposed that leads to a map between strings moving on
symmetric spaces with opposite curvatures. The transformation maps waves on a Riemannian
e This transformation
symmetric space M to waves on its dual Riemannian symmetric space M.
preserves the energy momentum tensor though it is not a canonical transformation. The preservation
of the energy momentum tensor has a natural geometrical interpretation. The transformation maps
particle-like solutions into static soliton-like solutions. The results presented here generalize
earlier results of E. Ivanov. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w; 03.50.-z; 02.40.-k
Keywords: Duality; Strings; Geometry
1. Introduction
We take spacetime to be two dimensional Minkowski space. Let = be the
standard lightcone coordinates and let be a massless free scalar field satisfying the wave
2 = 0. The standard abelian duality transformations (for a review see [1]) are
equation +
+ = ++ ,
(1.1)
= .
(1.2)
2 = 0.
The integrability conditions for the above are precisely the equations of motion +
This means that we can construct (
+ , ) and verify that as a consequence of the
2
above + = 0. A generalization of the above are the pseudochiral models of the type
introduced by Zakharov and Mikhailov [2]. Consider a standard sigma model with target
This work was supported in part by National Science Foundation grant PHY-9870101.
1 oalvarez@miami.edu
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 0 7 - 2
140
O. Alvarez / Nuclear Physics B 582 (2000) 139154
space a Lie group G and with equations of motions a (g 1 a g) = 0. Introduce the dual
Lie algebra valued field by
g 1 a g = a b b .
It was shown by Nappi that these two descriptions were not quantum mechanically
equivalent [3]. The correct dual models were found by Fridling and Jevicki [4] and by
Fradkin and Tseytlin [5]. It was eventually understood that the duality transformation
should be a canonical transformation [6,7] and the transformation in the pseudochiral
model is not. Nevertheless the pseudochiral model has a variety of interesting field theoretic
features [6,8]. Motivated by string theory, there is now a vast literature on nonabelian
duality [921] and PoissonLie duality [2227].
Following up on recent work [28] we consider a variant of the duality equations proposed
there. These equations are a generalization of the ones of Zakharov and Mikhailov. In light
of the introductory paragraph it will be interesting to study the physical and mathematical
properties of these equations. Here we point out that there is an interesting transformation
that maps solutions of the wave equation on a symmetric space into solutions of the wave
equation on a symmetric space with the opposite curvature (see the next paragraph). We
note that it was pointed out in [8] that the opposite signs found by Nappi in the beta
functions for the dual models of Zakharov and Mikhailov are explained by observing that
the generalized curvatures in the models have opposite signs.
The results presented here are a generalization 2 of results of E. Ivanov [29]. There is a
large body of literature discussing conserved currents in sigma models based on groups or
coset spaces. A seminal work was Pohlmeyers construction [30] for an infinite number
of conserved currents in sigma models with target space S n . This was generalized by
Eichenherr and Forger [31,32] who showed that the construction generalized to symmetric
spaces. Ivanov expanded on these ideas and in doing so introduced the notion the dual
algebra and the dual sigma model. He states,
. . . in which we show that the equations of any d = 2 model on a symmetric space
simultaneously describe a d = 2 model on some other, dual factor space. . . [29,
p. 475].
He explicitly works out the example of a sigma model with target a compact real Lie group
G (with zero B-field). He views G as a symmetric space G G/G and he explicitly
shows that the sigma model on G is dual to the sigma model on GC /G where GC is
the complexification of G. He subsequently asserts that the construction generalizes to a
generic symmetric space. Ivanovs construction for a symmetric space is given in Section 6.
The work here generalizes Ivanovs in that we assume a general Riemannian manifold and
show that it must be a symmetric space.
We first establish some notation. The sigma model with target space M, metric g and
2-form B will be denoted by (M, g, B) and has Lagrangian
i j

x i x j
x x
x i x j
1
(1.3)
+ Bij (x)
L = gij (x)
2

2 I am thankful to E. Ivanov for bringing his work to my attention.
141
with canonical momentum density

i =
L
= gij x j + Bij x 0j .
x i
(1.4)
The stress energy tensor for this sigma model is given by + = 0,

++ = gij (x)+ x i + x j
and = gij (x) x i x j .
(1.5)
e g,
e is a
In general a duality transformation between sigma models (M, g, B) and (M,
B)
canonical transformation between the respective phase spaces that preserves the respective
hamiltonian densities. We can study a less restrictive situation where we only have on
shell duality. By this we mean that we only require that a map exists between solutions to
e g,
e
the equations of motion of (M, g, B) and solutions of the equations of motion of (M,
B).
The on shell transformation proposed below will be referred to as pseudoduality. There
is a certain naturalness to the pseudoduality transformation because it preserves the stress
energy tensor.
As in [28] it is convenient to choose an orthonormal frame {i } with the antisymmetric
Riemannian connection ij . The Cartan structural equations are
di = ij j ,
1
dij = ik kj + Rij kl k l .
2
We also define the curvature 2-forms by ij = 12 Rij kl k l . Consider two sigma models
e g,
e Using the orthonormal frame we define the appropriate a
(M, g, B) and (M,
B).
e by 3
derivative of the maps x : M and x : M
i = x i a d a
and i = x i a d a .
(1.6)
In order to describe the equations of motion for the sigma model we need to consider
second derivatives. The covariant derivatives of x i a are x i ab and are defined by
dx i a + ij x j a = x i ab d b .
(1.7)
Taking the exterior derivative of i = x i a d a we learn that x i ab = x i ba . The equations of

motion coming from (1.3) are
1
x k + = Hkij x i + x j ,
2
(1.8)
where H = dB. There are similar definitions and equations for x i .

The Lagrangian version of the hamiltonian duality equations in [28] may be written as
+ ( ),
x+ ( ) = +T+ (x, x)x
(1.9)
( ).
x ( ) = T (x, x)x
(1.10)
3 More correctly these equations are pullbacks of the type x i = x i d a . In the field of exterior differential
a
systems [33] the pullback is implicit and not usually written. We adhere to that convention.
142
e SO(n) are not arbitrary but related

The orthogonal matrix valued functions T : M M
by
T+ (I + n) = T (I n),
(1.11)
e satisfies some PDEs given in [28]. In this

where the antisymmetric tensor nij on M M
article we relax such restrictions on T and consider orthogonal matrix valued functions
T : SO(n) with the constraint that solutions to the the sigma model (M, g, B) are
e g,
e and vice versa. We will refer to these equations as the
mapped into solutions of (M,
B)
e = .
pseudoduality equations. Note that the pseudoduality transformations satisfy
e In other words, we
This is very different than in [28] where T are functions on M M.
allow explicit dependence on in T where as in [28] the dependence occurs because x
and x depend on .
We specialize in this article to the case where T+ = T = T : SO(n) and H = 0,
e = 0. In this case the equations of motion become 4 x+ = x+ = 0 and the duality
H
equations are
x ( ) = T ( )x ( ).
(1.12)
e = 0 the sigma models are worldsheet parity and

We point out that when H = H
worldsheet time reversal invariant. The isometry T is odd under worldsheet parity and
e 6= 0 and therefore
under worldsheet time reversal. Note that in the pseudochiral model H
e will be treated
it is not covered in this paper. The general case with generic H and H
elsewhere [34].
Duality equations (1.12) are mathematically quite interesting. We digress a bit and
e
discuss the question of local Riemannian isometries. Assume we have a map f : M M
and we wish for this map to be an isometry. If {j } and { i } are local orthonormal frames
then we require that the metric be preserved: f ( i i ) = j j . The solution to
this equation is f = T , where T : M O(n). This Pfaffian system of equations is
integrable [33] if the Riemann curvature tensor and its higher order covariant derivatives
in the two spaces agree when identified via T . For more details see the discussion in the
e that
paragraph following (3.16). In our case we begin with maps x : M, x : M
satisfy hyperbolic Lorentz invariant equations. We fix a Lorentz frame and we observe
e | x+ = 0}
that we are interested in maps from {x : M | x+ = 0} into {x : M
that preserve the two independent components of the energy momentum tensor ++ and
. A linearity assumption and (1.5) tells us that the maps have to be of form (1.9)
and (1.10) with the more general T ( ). Given a map x : M there are two preferred
tangent vector fields / + and / . The conditions that require to be preserved
are very geometric: the lengths of / + and / are preserved by the map. If
denotes the Hodge duality operation on , our equations may be written as = (T )
where we interpret and as pull backs to . It is the desire to have the operation that
introduces a 1 in Eq. (1.10) even though in principle the 1 could be absorbed 5 into T .
4 We remind the reader that solutions to Laplaces equation are called harmonic functions yet solutions to the
wave equation are simply called waves.
5 We may have to allow T (n).
O
143
e
Here we show that there are interesting maps between waves on M and waves on M
that preserve natural geometrical structures. This may be of interest to researchers who
study two dimensional wave equations.
This paper is organized in the following way. In Section 2 we discuss pseudoduality
between the 2-sphere S 2 and the 2-dimensional hyperbolic space H 2 very concretely.
In particular we explicitly verify the necessity for the the orthogonal matrix valued
function T . In Section 3 we become a bit more abstract but still concrete by working with
explicit metrics and show that there is pseudoduality between S n and H n . In Section 4 we
e must be symmetric spaces
assume general metrics and show that the manifolds M and M
with the opposite curvatures. In Section 5 we construct many examples by discussing the
theory of dual symmetric spaces. Section 7 is a discussion of the results of this article.
2. A pedagogic example
It is worthwhile to be very concrete and to consider the pseudoduality between strings
moving on a 2-sphere S 2 and those moving on a 2-hyperboloid H 2 . This example
illustrates the necessity for the matrix T . We use a different coordinate version of the
constant curvature metric than in Section 3 to emphasize that everything is independent of
the choice of coordinates. The respective constant curvature metrics on S 2 and H 2 in polar
normal coordinates are
sin2 r 2
sinh2 r 2
2
2
d
,
d
s
=
d
r
+
d .
2
2
The equations of motion for the sigma models are:
ds 2 = dr 2 +
2
r=
+
sin r cos r
+ ,
(2.1)

sin r 2
+ = cos r + r + r+ ,
(2.2)
sinh r cosh r
2
+ ,
r =
(2.3)
+

sinh r 2
+ = cosh r + r + r + .
(2.4)
This elementary example illustrates the importance of the matrix T . Assume T = I then
two of the duality equations in this coordinate system would be + r = + r and r =
2 r = 0 and 2 r = 0 but these are not the equations of
r. This would imply that +
+
motion. We will see that with a T 6= I we can construct a pseudoduality transformation.
First we need an orthonormal frame. On S 2 we have r = dr and = (sin r/) d .
From this it follows that the connection and the curvature are respectively given by r =
cos(r) d and dr = 2 r . On H 2 we have r = dr and = (sinh r /) d .

From this we see that the connection and the curvature are respectively given by r =
cosh( r ) d and d r = 2 r .
We need a 2 2 orthogonal matrix that we parametrize by a rotation angle . The duality
equations are
144
r
sinh( r )/

=
cos sin
sin cos

r
sin(r)/

.
(2.5)
The integrability conditions on the above lead to the equation

d = cosh( r ) d + cos(r) d.
(2.6)
This is integrable if = where integrable means integrable modulo the equations of

motion.
What happens to the point particle geodesics on S 2 ? One easily verifies that for
constant a, the functions r = a ( + + ) = a (2 ) and = 0 give a solution of Eqs. (2.1)
and (2.2). This corresponds to particle geodesics on S 2 . We would like to understand
what are the dual solutions to the particle geodesics. If we note that r = a and = 0
we see that the duality equations become
r = a cos ,
sinh r
= a sin .
(2.7)
(2.8)
An immediate consequence of the above is that (+ + )r = 0 and (+ + ) = 0. We

conclude that r = r ( + ) = r (2 ) and = ( + ) = (2 ). Thus we get static
solutions on the hyperboloid. Note that (2.6) immediately tells us that = ( + ) =
(2 ). Since everything is a function of 2 = + the situation reduces to functions
of a single variable. The transformed solutions will be static solutions ( independent).
Note that Eqs. (2.7) and (2.8) lead to

sinh r 2
+ = a 2 .
(2.9)
(+ r )2 +
This is the conservation of energy equation associated with the particle Lagrangian

sinh r 2
1
2
+
.
L = (+ r ) +
2
This is a standard problem in classical mechanics. The canonical momentum

sinh r 2
L
+
=
J =
(+ )
is a constant of the motion. Thus the energy integral (2.9) may be written as
(+ r )2 +
2 J2
sinh2 r
= a 2,
(2.10)
and this problem is reducible to quadrature. This is interesting because we know that there
exists crystallographic subgroups of SL(2, R) such that H / is a genus g > 1 compact
Riemann surface. Such a surface has closed geodesics of minimal length and some of these
are the static soliton-like solutions (r (2 ), (2 )) we are constructing above.
145
3. Constant curvature metric

Before presenting the general theory we study the special case of constant curvature
spaces. Again we take T+ = T = T . The two dimensional nonlinear sigma model on a
space with constant positive curvature k > 0 is shown to be pseudodual to the nonlinear
sigma model on a space with constant negative curvature k. The pseudoduality equations
are used to map solutions of one model into solutions of the other model.
Locally, a metric on a space with constant curvature is may be written in the Poincar
form
dx i dx i
.
(1 + kx 2 )2
An orthonormal frame in this metric is
ds 2 =
dx i
.
1 + kx 2
The connection one-forms (with respect to the orthonormal frame) are
i =
ij = 2k
x i dx j x j dx i
.
1 + kx 2
(3.1)
(3.2)
(3.3)
The first Cartan structural equation is di = ij j . The second structural equation

gives the the curvature two forms
ij = dij + ik kj = 12 Rij kl k l = 4ki j .
(3.4)
Technically the curvature is 4k.

The Lagrangian for the sigma model is
2+ x i x i
.
(1 + kx 2 )2
The equations of motion associated with this Lagrangian are
L=
(3.5)
x j + x j x i + x j x j + x i x i + x j x j
.
(3.6)
1 + kx 2
e with respective curvatures k
Assume we have two constant curvature spaces M and M
and k. We would like to see if it is possible to have a duality transformation between them.
2
x i = 2k
+
Assume we have a solution x( ) of the sigma model on M. We attempt to construct a

e by requiring that
solution of the sigma model on M
+ x j
+ x i
= T ij
,
1 + kx 2
1 + k x 2
(3.7)
x j
x i
= T i j
,
1 + kx 2
1 + k x 2
(3.8)
where T is an orthogonal matrix, det T = 1. Note that the stress energy tensors satisfy
e .
=
The first step towards the integrability conditions for the above system is to differentiate
(3.7) with respect to and (3.8) with respect to + :
146
x j x j + x i
(+ x i )
2k
1 + k x 2
(1 + k x 2 )2
2
j
k
k
j
+ x j
i + x
i x x + x
+
T
2kT
,
j
j
1 + kx 2
1 + kx 2
(1 + kx 2 )2
x j + x j x i
+ ( x i )
+ 2k
1 + k x 2
(1 + k x 2 )2
= T i j
= + T i j
2
j
k
k
j
x j
i + x
i x + x x
+
T
2kT
.
j
j
1 + kx 2
1 + kx 2
(1 + kx 2 )2
(3.9)
(3.10)
By imposing the integrability conditions + ( x i ) = (+ x i ) we can add the above two

2 x and by imposing only the equations of motion for x i and the
equations to eliminate +
duality relations we obtain
+ x j
x j
i
+
T
+
j
1 + kx 2
1 + kx 2
j
i
i
j
j
i
i
j
k
x x x x j + x k
x + x x + x T j k x
= 2k
T k
+
2
k
1 + kx 2
1 + kx 2
1 + k x 2
1 + k x 2
T i j
k
j
j
k xk
x k x j x j x k + x k
i x + x x + x
2kT
. (3.11)
j
2
2
2
1 + kx
1 + kx
1 + kx
1 + kx 2
The particular combinations in the above follow from the general theory and the form
of (3.3). Note that at any point we can make x i ( ) arbitrary thus we need that
2kT i j
x j x i x i x j j
x k x j x j x k
T k 2kT i j
,
T i k = 2k
1 + kx 2
1 + k x 2
x j + x i x i + x j j
x k + x j x j + x k
T k 2kT i j
.
+ T i k = 2k
1 + kx 2
1 + k x 2
(3.12)
(3.13)
We interpret the above as the pullback under the map (x( ), x(

)) of the 1-form
dT i j = ik T k j + T i k kj .
(3.14)
We can immediately verify that if one inserts (3.12) into equation (3.9) we get the
equations for motion for x,
i.e., equations (3.6) with all quantities with tildes. Thus far we
have shown that if one starts with a solution of the sigma model defined by Lagrangian (3.5)
and we impose the generalized duality equations then we find a solution to the tilde sigma
model provided that we can satisfy equations (3.13) and (3.12).
The integrability conditions for the dT equation are found by taking the exterior
derivative leading to
eik T k j + T i k kj .
0 =
(3.15)
In our case the curvature is very simple and the above reduces to
4k i k T k j + 4kT i k k j = 0.
(3.16)
First we discuss what equation (3.15) is not and afterwards we discuss what it is. This
equation shows up in the study of local Riemannian isometries [33]. If we have two
147
e and a map f : M M
e then the conditions that f be
Riemannian manifolds M and M
a local isometry is that there exists an orthogonal transformation T such that i = T i j j .
When one works out the integrability conditions for this system one finds that (3.14)
must be satisfied along with its integrability condition which is (3.15). This tells us that
eij kl T k m T l n = T i k T j l Rklmn . There are further integrability conditions which tell us that
R
the covariant derivatives of the curvatures also satisfy similar relations. This is the classical
theorem of Christoffel on local isometries between Riemannian manifolds as reformulated
by E. Cartan. In our constant curvature example the requirement simply becomes k = k.

6
e
Our situation is very different. In effect our setup is a map from to M M. This
tells us that
(a x i ) d a
(a x i ) d a
and i =
.
i =
2
1 + kx
1 + k x 2
Substituting these into (3.16), using duality relations (3.7) and (3.8), and noting that at any
The change in sign
( + , ) we are free to arbitrarily specify x i ( ) leads to k = k.
is due to the negative sign in (3.8). Thus we discover that the duality equations can be
implemented if the constant curvature manifolds have the opposite curvature.
4. General theory
This problem is best analyzed in the bundle of orthogonal frames [34]. Because most
physicists are not familiar with this approach we work things out on the base manifold
e First thing to do is to take the exterior derivative of (1.12)
M M.
dx = (dT )x T dx
and use the definitions of the covariant derivative (1.7) to obtain
x + xa d a = (dT )x T x T xa d a .
If we use the duality equations (1.12) we have
T
x + xa d a = (dT )x T x T xa d a .
A little algebra shows that
)x T xa d a .
xa d a = (dT T + T
We wish to isolate the integrability conditions so wedge the above with
(4.1)
d .
)x d T x d d .
x d d = (dT T + T
We have two equations
)x+ d + + T x+ d d + ,
x+ d d + = +(dT T + T
)x d T x+ d + d .
x+ d + d = (dT T + T
In principle we wish that the integrability conditions x+ = x+ are satisfied if the
equations of motion x+ = 0 hold. Subsequently we would like that this implies that
6 We avoid discussing jet bundles.
148
x+ = 0. We might as well substitute x+ = 0 and x+ = 0 into the equations above

and find
0 = (dT T + T
)x d .
i
may be arbitrarily specified at any we have
Since x
0 = (dT T + T
) d .
e are maps of a two dimensional worldsheet we conclude
Since x : M and x : M
that on the worldsheet, the covariant derivative of T vanishes
dT T + T
= 0.
(4.2)
The reason is that the covariant differential of T is a 1-form. All our objects arise from
maps with domain so the covariant differential of T is a 1-form on . You are pulling
back both and to . Note that a covariantly constant tensor is determined its value
at one point on the worldsheet; the values elsewhere are determined by parallel transport.
In order to construct such a T we need to verify the integrability conditions for the above.
e Taking the exterior
These lead to important constraints on the geometry of M and M.
derivative and using the Cartan structural equations leads to
eik T k j = 0
T i k kj +
with special case (3.15). Expanding the above gives
eiklm T k j l m = 0.
12 T i k Rkj lm l m + R
If we now substitute (1.6) and use the pseudoduality equations (1.12) we see that
eij kl T k m T l n .
T i k T j l Rklmn = R
(4.3)
e have the opposite curvature. Next we

Thus we conclude that the manifolds M and M
take the exterior derivative of the above to look for further conditions. If the covariant
e we find
differential of R is defined by DRij kl = Rij kl;m m and similarly for R
eij kl;p T k m T l n p .
T i k T j l Rklmn;p p = R
If we now substitute (1.6) into the above and use the pseudoduality equations (1.12) we
obtain two equations since d + and d are independent:
eij kl;p T k m T l n T p q ,
T i k T j l Rklmn;q = R
eij kl;p T k m T l n T p q .
T i k T j l Rklmn;q = +R
The solution of the above is immediate
eij kl;p = 0.
Rklmn;q = 0 and R
(4.4)
e must be locally symmetric spaces with the opposite curvature.

The manifolds M and M
149
5. Dual symmetric spaces

There is a large class of examples of pairs of symmetric spaces with opposite curvature.
These pairs are called dual symmetric spaces. The simplest pair is the n-sphere S n and the
n-dimensional hyperbolic space H n . More complicated pairs are given by the Grassmann
manifolds 7 SO(p + q)/ SO(p) SO(q) and O0 (p, q)/ SO(p) SO(q).
The general theory of symmetric spaces is very extensive [35,36]. For our purposes we
take a lighter approach [37, Chapter 11] and a more restrictive view and consider what are
called normal symmetric spaces. These are specified by a triplet of data (G/H, , Q) where
G is a real Lie group, H is a closed subgroup of G and is an involutive automorphism
of G. Let g be the Lie algebra of G and denote the action of the automorphism on g
by s. Let h be the +1 eigenspace of s and let m be the 1 eigenspace of s. Since 12 (I s)
are projectors we have g = h m. The following are also required in a normal symmetric
space:
1. If F is the fixed point set of and F0 is its identity component then F0 H F .
This is a technical requirement that for our purposes we take it to mean that the Lie
algebra of H is h.
2. Q is an Ad(G)-invariant inner product on g. An ordinary symmetric space only
requires Q to be an Ad(H )-invariant inner product 8 on m. Here Ad(G) is the adjoint
action of the group G on its Lie algebra g.
3. Q is s-invariant. This is not required for an ordinary symmetric space.
It follows that [h, h] h, [h, m] m and [m, m] h. The s-invariance of Q tells us
that the direct sum decomposition is an orthogonal decomposition. Note that h and m
are orthogonal with respect to any s-invariant quadratic form such as the Killing form
Tr(ad(X) ad(Y )).
We pick an origin for the symmetric space G/H and associate m with the tangent space
at that point. For a normal symmetric space the sectional curvature associated to the 2-plane
spanned by X, Y m is given by the following simple formula [37, p. 319]
K(X, Y ) =
Q([X, Y ], [X, Y ])
Q(X, X)Q(Y, Y ) Q(X, Y )2
(5.1)
that requires the Ad(G)-invariance of Q in its derivation. We remind the reader that
knowing all sectional curvatures is equivalent to knowing the curvature tensor [37] and
that they are related by
K(X, Y ) =
Rij kl Xi Y j Xk Y l
.
Q(X, X)Q(Y, Y ) Q(X, Y )2
e H
e, , Q)
e are said to be dual symmetric
Normal symmetric spaces (G/H, , Q) and (G/
spaces if there exist
e
(1) Lie algebra isomorphism S : h h such that Q(SV
, SW ) = Q(V , W ) for all
V , W h;
7
O0 (p, q) is the component of O(p, q) connected to the identity.

8 An Ad(H )-invariant inner product on m leads to a G-invariant metric on M, see [37, p. 312].
150
e such that [T X, T Y ] = S[X, Y ] for all X, Y m.

(2) Linear isometry T : m m
In item (1) above the Lie algebra isomorphism tells us that brackets in h are the same as in
While the isometry in item (2) tells us that inner product on m is the same as in m
e.
h.
For dual symmetric spaces it is easy to see that the sectional curvatures are related by
e X, T Y ) = K(X, Y ).
K(T
It is worthwhile to work this out explicitly for the example of the dual symmetric spaces
SO(p + q)/ SO(p) SO(q) and O0 (p, q)/ SO(p) SO(q). For g G = SO(p + q) or
e = O0 (p, q) we take and to be given by
gG

Ip
Ip
0
0
g 7
g
.
0 Iq
0 Iq
One easily verifies that h = h = so(p) so(q) where the matrices are of the form

a 0
0 c
em
e are of the form
where a so(p) and c so(q). One also sees that X m and X

t
t
0 b
e= 0 b ,
and X
X=
b
0
b 0
where b is an arbitrary q p matrix. For the inner products we take Q(X, Y ) = 12 Tr(XY )
eY
e). These will be Riemannian dual symmetric spaces because
e X,
e Y
e) = + 1 Tr(X
and Q(
2
e to be
e are positive definite. We take the subgroups H and H
the metrics on m and m
SO(p) SO(q) and the Lie algebra isomorphism S : h h is the identity transformation.
e is taken to be
The isometry T : m m

0 bt
0 bt
7
.
T:
b
0
b 0
A brief computation shows that [T X, T Y ] = [X, Y ]. From this we see that SO(p +
q)/ SO(p) SO(q) is a space with positive sectional curvature and O0 (p, q)/ SO(p)
SO(q) is a space with negative sectional curvature.
A more extensive discussion of dual symmetric spaces requires a thorough discussion
of the theory of orthogonal involutive Lie algebras (orthogonal symmetric Lie algebras),
see [35,36].
6. Ivanovs construction
In this article we have limited ourselves to studying sigma models on an arbitrary
Riemannian manifold with vanishing 3-form Hij k . Ivanov analyzed a subset of those
models. He studied sigma models that are Lie group theoretic in origin. In this case we can
use special properties of Lie groups to simplify the analysis. Assume we have a normal
symmetric space G/H as in Section 5. We can choose an orthonormal basis that respects
the decomposition g = h m. If {Ti } is such a basis with Lie bracket relations
[Ti , Tj ] = f k ij Tk
151
then the Ad(G)-invariance of Q tells us that fij k are totally antisymmetric. The indices
a, b, c, . . . are associated to the basis elements that are in h and the indices , , , . . . are
associated to basis elements in m. With this notation the Lie algebra bracket relations are
[Ta , Tb ] = f c ab Tc ,
(6.1)
[Ta , T ] = f
a T ,
(6.2)
[T , T ] = f
Tc .
(6.3)
Let i be the the MaurerCartan forms for the Lie group G that satisfy the MaurerCartan
equations
1
d i + f i j k j k = 0.
2
Using the decomposition g = h m we may write the equations above as
1
1
d a + f a bc b c = f a ,
2
2
d + f b b = 0.
(6.4)
(6.5)
It is worthwhile discussing the geometric meaning of the above equations. It is well

known that G is a principal H -bundle over G/H . The H -connection a Ta on G defines
the horizontal tangent spaces. Its curvature is given by the right hand side of (6.4).
Equation (6.5) states that the covariant differential of the m-valued 1-form T is zero.
The equations of motion nonlinear sigma model with target space G/H may be
formulated in terms of a map g : G that satisfies the equations
1
1
d a + f a bc b c = f a ,
2
2
d + f b b = 0,
d( ) + f
( ) = 0.
(6.6)
(6.7)
(6.8)
Here we implicitly interpret i as the pullback under g : G/H . More properly we

should have written g i . We use to denote the Hodge duality operation on .
On 1-forms it is given by (d ) = d . The first two equations above are just
the pullbacks to of the MaurerCartan equations for G. The third equation (6.8) is
essentially the wave equation. For future reference we note that if , are 1-forms on
then
( ) ( ) = ,
(6.9)
and that ( )2 = .
Ivanov observed that if we define = and a = a then the equations above
become
1
1
(6.10)
d a + f a bc b c = + f a ,
2
2
(6.11)
d + f b b = 0,
(6.12)
d( ) + f b ( ) = 0.
152
These equations may be interpreted as the equations of motion for a sigma model on the
e
e is a real Lie group with real Lie algebra g = h (im).
symmetric space G/H
where G
This Lie algebra has Lie brackets given by
eb ] = f c ab Tec ,
[Tea , T
e ,
[Tea , Te ] = f a T
(6.13)
[Te , Te ] = f c Tec .
(6.15)
(6.14)
e = iT . Note that if you think of G

e as a principal H ea = Ta and T
In the above we have T
e given by (6.10), is opposite to that of G, given by (6.6).
bundle then the curvature of G,
7. Discussion
In [28] it was shown that a necessary condition for target space duality is that
the target spaces be parallelizable manifolds. In the scenario presented here where we
only require on shell pseudoduality and we see that the parallelizable requirement
is weakened substantially but we still find natural geometric restrictions on the target
e vanish we saw that
spaces. In the models considered here where the 3-forms H and H
pseudoduality requires that the target spaces be symmetric spaces with the opposite
curvature generalizing a result of Ivanov [29]. The class of Riemannian dual symmetric
spaces provides a wealth of examples. In particular we studied explicitly the example of
dual Grassmann manifolds SO(p + q)/ SO(p) SO(q) and O0 (p + q)/ SO(p) SO(q).
We also saw the importance of the isometry T . Note that we never explicitly solved for T
but we discovered conditions that guarantee its existence. Finally it should be emphasized
that T depends on the path since it comes from integrating (4.2) along the path.
We note that equations (1.12) may also be written as
x = T x ,
(7.1)
x = T x .
(7.2)
Thus we immediately see that particle-like solutions ( independent) on M get mapped

e and vice-versa. If say M
e has noncontractible loops
into static soliton-like solutions on M
then there will be stable soliton-like solutions.
Recently Evans and Mountain [38] constructed an infinite number of local conserved
commuting charges on sigma models with the target space being a compact symmetric
space. It would be interesting to apply the results of this paper to the construction of Evans
and Mountain.
Acknowledgements
I would like to thank T.L. Curtright, H. Fenderya, L.A. Ferreira, L. Mezincescu,
R. Nepomechie, J. Snchez Guilln, I.M. Singer and C. Zachos for comments and
discussions. I would also thank E. Ivanov for bringing to my attention reference [29]. This
work was supported in part by National Science Foundation grant PHY-9870101.
153
References
[1] A. Giveon, M. Porrati, E. Rabinovici, Target space duality in string theory, Phys. Rept. 244
(1994) 77202, hep-th/9401139.
[2] V.E. Zakharov, A.V. Mikhailov, Relativistically invariant two-dimensional models in field
theory integrable by the inverse problem technique, Sov. Phys. JETP 47 (1978) 10171027
(in Russian).
[3] C.R. Nappi, Some properties of an analog of the nonlinear sigma model, Phys. Rev. D 21 (1980)
418.
[4] B.E. Fridling, A. Jevicki, Dual representations and ultraviolet divergences in nonlinear sigma
models, Phys. Lett. B 134 (1984) 70.
[5] E.S. Fradkin, A.A. Tseytlin, Quantum equivalence of dual field theories, Ann. Phys. 162 (1985)
31.
[6] T. Curtright, C. Zachos, Currents, charges, and canonical structure of pseudodual chiral models,
Phys. Rev. D 49 (1994) 54085421, hep-th/9401006.
[7] E. Alvarez, L. Alvarez-Gaume, Y. Lozano, A canonical approach to duality transformations,
Phys. Lett. B 336 (1994) 183189, hep-th/9406206.
[8] C.K. Zachos, T.L. Curtright, The paradigm of pseudodual chiral models, in: K.C. Wali (Ed.),
PASCOS 94, 1995, pp. 381390; hep-th/9407044.
[9] E.B. Kiritsis, Duality in gauged WZW models, Mod. Phys. Lett. A 6 (1991) 28712880.
[10] M. Rocek, E. Verlinde, Duality, quotients, and currents, Nucl. Phys. B 373 (1992) 630646,
hep-th/9110053.
[11] A. Giveon, M. Rocek, Generalized duality in curved string backgrounds, Nucl. Phys. B 380
(1992) 128146, hep-th/9112070.
[12] X.C. de la Ossa, F. Quevedo, Duality symmetries from nonabelian isometries in string theory,
Nucl. Phys. B 403 (1993) 377394, hep-th/9210021.
[13] M. Gasperini, R. Ricci, G. Veneziano, A problem with nonabelian duality?, Phys. Lett. B 319
(1993) 438444, hep-th/9308112.
[14] A. Giveon, M. Rocek, Introduction to duality, hep-th/9406178.
[15] A. Giveon, M. Rocek, On nonabelian duality, Nucl. Phys. B 421 (1994) 173190, hepth/9308154.
[16] A. Giveon, E. Kiritsis, Axial vector duality as a gauge symmetry and topology change in string
theory, Nucl. Phys. B 411 (1994) 487508, hep-th/9303016.
[17] E. Alvarez, L. Alvarez-Gaume, Y. Lozano, On nonabelian duality, Nucl. Phys. B 424 (1994)
155183, hep-th/9403155.
[18] E. Alvarez, L. Alvarez-Gaume, J.L.F. Barbon, Y. Lozano, Some global aspects of duality in
string theory, Nucl. Phys. B 415 (1994) 71100, hep-th/9309039.
[19] Y. Lozano, Nonabelian duality and canonical transformations, Phys. Lett. B 355 (1995) 165
170, hep-th/9503045.
[20] Y. Lozano, Duality and canonical transformations, Mod. Phys. Lett. A 11 (1996) 28932914,
hep-th/9610024.
[21] O. Alvarez, C.-H. Liu, Target space duality between simple compact lie groups and lie algebras
under the Hamiltonian formalism: 1. Remnants of duality at the classical level, Commun. Math.
Phys. 179 (1996) 185, hep-th/9503226.
[22] C. Klimcik, P. Severa, Dual nonabelian duality and the Drinfeld double, Phys. Lett. B 351 (1995)
455462, hep-th/9502122.
[23] C. Klimcik, P. Severa, PoissonLie T-duality and loop groups of Drinfeld doubles, Phys.
Lett. B 372 (1996) 6571, hep-th/9512040.
[24] C. Klimcik, P. Severa, PoissonLie T-duality: Open strings and d-branes, Phys. Lett. B 376
(1996) 8289, hep-th/9512124.
154
[25] K. Sfetsos, Canonical equivalence of non-isometric sigma-models and PoissonLie T-duality,

Nucl. Phys. B 517 (1998) 549566, hep-th/9710163.
[26] K. Sfetsos, PoissonLie T-duality and supersymmetry, Nucl. Phys. Proc. Suppl. B 56 (1997)
302, hep-th/9611199.
[27] A. Stern, Hamiltonian approach to PoissonLie T-duality, Phys. Lett. B 450 (1999) 141, hepth/9811256.
[28] O. Alvarez, Target space duality I: General theory, hep-th/0003177, to be published in Nucl.
Phys. B (2000).
[29] E.A. Ivanov, Duality in d = 2 sigma models of chiral field with anomaly, Theor. Math. Phys. 71
(1987) 474484.
[30] K. Pohlmeyer, Integrable Hamiltonian systems and interactions through quadratic constraints,
Commun. Math. Phys. 46 (1976) 207.
[31] H. Eichenherr, M. Forger, On the dual symmetry of the nonlinear sigma models, Nucl.
Phys. B 155 (1979) 381.
[32] H. Eichenherr, M. Forger, Higher local conservation laws for nonlinear sigma models on
symmetric spaces, Commun. Math. Phys. 82 (1981) 227.
[33] R.L. Bryant, S.S. Chern, R.B. Gardner, H.L. Goldschmidt, P.A. Griffiths, Exterior Differential
Systems, Springer-Verlag, 1991.
[34] O. Alvarez, Pseudoduality: General theory. In preparation.
[35] S. Helgason, Differential Geometry, Lie Groups, and Symmetric Spaces, Academic Press, 1978.
[36] J.A. Wolf, Spaces of Constant Curvature, 5th edn., Publish or Perish, 1984.
[37] B. ONeill, Semi-Riemannian Geometry, Academic Press, 1983.
[38] J.M. Evans, A.J. Mountain, Commuting charges and symmetric spaces, hep-th/0003264.

Wilson loops in N = 4 supersymmetric

YangMills theory
J.K. Erickson a,1 , G.W. Semenoff b,2 , K. Zarembo a,c,d,3
a Department of Physics and Astronomy, University of British Columbia,
Vancouver, BC, Canada V6T 1Z1

b Institute for Advanced Study, Einstein Drive, Princeton, NJ 08540, USA
c Pacific Institute for Mathematical Sciences, University of British Columbia,
Vancouver, BC, Canada V6T 1Z1

d Institute of Theoretical and Experimental Physics, B. Cheremushkinskaya 25,
Moscow 117218, Russia

Received 29 March 2000; revised 9 May 2000; accepted 15 May 2000
Abstract
Perturbative computations of the expectation value of the Wilson loop in N = 4 supersymmetric
YangMills theory are reported. For the two special cases of a circular loop and a pair of antiparallel
lines, it is shown that the sum of an infinite class of ladder-like planar diagrams, when extrapolated
to strong coupling, produces an expectation
value characteristic of the results of the AdS/CFT
p
correspondence, hW i exp((constant) g 2 N ). For the case of the circular loop, the sump
is obtained
analytically for all values of the coupling. In this case, the constant factor in front of g 2 N also
agrees with the supergravity results. We speculate that the sum of diagrams without internal vertices
is exact for the circular loop and support this conjecture by showing that the leading corrections to the
ladder diagrams cancel identically in four dimensions. We also show that, for arbitrary smooth loops,
the ultraviolet divergences cancel to order g 4 N 2 . 2000 Elsevier Science B.V. All rights reserved.
Keywords: Supersymmetric YangMills theory; Conformal symmetry; AdS/CFT correspondence; Wilson loops
(or lines)
PACS: 11.15.Bt; 12.60.Jv
1. Introduction and summary

The AdS/CFT correspondence conjectures a duality between N = 4 supersymmetric
YangMills theory in four dimensions and type IIA superstring theory on an AdS5 S 5
background [14] (for other reviews, see [57]). There are three levels to this conjecture:
1 erickson@nbi.dk
2 semenoff@ias.edu
3 zarembo@physics.ubc.ca
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 0 0 - X
156
J.K. Erickson et al. / Nuclear Physics B 582 (2000) 155175
In its strongest version the correspondence asserts that there is an exact equivalence
between four-dimensional N = 4 supersymmetric YangMills theory and type IIB
superstring theory on the AdS5 S 5 background. This also contains the conjecture
that the AdS5 S 5 background is an exact solution of type IIB superstring theory.
A weaker version asserts a duality of the t Hooft limit of the gauge theory, where
N with the t Hooft coupling g 2 N held fixed, and the classical gs 0 limit
of type IIA superstring theory on AdS5 S 5 . In this correspondence, corrections to
0
classical supergravity theory from stringy effects which are of order
p would agree
2
with corrections to the large t Hooft coupling limit, of order 1/ g N , but higher
orders in gs on the supergravity side and nonplanar diagrams on the gauge theory side
could disagree.
An even weaker version is a duality between the t Hooft limit where one also takes
the strong coupling limit g 2 N and the low energy, supergravity limit of type
IIB superstring theory on AdS5 S 5 . In this case, there would be order 0 and
p gs
corrections to supergravity which might not agree with order 1/N 2 and 1/ g 2 N
corrections to N = 4 supersymmetric YangMills theory.
Even this last, weakest version of the correspondence has profound consequences. Previous
to it, the only quantitative tool which could be used to attack supersymmetric YangMills
theory was perturbation theory in g 2 , the YangMills coupling constant. This is limited to
the regime where g 2 is small. Furthermore, although some qualitative features of the large
N limit are known, it is not possible to sum planar diagrams explicitly. The AdS/CFT
correspondence enables one to compute correlation functions in the large N , large g 2 N
limit. This limit contains the highly nontrivial sum of all planar Feynman diagrams, and
emphasizes those diagrams which have infinitely many vertices.
Because they are inaccessible to perturbation theory, the predictions of the AdS/CFT
correspondence are very difficult to check in any direct way. The main evidence which
supports the correspondence comes from symmetry arguments. The global symmetries of
both N = 4 supersymmetric YangMills theory and type IIB string theory on AdS5 S 5 are
identical. They have the same global super-conformal group SU(2, 2 | 4) (whose bosonic
subgroup is SO(4, 2) SU(4)). Not only are the global symmetries the same, but some of
those objects which carry the representations of the symmetry group the spectrum of
chiral operators in the field theory and the fields in supergravity theory can, to some
degree, be matched [3]. Furthermore, both theories are conjectured to have an Montonen
Olive SL(2, Z) duality acting on their coupling constants.
The only correlation functions which can be checked directly are correlators which
are protected by non-renormalization theorems and thus do not depend on the coupling
constant. These correlators are related to anomalies in the SU(4) R-symmetry and
conformal current algebras. A number of them have been shown to match [4].
In this paper, we sum infinite classes of planar ladder diagrams in N = 4 supersymmetric
YangMills theory, and compare our results with the predictions of the AdS/CFT
correspondence. We are for the most part interested in the Wilson loop operator. All
computations are in the t Hooft limit where N is put to infinity while holding g 2 N
fixed. We shall not succeed in providing a direct test of the correspondence. However,
157
we will show that the generic dependence of the loop expectation value on the coupling
constant found in the AdS/CFT computations arises naturally when perturbation theory
is summed to all orders and extrapolated to strong coupling. We also compute the exact
weak coupling results to order g 4 N 2 which, if one takes the strongest version of the
AdS/CFT correspondence seriously, should tell us something about classical superstrings
on AdS5 S 5 in the strong curvature limit. We shall also find that leading order corrections
to the sum of ladder diagrams vanish for the two simple loops that we study.
1.1. The Wilson loop
The Wilson loop operator is a phase factor associated with the trajectory of a heavy
quark in the fundamental representation of the gauge group, which in our case is U (N).
The loop operator which couples to classical quantities in the superstring theory in the
simplest way provides a source for a classical string [811]. It is
I

1
(1)
W (C) = Tr P exp d iA (x)x + 8i (x)yi ,
N
C
where x ( ) is a parameterization of the loop, yi = x 2 i and i is a point on the

five-dimensional unit sphere ( 2 = 1). This operator measures the holonomy of a heavy
W-boson whose mass results from spontaneous breaking of U (N + 1) gauge symmetry to
U (N) U (1). In the AdS/CFT correspondence, it is computed by finding the area of the
world-sheet of the classical string in AdS5 S 5 whose boundary is the loop C, which in
turn lies on the boundary of AdS5 [4,8,9].
The expectation value of the Wilson loop operator is easy to compute in perturbation
theory
I

|x(
1 )||x(
2 )| x(
1 ) x(
2)
g2 N
d1 d2
+ .
(2)
W (C) = 1 +
16 2
|x(1 ) x(2 )|2
C
For a loop without cusps or self-intersections, this result is finite. This is because
a cancellation occurs between the contributions of the scalar and vector fields. The case
of cusps and self-intersections has been discussed in [10].
1.2. Summary of results
We work in Feynman gauge and consider specific, not necessarily gauge invariant,
classes of planar Feynman diagrams that do not contain internal vertices. In the following,
we call such graphs ladder diagrams. We will comment on the issue of gauge invariance
later. In all computations, we consider only planar diagrams, thus the coupling constant
dependence of all results is in the combination g 2 N .
1.2.1. Cancellation of ultraviolet divergences
N = 4 supersymmetric YangMills theory is a conformal field theory for any value
of its coupling constant. This conformal invariance is a result of the fact that the theory
158
has no dimensional coupling constants, so it is conformally invariant at the tree level. In

addition, the high degree of supersymmetry leads to a cancellation of loop corrections to
the coupling constant renormalization.
However, as we shall see, the theory does have divergent wave-function renormalization,
which does not contradict ultraviolet finiteness, since all divergences can be absorbed
by rescaling the fields. 4 The Wilson loop potentially contains its own short-distance
singularities which occur when the spacetime arguments of operators in the loop approach
each other. In (2), the divergent contributions of the gauge fields and the scalars mutually
cancel. We shall see in the following that, to order g 4 N 2 , part of such singularities survive
in order to compensate for the infinite wave function renormalization, and all ultraviolet
singularities cancel for a generic smooth (without cusps and self-intersections) loop.
1.2.2. Parallel lines
The Wilson loop operator for a single infinite straight line, W (C) where C is
parameterized by x(s) = (s, 0, 0, 0), is a BPS object: it commutes with half of the
sixteen supercharges [10]. This residual supersymmetry protects it from obtaining quantum
corrections. It is easy to verify that the first few orders in weak coupling perturbation theory
cancel. In the strong coupling limit of planar diagrams, the AdS/CFT correspondence
predicts that

W (C) = 1.
(3)
In fact, this is the case for any array of parallel straight lines with like orientations.
1.2.3. Circular loop
For the circular loop, conformal invariance predicts that the expectation value of the
loop operator is independent of the radius of the loop. 5 We find that the sum of all planar
Feynman diagrams which have no internal vertices (which includes both rainbow and
ladder diagrams) produces the expression

q

2
p
I1 g 2 N ,
(4)
W (C) ladders =
g2 N
where I1 is the Bessel function. Taking the large g 2 N limit gives

e gN
,
W (C) ladders =
(/2)1/2 (g 2 N)3/4
(5)
which has exponential behavior identical to the prediction of the AdS/CFT correspondence
[10,13]:
4 In fact, some divergences are necessary in order for the theory to solve the conformal bootstrap equations
[12].
5 Note the subtle fact that, since the circle could be obtained from an infinite straight line by a conformal
transformation, and this conformal transformation should be a symmetry of the vacuum, the circular loop should
have expectation value equal to that of the infinite straight line (3). We shall see explicitly that this is not the case.
This asymmetry is introduced by the boundary conditions which require that propagators go to zero at infinity.

2
W (C) AdS/CFT = e g N .
159
(6)
It is intriguing that this sum of a special class of diagrams produces the exact asymptotic
behavior that is predicted by the AdS/CFT correspondence, considering that it does not
include any diagrams which have internal vertices. Assuming that the AdS/CFT prediction
is indeed the correct asymptotic behavior, there are two possible reasons for this agreement.
First, the corrections to the sum of planar ladder diagrams couldpproduce a term which
would be added to (4) and which would grow no faster than exp( g 2 N ) for large g 2 N .
Such a term could modify the prefactor of the exponential but would not modify the
exponent. Second, it is possible that corrections to the sum of ladder diagrams cancel and
the result (4) is exact.
In order to explore this appealing possibility, we compute the leading order corrections
to (4) coming from diagrams with internal vertices. These occur at order g 4 N 2 . We find that
these diagrams do indeed cancel exactly when the spacetime dimension is four. Away from
dimension four, there is a residual term of order (D 4)g 4 N 2 . It is tempting to speculate
that all higher-order corrections from diagrams with internal vertices also cancel. At this
point we have not been able to check this possibility beyond order g 4 N 2 . One reason
we have to be optimistic is that the circular loop is related to the single straight line by
a conformal transformation. The conformal transformation must not be representable as
a unitary operator operating on the loop operator. Otherwise, the expectation value of the
circular loop would be one (identical to the straight line) and we know this is not the case.
However, for some classes of diagrams, the symmetry could survive. We know that, for the
straight line, the rainbow diagrams, which are what the ladder diagrams are mapped onto,
cancel identically. This clearly does not happen for their conformal images on the circle.
However it could still happen for the diagrams which have internal vertices which do in
fact cancel identically for the straight line Wilson loop. As stated above, we have checked
explicitly that this is indeed the case for the leading order g 4 N 2 corrections. This has not
yet been checked at higher-orders.
One important issue is gauge invariance. The computation of (4) is in Feynman gauge.
Since only a subset of the diagrams has been included, there is no guarantee that the
calculation is gauge invariant. In fact, summing diagrams without internal vertices in
a different gauge would give a different result. However, if the corrections indeed vanish
then the sum would be equal to the gauge invariant sum of all Feynman diagrams on the
Wilson loop. The cancellation of the diagrams with internal vertices would then be a special
property of the Feynman gauge. 6 We have explicitly demonstrated that corrections vanish
up to order g 4 N 2 .
1.2.4. Antiparallel lines
Consider a rectangular Wilson loop with length T and width L. In the limit that T L,
the ends can be ignored and the loop can be seen as a pair of antiparallel lines separated
6 In Feynman gauge, the vector propagator coincides with the scalar one, which makes, for instance,
cancellation of ultraviolet divergences more transparent than in other gauges: the second term in (2) vanishes
even before integration along the loop.
160
by a distance L. We find that the sum of all planar diagrams without internal vertices, the
ladder diagrams, is given by

2

g4 N 2
T
g N
1
ln 2 +
(7)
, g 2 N 1,
W (C) ladders = exp
3
4
8
g N
L
and by

p 2
1
T
g N
1+O p
,
W (C) ladders = exp
2
L
g N

g 2 N 1.
(8)
Here, we use the fact that T L to ignore any terms in the exponent which are of lower
order than T /L.
The logarithm in the exponent of the weak coupling limit (7) comes from an infrared
divergence which appears in the order g 4 N 2 ladder diagram. It was shown in [14] that
resummation of these logarithms to all orders replaces the logarithm of T /L by a logarithm
of the coupling constant. We will elaborate on those arguments in Section 5. Logarithms
of this kind are known to appear in the Wilson loop in QCD at order g 6 [15]. In the weak
coupling limit, the leading term and the coefficient of the next-to-leading term are indeed
gauge invariant. However, the strong coupling limit depends on the gauge parameter and
we have quoted the result (8)
in Feynman gauge only (while the particular constants vary,
2
the generic behavior is still e g N ).

In the large t Hooft coupling limit, the AdS/CFT computation 7 [8,9] yields:
2p 2

4 g N T
, g 2 N 1.
W (C) AdS/CFT = exp
0 4 (1/4) L
(9)
Comparing this with (8), we see that summing only ladder diagrams does not produce (9).
The failure of this result to agree with the supergravity computation is not very interesting:
since the antiparallel lines are not as closely related to a BPS object as a circle, we have
no reason to speculate that the sum of ladder diagrams would reproduce the exact strongcoupling behavior.
We also find that the diagrams with internal vertices cancel exactly to the leading
order g 4 N 2 . In this case, as opposed to the circular loop where they only cancel in four
dimensions, the cancellation occurs in any number of spacetime dimensions.
The sum over an infinite number
p of diagrams that we have done produces the strong
coupling behavior exp[(constant) g 2 N ] which is characteristic of the Wilson loops
computed using the AdS/CFT correspondence. This seems to be the generic behavior of
infinite sums of planar ladder diagrams.
1.3. Outline
In the rest of this paper we give a detailed derivation of the above results. In Section 2,
we give the results of some loop calculations. In Section 3, we prove that ultraviolet
7 Note that the g 2 in [8] differs by a factor of 2 from our definition.
161
singularities cancel to order g 4 N 2 for any smooth loop. In Section 4, we consider the
circular loop. We compute the sum over ladder diagrams and show that the diagrams with
internal vertices cancel to order g 4 N 2 . In Section 5, we give similar arguments for two
antiparallel lines. In Section 6, we summarize some speculations about the implications of
our results. In Appendix A, we summarize our notation and conventions and record some
useful formulae.
2. Perturbation theory
With the exception of Section 3, we use regularization by dimensional reduction
throughout this paper. This procedure considers supersymmetric YangMills theory in 2
dimensions as a dimensional reduction of N = 1 supersymmetric YangMills theory in
ten dimensions. In this scheme, the gauge field Aa (x) is a 2 component vector field.
The index of the scalar field runs over 10 2 values, i = 1, . . . , 10 2. In every
dimension, the fermion field has sixteen real components. Regularization by dimensional
reduction preserves the sixteen supersymmetries of the ten-dimensional YangMills theory.
These lead to four conserved four-component Majorana spinor supercharges in four
dimensions. This regularization scheme provides a supersymmetric regularization of the
four-dimensional theory. Since the gauge coupling becomes dimensional in any spacetime
dimension other than four, the regularization breaks conformal symmetry explicitly. It also
modifies the R-symmetry.
2.1. One loop self energy of the vector and scalar fields
Consider the one loop self-energies of the vector and scalar field. The dimension of
spacetime is D = 2. All of the loop integrals are elementary and can be found in
reference books such as [16]. We parameterize the Wilson loop by x( ), and abbreviate
x (i) = x( i ).
The vector field obtains self-energy corrections from:
N 2 colors of vector fields and ghost fields:
+
= ab g 4 N
p p /p2
0(2 )0()0( 1)
2(3
1)
;
(4) 0(2)
p22
10 2 real scalar fields in the adjoint representation:

= ab g 4 N
p p /p2
0(2 )0()0( 1)
(10 2)
;
(4) 0(2)
p22
four flavors of four-component Majorana fermions in the adjoint representation:

= ab g 4 N
p p /p2
0(2 )0()0( 1)
16(
1)
.
(4) 0(2)
p22
162
Note that these are the negative of the conventionally defined self-energies. Thus, to one
loop order, the propagator for the unrenormalized gluon, in Feynman gauge is
2 ab
ab
= g
4 0(2 )0()0( 1)
ab p p /p
g
N
.
4(2
1)
p2
(4) 0(2)
p62
(10)
Similarly, we can compute the one loop correction to the scalar propagator. It obtains
corrections from:
the scalar-vector intermediate state:
ij
0(2 )0()0( 1)
4(2 1) 62 ,
= ab g 4 N
(4) 0(2)
p
and the fermion loop:
= ab g 4 N
ij
0(2 )0()0( 1)
8(2 1) 62 .
(4) 0(2)
p
Thus, to one-loop order, the (unrenormalized) scalar propagator is

Dijab = g 2 ab
ij ab
ij
0(2 )0()0( 1)
4(2 1) 62 .
g4 N
2
p
(4) 0(2)
p
(11)
Note that, aside from vector indices, the scalar and vector propagators are identical. Also,
note that the self-energy corrections have poles at = 2 which arise from an ultraviolet
divergence. If we were to compute correlators of local renormalized fields, it would be
necessary to add a counterterm to the action in order to cancel these ultraviolet singularities.
Here, for purpose of computing the Wilson loop, we leave them unrenormalized.
2.2. Ladder diagrams
The ladder-like diagrams to order g 4 N 2 contribute:
I
g4 N 2
d1 d2 d3 d4
1 =
32 4
1 >2 >3 >4
(1) (2)

x x | x (1) x (2) x (3)x (4) x (3) x (4)
.

x (1) x (2) 2 x (3) x (4)2
This result is finite when x( ) is a smooth curve.
2.3. Insertion of one-loop corrections to propagators
Using the one-loop self-energies of the vector and scalar fields that we found in
Section 2.2, we can find the corrections to the expression (2) resulting from insertion of
one-loop into the vector and scalar field propagators.
We begin with the correction to the scalar propagator in momentum space. From Eq. (11)
it is
ab g 4 N
163
1
0(2 )0()0( 1)
4(2 1) 2 3 .
(4) 0(2)
[p ]
The Fourier transform of this expression, computed using (45), is

ab g 4 N
0 2 ( 1)
25 2 (2 )(2 3)
1
.
[x 2 ]23
If we now combine this with the analogous expression for the vector field propagator and
compute the correction to the circular loop the result is
(1) (2)
I
x x x (1) x (2)
0 2 ( 1)
4 2
d1 d2
(12)
2 = g N 7 2
2 23 ,
2 (2 )(2 3)
x (1) x (2)
where a factor of N 2 /2 came from taking the trace over gauge group generators, a factor
of 1/N came from the normalization of the Wilson loop, and an additional factor of 1/2
came from the combinatorics of expanding the Wilson loop operator to second order. We
see that the integrand is identical to (2) with a correction to the coefficient
g2 N
g 4 N 2 0 2 ( 1)
g2 N
.
7
2
2
4
4
128 2 (2 )(2 3)
The coefficient diverges in four dimensions (at = 2).
2.4. Diagrams with one internal vertex
The order g 4 N 2 contributions with one internal vertex come when we Taylor expand
W (C) to third order in A and 8 and Wick contract it with the relevant vertices to obtain
the quantities:

Z
Z

i3
4
abc
a
b
c
i (y) A (y) i (y) ,
d1 d2 d3 Tr P A(1 )A(2 )A(3 ) d y f
3!
i
2!1!
Z
d1 d2 d3
Z

4
abc
a
b
c
A (y) A (y) A (y) ,
Tr P 8(1 )A(2 )8(3 ) d y f
a . The minus signs in both vertices

where A( ) = Aa (x)x ( )T a and 8( ) = 8a (x)|x|T
S
(we work in Euclidean space). The sum of the two
come from the expansion of e
diagrams is
I

g4 N 2
d1 d2 d3 (1 2 3 ) x (1)x (3) x (1) x (3)
3 =
4
x
(2)
(1)
x

d2 w 1 x (1) w 1 x (2) w 1 x (3) w .
(13)
Here, is the antisymmetric path ordering symbol: we define (1 2 3 ) = 1 for 1 >

2 > 3 and let be antisymmetric under any transposition of i . It is straightforward to
introduce Feynman parameters and do the integration over w to obtain:
164
3 = g 4 N 2
0(2 2)
27 2
I
Z1
d d d ( )2 (1 )
d1 d2 d3 (1 2 3 )
(14)
(1) (3)

x x x (1) x (3) (1 )x (2) x (1) x (2) x (3) x (2) x (2)
.
2
2
2 22

x (1) x (2) + x (1) x (3) + x (3) x (2)
3. Cancellation of divergences to order g 4 N 2
A logarithmic divergence arises in the integral (13) from where 1 is coincident with 2 .
This divergence should cancel with the divergence in the coefficient of 2 in (12) so that
the order g 4 N 2 contribution 1 + 2 + 3 is finite. In extracting the divergences from
(13), we must consider the integral
Z

(15)
G(i ) = d4 w 1 w x (1) 1 w x (2) 1 w x (3) ,
in detail. This integral is singular in the limit x (2) x (1) . The divergent contribution
comes from the integration over w close to x (1) and x (2) . We can approximate |w x (3)|
|x (1) x (3)| introducing simultaneously an infrared cutoff , so that the divergent part of
(15) takes the form
G(i ) 1 x
(1)
(3)

d4 w 1 w x (1) 1 w x (2) .
(16)
As follows from dimensional counting, this integral depends only on the dimensionless
ratio |x (1) x (2) |/. The limit where x (1) and x (2) are coincident is then equivalent to the
limit of infinite . Without the infrared cutoff the integral would diverge logarithmically,
so up to terms regular in the limit , we get:
2

Z 4
1
1 logx (1) x (2) 2
d w
1
=
.
(17)
G(i )

64 6 x (1) x (3)2
w4
64 4
x (1) x (3)2
Since G only goes as a logarithm as points approach each other, (13) receives divergent
contributions only from x (1) near x (2) ; the parts of the integral with x (1) or x (2) near x (3)
are finite. This divergence is regularized by cutting off the integral over 1 . Since the overall
divergence is logarithmic, the result is independent of the method of regularization. Writing
= (1) (2) and Taylor expanding x (1) = x (2) + x (2) + , we see that the divergent
part of (13) is
(2) (3)

I
I
x x x (2) x (3) Z
x (2) x (1) x (2)
g4 N 2
d
d
d
sign
3

2
3
128 4
x (2) x (3)2
x (1) x (2)2
(2) (3)
I
I
x x x (2) x (3)
g4 N 2
=
log .
d
d

2
3
64 4
x (2) x (3)2
165
This cancels exactly against (12), for one should replace the pole 1/(2 ) at = 2 by
2 log .
4. The circular loop

In this section, we consider a circular Wilson loop, whose radius we can assume to be
unity. A convenient parameterization of this loop is
x( ) = (cos , sin , 0, 0).
(18)
4.1. Summing the planar ladder graphs

First, we will see how to sum all planar diagrams which have no internal vertices. These
include all ladder and rainbow diagrams. Our strategy is the following: the large N limit
with g 2 N held fixed is given by planar diagrams. It is well known that each planar graph
will contain the same group theoretical factors [17]. We observe that, in fact, each planar
diagram without internal vertices gives an identical contribution to the loop expectation
value. We choose a fixed ordering of the times and compute a particular, convenient
diagram. Then we multiply by the number of diagrams that occur to that order and sum
over all orders.
First, consider the 2nth order term in the Taylor expansion of the loop
1
N
Z2
Z1

d2n Tr iA(1 ) + 8(1 ) iA(t2n ) + 8(t2n ) .
d2
d1
0
Z2n1
Here we have chosen a particular time ordering, which cancels the factor of 1/(2n)! which
would come from the Taylor expansion of the exponential. We are interested in all Wick
contractions which represent planar diagrams. Note that for the circular loop

b

a
g 2 ab x (1)x (2) x (1) x (2)
a
b
iA (1 ) + 8 (1 ) iA (2 ) + 8 (2 ) 0 =

4 2
x (1) x (2)2
g 2 ab
.
(19)
8 2
Thus, the contributions of all (free-field) Wick contractions giving planar diagrams are
identical (if we sum over all ways of choosing scalar or gluon lines). The color factor can
be computed by repeated application of the identity:
=
N
1.
2
Thus, for the sum of ladder-like diagrams with n propagators we obtain
n
g 2 N/4
(# of planar graphs with n internal lines),
(2n)!
T aT a =
where the factor 1/(2n)! accounts for the integral over the (i) .
(20)
166
We now have the task of counting the number of planar graphs with n internal lines. Any
such diagram with n + 1 propagators can be uniquely decomposed as
If we define An+1 as the number of such diagrams planar diagrams without internal
vertices then An+1 satisfies the recursion relation
An+1 =
n
X
Ank Ak ,
k=0
with A0 = 1. If we define a generating function f by

f (z) =
An z n ,
n=0
then f satisfies zf 2 (z) = f (z) 1. So,
1 1 4z X (2n)!
=
zn .
f (z) =
2z
(n + 1)!n!
(21)
n=0
The sign of the square root is chosen by requiring that f be finite at z = 0. Hence
An =
(2n)!
,
(n + 1)!n!
(22)
so the sum of all planar diagrams without vertices on the loop is, from (20) and (22)
n

q
X

g 2 N/4
2
=p
I1 g 2 N .
(23)
W (C) ladders =
(n + 1)!n!
g2 N
n=0
Thus, the large g 2 N behavior is

e gN
.
W (C) ladders
(/2)1/2 (g 2 N)3/4
The supergravity prediction is that

2
W (C) AdS/CFT e g N ,
(24)
(25)
so the ladder diagrams have the same leading behavior as the prediction of the AdS/CFT
correspondence.
It is worth mentioning that the cancellation of coordinate dependence in the Wick
contraction (19) maps the problem of summing the ladder-like diagrams to the zerodimensional theory. In particular, the number of planar graphs with no vertices and n
propagators can be calculated from the infinite N limit of the matrix integral [18]:

Z
1
N
1
1
2n
2n
2
Tr M
[dM] Tr M exp Tr M , where
=
An =
N
Z
N
2
Z
Z=
167

N
[dM] exp Tr M 2 .
2
This can be evaluated by extracting the term of order z2n in the Taylor expansion of the
zero-dimensional Wilson loop [19]:

1
1
Tr
.
(26)
(z) =
N
1 zM
Using the identity
I
dz ez
1
=
(2n)!
2i z2n+1
C
for a positively oriented contour C containing the origin, we can represent the sum of the
ladder diagrams as
I

dz ez
g 2 N/4z ,
(27)
W (C) ladders =
2i z
C
where C must be chosen large enough to encircle all singularities of the integrand. The
function (z) satisfies a zero-dimensional loop equation [19] that follows from the identity

Z
N
1
2
exp Tr M
0 = [dM]
Mij
1 zM ij
2
and large N factorization. In the infinite N limit, this equality reduces to the algebraic
equation for (z):
1
1
(z) + = 0.
z
z
This has solution
1 1 4z2
.
(z) =
2z2
z2 (z)
Substituting z2 z, this is the same as (21). Shrinking the contour of integration in (27)
to the branch cut of (z), we obtain
Z1
W (C) ladders = 4
1
dx g 2 N x p
e
1 x 2,
2
which is just the Bessel function (23).

4.2. Diagrams with vertices
The sum of the diagrams with one internal vertex attaching to three points on the
Wilson loop is given by the expression (15). It is convenient to abbreviate ij = i j .
For a circular loop, |x (i)|2 = 1, |x (i) x (j )|2 = 2(1 cos ij ), x (i) x (j ) = sin ij , and
x (i) x (j ) = cos ij . Thus, from (13)
168
3 = g
0(2 2)
N 2 2+5 2
2
d d d ( )2 (1 )
(28)
Z1
d1 d2 d3
(1 2 3 )(1 cos 13 )((1 ) sin 12 + sin 23 )

.
[(1 cos 12 ) + (1 cos 23 ) + (1 cos 13 )]22
We are going to use integration by parts to rewrite (29) as a sum of a term which will cancel
with the order g 4 N 2 diagrams with internal vertices in (12) and a term which vanishes
when = 2. For compactness, write the denominator
= (1 cos 12 ) + (1 cos 23 ) + (1 cos 13 ).
Consider the identity
I
(1 2 3 )(1 cos 13 )
= 0.
d1 d2 d3
1
23
(29)
Using
(1 2 3 ) = 2(12 ) 2(13 ),
1
(30)
that + + = 1, and the fact that the integrand in vanishes when 1 = 3 , we get

I
sin 13 (1 cos 12 ) + (1 cos 23 ) + (1 cos 13 )
d1 d2 d3
22

(1 cos 13 )( sin 12 + sin 13 )
+ (2 3)
(1 2 3 )
22
I
1
1
.
(31)
= 2 d1 d2
23
[ (1 )]
[1 cos 12 ]24
Now, change variables in the first term on the left hand side so that the cosines in the
denominator all have argument 13 (note that this involves permuting , , and as
well, so the following holds only after inserting (1 ) and integrating these
parameters over the unit cube). Then add part of the second term, so that the remaining
part is proportional to (2 4) to obtain

I
(1 cos 13 )((1 ) sin 12 + sin 23 )
d1 d2 d3 (1 2 3 )
22

(1 cos 13 )( sin 12 + sin 13 )
+ (2 4)
22
I
1
1
.
= 2 d1 d2
[ (1 )]23 [1 cos 12 ]24
The first term on the left hand side is precisely the term occurring in (29). Note that the
second term can be rewritten as
I
1
2 4
d1 d2 d3 (1 2 3 )(1 cos 13 )
23
2 3
1
2 4
2 3
169
I
sin 13
d1 d2 d3 (1 2 3 ) 23

I
1
1
+ 2 d1 d2
,
[ (1 )]23 [1 cos 12 ]24
using integration by parts. Finally, we have
I
(1 cos 13 )((1 ) sin 12 + sin 23 )
d1 d2 d3 (1 2 3 )
22
I
2 4
sin 13
=
d1 d2 d3 (1 2 3 ) 23
2 3
I
2
1
1
+
.
d1 d2
23
2 3
[ (1 )]
[1 cos 12 ]24
If we symmetrize, the integral in the first term on the right hand side may be rewritten
(setting = 2) as
I
sin 12 + sin 23 + sin 31
1 2 4
.
d1 d2 d3 (1 2 3 )
3 2 3
(1 cos 12 ) + (1 cos 23 ) + (1 cos 13 )
This integration is completely finite. Since this term appears with a coefficient that vanishes
when = 2, it vanishes in four dimensions. Inserting this expression into (29), we have
the result:
I
0 2 ( 1)
1
4 2
d1 d2
+ O(2 4). (32)
3 = g N 2+4 2
2
(2 3)(2 )
[1 cos 12 ]24
If we specialize (12) to the case of a circular loop, we obtain
I
0 2 ( 1)
1
.
(33)
d1 d2
2 = g 4 N 2 2+4 2
2
(2 )(2 3)
[1 cos 12 ]23
The two contributions (32) and (33) cancel exactly when 2 = 4:
=
2 + 3 = 0.
5. Antiparallel lines
5.1. Summing ladder diagrams
It is possible to sum the planar ladder diagrams for antiparallel lines separated by
a distance L [14]. The calculation can be done by considering the more general case of
a sum 0(S, T ) of ladder diagrams for two finite antiparallel lines of lengths S and T (we
take S, T L). The result we are interested in arises in studying the large S = T behavior.
The sum of ladder diagrams satisfies the recursion relation
ZS
0(S, T ) = 1 +
ZT
dt 0(s, t)
ds
0
g2 N
,
4 2 (s t)2 + L2
(34)
where the second factor in the integrand is the sum of vector and scalar propagators.
0(S, T ) satisfies the boundary conditions
170
0(S, 0) = 0(0, T ) = 1.
(35)
Taking derivatives of (34) we obtain the differential equation

g2 N
2 0(S, T )

0(S, T ).
=
S T
4 2 (S T )2 + L2
(36)
This equation is separable in the variables x = (S T )/L and y = (S + T )/L. (However,

as we shall see in the following, the boundary conditions are not conveniently expressed in
terms of these variables.) The separated ansatz is
X
cn n (x) exp(n y/2),
(37)
0[x, y] =
n
where cn are constants which must be determined so that the boundary condition (35) is
satisfied and n (x) are solutions of the Schrdinger equation

g2 N
2n
d2
n (x).
(x)
=
(38)
2
n
dx
4 2 (x 2 + 1)
4
The largest n with nonvanishing contribution in (37) dominates the large y for fixed x
(large S and T ) asymptotics of 0. In order to see that the lowest bound state of the
Schrdinger operator in (38) contributes, we consider the Laplace transform of 0[x, y].
Accounting for the boundary condition (35) which implies that 0[x, y] vanishes when
y = |x|, we see that the Laplace transform is the resolvent of the Schrdinger equation,
Z
dy epy 0(x, y) = 2
X n (0)n (x)
n
|x|
p2 2n /4
Since the potential is symmetric and the ground state has no nodes (0 (0) 6= 0), the ground
state eigenvalue gives the largest n contributing to (37).
It is possible to find the ground state eigenvalues of (38) for small and large g 2 N . We
find

T
g4 N 2
1
, for g 2 N 1 and
+
ln
(39)
ln 0(T , T ) = g 2 N/4
3
2
L
8
g N
ln 0(T , T ) =
q
q
T
,
g 2 N / 1 + O 1/ g 2 N
L
for g 2 N 1.
(40)
5.2. Diagrams with vertices

The sum of the diagrams with one internal vertex and three lines going to the Wilson
loop is given by (15). For antiparallel lines, 1 and 3 must be on opposite lines so that the
factor |x (1)||x (3)| x (1) x (3) is nonzero. In that case, it provides a factor of 2. Furthermore,
we will find a nonvanishing result only when 2 is on the same line as 1 . There are
two possible configurations like this, which provides a further factor of 2. Taking the
parameterization of the lines to be
x (1) = (1 , L/2, 0, 0),
x (2) = (2 , L/2, 0, 0),
171
x (3) = (3 , L/2, 0, 0)
we obtain
0(2 2)
25 2
3 = g 4 N 2
Z1
d d d ( )2 (1 )
d1 d2 d3 (1 2 3 )
(1 2
)2
[(1 2 ) + (1 + 3 )]
2
2 22 ,
+ (2 + 3 )2 + L2 + (3 + 1 )2 + L2
which can be written as

3 = g 4 N 2
0(2 3)
26 2
Z
Z1
d d d ( )
(1 )
d1 d2 d3 (1 2 3 )
.

1 (1 2 )2 + (2 + 3 )2 + L2 + (3 + 1 )2 + L2 23
Then, using (30) we can integrate over 1 to obtain

3 = g
0(2 3)
N2
25 2
Z1
d d d
0
d2 d3
( )2 (1 )
[ (1 )]23
1
(3 + 2
)2
+ L2
23 ,
which yields
0 2 ( 1)
3 = g N
32 2 (2 3)(2 )
4
d2 d3
1
(3 + 2 )2 + L2
23 .
This result is to be added to the diagrams which come from exchange of a vector and scalar
field, each with internal loop corrections. The sum of those two diagrams is
Z
0 2 ( 1)
1
4 2
d2 d3
(41)
2 = g N
23 ,
2
2
32 (2 3)(2 )
(3 + 2 ) + L2
and, in the end,
2 + 3 = 0.
We see that the quantum corrections to the Wilson loop cancel identically in all spacetime
dimensions less than ten.
172
6. Conclusions
We have calculated contribution of planar diagrams without internal vertices to Wilson
loops in N = 4 supersymmetric YangMills theory. For the two particular types of loops
we have considered, circular and rectangular, these diagrams exponentiate. This fact is
rather unexpected, given that the combinatorics of planar diagrams without vertices is
different from that of diagrams in the free U (1) theory [2022], which exponentiate for
obvious reasons. Because the ladder-like diagrams exponentiate, the static potential is well
defined in the ladder approximation and is easy to extract from the expectation values of
the rectangular Wilson loop, (39) and (40) [14]. The dependence of the potential on the
coupling constant, when extrapolated to the strong coupling regime, is very similar to, but
not exactly the same as the one predicted by the AdS/CFT correspondence. We have no
explanation for this similarity, but the result for the circular loop indicates that this is not
a mere coincidence.
The planar diagrams without internal vertices for the circular Wilson loop sum up
into an expression for which the strong coupling limit coincides with the prediction of
the AdS/CFT correspondence. We have conjectured that diagrams with internal vertices
cancel for this Wilson loop in N = 4 supersymmetric YangMills theory. This conjecture
is supported both by a direct calculation at order g 4 N 2 and by agreement with the strong
coupling results of the AdS/CFT correspondence. It seems that conformal symmetry plays
a key role here, as the conformal symmetry of the theory would navely suggest that all
higher orders of perturbation theory should cancel. While this is not the case, for reasons
discussed above, it may be true that the diagrams with internal vertices are constrained
by conformal symmetry to cancel amongst themselves. This is supported by the fact that
the remarkable cancellation occurs only in four dimensions, the-dimensionality in which
conformal symmetry is present.
Acknowledgements
This work is supported in part by NSERC of Canada. G.W.S. is supported by the the
Ambrose Monell Foundation. The work of K.Z. is supported by PIMS. G.W.S. thanks
Steve Adler, Ken Intriligator, Sasha Polyakov and Frank Wilczek for helpful discussions.
J.K.E. and K.Z. thank Yuri Makeenko for helpful discussions.
Appendix A. Useful formulae and notation

The Euclidean space action of four-dimensional N = 4 supersymmetric YangMills
theory is

Z
2

1 1 a 2
4
F + ia + f abc Ab ic + a i a + f abc Ab c
S= d x 2
2g 2
+ if abc a 0 i ib c
173
f abc f ade ib jc id je
i<j

2
+ ca ca + f abc Ab cc + Aa
,
(42)
where
a
= Aa Aa + f abc Ab Ac
F
and f abc are the structure constants of the U (N) Lie algebra,
a b
T , T = if abc T c .
The generators are normalized as
Tr T a T b = 12 ab ,
and obey the identity
X
f acd f bcd = N ab .
c,d
Now, a is a sixteen component spinor obeying the Majorana condition

(x) = C (x),
(43)
where C is the charge conjugation matrix. 0 A = ( , 0 i ), for = 1, . . . , 4 and

i = 5, . . . , 10 are ten real 16 16 Dirac matrices (in the 10-dimensional Majorana
representation with the Weyl constraint) obeying

Tr 0 A 0 B = 16 AB .
We have chosen the covariant gauge fixing condition,
Aa = 0,
(44)
and we work in Feynman gauge, where the gauge parameter is chosen as = 1. The
appropriate action for ghost fields, ca (x) has been included.
In Feynman gauge, the vector field propagator is
1ab
(p) =
= g 2 ab
,
p2
= g 2 ab
ij
,
p2
the scalar propagator is

Dijab (p) =
the fermion propagator is

S ab (p) =
= g 2 ab
p
,
p2
and the ghost propagator is

C ab (p) = = g 2 ab
1
.
p2
174
The vertices can be easily deduced from the non-quadratic terms in the action (42). Each
vertex carries a factor of 1/g 2 and each propagator carries a factor of g 2 .
We use the position-space propagators in 2-dimensions. These can be deduced from
the Fourier transform
Z
1
0( s)
d2 p eipx
= s
.
(45)
2
2
s
2
4 0(s) [x ]s
(2) [p ]
By setting s = 1 we find the Green function in 2 dimensions:
1(x) =
1
0( 1)
4 [x 2 ]1
which satisfies
2 1(x) = 2 (x).
(46)
We record some formulae which are useful in reproducing the computations in this
paper. Eulers B function is defined as
Z1
dx x 1 (1 x)1 =
B(, ) =
0
0()0()
0( + )
and the 0 function is

Z
0(n + 1) =
dt t n et ,
which satisfies 0(n + 1) = n!, 0(1/2) = , and the combinatorial formulae

0 (2n + 1)/2 = (n 1/2)(n 3/2) . . . (1/2) ,

0(n)0(1/2) = 2n1 0(n/2)0 (n + 1)/2 .
One loop integrals in 2 dimensions can be computed using the dimensional regularization
formulae [16]:
Z
s
s
0(s ) 2
m p2
=
,
d2 k k 2 + 2p k + m2
(s)
Z
s
s
0(s ) 2
m p2
= p
,
d2 k k k 2 + 2p k + m2
(s)
Z
d2 k k k k 2 + 2p k + m2
=
s
s

1
m2 p 2
p p 0(s ) 12 g 0(s 1) p2 + m2 ,
0(s)
and the Feynman parameter formula,

Y
i
i
An
i
Z1
P
xi
ni
n1 1
nk 1 1
P .
dx1 dxk x1
xk
P

ni
i 0(ni )
Ai xi
0
=Q
175
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.

S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, hep-th/9905111, and references
therein.
E.T. Akhmedov, hep-th/9911095.
P. DiVecchia, Lectures given at ICTP Trieste Spring Workshop on Superstrings and Related
Matters, Trieste, Italy, 2230 March 1999, hep-th/9908148.
J.L. Petersen, Int. J. Mod. Phys. A 14 (1999) 35973672, hep-th/9902131.
J. Maldacena, Phys. Rev. Lett. 80 (1998) 4859, hep-th/9803002.
S. Rey, J. Yee, hep-th/9803001.
N. Drukker, D.J. Gross, H. Ooguri, Phys. Rev. D 60 (1999) 125006, hep-th/9904191.
J. Sonnenschein, hep-th/0003032.
W. Chen, Y. Makeenko, G.W. Semenoff, Ann. Phys. 228 (1993) 341364, hep-th/9301069.
D. Berenstein, R. Corrado, W. Fischler, J. Maldacena, Phys. Rev. D 59 (1999) 105023, hepth/9809188.
J.K. Erickson, G.W. Semenoff, R.J. Szabo, K. Zarembo, hep-th/9911088, to appear in Phys.
Rev. D.
T. Appelquist, M. Dine, I.J. Muzinich, Phys. Rev. D 17 (1978) 2074.
P. Ramond, Field Theory: A Modern Primer, Addison-Wesley, 1989.
G. t Hooft, Commun. Math. Phys. 86 (1982) 449.
E. Brezin, C. Itzykson, G. Parisi, J.B. Zuber, Commun. Math. Phys. 59 (1978) 35.
Y. Makeenko, Mod. Phys. Lett. A 6 (1991) 1901, and references therein.
P. Cvitanovic, Phys. Lett. B 99 (1981) 49.
P. Cvitanovic, P.G. Lauwers, P.N. Scharbach, Nucl. Phys. B 203 (1982) 385.
R. Gopakumar, D.J. Gross, Nucl. Phys. B 451 (1995) 379, hep-th/9411021.

Classification of 5-dimensional spacetime with

parallel 3-branes
Tianjun Li
Department of Physics, University of Wisconsin, Madison, WI 53706, USA
Received 22 December 1999; revised 17 April 2000; accepted 22 May 2000
Abstract
If the fifth dimension is a one-dimensional connected manifold, up to diffeomorphism, the only
possible spacetimes will be M 4 R 1 , M 4 R 1 /Z2 , M 4 S 1 and M 4 S 1 /Z2 . And there exist
two possibilities for the cosmological constant along the fifth dimension: the cosmological constant
is constant, and the cosmological constant is sectionally constant. We construct general models with
parallel 3-branes and constant/sectionally constant cosmological constant along the fifth dimension
on those kinds of the spacetimes. Moreover, we point out that for a compact fifth dimension, the
sum of the brane tensions is zero, and for a non-compact fifth dimension, the sum of the brane
tensions is positive. Furthermore, we assume the observable brane which includes our world should
have positive tension, and obtain that the gauge hierarchy problem can be solved in those scenarios.
We also discuss some simple models. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Mj; 04.65.+e; 11.30.Pb; 12.60.Jv
Keywords: AdS5 ; Compactification; Brane; Scale; Hierarchy
1. Introduction
Experiments at LEP and Tevatron have given the strong support to the Standard Model
SU(3)c SU(2)L U (1). However, the Standard Model has some unattractive features
which may imply new physics. One of these problems is that the gauge forces and
the gravitational force are not unified. Another is the gauge hierarchy problem between
the weak scale and the 4-dimensional Planck scale. Previously, two solutions to the
gauge hierarchy problem have been proposed: one is the idea of the technicolor and
compositeness which lacks calculability, and the other is the idea of supersymmetry.
More than one year ago, it was suggested that the large compactified extra dimensions
may also be the solution to the gauge hierarchy problem [1,2]. Furthermore, about half
year ago, Randall and Sundrum [3] proposed another scenario that the extra dimension is
E-mail: li@pheno.physics.wisc.edu; phone: (608) 262-9820; fax: (608) 262-8628
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 1 6 - 3
T. Li / Nuclear Physics B 582 (2000) 176202
177
an orbifold, and the size of the extra dimension is not large but the 4-dimensional mass
scale in the standard model is suppressed by an exponential factor from the 5-dimensional
mass scale. In addition, they suggested that the fifth dimension might be infinite [4] and
that there may exist only one brane with positive tension at the origin, but, the gauge
hierarchy problem persists. The remarkable aspect of the second scenario is that it gives
rise to a localized graviton field. Combining those results, Lykken and Randall obtained
the following physical picture [5]: the graviton is localized on Planck brane, we live on a
brane separated from the Planck brane about 30 Planck lengths along the fifth dimension.
On our brane, the mass scale in the Standard Model is suppressed exponentially, which
gives the low energy scale. We generalized those scenarios and obtained the scenario with
the following property [19]: the 4-dimensional Planck scale is generated from the low
5-dimensional Planck scale by an exponential hierarchy, and the mass scale in the Standard
model, which is contained in the observable brane, is not rescaled. In short, this kind of
compactification or similar idea has attracted a lot of attention recently [633]. By the way,
supergravity domain walls were discussed previously in 4-dimensional spacetime [34].
In the model building, there exist several approaches: (I) The 3-branes are parallel along
the fifth dimension and the observable brane is one of those branes: Oda constructed
a model on the spacetime M 4 S 1 where M 4 is the four-dimensional Minkowski
space, and cosmological constant is constant [13]. However, his solution is not general.
H. Hatanaka, M. Sakamoto, M. Tachibana, and K. Takenaga constructed a model on the
spacetime M 4 S 1 in which the cosmological constant is sectionally constant along
the fifth dimension [31]. We also constructed a model with only parallel positive tension
3-branes on M 4 R 1 and M 4 R 1 /Z2 [20]. (II) The observable sector or our world lives
in a brane intersection/junction [7,28,29]. (III) Many brane intersections/junctions where
our world lives at one intersection/junction, which might be thought to be the combination
of the approach (I) and (II) [21,32]. In this paper, we classify the models of the first
approach, or in other words, we classify the models on the 5-dimensional spacetimes with
parallel 3-branes. From differential topology/manifold, up to diffeomorphism, there is only
one connected non-compact one-dimensional manifold: R 1 , there is only one connected
non-compact one-dimensional manifold with boundary: H1 or R 1 /Z2 (the equivalence
class is y y), there is only one connected compact one-dimensional manifold S 1 , and
there is only one connected compact one-dimensional manifold with boundary S 1 /Z2
or I . Therefore, the spacetimes we consider are M 4 R 1 , M 4 R 1 /Z2 , M 4 S 1 ,
M 4 S 1 /Z2 . Furthermore, there are two possibilities for the cosmological constant along
the fifth dimension: the cosmological constant is constant, and the cosmological constant
is sectionally constant.
Before our discussion, we would like to explain our assumption and notation which are
similar to those in [19,20]. We assume that all the gauge forces are unified on each 3-brane
(i)
if gauge forces exist. The 5-dimensional GUT scale on the ith 3-brane M5GUT and the
5-dimensional Planck scale MX are defined as the GUT scale and Planck scale in the
5-dimensional fundamental metric, respectively. The 4-dimensional GUT scale on the
(i)
and the 4-dimensional Planck scale Mpl are defined as the GUT
ith 3-brane MGUT
scale and Planck scale in the 4-dimensional Minkowski Metric ( ). In order to avoid
178
the gauge hierarchy problem between the weak scale and the 4-dimensional GUT scale
MGUT on the observable brane which includes our world, we assume low energy unification. 1 The key ansatz is that the 5-dimensional GUT scale on each brane is equal to the
5-dimensional Planck scale. In addition, as we know, an object with negative tension cannot be stable. However, we do not need to worry about that if the anti-brane is orientable,
so, we consider models with both positive and negative tension 3-branes. The brane with
negative brane tension may be an anti-gravity world [17], so, we assume that the observable brane which includes our world has positive tension. By the way, there are positive
energy objects, namely D-branes and NS-branes, that are well understood and on which
the gauge fields and matter fields can be localized so that the Standard Model fields can be
placed there.
In this paper, we construct the general models with parallel 3-branes and constant/sectionally constant cosmological constant along the fifth dimension on spacetimes
M 4 R 1 , M 4 R 1 /Z2 , M 4 S 1 , and M 4 S 1 /Z2 . In general, for a compact fifth dimension, the sum of the brane tensions is zero, and for a non-compact fifth dimension, the
sum of the brane tensions is positive if one requires that the four-dimensional Planck scale
is finite. In addition, if the cosmological constant is constant along the fifth dimension, we
can define the index of the model: the sum of the brane tensions divided by 6kMX3 , where k
is defined in the following section. If the spacetime is M 4 R 1 , the index is +1, if space
time is M 4 R 1 /Z2 , the index is + 12 , and if the fifth dimension is compact, the index is 0.
Moreover, for any point in M 4 R 1 , M 4 R 1 /Z2 , M 4 S 1 , and M 4 S 1 /Z2 , which does
not belong to any brane or the section where the cosmological constant is zero, there is a
neighborhood which is diffeomorphic to (or a slice of) AdS5 space. It seems to us the gauge
hierarchy problem can be solved easily in these scenarios. Assume we have l + m + 1 brane
with position yl < yl+1 < < ym1 < ym . (y), which is defined in the following sections, will have minimal value at the brane with positive tension. Without loss of generality,
2 /M 3 will
we assume (yj ) is minimal. If (yi ) (yj ) is not small for all i 6= j , then, Mpl
X
be proportional to e2 (yj ) . And if the observable brane is at y = yio , the 4-dimensional
GUT scale in our world will be proportional to Mpl e( (yio ) (yj )) under some conditions. Therefore, if (yio ) (yj ) = 34.5 or 30, we will push the four-dimensional GUT
scale in our world to the TeV or 105 GeV range, respectively. The 4-dimensional GUT
scale on the j th brane is the highest, but, that does not mean the j th brane has larger
brane tension. Furthermore, if (yj ) > 0 and the j th brane is the observable brane, we can
also solve the gauge hierarchy problem, although the 5-dimensional Planck scale will be
very large. In short, the 5-dimensional GUT scale on each brane can be identified as the
5-dimensional Planck scale, but, the 4-dimensional Planck scale is generated from the low
4-dimensional GUT scale exponentially in our world.
We also give some simple models to show how to solve the gauge hierarchy problem.
1 We will not explain why M
GUT can be at low energy scale here, but, it is possible if one considers additional
particles which change the RGE running. Of course, proton decay might be the problem, but we do not discuss it
here. Moreover, using extra dimensions, low-scale gauge coupling unification can occur due to power-law running
induced by the presence of KaluzaKlein states [22,35], and proton-decay constraints might be satisfied in this
framework.
179
This paper is organized in the following way. In Section 2, we discuss the general
models with constant cosmological constant along the fifth dimension on the spacetimes
M 4 R 1 , M 4 R 1 /Z2 , M 4 S 1 and M 4 S 1 /Z2 . In Section 3, we discuss the general
models with sectionally constant cosmological constant along the fifth dimension on the
spacetimes M 4 R 1 , M 4 R 1 /Z2 , M 4 S 1 and M 4 S 1 /Z2 . The conclusion is given
in Section 4.
2. General models with constant cosmological constant along the fifth dimension
In this section, we consider the cosmological constant to be constant along the fifth
dimension. The models on the spacetimes M 4 R 1 , M 4 R 1 /Z2 , M 4 S 1 and M 4
S 1 /Z2 will be discussed in the Subsections 2.1, 2.2, 2.3, 2.4, respectively. Generally
speaking, for any point in M 4 R 1 , M 4 R 1 /Z2 , M 4 S 1 and M 4 S 1 /Z2 , which
does not belong to any brane, there is a neighborhood which is diffeomorphic to (or a slice
of) AdS5 space. We assume k, which is defined in the following subsections, to be greater
than 0.
2.1. Models on the spacetime M 4 R1
In this subsection, we consider that the fifth dimension is R 1 . Assume we have m
parallel 3-branes, and their fifth coordinates are: < y1 < < ym1 < ym < +. The
5-dimensional metric in these branes are:

(i)
(x ) G x , y = yi ,
(1)
g
where GAB is the five-dimensional metric, i = 1, . . . , m and A, B = , y. 2
The classical Lagrangian is given by:
S = Sgravity + SB ,
Z

Sgravity = d4 x dy G + 12 MX3 R ,
SB =
m Z
X
d4 x
g (i) {Li Vi },
(2)
(3)
(4)
i=1
where MX is the 5-dimensional Planck scale, is the cosmological constant, and Vi where
i = 1, . . . , m is the ith brane tension.
The 5-dimensional Einstein equation for the above action is:

G RAB 12 GAB R
#
"
m
q
X
(i)
Vi g (i) g M N (y yi ) .
(5)
= 3 G GAB +
MX
i=1
2 We assume that G
5 = 0 here. In addition, if the fifth dimension has Z2 symmetry, i.e., the Lagrangian is
invariant under the transformation y y, then, G5 = 0.
180
Assuming that there exists a solution that respects 4-dimensional Poincar invariance in
the x -directions, one obtains the 5-dimensional metric:
ds 2 = e2 (y) dx dx + dy 2.
(6)
With this metric, the Einstein equation reduces to:

02 =
00 =
,
6MX3
m
X
Vi
3MX3
i=1
(7)
(y yi ).
(8)
Now, we consider the general solutions. For m odd, i.e., m = 2n + 1 for n > 0, we obtain
two solutions, A and B:
(y)A =
2n+1
X
(1)i+1 k|y yi | + c,
(9)
i=1
(y)B =
2n+1
X
(1)i k|y yi | + c.
(10)
i=1
For m even, i.e., m = 2n for n > 1, we also obtain two solutions, C and D:
(y)C =
2n
X
(1)i+1 k|y yi | ky + c,
(11)
i=1
(y)D =
2n
X
(1)i k|y yi | + ky + c.
(12)
i=1
And the cosmological constant is:

= 6k 2MX3 .
(13)
For solutions A and C, the brane tensions are:

Vi = (1)i+1 6kMX3 ,
(14)
and for solutions B and D, the brane tensions are:

Vi = (1)i 6kMX3 ,
(15)
where i = 1, . . . , m in Eqs. (14) and (15).

For the models with constant cosmological constant along the fifth dimension, we can
define an index as the sum of brane tensions divided by 6kMX3 . Therefore, solution A has
index +1, solution B has index 1, the solution C and D have index 0. If one requires
that the 4-dimensional Planck scale is finite, only solution A is reasonable. Therefore, we
obtain the five-dimensional metric:
ds 2 = e2
P2n+1
i=1
(1)i+1 k|yyi |2c
dx dx + dy 2 .
(16)
181
The corresponding four-dimensional Planck scale is:

!
2n+1
1 3 X
2
i+1 2 (yi )
(1) e
.
Mpl = MX
k
(17)
i=1
(i)
In addition, the four-dimensional GUT scale on the ith brane MGUT is related to the
five-dimensional GUT scale on the ith brane M5(i)
GUT :
MGUT = M5GUT e (yi ) .
(i)
(i)
(18)
(i)
Our general assumption is that M5GUT MX , for i = 1, . . . , 2n + 1.

These models can be generalized to the models with Z2 symmetry: one just requires that
yn+1 = 0 and yi = y2n+2i for 1 6 i 6 n.
Now, we give two simple models.
(I) Three brane model: the branes fifth coordinates are: y1 , y2 , y3 , and the brane tensions
divided by 6MX3 are: k, k, k, respectively. Therefore, we obtain:
(y) =
3
X
(1)i+1 k|y yi | + c,
(19)
i=1
(y1 ) = k(y3 y2 ) + c,
(y2 ) = k(y3 y1 ) + c,
(y3 ) = k(y2 y1 ) + c.
(20)
(21)
Without loss of generality, we assume y2 y1 < y3 y2 . The four-dimensional Planck

scale is

MX3 2 (y3 ) 2( (y1) (y3 ))
e
e2( (y2) (y3 )) + 1 .
(22)
e
k
If the brane at y = y1 is the observable brane which includes our world, we can solve the
gauge hierarchy problem. Assuming that e2( (y1) (y3 )) 1, and MX = k, we obtain:
2
=
Mpl
MGUT = Mpl ek(y3 +y1 2y2 ) .

(1)
(23)
So, we can push the GUT scale in our world to TeV scale or 105 GeV scale range if
k(y3 + y1 2y2) = 34.5 or 30, respectively. And the value of (y3 ) = k(y2 y1 ) + c
determines the relation between Mpl and MX .
If the brane at y = y3 is the observable brane which includes our world, we can solve
the gauge hierarchy problem, too. Assuming that Mpl = k and e2( (y1) (y3 )) 1, we
obtain:
1
MGUT = Mpl e 3 k(y2 y1 ) 3 c ,

(3)
(24)
with k(y2 y1 ) + c = 103.5 or 90, we can have the GUT scale in our world at TeV scale
or 105 GeV scale, respectively. Obviously, in this case, because of the exponential factor,
the five-dimensional Planck scale is very high, 1048 GeV or 1044 GeV, respectively.
(II) Five brane model with Z2 symmetry: the branes fifth coordinates are: y1 = y5 ,
y2 = y4 , y3 = 0, y4 , y5 , and the brane tensions divided by 6MX3 are: k, k, k, k, k,
respectively. Therefore, we obtain:
182
(y) = k|y| +
5
X
(1)i+1 k(|y yi | + |y + yi |) + c,
(25)
i=4
(y1 ) = (y5 ) = ky5 + c,
(26)
(y2 ) = (y4 ) = 2ky5 ky4 + c,
(27)
(y3 ) = 2k(y5 y4 ) + c.
(28)
Without loss of generality, we assume that, y5 > 2y4 . The four-dimensional Planck scale
is:

MX3 2ky5 2c
e
2 2 e2k(y5y4 ) + e2k(y52y4 ) .
(29)
k
If the brane at y = y3 is the observable brane, the gauge hierarchy problem can be solved.
Assuming that e2k(y52y4 ) 1, and MX = k, we obtain:
2
=
Mpl
(3)
= Mpl ek(y5 2y4 ) .
MGUT
(30)
So, the GUT scale in our world will be pushed to TeV scale or 105 GeV scale range if
k(y5 2y4 ) = 34.5 or 30, respectively. And the value of ky5 + c determines the relation
between Mpl and MX .
If the brane at y = y1 is the observable brane, one can also solve the gauge hierarchy
problem. Assuming that Mpl = k and e2k(y5 2y4 ) 1, one obtains:
1
(1)
= Mpl e 3 ky5 3 c ,
MGUT
(31)
with ky5 + c = 103.5 or 90, one obtains that the GUT scale in our world is at TeV scale or
105 GeV scale, respectively. The five-dimensional Planck scale is 1048 GeV or 1044 GeV,
respectively.
2.2. Models on the spacetime M 4 R1 /Z2
Now, we consider the models on the spacetime M 4 R 1 /Z2 . The models with Z2
symmetry in the Subsection 2.1 can be generalized to the models in which the fifth
dimension is R 1 /Z2 . For the models with Z2 symmetry in the Subsection 2.1, we introduce
the equivalence classes: y y and ith brane (2n + 2 i)th brane, and modulo the
equivalence classes, then, we obtain the models on M 4 R 1 /Z2 . The tricky point is the
tension of the brane at the origin which is the fixed point. We would like to explain the
technical detail in this way: we split the brane at origin, which is the (n + 1)th brane with
tension (1)n 6kMX3 , into two branes with equal tension (1)n 3kMX3 , and with the fifth
coordinates: y = and y = +, respectively. So, d (y)/dy is zero at y = 0. In the limit,
0, we obtain the original case. After we modulo the equivalence classes, the brane at
origin becomes the boundary brane, and its brane tension Vn+1 becomes half of the original
value, i.e., Vn+1 = (1)n 3kMX3 . In general, using this method, we obtain d (y)/dy is zero
at boundary. The index of this kind of the model is 12 .
For n odd, we obtain:
(y) =
2n+1
X
i=n+2
(1)i+1 k|y yi | + c0 ,
(32)
183
and for n even, we obtain:

2n+1
X
(y) = k|y| +
(1)i+1 k|y yi | + c0 ,
(33)
i=n+2
where
c0 = c +
n
X
(1)i+1 ky2n+2i .
(34)
i=1
Because c0 is just a constant, we will write it as c in the following part. Similar conventions
will be used in the following subsections.
For simplicity, we renumber the above subscript i where i = n + 1, . . . , 2n + 1 as
i (n + 1). In short, we have a model with n + 1 parallel 3-branes whose fifth coordinates
are y0 = 0 < y1 < < yn1 < yn < +. And (y) can be written as:
X

1
1 + (1)n k|y| +
(1)i+n k|y yi | + c.
2
n
(y) =
(35)
i=1
One can easily obtain the cosmological constant

= 6k 2MX3 ,
(36)
and the brane tensions are:

V0 = (1)n 3kMX3 ,
Vi = (1)i+n 6kMX3 .
(37)
The corresponding 4-dimensional Planck scale is:

2
Mpl
!
n
1 3 (1)n 2 (y0 ) X
i+n 2 (yi )
e
= MX
+
(1) e
.
k
2
(38)
i=1
(i)
is related to the
In addition, the four-dimensional GUT scale on the ith brane MGUT
(i)
five-dimensional GUT scale on the ith brane M5GUT :
(i)
(yi )
= M5(i)
MX e (yi ) .
MGUT
GUT e
(39)
Now, we discuss an example. Because the model with three branes is equivalent to the
model with five 3-branes and Z2 symmetry in Subsection 2.1, we do not discuss it here.
We discuss the model with four 3-branes: the branes fifth coordinates are: y0 , y1 , y2 , y3 ,
and the brane tensions divided by 6MX3 are: 12 k, k, k, k, respectively. Therefore, we
obtain:
(y) =
3
X
(1)i+1 k|y yi | + c,
(40)
i=1
(y0 ) = k(y3 + y1 y2 ) + c,
(y2 ) = k(y3 y1 ) + c,
(y1 ) = k(y3 y2 ) + c,
(y3 ) = k(y2 y1 ) + c.
(41)
(42)
184
Without loss of generality, we assume y2 y1 > y3 y2 . The four-dimensional Planck

scale is

MX3 2k(y3 y2 )2c 1 2ky1
e
+ 1 e2k(y2 y1 ) + e2k(2y2y1 y3 ) .
2 e
(43)
k
If the brane at y = y3 is the observable brane, we can solve the gauge hierarchy problem.
Assuming that e2k(2y2y1 y3 ) 1 and MX = k, we obtain:
2
=
Mpl
(3)
= Mpl ek(2y2 y1 y3 ) .
MGUT
(44)
k(2y2 y1 y3 ) = 34.5 or 30, respectively. And the value of (y1 ) = k(y3 y2 ) + c
determines the relation between Mpl and MX .
If the brane at y = y1 is the observable brane, we can solve the gauge hierarchy problem,
too. Assuming that Mpl = k and e2k(2y2y1 y3 ) 1, we obtain:
MGUT = Mpl e 3 k(y3 y2 ) 3 c ,
1
(1)
(45)
with k(y3 y2 ) + c = 103.5 or 90, we can have the GUT scale in our world at TeV scale or
respectively.
2.3. Models on the spacetime M 4 S1
In this subsection, we consider models on the spacetime M 4 S 1 . Because the adjacent
two branes will have opposite brane tensions, i.e., one has positive tension, and the other
one has negative tension, we will have even number of the branes and the index is 0. Let us
assume that we have 2n branes, and their fifth coordinates are 0 6 y1 < y2 < < y2n1 <
y2n < 2, where is the radius of the fifth dimension. The Lagrangian, Einstein equation,
and the differential equations of (y) are similar to those in Subsection 2.1 except we must
have an even number of branes, 0 6 y 6 2, and (0) = (2).
If y1 = 0, then, the two solutions A and B of (y) are
(y)A =
2n
X
(1)i+1 k|y yi | + c,
(46)
(1)i k|y yi | + c.
(47)
i=2
(y)B =
2n
X
i=2
If y1 6= 0, then, the two solutions C and D are:

(y)C =
2n
X
(1)i+1 k|y yi | ky + c,
(48)
i=1
(y)D =
2n
X
(1)i k|y yi | + ky + c.
i=1
(49)
185
And the cosmological constant is:

= 6k 2MX3 .
(50)
For solution A and C, the brane tensions are:

Vi = (1)i+1 6kMX3 ,
(51)
and for solution B and D, the brane tensions are:

Vi = (1)i 6kMX3 ,
(52)
where i = 1, . . . , 2n in Eqs. (51) and (52).

The constraint is (0) = (2). So, we obtain the constraint equations on solutions
A, B, C, D, respectively:
2n
X
(1)i+1 yi = ,
(53)
i=2
2n
X
(1)i yi = ,
(54)
i=2
2n
X
(1)i+1 yi = ,
(55)
i=1
2n
X
(1)i yi = .
(56)
i=1
The corresponding 4-dimensional Planck scale for solutions A and C is:

!
2n
1 3 X
2
i+1 2 (yi )F
(1) e
,
Mpl = MX
k
(57)
i=1
where F = A, or C. And the corresponding 4-dimensional Planck scale for solutions B

and D is:
!
2n
1 3 X
2
i 2 (yi )F
(1) e
,
(58)
Mpl = MX
k
i=1
where F = B or D.
(i)
GUT :
MGUT = M5GUT e (yi )F MX e (yi )F ,
(i)
(i)
(59)
where F = A, B, C, D.
We can also consider the models with Z2 symmetry. The Z2 symmetry is defined by
the transformation of y 2 y, and we require that the metric and the Lagrangian are
invariant under this transformation. With Z2 symmetry, we must have a brane at y = 0,
and a brane at y = . Therefore, only solutions A and B are possible. In short, the fifth
186
coordinates of branes should satisfy: y1 = 0, yn+1 = , yi = 2 y2n+2i for i =

2, . . . , n.
Let us give an explicit model. We discuss a four 3-brane model: the branes fifth
coordinates are: y1 = 0, y2 , y3 , y4 , and the brane tensions divided by 6MX3 are: k, k,
k, k, respectively. Therefore, we obtain:
(y) =
4
X
(1)i+1 k|y yi | + c,
(60)
i=2
with the constraint equation:

y3 = y2 + y4 .
(61)
So, one can easily obtain:

(y1 ) = k + c,
(y2 ) = k( y2 ) + c,
(y3 ) = k(y4 y2 ) + c,
(y4 ) = k(y4 ) + c.
(62)
(63)
Without loss of generality, we assume y4 y2 < . The four-dimensional Planck scale

is

MX3 2k2c
e
1 e2ky2 + e2k(y4 +y2 ) e2k(2y4 ) .
(64)
k
Assuming that e2k(y4 +y2 ) 1, e2ky2 1, and MX = k, we obtain:
2
=
Mpl
(3)
= Mpl ek(y4 +y2 ) .
MGUT
(65)
k( y4 + y2 ) = 34.5 or 30, respectively. The value of k c determines the relation
2.4. Models on the spacetime M 4 S1 /Z2
The models with Z2 symmetry in the Subsection 2.3 can be generalized to the models
on spacetime M 4 S 1 /Z2 . For the models with Z2 symmetry in the Subsection 2.3, we
introduce the equivalence classes: y 2 y, and ith brane (2n + 2 i)th brane,
and modulo the equivalence classes, then, we obtain the models with n + 1 3-branes on
M 4 S 1 /Z2 . Using our method of splitting the branes at fixed points as before, after we
modulo the equivalence classes, we obtain the tensions of the boundary branes will become
half of their original values. Thus, the index in this kind of model is zero, too, which is the
only constraint. In short, we have a model with n + 1 branes whose fifth coordinates are:
y0 = 0 < y1 < < yn1 < yn = .
The general solution of (y) is
!
n1
X

1
i
n+1
(1) k|y yi | + 1 + (1)
ky + c,
(66)
(y) = sign(V0 )
2
i=1
where sign(Vi ) = |Vi |/Vi , for i = 1, . . . , n.
187

2
Mpl
!
n1
sign(V0 ) 3 1 2 (y0 ) X
(1)n 2 (yn )
i 2 (yi )
MX
e
e
=
+
(1) e
+
.
k
2
2
(67)
i=1
(i)
GUT :
MGUT = M5GUT e (yi ) MX e (yi ) .
(i)
(i)
(68)
Here, we will give two simple models.

(I) Although the model with three branes is equivalent to the model with four
3-branes and Z2 symmetry in Subsection 2.3, we also discuss it, because of its simplicity
and phenomenological interest. Assuming we have three branes whose fifth coordinates
are: y0 = 0, y1 , y2 = , and whose brane tensions divided by 6MX3 are: 12 k, k, 12 k,
respectively, we obtain:
(y) = k|y y1 | + c,
(y0 ) = ky1 + c,
(69)
(y1 ) = c,
(y2 ) = k( y1 ) + c.
(70)
(71)
Without loss of generality, we assume < 2y1 . The four-dimensional Planck scale is
2
=
Mpl

MX3 2ky1 2c
e
1 2 e2ky1 + e2k(2y1) .
2k
(72)
Assuming that e2k(2y1) 1, and MX = 2k, we obtain:
(2)
= Mpl ek(2y1 ) .
MGUT
(73)
k(2y1 ) = 34.5 or 30, respectively. The value of ky1 c determines the relation
(II) Model with four 3-branes: the branes fifth coordinates are: y0 = 0, y1 , y2 , y3 = ,
and the brane tensions divided by 6MX3 are: 12 k, k, k, 12 k, respectively. Therefore, we
obtain:
(y) = ky +
2
X
(1)i k|y yi | + c,
(74)
i=1
(y0 ) = k(y2 y1 ) + c,
(y2 ) = ky1 + c,
(y1 ) = ky2 + c,
(75)
(y3 ) = k( y2 + y1 ) + c.
(76)
Without loss of generality, we assume 2y1 < y2 . The four-dimensional Planck scale is

MX3 2ky1 2c 1 2k(y2 2y1 )
1 2k(y2 )
2
2k(y2 y1 )
e
e
e
+1 e
.
(77)
Mpl =
k
2
2
188
Assuming that e2k(y22y1 ) 1, e2k(y2 ) 1, and MX = k, we obtain:
(0)
= Mpl ek(y2 2y1 ) .
MGUT
(78)
So, the GUT scale in our world will be pushed to TeV scale or 105 GeV scale range
if k(y2 2y1 ) = 34.5 or 30, respectively. The value of (y2 ) = ky1 + c determines the
relation between Mpl and MX .
If the brane at y = y2 is the observable brane, one can also solve the gauge hierarchy
problem. Assuming that Mpl = k, e2k(y2 2y1 ) 1, and e2k(y2 ) 1, one obtains:
1
MGUT = Mpl e 3 ky1 3 c ,

(2)
(79)
with ky1 + c = 103.5 or 90, one will have the GUT scale in our world at TeV scale or
105 GeV scale range, respectively. The five-dimensional Planck scale is 1048 GeV or
1044 GeV, respectively.
3. General models with sectionally constant cosmological constant along the fifth
dimension
In this section, we consider the general models with sectionally constant cosmological
constant along the fifth dimension. The models on the spacetimes M 4 R 1 , M 4
R 1 /Z2 , M 4 S 1 and M 4 S 1 /Z2 will be discussed in the Subsections 3.1, 3.2, 3.3,
3.4, respectively. In addition, we assume that i,i+1 6= 0 in the discussions of the explicit
models, where i,i+1 is defined in the following subsections.
3.1. Models on the spacetime M 4 R1
First, we consider the fifth dimension is R 1 . Assume we have l + m + 1 parallel 3-branes,
and their fifth coordinates are: < yl < yl+1 < < y1 < y0 < y1 < < ym1 <
ym < +. The 5-dimensional metric in these branes are:

(i)
(x ) G x , y = yi ,
(80)
g
where GAB is the five-dimensional metric, i = l, . . . , m and A, B = , y.
The classical Lagrangian is given by:
S = Sgravity + SB ,
Z

Sgravity = d4 x dy G (y) + 12 MX3 R ,
SB =
m Z
X
d4 x

g (i) Li Vi ,
(81)
(82)
(83)
i=l
where MX is the 5-dimensional Planck scale, (y) is cosmological constant, and Vi where
i = l, . . . , m is the ith brane tension. (y) is defined as the following:
(y) =
m
X
189

i (y yi1 ) (y yi ) + + (y ym )
i=1
0
X

i (y + yi ) (y + yi1 ) + (y + yl ),
(84)
i=l+1
where (x) = 1 for x > 0 and (x) = 0 for x < 0. From advanced calculus, we know that
(y) is piece-wise continuous or sectionally continuous. (y) is sectionally constant for
(y) is constant in each section.
The 5-dimensional Einstein equation for the above action is:

1
G RAB GAB R
2
#
"
m
q
X
(i)
Vi g (i) g M N (y yi ) .
(85)
= 3 (y) G GAB +
MX
i=l
Assuming that there exists a solution that respects 4-dimensional Poincar invariance in
the x -directions, one obtains the 5-dimensional metric:
ds 2 = e2 (y) dx dx + dy 2.
(86)
With this metric, the Einstein equation reduces to:

02 =
(y)
6MX3
m
X
Vi
00 =
i=l
3MX3
(y yi ).
(87)
So, 0 is also sectionally continuous or sectionally constant.

The general solution to the above differential equations is:
(y) =
m
X
ki |y yi | + kc y + c,
(88)
i=l
where kc and c are constants, and ki 6= 0 for i = l, . . . , m.

The brane tension Vi , the sectionally cosmological constant i and are:
Vi = 6ki MX3 ,
i = 6MX3
m
X
kj
j =i
= 6MX3
m
X
j =l
i1
X
(89)
!2
kj kc
(90)
j =l
!2
kj kc
+ = 6MX3
m
X
!2
kj + kc
(91)
j =l
Therefore, the cosmological constant is negative or zero in each section. Then, for any point
in M 4 R 1 , which does not belong to any brane and the section(s) with zero cosmological
constant, there is a neighborhood which is diffeomorphic to (or a slice of) AdS5 space.
Similarly, this statement is correct for the models on the spacetimes M 4 R 1 /Z2 ,
M 4 S 1 and M 4 S 1 /Z2 . Moreover, the cosmological constant and brane tensions should
190
satisfy the above equations. In order to obtain a finite 4-dimensional Planck scale, we
P
obtain the constraint: m
j =l kj > |kc |. So, the sum of the brane tensions is positive.
The general bulk metric is:
ds 2 = e2
Pm
i=l ki |yyi |2kc y2c
dx dx + dy 2 .

!
m1
X
2
3
Ti,i+1 ,
Mpl = MX T,l + Tm,+ +
(92)
(93)
i=l
where
1
e2 (yl ) ,
2
if i,i+1 6= 0, then
T,l =
Ti,i+1 =
1
2i,i+1
Tm,+ =
1
e2 (ym ) ,
2+

e2 (yi+1) e2 (yi ) ,
(94)
(95)
and if i,i+1 = 0, then

Ti,i+1 = (yi+1 yi )e2 (yi ) ,
(96)
where
=
m
X
kj kc ,
+ =
j =l
i,i+1 =
m
X
j =i+1
m
X
kj + kc ,
(97)
j =l
kj
i
X
kj kc .
(98)
j =l
One can easily prove that Ti,i+1 is positive, which assures that the 4-dimensional Planck
scale is positive.
(i)
The four-dimensional GUT scale on the ith brane MGUT is related to the fivedimensional GUT scale on the ith brane M5(i)
GUT by:
(i)
(yi )
= M5(i)
.
MGUT
GUT e
(99)
M5(i)
GUT
MX , for i = l, . . . , m.
In this paper, we assume that
Now, we would like to explain how to solve the gauge hierarchy problem in general.
Using fact that T,l > 0, Tm,+ > 0, Ti,i+1 > 0 where i = l, . . . , m 1, and noticing
P
that m
j =l kj > |kc |, we can prove that if (yj ) is the minimal value of (yi ) for all i,
then the j th brane will have positive brane tension. If e2( (yi ) (yj )) 1 for all i 6= j ,
then, we obtain:
MX3 2 (yj )
e
,
(100)
mc
where mc is a function of kc and ki for all i. Assuming the observable brane is at y = yio
and MX = mc , we obtain:
2
=
Mpl
(io )
= Mpl e( (yio ) (yj )) .
MGUT
(101)
191
So, if (yio ) (yj ) = 34.5 or 30, we will push the four-dimensional GUT scale in our
world to the TeV or 105 GeV range, respectively. Of course, the 4-dimensional GUT scale
on the j th brane is the highest, but, the j th brane does not need to have largest brane
tension. Furthermore, if (yj ) > 0, and if the j th brane is the observable brane, we can
also solve the gauge hierarchy problem. Assume that Mpl = mc , we obtain:
1
MGUT = Mpl e 3 (yj ) ,

(j )
(102)
with (yj ) = 103.5 or 90, we can have the GUT scale in our world at TeV scale or 105
GeV scale, respectively. Obviously, in this case, because of the exponential factor, the fivedimensional Planck scale is very high, 1048 GeV or 1044 GeV, respectively.
These models can be generalized to models with Z2 symmetry. Because of Z2 symmetry,
kc = 0. There are two kinds of such models, one kind with an odd number of branes, and
the other kind with an even number of the branes. For the first kind, one requires that
ki = ki , yi = yi , and m = l. For the second kind, one requires that ki = ki , yi =
yi , m = l, and k0 = 0 (no number 0 brane).
Now, we discuss three models:
(I) First, we consider a model with only one brane whose fifth coordinate is y0 and
whose brane tension divided by 6MX3 is k0 . Therefore, we obtain:
(y) = k0 |y y0 | + kc y + c.
(103)
And the constraint is k0 > |kc |. One can easily obtain:

(y0 ) = kc y0 + c.
(104)
The four-dimensional Planck scale is

k0
2
= MX3 2
e2kc y0 2c .
Mpl
k0 kc2
(105)
Assuming that Mpl = (k02 kc2 )/k0 , we obtain:

1
(0)
= Mpl e 3 kc y0 3 c ,
MGUT
(106)
with kc y0 + c = 103.5 or 90, the GUT scale in our world will be pushed to the TeV scale
or 105 GeV scale range, respectively. The five-dimensional Planck scale is 1048 GeV or
1044 GeV, respectively. By the way, if kc = 0, the constant c is the key factor, although it is
often thought to be physically irrelevant. The model with kc = c = 0 was discussed in [4].
(0)
In addition, we can obtain the exact relation between Mpl and MGUT :
s
MX k0
(0)
.
(107)
Mpl = MGUT 2
k0 kc2
In order to solve the gauge hierarchy problem, in general, we might require that MX is very
large compare to k0 and kc , or k02 kc2 is very small.
(II) The model with two 3-branes: the branes fifth coordinates are: y0 and y1 , and the
brane tensions divided by 6MX3 are: k0 and k1 . We obtain:
(y) = k0 |y y0 | + k1 |y y1 | + kc y + c,
The constraint is k0 + k1 > |kc |. One can easily obtain:
(108)
192
(y0 ) = k1 (y1 y0 ) + kc y0 + c,
(109)
(y1 ) = k0 (y1 y0 ) + kc y1 + c.
(110)

k1
k0
2
2 (y0 )
2 (y1 )
= MX3 2
e
+
e
.
Mpl
k12 (k0 + kc )2
k0 (k1 kc )2
(111)
If k0 > 0, k1 > 0, without loss of generality, assuming that (y0 ) < (y1 ) and
e2( (y1) (y0 )) 1, we obtain
2
= MX3
Mpl
k0
k02 (k1 kc )2
e2 (y0 ) .
(112)
If the brane at y = y1 is the observable brane, assuming MX = (k02 (k1 kc )2 )/k0 , we

obtain:
(1)
= Mpl e( (y1 ) (y0 )) .
MGUT
(113)
(k0 k1 + kc )(y1 y0 ) = 34.5 or 30, respectively. And the value of (y0 ) determines the
relation between Mpl and MX . Explicit example: k0 = k1 = kc > 0, kc (y1 y0 ) = 34.5
or 30. From this explicit example, we conclude that kc is also an important factor to solve
the gauge hierarchy problem.
If the brane at y = y0 is the observable brane, we can solve the gauge hierarchy problem
only when (y0 ) > 0. Assuming that Mpl = (k02 (k1 kc )2 )/k0 and e2( (y1 ) (y0))
1, we obtain:
MGUT = Mpl e 3 (y0 ) ,
1
(0)
(114)
with (y0 ) = 103.5 or 90, we can have the GUT scale in our world at TeV scale or
respectively.
If k0 > 0, k1 < 0, the solution to the gauge hierarchy problem is similar to that in the
above paragraph.
(III) Model with three 3-branes: the branes fifth coordinates are: y1 , y2 , y3 , and the
brane tensions divided by 6MX3 are: k1 , k2 , k3 , respectively. The constraint is k1 + k2 + k3
> |kc |. Therefore, we obtain:
3
X
ki |y yi | + kc y + c,
(115)
(y1 ) = k2 (y2 y1 ) + k3 (y3 y1 ) + kc y1 + c,
(116)
(y2 ) = k1 (y2 y1 ) + k3 (y3 y2 ) + kc y2 + c,
(117)
(y3 ) = k1 (y3 y1 ) + k2 (y3 y2 ) + kc y3 + c.
(118)
(y) =
i=1
2
Mpl
M3
= X
2
193
1
e2 (y1 )
k1 + k2 + k3 kc

1
+
e2 (y2 ) e2 (y1)
k2 + k3 kc k1

1
e2 (y3 ) e2 (y2)
+
k3 kc k1 k2

1
2 (y3 )
e
.
+
k1 + k2 + k3 + kc
(119)
If there exist at least two branes which have positive tensions, without loss of generality,
we assume the brane at y = y1 has positive tension, (y1 ) < (y2 ), and (y1 ) < (y3 ). If
e2( (y2) (y1 )) 1 and e2( (y3) (y1 )) 1, we obtain
k1
2
= MX3 2
e2 (y1) .
(120)
Mpl
2
k1 (k2 + k3 kc )
If the brane at y = yj where j = 2 or 3 is the observable brane, and assuming MX =
(k12 (k2 + k3 kc )2 )/k1 , we obtain:
MGUT = Mpl e( (yj ) (y1 )) .
(j )
(121)
GeV scale range if (yj )

So, the GUT scale in our world will be at TeV scale or
(y1 ) = 34.5 or 30, respectively. The value of (y1 ) determines the relation between Mpl
and MX . For an explicit example, y1 = y3 , y2 = 0, and k1 = k3 = kc , k2 = 12 kc .
105
1
(y1 ) = kc y3 + c,
2
5
(y3 ) = kc y3 + c.
2
(y2 ) = 2kc y3 + c,
(122)
(123)
If MX = 34 kc and the brane y = y3 is the observable brane, we obtain

(3)
= Mpl e3kc y3 .
MGUT
(124)
So, one can easily solve the gauge hierarchy problem.

only if (y1 ) > 0. Assuming that Mpl = (k12 (k2 + k3 kc )2 )/k1 and e2( (yj ) (y1))
1 where j = 2, 3, we obtain:
(1)
= Mpl e 3 (y1 ) ,
MGUT
1
(125)
GeV,
if (y1 ) = 103.5 or 90, the GUT scale in our world will be about TeV or
respectively. The five-dimensional Planck scale is 1048 GeV or 1044 GeV, respectively.
If only one brane has positive tension, the discussion of the gauge hierarchy problem is
similar to that in the above paragraph.
105
3.2. Models on the spacetime M 4 R1 /Z2

Now, we consider the models on the spacetime M 4 R 1 /Z2 . Similar to Section 2,
the models with Z2 symmetry and odd number of 3-branes in the Subsection 3.1 can be
194
generalized to the models in which the fifth dimension is R 1 /Z2 by this way: we introduce
the equivalence classes: y y and ith brane (i)th brane, and then, modulo the
equivalence classes. Using our method of splitting brane at fixed point as before, we obtain
that the tension V0 of the brane at the origin (or the boundary brane) becomes half of its
original value, i.e., V0 = 3k0 MX3 . So, the sum of the brane tensions is positive. And we will
have m + 1 brane whose fifth coordinates are: y0 = 0 < y1 < y2 < < ym < +.
P
Assuming that kT = m
j =1 kj , we obtain the general solution to (y) is:
(y) =
m
X
ki |y yi | + kT y + c.
(126)
i=0
And the corresponding 4-dimensional Planck scale is:

!
m1
X
2
3
Ti,i+1 ,
Mpl = MX Tm,+ +
(127)
i=0
where
Tm,+ =
1
e2 (ym ) ,
2+
(128)
if i,i+1 6= 0, then
Ti,i+1 =
1
2i,i+1

e2 (yi+1) e2 (yi ) ,
(129)

Ti,i+1 = (yi+1 yi )e2 (yi ) ,
(130)
where
+ =
m
X
kj + kT ,
(131)
j =0
i,i+1 =
m
X
j =i+1
kj
i
X
kj kT .
(132)
j =0
One can easily prove that Ti,i+1 is positive, which assures that the 4-dimensional Planck
scale is positive.
(i)
GUT :
(i)
(yi )
= M5(i)
MX e (yi ) .
MGUT
GUT e
(133)
Now, we discuss a simple model. The model with two 3-branes is equivalent to the model
with three 3-branes and Z2 symmetry in the Subsection 3.1, so, we do not discuss it here.
We give a model with three 3-branes: the branes fifth coordinates are: y0 = 0, y1 , y2 , and
the brane tensions divided by 6MX3 are: 12 k0 , k1 , k2 , respectively. Noticing kT = k1 + k2 ,
we obtain:
(y) =
2
X
ki |y yi | + (k1 + k2 )y + c,
195
(134)
i=0
(y0 ) = k1 y1 + k2 y2 + c,
(135)
(y1 ) = k0 y1 + k2 (y2 y1 ) + (k1 + k2 )y1 + c,
(136)
(y2 ) = k0 y2 + k1 (y2 y1 ) + (k1 + k2 )y2 + c.
(137)
The four-dimensional Planck scale is:

M3
1
1 2 (y1 )
2
= X
e2 (y0)
e
e2 (y2 ) e2 (y1 )
Mpl
2
k0
k + 2k1
0
1
e2 (y2 ) .
+
k0 + 2k1 + 2k2
(138)
If there exist at least two branes which have positive tensions, without loss of generality,
we assume the brane at y = y0 has positive tension, (y0 ) < (y1 ), and (y0 ) < (y2 ). If
e2( (y1) (y0 )) 1 and e2( (y2) (y0 )) 1, we obtain
2
=
Mpl
MX3 2 (y0 )
e
.
2k0
(139)
If the brane at y = yj where j = 1 or 2 is the observable brane, we can solve the gauge
hierarchy problem. Assuming MX = 2k0 , we obtain:
MGUT = Mpl e( (yj ) (y0 )) .
(j )
(140)
(yj ) (y0 ) = 34.5 or 30, respectively. The value of (y0 ) determines the relation
If the brane at y = y0 is the observable brane, the gauge hierarchy problem can be solved
only if (y0 ) > 0. Assuming that Mpl = 2k0 and e2( (yj ) (y0 )) 1 where j = 1, 2, we
obtain:
1
MGUT = Mpl e 3 (y0 ) ,

(0)
(141)
with (y0 ) = 103.5 or 90, we can have the GUT scale in our world at TeV scale or 105
GeV scale, respectively. The five-dimensional Planck scale is 1048 GeV or 1044 GeV,
respectively.
For just one brane with positive tension, the discussion of the gauge hierarchy problem
is similar to that in the above paragraph.
3.3. Models on the spacetime M 4 S1
In this subsection, we consider the models on the spacetime M 4 S 1 . Assume we have
l + m + 1 parallel 3-branes, and their fifth coordinates are: 6 yl < yl+1 < <
y1 < y0 < y1 < < ym1 < ym < . The Lagrangian, the Einstein equation and the
differential equations of (y) are similar to those in Subsection 3.1, except we require that
|y| 6 , () = (), and (y) is defined as the following:
196

m
X
(y) =

i (y yi1 ) (y yi ) + (y ym )
i=l+1

+ (y + yl ) .
(142)
The general solution to the differential equations of (y) is:

(y) =
m
X
ki |y yi | + kc y + c.
(143)
i=l
The brane tension Vi , the sectionally cosmological constant i and are:

Vi = 6ki MX3 ,
i = 6MX3
m
X
j =i
kj
i1
X
(144)
!2
kj kc
(145)
j =l
= 6MX3 kc2 ,
(146)
d (y)/dy is a piece-wise continuous function or sectionally continuous function for it has

finite jumps because of finite branes. The jump is 2ki when it passes the ith brane. In
fact, d (y)/dy is sectionally constant. In addition, if starting at a point not belonging
to any brane, we wind around the circle one time, the sum of the total jumps for the
P
function d (y)/dy is 2 m
i=l ki , which is zero because d (y)/dy is sectionally constant.
Therefore, we obtain the sum of the brane tensions should be zero:
m
X
kj = 0.
(147)
j =l
In addition, from () = (), we obtain the constraint:

m
X
kj yj = kc .
(148)
j =l

!
m1
X
2
3
Ti,i+1 ,
Mpl = MX Tm,l +
(149)
i=l
where if kc 6= 0, then
Tm,l =

1 2 (yl )
e2 (ym) ,
e
2kc
(150)
and if kc = 0, then
Tm,l = (2 + yl ym )e2 (yl ) ,
(151)
if i,i+1 6= 0, then
Ti,i+1 =
1
2i,i+1

e2 (yi+1) e2 (yi ) ,
(152)
197

Ti,i+1 = (yi+1 yi )e2 (yi ) ,
(153)
where
i,i+1 =
m
X
kj
j =i+1
i
X
kj kc .
(154)
j =l
As before, one can easily prove that Ti,i+1 is positive, which assures that the 4-dimensional
Planck scale is positive.
(i)
GUT :
(i)
(yi )
= M5(i)
MX e (yi ) .
MGUT
GUT e
(155)
The models can be generalized to models with Z2 symmetry. There are four kinds of
such models, two kinds with an odd number of the branes, and two kinds with an even
number of the branes. For the models with an odd number of branes: (I) one requires that:
ki = ki , yi = yi , m = l and kc = 0; or (II) one requires that: ki = ki , yi = yi for
1 6 i 6 m, l = m + 1, k0 = 0 (no number 0 brane), yl = and kc = kl . For the
models with an even number of branes; (I) one requires that ki = ki , yi = yi , m = l,
k0 = 0 (no number 0 brane) and kc = 0; or (II) one requires that ki = ki , yi = yi , for
1 6 i 6 m, l = m + 1, k0 6= 0, yl = and kc = kl .
Let us discuss two simple models.
(I) Model with three 3-branes: the branes fifth coordinates are: y1 , y2 , y3 , and the brane
tensions divided by 6MX3 are: k1 , (k1 + k3 ), k3 , respectively, where we assume k1 > 0,
k3 > 0. Therefore, we obtain:
(y) = k1 |y y1 | (k1 + k3 )|y y2 | + k3 |y y3 | + kc y + c.
(156)
The constraint equation is

k1 y1 (k1 + k3 )y2 + k3 y3 = kc
(157)
(y1 ) = kc ( + y1 ) + c,
(158)
(y2 ) = 2k1 (y2 y1 ) + kc ( + y2 ) + c,
(159)
(y3 ) = kc (y3 ) + c.
(160)
and
Without loss of generality, assuming kc > 0, we obtain (y3 ) < (y1 ) < (y2 ).

M3
1
1 2 (y1 )
2
= X
e2 (y3)
e
e2 (y2) e2 (y1 )
Mpl
2
kc
2k1 + kc

1
2 (y3 )
2 (y2 )
e
e
.
+
2k3 kc
(161)
198
If e2( (y1) (y3 )) 1, we obtain:

2
= MX3
Mpl
k3
e2kc (y3 )2c .
(2k3 kc )kc
(162)
Because of the constraint equation, 2k3 kc > 0, if the brane at y = y1 is the observable
brane, we can solve the gauge hierarchy problem. Assuming MX = ((2k3 kc )kc )/k3 , we
obtain:
(1)
= Mpl ekc (2+y1 y3 ) .
MGUT
(163)
kc (2 + y1 y3 ) = 34.5 or 30, respectively. And the value of (y3 ) determines the
relation between Mpl and MX .
only if (y3 ) > 0. Assuming that Mpl = ((2k3 kc )kc )/k3 and e2( (y1) (y3 )) 1, we
obtain:
1
(3)
= Mpl e 3 (y3 ) ,
MGUT
(164)
with (y3 ) = 103.5 or 90, we can have the GUT scale in our world at TeV scale or 105
GeV scale, respectively. The five-dimensional Planck scale is 1048 GeV or 1044 GeV,
respectively.
(II) Model with four 3-branes and Z2 symmetry. The branes fifth coordinates are: y2 =
, y1 = y1 , y0 = 0, y1 , and the brane tensions divided by 6MX3 are: 2k2 , (k0 + k2 ),
2k0 , (k0 + k2 ), respectively, where we assume k0 > 0, k2 > 0. Therefore, we obtain:
(y) = (k2 + k0 )(|y y1 | + |y + y1 |) + 2k0|y| + c,
(165)
(y2 ) = 2k2 + c,
(166)
then,
(y0 ) = 2(k2 + k0 )y1 + c,
(y1 ) = (y1 ) = 2k2 y1 + c.


1 2 (y )

M3
1 2 (y1)
2
0 e2 (y1 )
= X
e2 (y2) +
e
e
Mpl
4
k2
k0

1
1 2 (y1 )
e2 (y0) +
e
e2 (y2 ) e2 (y1 ) .
k0
k2
(167)
(168)
Without loss of generality, we assume that (y2 ) < (y0 ) or k2 (k2 + k0 )y1 > 0. If
the brane at y = y0 is the observable brane, the gauge hierarchy problem can be solved.
Assuming that e2( (y0) (y2)) 1, and MX = 2k2 , one obtains:
MGUT = Mpl e( (y0 ) (y2 )) .
(0)
(169)
So, one can push the GUT scale in our world to TeV scale or 105 GeV scale range if
(y0 ) (y2 ) = 2(k2 (k2 + k0 )y1 ) = 34.5 or 30, respectively. And the value of
(y2 ) determines the relation between Mpl and MX .
199
3.4. Models on the spacetime M 4 S1 /Z2

The models with Z2 symmetry in the last Subsection 3.3 can be generalized to the
models in which the fifth dimension is S 1 /Z2 . For the models with Z2 symmetry in the
Subsection 3.3, we require that ki = ki , yi = yi , for 1 6 i 6 m, l = m + 1, yl = ,
and kc = kl , then, introduce the equivalence classes: y y and ith brane (i)th
brane for i = 1, . . . , m. After we modulo the equivalent classes, we obtain the models on
M 4 S 1 /Z2 . For convenience, we renumber (l)th brane as (m + 1)th brane, so, ym+1 =
. Using the splitting brane method for the branes at fixed points, we obtain the boundary
brane tensions V0 and Vm+1 are half of their original values, i.e., V0 = 3k0 MX3 , Vm+1 =
P
3km+1 MX3 . So, the sum of the brane tensions is zero as well, i.e., k0 + km+1 + 2 m
i=1 ki =
0. In short, we will have a model with m + 2 branes whose fifth coordinates are y0 = 0 <
y1 < < ym < ym+1 = .
The general solution of (y) is:
(y) = k0 |y| +
m
X
ki (|y yi | + |y + yi |) + c.
(170)
i=1
We can rewrite it as
(y) =
m
X
ki |y yi | + kT y + c,
(171)
i=1
where kT = 12 (k0 km+1 ).

!
m
X
2
= MX3
Ti,i+1 ,
Mpl
(172)
i=0
where if i,i+1 6= 0, then

Ti,i+1 =

1
e2 (yi+1) e2 (yi ) ,
2i,i+1
(173)

Ti,i+1 = (yi+1 yi )e2 (yi ) ,
(174)
where
0,1 = k0 ,
m,m+1 = km+1 ,
(175)
and for i = 1, . . . , m 1,
i,i+1 =
m
X
j =i+1
kj
i
X
kj kT .
(176)
j =1
One can easily prove that Ti,i+1 is positive, which makes sure that the 4-dimensional
Planck scale is positive.
200
(i)
The four-dimensional GUT scale on the ith brane MGUT
is related to the five(i)
dimensional GUT scale on the ith brane M5GUT :
MGUT = M5GUT e (yi ) MX e (yi ) .

(i)
(i)
(177)
Now, we discuss a simple model. The model with three 3-branes is equivalent to the
model with four 3-branes and Z2 symmetry in Subsection 3.3. So, we will not discuss it
here. We discuss a model with four 3-branes: the branes fifth coordinates are: y0 = 0, y1 ,
y2 , y3 = , and the brane tensions divided by 6MX3 are: 12 (k1 + k2 + kT ), k1 , k2 ,
12 (k1 k2 + kT ), respectively, where we assume k1 > 0, k2 > 0. Therefore, we obtain:
(y) = kT y + k1 |y y1 | k2 |y y2 | + c,
(178)
(y0 ) = k1 y1 k2 y2 + c,
(179)
(y1 ) = k2 (y2 y1 ) + kT y1 + c,
(180)
(y2 ) = k1 (y2 y1 ) + kT y2 + c,
(181)
(y3 ) = (k1 k2 + kT ) (k1 y1 k2 y2 ) + c.

M3
1
2
= X
e2 (y1 ) e2 (y0 )
Mpl
2 k2 + kT k1

1
e2 (y2 ) e2 (y1 )
+
k1 + k2 + kT

1
2 (y3 )
2 (y2 )
e
e
.
+
k1 k2 + kT
(182)
(183)
Without loss of generality, we assume that kT > 0 and k2 + kT > k1 . Then, the brane at y =
y0 has positive tension and (y0 ) < (y1 ). Assuming e2( (yj ) (y0 )) 1 for j = 1, 2, 3,
MX = 2(k2 + kT k1 ), and the brane at y = y1 is the observable brane, we can solve the
gauge hierarchy problem:
(1)
= Mpl e(k2 k1 +kT )y1 .
MGUT
(184)
(k2 k1 + kT )y1 = 34.5 or 30, respectively. And the value of (y0 ) determines the relation
If the brane at y = y0 is the observable brane, one can solve the gauge hierarchy
problem only if (y0 ) > 0, i.e., k1 y1 > k2 y2 . Assuming that Mpl = 2(k2 + kT k1 ) and
e2(k2 k1 +kT )y1 1, one obtains:
(0)
= Mpl e 3 (y0 ) ,
MGUT
1
(185)
with (y0 ) = 103.5 or 90, one can push the GUT scale in our world to TeV scale or 105
GeV scale range, respectively. The five-dimensional Planck scale is 1048 GeV or 1044 GeV,
respectively.
201
4. Conclusion
If the fifth dimension is a one-dimensional connected manifold or one-dimensional
connected manifold with boundary, up to diffeomorphism, the only possible spacetimes
will be M 4 R 1 , M 4 R 1 /Z2 , M 4 S 1 and M 4 S 1 /Z2 . There exist two possibilities
on cosmological constant along the fifth dimension: the cosmological constant is constant,
or the cosmological constant is sectionally constant. For any point in M 4 R 1 , M 4
R 1 /Z2 , M 4 S 1 and M 4 S 1 /Z2 , which does not belong to any brane or the section(s)
where the cosmological constant is zero, there is a neighborhood which is diffeomorphic
to (or a slice of) AdS5 space. We construct the general models with parallel 3-branes
and constant/sectionally constant cosmological constant along the fifth dimension on the
spacetimes M 4 R 1 , M 4 R 1 /Z2 , M 4 S 1 and M 4 S 1 /Z2 . Moreover, we point
out that for a compact fifth dimension, the sum of the brane tensions is zero, and for
a non-compact fifth dimension, the sum of the brane tensions is positive if we require
that the 4-dimensional Planck scale is finite. Furthermore, we assume the observable
brane which includes our world should have positive tension, and conclude that in those
general scenarios, the 5-dimensional GUT scale on each brane can be identified as the
5-dimensional Planck scale, but, the 4-dimensional Planck scale is generated from the low
4-dimensional GUT scale exponentially in our world. We also give some simple models to
show explicitly how to solve the gauge hierarchy problem.
Note added
During the preparation of this paper, Ref. [32] appeared, which overlapped part of
this paper. In Ref. [32], the author derived the crystal brane world solutions, comprising
of intersecting families of parallel n + 2 branes in a (4 + n)-dimensional AdS space
with the constant cosmological constant in the whole spacetime. Therefore, the models
in Section 2 may be considered as specific models in Ref. [32], although we discuss
the models in detail. However, we also discuss the models with sectionally constant
cosmological constant along the fifth dimension in Section 3. By the way, when we
construct models with many brane intersections/junctions on the spacetime M 4 M n ,
the cosmological constant might not be constant in the whole spacetime, and M n should
be n-dimensional manifold or n-dimensional manifold with boundary [21].
Acknowledgements
We would like to thank T. Coleman and M. Olsson for reading the manuscript and
comments. And we would like to thank S. Nam for bringing my attention to his paper. This
research was supported in part by the U.S. Department of Energy under Grant No. DEFG02-95ER40896 and in part by the University of Wisconsin Research Committee with
funds granted by the Wisconsin Alumni Research Foundation.
202
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263.

I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257.
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370.
J. Lykken, L. Randall, hep-th/9908076.
M. Gogberashvili, hep-ph/9812296, Eur. Phys. Lett. 49 (2000) 396; Mod. Phys. Lett. A 14
(1999) 2025; hep-ph/9908347.
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, N. Kaloper, Phys. Rev. Lett. 84 (2000) 586.
W.D. Goldberger, M.B. Wise, Phys. Rev. Lett. 83 (1999) 4922; Phys. Lett. B 475 (2000) 275.
T. Nihei, Phys. Lett. B 465 (1999) 81.
C. Csaki, M. Grasesser, C. Kolda, T. Terning, Phys. Lett. B 462 (1999) 34.
N. Kaloper, Phys. Rev. D 60 (1999) 074027.
H. Verlinde, hep-th/9906182.
I. Oda, hep-th/9908104.
A. Brandhuber, K. Sfetsos, JHEP 9910 (1999) 013.
A. Kehagias, Phys. Lett. B 469 (1999) 123.
H.B. Kim, H.D. Kim, Phys. Rev. D 61 (2000) 064003.
T. Shiromizu, K. Maeda, M. Sasaki, gr-qc/9910076.
K. Behrndt, M. Cvetic, Phys. Lett. B 475 (2000) 253.
T. Li, Phys. Lett. B 471 (1999) 20.
T. Li, hep-th/9911234, Phys. Lett. B, in press.
T. Li, in preparation.
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 567 (2000) 111.
O. DeWolfe, D.Z. Freedman, S.S. Gubser, A. Karch, hep-th/9909134.
S. Chang, M. Yamaguchi, hep-ph/9909523.
A. Chodos, E. Poppit, Phys. Lett. B 471 (1999) 119.
C. Grojean, J. Cline, G. Servant, Phys. Rev. Lett. 83 (1999) 4245; Phys. Lett. B 472 (2000) 302,
hep-th/9910081.
C. Csaki, M. Grasesser, L. Randall, T. Terning, hep-ph/9911406, and references therein.
C. Csaki, Y. Shirman, Phys. Rev. D 61 (2000) 024008.
A.E. Nelson, hep-th/9909001.
I. Oda, Phys. Lett. B 472 (2000) 59.
H. Hatanaka, M. Sakamoto, M. Tachibana, K. Takenaga, Prog. Theor. Phys. 102 (1999) 1213.
N. Kaloper, Phys. Lett. B 474 (2000) 269.
S. Nam, hep-th/9911237.
M. Cvetic, H.H. Soleng, Phys. Rep. 282 (1997) 159, and references therein.
K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Lett. B 436 (1998) 55; Nucl. Phys. B 537 (1999)
47, hep-ph/9807522.

D-branes on K3-fibrations
P. Kaste, W. Lerche, C.A. Ltken 1 , J. Walcher 2
CERN, CH-1211 Geneva 23, Switzerland
Abstract
B-type D-branes are constructed on two different K3-fibrations over P1 using boundary
conformal field theory at the rational Gepner points of these models. The microscopic CFT charges
are compared with the Ramond charges of D-branes wrapped on holomorphic cycles of the
corresponding CalabiYau manifold. We study in particular D4-branes and bundles localized on
the K3 fibers, and find from CFT that each irreducible component of a bundle on K3 gains one
modulus upon fibration over P1 . This is in agreement with expectations and so provides a further test
of the boundary CFT approach to D-brane physics. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Mj
Keywords: D-branes; Boundary CFT; CalabiYau compactification
1. Introduction
In the large radius region of moduli space, strings can be viewed as moving in a
fixed extrinsic background spacetime whose geometry is determined by the closed string
sector of the theory itself; in particular it must be a CalabiYau (CY) space. At string
tree level these classical geometric data and some of the quantum corrections are encoded
in N = 2 superconformal field theory on the plane, hereafter called bulk CFT. At
special points in its moduli space the CFT is rational and can therefore be exhaustively
studied using Gepners construction. The complementary viewpoints provided by the
intrinsic, microscopic CFT picture, and the extrinsic, macroscopic CY sigma-model,
have provided a wealth of information about the closed string sector, most notably the
phenomenon of mirror symmetry, amounting to a profound redirection of our thinking
about the structure of spacetime at small scales.
Our understanding of the open string sector is currently undergoing a parallel evolution
which may ultimately provide a similar redirection of our thinking about the gauge
1 On leave from Department of Physics, University of Oslo, N-0316 Oslo, Norway.
2 Also at Institut fr Theoretische Physik, ETH-Hnggerberg, CH-8093 Zrich, Switzerland.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 0 6 - 0
204
P. Kaste et al. / Nuclear Physics B 582 (2000) 203215
structure of the theory. Indeed, a major effort is currently underway to extend the mirror
map to the gauge bundles that may decorate various cycles in CY manifolds (see, e.g.,
[1,2]).
For the open string sector the appropriate microscopic description is provided by
boundary conformal field theory on the upper half plane (BCFT), and the corresponding
macroscopic objects are D-branes wrapped on cycles of the CY target spaces embodied in
the closed string sector. Since open strings can end on D-branes, these should be intimately
related to the boundary states of the (microscopic) BCFT. In fact they are supposedly
nothing but manifestations of the boundary states at large radii of the compactification
manifold, where naive geometrical concepts apply. On the other hand, at small radii
instanton corrections can swamp out any classical picture, to an extent where the very
notion of spacetime, as well as the vector bundles it supports, evaporates.
However, it is conceivable that the microscopic BCFT still encodes some of the
geometric information pertaining to the D-branes at large radii, like the intersection
properties of the cycles around which they wrap. This can be tested at the Gepner points
in moduli space where the theory is exactly solvable, and thus opens a window on Dbrane physics in a regime where quantum corrections are strong. This approach has been
pioneered in Refs. [36], building on earlier work, for example, [711].
Our purpose in this letter is to apply the methods developed in these papers to study
properties of D-branes on K3-fibrations. The motivation is twofold. First, K3-fibrations
are important in the context of heterotic-type II duality [12,13], and their quantum algebraic
geometry is well understood. In particular, for some cases the periods, monodromy and
analytic continuation to Gepner points are either directly available from the literature [14],
or can be obtained with a reasonable amount of effort. Second, for generic threefolds the
quantum corrections at the Gepner point are strong, and for some states [36] there is no
good match between BCTF data at the Gepner point and geometrical D-brane data at large
radii for these cases it may not always be clear whether the quantum geometries really
differ so much, or whether the method fails. However, for K3 fibrations we may expect
that some quantities are protected by non-renormalization properties of the K3, while the
whole theory is not as trivial as if we would be looking just at an isolated K3. Therefore,
studying K3-fibrations might not only offer new insights in D-brane physics at small radii,
but also provide further tests of the validity of the BCFT method.
In Section 2, we review the BCFT approach at the Gepner point of generic n-folds, but
also improve and clarify some steps in the papers [36]. Where our work overlaps, we
are in agreement, and we explain why this is so. In Section 3 we then discuss some basic
D-geometric features of the two K3-fibrations we study. In Section 4, we compare the
BCFT and the geometrical data, producing lists of D-brane charges that correspond to the
CFT boundary states. Finally, in Section 5 we discuss some features of these data and find
agreement with expectations.
205
2. Gepner models and boundary conformal field theory

The Gepner construction is a GSO-projected tensor product of r minimal models at
levels ki , whose central charges ci = 3ki /(ki + 2) add up to 3n(n = 1, 2, 3, . . .), suitably
twisted to enforce modular invariance of the partition function. An explicit connection with
the geometric data at large radius is provided by the LandauGinzburg representation of
minimal models, provided that the tensor product is extended by a trivial factor kr+1 = 0
when n + r is odd. Let K = lcm{h1 , . . . , hr }, where hi = ki + 2 are called the heights of the
minimal models. The wi = K/ hi will here be called the (relative) weights of the minimal
models in the given product. Since K is even when n + r is odd, hr+1 = 2 is automatically
a divisor of K and the weight of the trivial extension is wr+1 = K/2. The smooth CY
n-fold is then obtained by desingularizing the degree K quasi-homogeneous hypersurface
in the weighted projective space of dimension dim(w) 1 with scaling weights wi .
We consider here only the two Gepner models 10242 0/Z12 = 10 10 4 4 0/Z12
and 62 23 /Z8 . By the above algorithm, the first model contains the CY manifold which
is defined classically by resolving the singularities of a quasi-homogeneous polynomial
hypersurface of total degree 12 in a four-dimensional projective space weighted by
(1, 1, 2, 2, 6), hereafter denoted P(1, 1, 2, 2, 6) [14] for short. The second one contains the
geometric data for a degree 8 projective variety, denoted P(1, 1, 2, 2, 2) [8] for short. Both
these varieties are K3-fibrations over P1 . Their quantum geometry has been extensively
investigated using mirror symmetry [14,15], and we will draw heavily on this information.
Open string states satisfying Neumann or Dirichlet boundary conditions are built from
generalized coherent states (Ishibashi states) of a boundary conformal field theory. The
P
coherent basis |iii = n |i, ni U |i, ni is obtained by tracing over the modules labeled
by the primary fields |ii = |i, 0i. The operator U=A,B selects so-called A- or B-type
states by enforcing charge relations between the left and right sectors. Acceptable
string states must satisfy Cardys modular consistency conditions, in this case on the
P
annulus. For A-type states Cardys solutions kiiiA = j Sij /(S0j )1/2|j iiA apply. They
are built from the S-matrices, which for minimal models at height h (whose primary
fields are labelled |ii = |l, m, si), are simple trigonometric functions, whose SU(2)k part
is just Sl,l 0 = 2/ h sin((l + 1)(l 0 + 1)/ h). The construction can be adapted to B-type
states by restricting the sum to run over labels appropriate for this case, and extended to
Gepner models by tensoring such minimal model states [7,8]. Consequently, every detail
about these classes of rational boundary states in Gepner models is known. A- and B-type
boundary states kLMSii are labeled by vectors L, M of SU(2) quantum numbers from
the minimal models in the tensor product, restricted to Li 6 ki /2 by field identification,
as well as a vector S which distinguishes the NS- and R-fields. We shall return to the
classification of boundary states in Gepner models elsewhere, since this will be important
in establishing a complete correspondence between non-perturbative string theory and the
geometry of D-branes in CY spaces.
The microscopic version of the intersection numbers in classical geometry should
be overlap integrals (open string amplitudes) between boundary states kLMSii and
ee
ke
LM
Sii . As explained in Ref. [3] these can be computed by evaluating the Witten index
206
in the Ramond sector. The result depends on the parity of n + r. When n + r is even the
intersection matrix for B-type boundary states is:
I
Be
=B
2hi
r X
r
Y
Y
e

l
(K)

e
e
e
l/2,(
NLj
LM S I LMS B = Ce
e
MM)/2
i=1 li =1
j =1
Lj
je
(1)
P
P
S) is a known constant, M = j wj Mj = 1, . . . , 2K, l = j wj lj , and
where Ce (S e
the SU(2)k fusion coefficients are given by the Verlinde formula
=
NLl e
L
X Sl 0 ,L Sl 0 ,l Se
L,l 0
Sl 0 ,0
l0
The clever trick [3] is to extend the definition of the fusion coefficients in a formal but
natural way (compatible with the Verlinde formula) so that they become periodic over the
extended range of labels. By virtue of the properties of the S-matrices and the Verlinde
formula we have

L ,
L + 2, . . . , L + e
L, L e
1 for l = L e

l
e 2, L e
(2)
= 1 for l = L L
NL,
L+2 ,
L 4, . . . , L + e
e
L
0 otherwise.
Eq. (1) coincides with the formula derived in Ref. [3]. When n + r is odd, we find a similar
expression

ee
I Bo = B e
LM
S I o LMS B
= Co
2hi
r X
Y
i=1 li =1
(K/2)
(1)(MM+l)/K l/2,(MM)/2
e
r
Y
j =1
NLj
ej
jL
(3)
but with half the period in the -function and an additional sign factor.
Gepner models enjoy a large group of discrete symmetries which play a fundamental
role in both the structure of the CFT and its geometrical limits. Since the intersection
matrices are computed at the Gepner points they inherit many of these symmetries. This
can be exploited to give a much simpler representation of the intersection matrix in terms
of two (highly reducible) representations of the abelian group ZK . Define the fundamental
K-dimensional shift matrices:
0 1 0 0 0
(K) =
0
..
.
0
..
.
1
..
.
0 0
1 0
0
0
..
.
0 0
.. ..
. .
0 1
0 0
(4)
KK
and note that + (K) has period K while (K) has period 2K, because (K)K = 1.
For fixed values of L, e
L the intersection form can be regarded as a matrix with rows
e and M. When n + r is even we find that I B is most simply
and columns labelled by M
expressed as a polynomial in g = + (2K)2 , while when n + r is odd we find that I B is
207
most simply expressed as a polynomial in g = (K)2 . Introducing the notation gj = g wj

and the fundamental step-operators from Ref. [6],
Lj /2
tLj = tLt j =
gjl ,
(5)
l=Lj /2
it is possible to prove [6] that the nasty looking expression for I Be displayed above reduces
to:
Y
1
=
nLj e
with
nLj e
.
(6)
ILBe
Lj 1 gj
Lj
Lj = tLj te
L
j
We find exactly the same form also for the odd case, even if the representation involving
is used. The particular representation of the intersection form chosen here emerges
naturally from the BCFT, through the doubling trick for fusion coefficients described
above, but it is highly redundant. There are many non-trivial relations between the set of
boundary states kLMSiiB , and this is reflected in the redundancy encoded in I B . Similarly,
the boundary state itself is redundantly labelled by M, or equivalently the U (1) charges
qj = Mj / h of the minimal model states from which it is built. For the spinless sector we
write: kqiiB = kL = 0; qiiB , while higher spin states (not all independent) are again given
Q
by the step-operators: kL; qiiB = j tLj kqiiB .
The number of moduli of these boundary states are also given by the (extended) fusion
coefficients [3]
B
L =

1 Y
nLj Lj |g| ,
2
(7)
where subtracting compensates for the counting of vacuum states. If n + r is odd, = 2` ,

where ` is the number of Lj s equal to kj /2 [6]. However, if n + r is even, we find that
= 2`1 when ` > 0, and = 1 if ` = 0. Hence if all kj are odd, is always equal to one.
Note that from a CFT point of view we are free to extend the tensor product with
trivial (kj = 0) factors, if we so desire. Thus, although (K)2 provides the simplest and
most obvious representation when n + r is odd, we can also work in the n + r + 1 even
channel where computing the intersection form and number of moduli requires appending
a factor of P (g) = (1 g K/2 )/2, appropriate for a minimal model with no central charge.
Since P+ (g) and P (g) are (orthogonal) projection operators the number of such factors is
immaterial. This appears to be the reason why the calculations reported in [6] are correct.
As already explained, it is the even channel which offers the most direct down-link to
geometry. It is a useful consistency check to compare results from computations in both
channels, and we indeed found agreement.
We now want to transport the CFT data, in particular the charges of the B-type boundary
states, to large radius points on the boundary of moduli space, where they can be compared
with the Ramond charges of D-branes wrapping cycles of large CY manifolds. This can
not be done without specifying some geometric data which select a CY space.
208
3. Bundle data and brane charges

We summarize the relevant D-brane geometry for the K3-fibrations under examination
here. P(1, 1, 2, 2, 2) [8] has been analyzed in great detail using mirror symmetry in
Ref. [14], who also discussed P(1, 1, 2, 2, 6) [14] in somewhat less detail; further explicit
properties of the latter were worked out in Ref. [15]. For each of the two compactifications
the N = 2 special geometry is encoded in a prepotential whose classical part can be written
in the form:
1
1
1
(8)
F (t1 , t2 ) = c111 t1 3 c112t1 2 t2 + (b1 t1 + b2 t2 ) + const,
3!
2
24
where t1 and t2 parametrize the complexified Khler cone: K = t1 J1 + t2 J2 . Here J1 and
J2 are the Khler classes of the K3-fiber and P1 base, respectively. 3 The coefficients are
given in terms of the following topological intersections:
Z
Z
bi = c2 (X) Ji , i = 1, 2.
(9)
cij k = Ji Jj Jk ,
For the threefolds at hand, we have [14]:
P(1, 1, 2, 2, 6) [14] :
c111 = 4,
c112 = 2,
b1 = 52,
b2 = 24,
P(1, 1, 2, 2, 2) [8] :
c111 = 8,
c112 = 4,
b1 = 56,
b2 = 24.
(10)
From the prepotential one derives the symplectic period vector (F0 2F
ti i F , F1 , F2 , 1, t1 , t2 )t , which for P(1, 1, 2, 2, 6) [14] reads:
13 t1 2 t1 3
2
6 + 3 + t2 + t1 t2
13 2t1 2 2 t1 t2
6
1 t1 2
.
(11)
=
t1
t2
The entries (from the top down) correspond to integrations over 6-, 4-, 4-, 0-, 2- and 2cycles, respectively.
We now want to obtain an explicit map between the topological invariants of K-theory
(characteristic classes of the ChanPaton sheaf V ), and brane charges. The latter are
defined as the coefficients of the N = 2 central charge given by
Z(ni ) = n6 F0 + n14 F1 + n24 F2 + n0 + n12 t1 + n22 t2 .
(12)
Following Ref. [6], the topological invariants can be obtained by comparing this to the
central charge of a D-brane configuration
Z
Z
q

K
K
F
(13)
Z Q(V ) = e Q = e Tr e A(X),
r
1
c2 (X), ch3 (V )+ 24
c1 (V )c2 (X)) H 2(X) is the Mukai
where Q = (r, c1 (V ), ch2 (V )+ 24
vector of effective brane charges [16,17]. Specifically, r is the rank of the bundle and
3 More precisely, J = 2L + E and J = L, where E is an exceptional divisor coming from blowing up a curve
1
2
of singularities, and the linear system |L| is a pencil of K3s.
209
chi (V ) are its Chern characters. Inserting the classical periods (11) and substituting the
values of the intersections given in (10), equating (12) and (13) yields:

1
c1 (V ) = n14 J1 + n24 n14 J2
r = n6 ,
2
1
ch2 (V ) = n12 h + n22 `,
ch3 (V ) = n0 n14 2n24 ,
(14)
3
where h = 12 J1 J2 and ` = 12 J1 J1 J1 J2 .
We will be mainly interested in D-branes wrapping the K3-fiber (called fiber branes in
Ref. [17]). This corresponds to considering only states with n14 = n22 = 0, and accordingly
we will set n4 n24 and n2 n12 . It follows from Ref. [6] that the topological invariants of
the torsion sheaf V supported on the K3-fiber can be written in terms of its extension i V
to the threefold as:

(15)
Q = 0, rJ2 , i c1 (V ), i ch2 (V ) + r .
That the vector of induced charges takes this naive form is a consequence of the triviality
of the normal bundle, c1 (K3) = 0, and c2 (K3) = 24; otherwise, there would be extra
correction terms. Comparing the central charges as before, we thus obtain a simple
truncation of (16):
r = n4 ,
i c1 (V ) = n2 h,
i ch2 (V ) = n0 2n4 .
(16)
We are particularly interested in the dimensions of the moduli spaces M of D-brane

configurations on the K3-fiber. Unlike for threefolds, for an isolated K3 such dimensions
(for irreducible bundles) are completely determined by the Mukai vector [18,19]:
K3 (Q) dimc M(Q) = hQ, Qi + 2.
(17)
To write this in terms of the brane charges ni , we need to know what the precise form of
the inner product in (17) is. For this, note that the intersection form of the K3-fiber is easily
exhibited by transforming the periods (11) in the t2 limit to a more suitable basis,
et = (1, t1 , t1 2 , t2 , t2 t1 , t2 t1 2 )t . The symplectic intersection form on the threefold
given by
0 1
then turns into I geom Igeom = 1
0 IK3 , where
0 0 1
1
0 .
(18)
IK3 = 0
2
1 0 0
Accordingly, the Mukai charge vector on K3 takes the form QK3 = (r, c1 (V ), ch2 (V ) +
r) = (n4 , n2 , n0 n4 ), whence:
1
(19)
K3 (ni ) = 2 + n2 2 + 2n0 n4 + 2n4 2 .
2
Analogously, for the threefold P(1, 1, 2, 2, 2) [8] we find that the intersection form on the
fiber is half of the one given in (18), and QK3 = (2r, c1 (V ), ch2 (V ) + r) = (2n4 , n2 , n0
n4 ). We therefore get in this case:
1
K3 (ni ) = 2 + n2 2 + 2n0 n4 + 2n4 2 .
4
(20)
210
4. Matching of geometric and BCFT data

We now relate the BCFT data of Section 2 to the geometrical data of the previous section.
This first of all involves an analytic continuation of the large-radius periods (11) to the
E G ). The continuation matrix m is path
E = (E
Gepner point, i.e., Z = nE
nG m1 ) (m
dependent and thus only defined up to Sp(6, Z) monodromy transformations. It is most
natural to choose, similar to the case in Refs. [6,14], a preferred basis such that the conifold
singularity corresponds to a single wrapped D6-brane.
By applying the method for analytic continuation described in Ref. [14] to the threefold
P(1, 1, 2, 2, 6) [14], we find after some work that for our choice of periods (11) the
continuation matrix takes the form
1 1
0
0
0
0
3
1
1
1
1
32
2
2
2 2 2
1 0
1
0
0
0
.
(21)
m=
1 0
0
0
0
0
1
1
1 0
0
0
2
1
12 21 12
2
The geometric intersection form in the Gepner basis is therefore:
0
1
0 2 0
1
1 0
1
0 2 0
0 1 0
1
0 2
.
I G = m1 I geom m1,t =
2
0 1 0
1
0
0
2
0 1 0
1
1 0
2
0 1 0
1
2
1
2
On the other hand, for P(1, 1, 2, 2, 2) [8] we get 4
1 1
0
0
0
0
3
3
1
1
2
0
0 2 2
2
1 0
1
0
0
0
m=
1 0
0
0
0
0
1
1
1 0
0
0
4
2
4
3
1
1
1
1
1
4
4 2
2 4
4
(22)
(23)
and an expression for I G similar to (22).

Since there is no way the BCFT can know how we will choose the continuation path
for a given large-radius endpoint threefold, the monodromy basis selected by the above
construction of boundary states need not be the same as the Gepner basis associated with
our choice of m. A further change of basis is necessary to link the BCFT boundary state
charges to the geometrical charges at the Gepner point. The precise form of the required
transformation can be determined by comparing the geometric intersection matrix at the
Gepner point (22) with the projection of the Li = 0 BCFT intersection matrix (1), (3) onto
the non-redundant basis.
4 Note that we use a basis different to the one of Ref. [14]; moreover, we found that there are some typing errors
in their monodromy matrices.
211
Table 1
CFT and bundle data of states localized on the K3-fiber of P(1, 1, 2, 2, 6) [14]. On the left we list
the CFT labels Li and the number of vacuum states , as defined in the text after Eq. (7) (note that
in some cases there are two sets of such labels that lead to the same states, and we show only one
of them). In the middle we have three sets of charges, which have the same properties because they
belong to the same Z12 orbit. On the right, B
L is the number of BCFT moduli in the threefold as
given by Eq. (7). Finally, B
is
the
excess number of moduli of the boundary state as
K3
L
compared to the number of K3 bundle moduli K3 given by Eq. (19)
Li
[1, 0, 0, 0]
[3, 0, 0, 0]
[3, 0, 1, 0]
[3, 0, 1, 1]
[5, 0, 0, 0]
[5, 0, 1, 0]
[5, 0, 1, 1]
[5, 0, 2, 0]
[5, 0, 2, 1]
[5, 0, 2, 2]
1
1
1
1
2
2
2
4
4
8
(r, ch2 (V ), c1 (V ))
(1, 0, 0)
(1, 2, 0)
(1, 0, 4)
(3, 6, 0)
(2, 2, 0)
(2, 4, 0)
(2, 0, 8)
(4, 4, 4)
(4, 8, 0)
(4, 0, 12)
(1, 1, 2)
(1, 1, 2)
(1, 3, 2)
(3, 3, 6)
(2, 0, 4)
(2, 2, 4)
(2, 6, 4)
(0, 4, 4)
(4, 4, 8)
(4, 8, 4)
(2, 1, 2)
(0, 1, 2)
(2, 3, 2)
(0, 3, 6)
(0, 2, 0)
(0, 2, 4)
(4, 6, 4)
(0, 0, 4)
(0, 4, 8)
(4, 8, 4)
B
L
1
5
9
21
6
14
30
20
44
64
1
1
1
1
4
4
4
10
10
22
Table 2
Analogous data as in Table 1 for D-branes wrapped on the K3-fiber of P(1, 1, 2, 2, 2) [8]
Li
[1, 0, 0, 0, 0]
[3, 0, 0, 0, 0]
[3, 0, 1, 0, 0]
[3, 0, 1, 1, 0]
[3, 0, 1, 1, 1]
1
1
2
4
8
(r, ch2 (V ), c1 (V ))
(1, 0, 0)
(3, 0, 8)
(2, 2, 4)
(2, 4, 0)
(4, 4, 8)
(4, 8, 0)
(1, 2, 4)
(0, 2, 4)
(2, 0, 8)
(0, 4, 8)
(4, 0, 16)
(3, 2, 4)
B
L
1
7
14
28
56
1
1
4
10
22
In all cases studied in Refs. [3,6] and by us so far, it turns out that I B is to be compared
with (1 g)T I G T t (1 g)t , where T is the appropriate intertwining matrix. Taking
everything together, the geometric brane configurations |ni i are given in terms of the
boundary CFT charges by |ni i = (m1 )t T t (1 g)t kL; qiiB , where ni can be thought of
as functions of the BCFT labels {L; q}.
What we find for the two models considered here are lists of charges that are similar to
those of Ref. [6]. Indeed a similar bundle interpretation can be given for some of them,
however in order not to be repetitious, we refrain from presenting this here. There are
also certain states, like the one with charges (1, 2, 0), for which there is no conventional
bundle interpretation, and this is analogous to the findings of [3].
The major difference of our work and Ref. [6] is that we deal with K3-fibrations rather
than elliptic fibrations. We will thus focus here on the brane features which are specific to
K3 fibrations, and in particular on brane configurations supported on the K3 fiber. Tables 1
212
and 2 list all brane states we have found within this construction that have n6 = n14 = n22 =
0, together with some of their essential characteristics.
5. Discussion
The data in Tables 1 and 2 provide support for the conjectured equivalence of boundary
states and D-branes. We confine ourselves here to the following remarks.
(i) The entries of each row belong to the same ZK orbit (which also contains other
states not localized only on the K3 fiber that we dont show). They thus have the
same properties with regard to the moduli space, even though the brane charges can
be very different. All charges satisfy the BPS condition [17] hQK3 , QK3 i > 2,
with r > 0 or r = 0, c1 > 0 or r = c1 = 0, ch2 (V ) 12 c1 2 c2 < 0. 5
(ii) Consider the charge vectors of the form QK3 = (r, c2 , 0), which can be
interpreted as SU(r) bundles with c1 = 0. For such bundles c2 must obey c2 > 2r.
We indeed find, as a first consistency test, that for all such charges, c2 is equal to
2r, except for r = 2 where we find in addition also c2 = r. However, this can be
given an interpretation as an SO(3) bundle instead of SU(2), so all is well.
(iii) More interestingly, note that for boundary states with one vacuum ( = 1 in (7)), the
number of CFT moduli is one larger than the number of classical K3 moduli:
B
L K3 = 1. This is indeed exactly as expected from geometrical considerations:
the brane system should gain a modulus from the one-dimensional base space (P1
here) when it is embedded as a fiber in a threefold.
(iv) On the other hand, if the number of CFT vacua gets larger, > 1, there is a growing
discrepancy between the numbers of threefold BCFT moduli and classical K3
moduli. However, as suggested in [6], such boundary states may describe not single
branes but collections of several branes, and this would correspond to reducible
bundles or sheaves. For such bundles the geometry of the moduli space is more
intricate than for irreducible ones, as the relative positions of the configurations
correspond to Coulomb branches of additional moduli [20].
By studying the tables one finds consistency between the charge vectors and the values
of . That is, configurations with > 1 can be decomposed into configurations with
smaller i , in a way such that i and the charges add up correctly. 6 For example,
the configurations with charges (2, 2k, 0), k = 1, 2 have = 2, and indeed can be
decomposed [20] into two branes with charges (1, 12 c1 2 (L), c1 (L)). These correspond
to reducible bundles of the form L L1 , where L are line bundles with c1 2 (L) =
2k. The Higgs branches of these U (1) U (1) theories happen to coincide [20] with
the moduli spaces of SU(2) bundles with charges (2, 2k, 0). This illustrates that the
charge label alone does not completely specify the configuration, rather it is that tells
that we have a reducible U (1) U (1) bundle rather than an SU(2) bundle (however,
5 Note that ch (V ) < 0 corresponds to an anti-selfdual gauge connection, and positive brane charges, cf.
2
Eq. (16).
6 Details will be discussed elsewhere.
213
these configurations are continuously connected, so that a distinction is not fundamentally

important.) In contrast, the configuration (3, 6, 0) in Table 1 is irreducible because = 1.
Therefore, it corresponds to an SU(3) bundle with no Coulomb branch.
Some interesting observations can also be made by comparing both tables: for a
given number of CFT vacuum states, we find that the number of excess moduli is
always the same. For example, the state with charges (4, 4, 8) has 44 CFT moduli on
P(1, 1, 2, 2, 6) [14], which is 10 more than what the Mukai formula gives for the K3 alone.
On the other hand, it has 28 moduli 7 on P(1, 1, 2, 2, 2) [8], which is again 10 more than
one would have for the corresponding K3.
The point is that while the absolute numbers of moduli may differ, the excess number of
moduli coming from the fibration is universal and given by 8 = 3 2. This can be given
the following interpretation.
From index theorems we know that K3 = Nm 2(Ng 1),
R
where Nm = 2r c2 is the number of matter fields in six dimensions and Ng is the total
number of gauge fields. For an irreducible bundle we have Ng = dimG + 1, where the shift
by one is due to the U (1) factor that corresponds to the overall center-of-mass modulus of
the brane configuration [3]. For a reducible bundle, which corresponds to a collection of
branes, we have several center-of-mass moduli and thus Ng = dimG + NU (1) , where NU (1)
is equal to the number of irreducible components. Hence we can identify NU (1) = and
express the excess number of moduli as:

B
B
L K3 = L Nm 2(Ng 1)

= B
L (Nm 2 dim G) + 2( 1)
e + 2( 1).
=:
(24)
R
The second term may be thought of as a correction to the Mukai formula K3 = 2r c2
2 dim G, taking into account that we have a reducible bundle if > 1. A comparison with
the CFT result = 3 2 then yields
e=
(i) = ,
(25)
irreducible
components
which means that each irreducible component of the K3 bundle or sheaf gains one extra
modulus upon fibration which is precisely as expected! This is because each such
component corresponds to an independent brane configuration that can sit anywhere over
the P1 base.
Concluding, we note that there is a high degree of consistency and universality in our
results. We therefore believe that these provide a successful test of the BCFT approach to
D-branes.
Our findings suggest in particular that the moduli space of the K3 remains protected even
when embedded in a threefold with less supersymmetries, and its dimension is still given by
7 Note that the two K3s have different intersection forms, so the Mukai formulas (19) and (20) yield different
numbers of moduli.
8 This is a general property of the BCFT, as will be shown elsewhere.
214
the Mukai formula (17), if we subtract the degrees of freedom associated with the fibration
over the P1 base. While this robustness may not be too surprising in an adiabatic, large base
limit of the threefold, it is more impressive here as we look into a region of the threefold
moduli space (the Gepner point) where the classical geometry is maximally distorted by
quantum corrections. It appears that nothing serious happens to the branes supported on
the K3-fiber at large radius when we transport them all the way to the Gepner point of the
CY moduli space. One may argue that there is an N = 4 subsector in the N = 2 BCFT that
protects these states from instanton corrections also in the non-geometric regime.
Acknowledgements
We would like to thank Jrg Frhlich, Jrgen Fuchs, Michael Gutperle, Graham Ross,
Christoph Schweigert, and especially Peter Mayr for useful discussions, and Emanuel
Diaconescu for illuminating correspondence.
References
[1] H. Ooguri, Y. Oz, Z. Yin, D-branes on CalabiYau spaces and their mirrors, Nucl. Phys. B 477
(1996) 407, hep-th/9606112.
[2] C. Vafa, Extending mirror conjecture to CalabiYau with bundles, hep-th/9804131.
[3] I. Brunner, M.R. Douglas, A. Lawrence, C. Rmelsberger, D-branes on the quintic, hepth/9906200.
[4] M.R. Douglas, Topics in D-geometry, hep-th/9910170.
[5] D. Diaconescu, J. Gomis, Fractional branes and boundary states in orbifold theories, hepth/9906242.
[6] D. Diaconescu, C. Rmelsberger, D-branes and bundles on elliptic fibrations, hep-th/9910172.
[7] A. Recknagel, V. Schomerus, D-branes in Gepner models, Nucl. Phys. B 531 (1998) 185, hepth/9712186.
[8] J. Fuchs, C. Schweigert, Branes: From free fields to general backgrounds, Nucl. Phys. B 530
(1998) 99, hep-th/9712257.
[9] M. Gutperle, Y. Satoh, D-branes in Gepner models and supersymmetry, Nucl. Phys. B 543
(1999) 73, hep-th/9808080.
[10] M. Gutperle, Y. Satoh, D0-branes in Gepner models and N = 2 black holes, Nucl. Phys. B 555
(1999) 477, hep-th/9902120.
[11] S. Govindarajan, T. Jayaraman, T. Sarkar, World sheet approaches to D-branes on supersymmetric cycles, hep-th/9907131.
[12] A. Klemm, W. Lerche, P. Mayr, K3 fibrations and heterotic type II string duality, Phys. Lett.
B 357 (1995) 313, hep-th/9506112.
[13] P.S. Aspinwall, J. Louis, On the ubiquity of K3 fibrations in string duality, Phys. Lett. B 369
(1996) 233, hep-th/9510234.
[14] P. Candelas, X. De La Ossa, A. Font, S. Katz, D.R. Morrison, Mirror symmetry for two
parameter models. I, Nucl. Phys. B 416 (1994) 481, hep-th/9308083.
[15] S. Kachru, A. Klemm, W. Lerche, P. Mayr, C. Vafa, Nonperturbative results on the point
particle limit of N = 2 heterotic string compactifications, Nucl. Phys. B 459 (1996) 537, hepth/9508155.
[16] M.B. Green, J.A. Harvey, G. Moore, I -brane inflow and anomalous couplings on D-branes,
Class. Quant. Grav. 14 (1997) 47, hep-th/9605033.
215
[17] J.A. Harvey, G. Moore, On the algebras of BPS states, Commun. Math. Phys. 197 (1998) 489,
hep-th/9609017.
[18] S. Mukai, Symplectic structure of the moduli of sheaves on an abelian or K3 surface, Invent.
Math. 77 (1984) 101.
[19] S. Mukai, On the moduli space of bundles on K3 surfaces, I, in: Vector Bundles on Algebraic
Varieties, Tata Institute of Fundamental Research.
[20] M. Bershadsky, V. Sadov, F -theory on K3 K3 and instantons on 7-branes, Nucl. Phys. B 510
(1998) 232, hep-th/9703194.

On seven-brane and instanton solutions of type IIB

Martin B. Einhorn 1 , Leopoldo A. Pando Zayas 2
Randall Laboratory of Physics, The University of Michigan, Ann Arbor, Michigan 48109-1120, USA
Abstract
It is shown that magnetic seven-branes previously considered as different objects are members of
a one-parametric family of supersymmetric seven branes. We enlarge the class of seven-branes by
constructing new magnetically and also electrically charged seven-branes. The solutions display a
kink-like behavior. We also construct a solution that naturally generalizes the D-instanton. 2000
Elsevier Science B.V. All rights reserved.
PACS: 11.27.+d; 11.25.Sq
Keywords: D-branes; Solitonic; Supersymmetric; Extended solutions; Instantons
1. Introduction
Solitonic solutions of supergravity have played a crucial role in achieving our current
understanding of nonperturbative string theory. The identification of a subset of these
solutions with the strong coupling limit of D-branes was central to the resolution of
longstanding puzzles in black hole physics and large N gauge theories. Although in general
the properties of these solutions are well known (see reviews [14]), it is fair to say that
some solutions have been explored less than others. Such is the case of 7-branes and
instantons in type IIB. Seven-branes play a central role in different recent developments
in nonperturbative string theory. Two examples are the instrumental role in the formulation
of F-theory [6] and in the construction of supersymmetric gauge theories [7,8]. The role
of the type IIB instanton has been widely discussed in the context of the AdS/CFT
correspondence.
In this paper we attempt to remedy this situation by widening the class of seven-brane
solutions 3 and showing that the D-instanton of Gibbons, Green and Perry [5] admits a
very natural generalization. We also hope to clarify some misconceptions in the literature
1 meinhorn@umich.edu
2 lpandoz@umich.edu
3 Some papers considering different aspects of seven-branes include [9,10].
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 0 2 - 3
M.B. Einhorn, L.A. Pando Zayas / Nuclear Physics B 582 (2000) 216230
217
by showing how some solutions that have been regarded as independent of each other
are actually particular cases of a one-parametric family of supersymmetric seven-brane
solutions.
In Section 2 we study various seven-brane constructions. After reviewing the most
general solution of the seven-brane ansatz and showing how it reduces to previously known
solutions we study seven branes from a holomorphic point of view. This is a distinguishing
feature of seven-branes that have been widely exploited in nonperturbative constructions.
We show that holomorphic seven branes are very similar to the nonlinear O(3) model used
to describe isotropic ferromagnets. In particular we show explicitly that by imposing very
reasonable asymptotic conditions on the solution, one is led to a homotopical classification
for the charge of the seven-brane, which is not possible for D-branes in general. In the two
last subsections of Section 2 we construct new seven-brane solutions; we consider both
electric and magnetic seven-branes. The gauge field of the electrically charged seven-brane
displays a kink-like behavior. In Section 3 we discuss the supersymmetry of seven branes
and show that these equations admit the one-parametric family of solutions constructed at
the beginning of Section 2. In Section 4 we construct generalizations of the D-instanton
of [5].
2. On seven-branes
2.1. The various D7-branes
In this subsection we review the known seven-brane solutions and clarify a common
misconception about the D7-brane. We argue that three seven-brane constructions found in
the literature are basically the same object. The seven-brane constructed implicitly in [1,3]
along the lines of the traditional p-brane solution, the GGP seven brane [5] and a circular
seven-brane constructed in [12] are all, to an extent to be specified shortly, particular cases
of a unique construction. Let us start by reviewing the general p-brane construction in
10-dimensions. One starts with the following Einstein-frame action [1,3]

Z
3p
1
1
1
10
2
2
e 2 Fp+2 .
d x g R ()
(1)
SI I,p =
16G
2
2(p + 2)!
The equations of motion (EOM) for p = 1 are:
RMN =

1
M N + e2 FM FN ,
2
1
M g g MN N = e2 g MN FM FN ,
g
M ( g g MN e2 FN ) = 0.
(2)
Here, FM = M a, where a is the zero-form, RR potential. Unlike the quantum theory, the
classical theory is invariant under a rescaling of the action, and, as the Lagrangian has no
mass parameter, this feature is critical for understanding some of the properties of classical
solitons, such as D-branes, in the absence of other scales.
218
One expects the D7-brane to be a singular, magnetic solution of SII,1 and the
corresponding ansatz for the metric and field strength is [13]
ds 2 = e2A(r) dx dx + e2B(r)mn dy m dy n ,
yn
(3)
Fm = mn 2 or a = ,
r
q
with , = 0, . . . , 7, m, n = 1, 2; 12 = 1; r = y12 + y22 ; is the angular coordinate in
the transverse space. This ansatz for the metric is dictated by preserving P8 SO(2), that
is, Poincar invariance in the longitudinal directions and SO(2) in the perpendicular ones.
One very useful relation, which is true for all Dp-branes and is required by supersymmetry
is Ad + B d = 0, where d = p + 1 and d = 10 p 3 [1] (see appendix for the general
definition). Applied to seven-branes (d = 10 8 2 = 0), one has A = 0. The Ricci tensor
in this case becomes Rmn = mn 2 B (see Appendix). Einsteins equations then imply
e
,
r
1
B0
= ( 0 )2 .
B 00 +
(4)
r
2
The first of these equations is characteristic of magnetic p-brane solutions and says that
exp() is a harmonic function in the transverse coordinates. The general solution to the
second equation can be written as
0 =
1
ln(1 + ),
(5)
2
where we have defined ln(r/r0 ), a notation that we shall find useful throughout the rest
of this paper. By a rescaling of y m , we may choose B0 = 0, whence the entire background
takes the form

ds 2 = dx dx + e2(B1 +1) (1 + ) d 2 + d 2 ,
B = B0 + B1 +
e = 1 + ,
yn
(6)
Fm = mn 2 .
r
For sufficiently small r, viz., r < r1 r0 exp (1/), the quantity 1 + can become
negative, and the solution breaks down, since of course exp() must be nonnegative. This
also shows up as a singularity in the metric and, more importantly, as a naked singularity
in the curvature:
R=
2
.
r 2(B1 +1) (1 + )3
(7)
In this respect, the 7-brane differs from lower p-branes, whose singularities do not
occur at finite r. This disturbing feature of the solution can and should be interpreted
as the appearance of new physics in a natural way: the SUGRA Lagrangian, being
nonrenormalizable, must be understood as the first term of an effective field theory
involving an expansion in powers of the curvature so, as the curvature becomes large,
219
the terms ignored become as important as the leading term. The value of 1 is arbitrary
because the Lagrangian is scale-invariant. However, by reference to other forms of matter,
1/8
it is natural to expect such a breakdown on a scale
r G . Similarly, from the superstring
point of view, one expects corrections at r 0 . Moreover, since exp() is the (local)
string coupling constant, this becomes large as r approaches r1 , so that one cannot trust
the classical approximation anyway. Therefore, even though in principle r0 is simply an
integration constant, we assume that it is sufficiently small so that our solution makes
sense physically at distances r r1 .
For B1 = 0, one has the typical harmonic function representation for p-branes [1
3]. For B1 = 1, one recovers the circular seven-brane of [12]. 4 It is not possible to
establish a one-to-one relation with the solution of [5]. One can however show that the
asymptotic behavior of the solution of [5] can be obtained from Eq. (6). 5 It is very natural
to concentrate on the large r asymptotic behavior because most of the constraints that string
theory imposes through the boundary state formalism on supergravity solutions depend
only on the asymptotic behavior. The notational correspondence between the asymptotic
behavior of the dilaton and the axion field given in [5] and the fields of Eq. (6) are:
1
1
,
b= .
(8)
=
2
r0
The asymptotic value of the dilaton b appears in Eq. (28) of [5].
To establish the relation between the metrics we now turn to Eq. (27) in [5] (see next
subsection)
1
(9)
ln = ln 2 ,
2
where, in our notation, ln = B and 2 = exp (). The general solution to this equation
is e2B = 2 exp(F (z) + F (z)), where F is an arbitrary holomorphic function. Rotational
invariance in the transverse dimensions leads to the choice (6). Further specification
within this class amounts to different choices of the parameter B1 , as shown before.
The conditions imposed in [5] were that the metric must be real, modular invariant and
nondegenerate (for an extensive discussion see [14]). This brings us to exp(2B) = (1 +
ln(r/r0 ))2 2 , where is the Dedekind eta function, which is perfectly compatible with
(6) for asymptotically large r. To guarantee nondegeneracy of the metric in the presence of
multiple seven-branes, other powers of r must be included but this is still compatible with
(6).
2.2. Comments on holomorphic seven-branes
In the case p = 1 the action (1) can be rewritten as

Z
1
1
10
d x g R 2 ,
SII,1 =
16G
22
(10)
4 To obtain precisely Eq. (6.14) in [12] it is necessary to make the following change: ln r r and to identify
= m.
5 The limit we consider here has been previously discussed in the literature (see, for example, [13]). This limit
is known as the weak coupling limit of the GGP solution.
220
where 1 + i2 = a + ie . One interesting property of the equation of motion

following from this action is that, for the type of metric we are considering here (A = 0) the
dilaton and axion equations are independent of the metric. This decoupling from the metric
signals a complex structure that has been exploited in various contexts starting from [14].
The equation of motion for is
+

2
= 0,

(11)
where z y 1 + iy 2 and z . One can easily see that any holomorphic or antiholomorphic function is a solution to this equation. This particular class of solutions is required
for supersymmetric solutions (see below). Einsteins equations take the form

1
+
= ln 2 ,
(12)
2( )2
2
where the last equation is true only for holomorphic or antiholomorphic . At this stage
one can conclude that the seven-brane-like solutions to the EOM obtained from (10) are
characterized by an arbitrary holomorphic function F (z)
=
B
e(2B) = 2 e2(F +F ) .
(13)
This situation is familiar, it is typical of systems with two spatial dimensions, such is
the case of the nonlinear O(3) model [15] that describes isotropic ferromagnets. In our
problem, however, we have to go a step further because not any holomorphic function is
acceptable as opposed to the case of the O(3) model.
The domain in which is defined presumably carries some information about the
symmetries of the theory. The idea follows from the fact that in supergravity belongs
to the upper half plane H and the symmetry of the theory is SL(2, R). This symmetry is
expected to be broken at the string theory level due to charge quantization to a discrete
subgroup, SL(2, Z), which is presumed to be a discrete gauge symmetry. In that case,
must be restricted to the fundamental domain; this was the path taken in [5] which
is asymptotically compatible with Eq. (6) and raises the natural question of how much
supergravity solutions know about the full string theory. We will explore the restrictions
that must be applied to such that a well-defined magnetic charge exists at the end of this
subsection.
If one insists on a metric preserving P8 SO(2), which is the usual ansatz for the pbrane metric, then using Einstein Eq. (13), one has that 2 = 2 (r 2 ). The condition that the
sum F + F be a function of r 2 = zz poses a functional equation stating that the sum of
F and F must be expressed as a product of their arguments. This equation is solved for
= 0 we get: i a/
z + a 0 z + i20 z = 0, where
F (z) = B1 ln z. Using holomorphicity of :
the dot means derivative with respect to and the prime with respect to r 2 . For purely
z). Since 20 is a function of only
magnetic solutions (a 0 = 0), one has that 20 = a/(z
2
r = zz and a is a function of only = (i/2) ln z/z, the only solution is a = which
implies 2 = 1 + ln r. This is essentially the solution presented in Eq. (6).
The converse is however not true. For any magnetically charged holomorphic sevenbrane: i a/(2
z) + i20 z 2 /(2z) = 0. Since a and 2 are real, the only solution to this
221
equation is 2 = 0 (i.e., 2 is independent of the angle). For the same reasons, a = const =
and 2 = exp() = 1 + . This, however, is not enough to fix the form of F, and the
solution will in general depend on an arbitrary holomorphic function.
The analogy with the O(3) model can be pursued a step further to establish that the
charge is equal to the degree of the map z. For exp(2i ) = bzn + c, one has that
for asymptotically large r, a = n/(2) which implies q = n, the same relation one
has for the O(3) model. This means that the charge is determined by the homotopy class
of the map z and provides the homotopic sector classification known for solitons and
instantons in gauge theories. This is a distinctive feature of seven-branes, the gauge fields
in D-branes are U (1) gauge fields and therefore the only chance to get some nontrivial
homotopy is for two-dimensional transverse space since k (U (1)) = 0 for k > 1. It is
interesting to note that, as in solitons and instantons in gauge theory, the long distance
behavior determines the topological class of the solution on very general grounds and that
one need not know the details of the solution for all values of the radius.
2.3. Electric kink-like seven brane
In this section we construct an electric 6 seven-brane with a metric somewhat resembling
the one of the circular seven-brane constructed in [12].
We know from general considerations [5,11] that the charge dual to the D7-brane charge
(called magnetic here) is the D(-1)-brane charge, i.e., the Type IIB-instanton charge that
couples locally to the zero-form a field. The instanton, as a classical solution of the EOM,
only exists in Euclidean spacetime, although of course it represents a quantum tunneling
amplitude in a Hilbert space built up in a spacetime with Lorentz signature. The charge
that the solution we present here carries plays a prominent role in the D-instanton solution
of [5]. One interpretation of this charge will be evident at the end of this section.
We will look for solutions of the truncated IIB supergravity Lagrangian discussed in
the previous sections. The main difference will be that, in the ansatz for the metric, we
will not assume A = 0 as we did previously, so, as mentioned earlier, such a solution will
not be supersymmetric and not BPS. 7 The reason for starting with this general ansatz is
that for A = 0, there are no electric solutions. From the Einstein equations Eq. (2) and
the general form of the Ricci tensor (see Appendix) one has, in the transverse directions,
Rmn = mn 2 B, but the right hand side of Einsteins equations is ym yn /r 2 (( 0 )2 +
e2 (a 0 )2 ) which has a different tensor structure and is signaling that the equation can not
be nontrivially satisfied. We will, therefore, consider the following ansatz
ds 2 = e2A(r) dx dx + e2B(r)mn dy m dy n .
(14)
The EOM are given by Eq. (2). The Einstein equation for g implies that
2 e8A = 0, or more explicitly,
6 Here the term electric is used exclusively in the sense that a = a(r) and, unlike the previous case, the
R
conserved charge S 1 e2 (r a) is nonzero.
7 Hence, for strong coupling, we can expect quantum corrections to be large.
222
1
A00 + A0 + 8(A0 )2 = 0.
r
The solution of this harmonic equation is
(15)
1
ln ,
(16)
8
where, as before, = ln(r/r0 ). The Einstein equation in the direction transverse to the
brane has two different tensor structures: mn and y m y n /r 2 . The mn equation gives
A A0 =
1
8
B 00 + B 0 + 8A0 B 0 + A0 = 0.
r
r
Solving this equation is straightforward, giving [16]
B B0 = + B1 ln ,
(17)
(18)
here B1 is an arbitrary constant. Using the axion equation, one can rewrite the remaining
Einstein equation (for y m y n /r 2 ) in terms of the dilaton only

1 + 1 0
7 2
+ ( 0 )2 = 4B1 +
.
(19)
00 +
r
4 r2
This equation may be linearized by replacing by exp() and further simplified by
defining t ln , yielding
d2
e = 2 e .
(20)
dt 2
Remarkably, this is invariant under translation of ln(r/r0 ) (or rescaling of ), but this
is clearly not a symmetry of the other equations. The solution of this equation may be
expressed in a variety of ways, e.g.,

1
0
+
,
(21)
e = cosh((t t0 )) =
2 0
where = 4B1 + 7/4, 0 is a constant and t = ln . Up to this point, we actually have

not found it necessary to have 2 > 0, but from this solution, we see that this is required in
order to have exp() > 0 for all t. 8
The general solution for the metric and the axion is, up to a global SL(2, R)
transformation,
ds 2 = ()1/4 dx dx + r02 2B1 (d 2 + d 2 ),

,
a = tanh ln
0
(22)
where 0 is a constant. This has a form very similar to the circular seven-brane of [12].
It is easy to see that | | = 1, so that the solution corresponds to a classical solution in
the -plane that starts at = 1 and traverses a semicircle until arriving at = +1.
We would like to stress at this point, although it will be completely clear in what
follows, that the solution presented here, which we called electric seven brane because
8 This corresponds to the restriction B > 7/16.
1
223
the charge that it carries is a Noether charge, i.e., it is conserved as a consequence of the
equations of motion, is not a supersymmetric solution. In the absence of supersymmetry
or a topologically conserved charge, it is not at all clear that this solution is stable. One
of the properties of the charge of the electric seven brane presented here is that it is
invariant under the SL(2, R) transformation acting on the dilaton and action fields, viz.,
(a + b)/(c + d). This means that the electric charge can be viewed as an SL(2, R)
charge. A similar situation takes place for the instanton solution of [5]. In this context the
electric seven brane and the D-instanton are two SL(2, R)-charged objects.
2.4. Magnetic seven-branes
In this subsection we will construct general magnetic seven-branes. The ansatz for the
metric and the field strength is exactly as at the beginning of this section, Eq. (3). The
Einstein equation of motion in the longitudinal directions to the brane is as in the previous
subsection and therefore the solution for A is as for the electric 7-brane Eq. (16), A A0 =
(1/8) ln ln r/r0 . The off-diagonal Einstein equation

e2 2
A0
00
16(A0 )2 + 32A0 B 0 .
=
16
A
(23)
( 0 )2
r2
r
The two diagonal equations are identical
B0
1 0 2
( ) + 8A00 + B 00 +
+ 8(A0 )2 8A0 B 0 = 0.
(24)
2
r
The axion equation, as is always the case for magnetic solutions, is identically satisfied.
The dilaton equation becomes
0
2 e2
+ 8A0 0 =
.
(25)
r
r2
A convenient way to solve these equations is, once again, to change variables to =
ln r/r0 . In this new coordinate the dilaton equation becomes
00 +
00 +
0
= 2 e2 ,
(26)
where the prime now denotes the derivative with respect to . Using Eqs. (23), (24) takes
the following simple form in terms of
B 00 +
1 2 e2
B0
+ +
= 0.
(27)
The previous two equations can be combined using the identity ( 0 )0 / = 00 + 0 / into
B0 +
0 t0
= 1,
2
(28)
where t0 is an integration constant. These results can be plugged into Eq. (23) giving
00
0
(16t0 + 7)
( 0 )2 =
.
4 2
(29)
224
A linearization procedure, similar to the one discussed in the electric case, can be applied
to this equation with the following modification
d 2 t
e
= 2 et ,
dt 2
where again t = ln 0 .
The complete solution takes the following form

(d 2 + d 2 ),
ds 2 = 1/4 dx dx + r02 2t0 +1 sinh ln
0

e = eo sinh ln
,
0
a = ,
(30)
(31)
with the following algebraic constraint, e0 = , 2 = 4t0 + 11/4. We could just set
0 = 0 so that becomes . We shall now show that there is a limit in which this
solution becomes the solution presented in Eq. (6). In order to do so let us consider the
following substitution 8 ln(r0 ) = 1/CO . Using this substitution and renaming some of
the integration constants we can rewrite
A A0 =
1
ln(1 + C0 ln r),
8
1
ln sinh(x0 + ln(1 + C0 ln r)),
2
(32)
0 = ln(1 + Co ln r) ln sinh(x0 + ln(1 + C0 ln r)).
B B0 = ln r + (t0 + 1/2) ln(1 + C0 ln r)
The limit to the extremal case Eq. (6) is: C0 , x0 0 and , e2B0 , e0 in such a
way that C0 2 , C0 /x0 , e2B0 x0 and exp(0 x0 ) are constant. This can be easily arranged
by sending to zero in the following relations
p
1
B1 + 1
. (33)
, x0 = B1 + 1 , e2B0 = e0 =
C0 = 2 , 2 =
2
B1 + 1
This way we recover precisely the solution of Eq. (6).
3. Comments on supersymmetric seven-branes

In this section we will analyze the condition under which a seven-brane solution admits a
supersymmetric generalization. The supersymmetry transformations of the corresponding
fermionic fields in type IIB supergravity can be presented in an SU(1, 1) [17,18] (see also
the recent review [19]) or SL(2, R) invariant formulation [2023].
The dilaton and axion fields of type IIB theory parametrize the upper half of the complex
plane using the traditional parametrization = a + ie . Following [1722], one can
introduce the zweibein V ,

1
V V+1
,
(34)
V=
V2 V+2
and define the local U (1) action as

i
0
e
,
V V
0
ei
225
(35)
where is a U (1) phase. It is convenient to parametrize the matrix V following [21,22]

i

1
e
ei
.
(36)
V=
ei
2i2 ei
One can fix the local U (1) gauge symmetry by setting the scalar field to be a function of
: = ( ).
To write the type IIB supergravity EOM, it is convenient to introduce two SL(2, R)
singlet currents [1722],
i M 2i
e ,
2 2
1 M 1
.
QM = i V M V+ = M
2 2
PM = V+ M V+ =
(37)
Under the U (1) gauge symmetry (35), they transform as

PM PM e2i ,
QM QM + M .
(38)
These transformations show that QM is a composite gauge potential and that the U (1)
charge of PM equals 2.
The equations that result from setting all the fermions to zero as well as all bosonic fields
except the relevant ones are [1720]
PN ,
RMN = PM PN + PM
D M PM = (M 2iQM )PM = 0.
(39)
Substituting (37) into these, the EOM can be expressed in terms of and a as

1
M N + e2 M aN a ,
2
1a + 2 M M a = 0,
RMN =
1 e2 (a)2 = 0.
(40)
Note that does not appear in the EOM. In fact it appears as a multiplier in the dilaton and
the axion equations but this does not affect the EOM. It should be taken as a bonus that
these equations can be derived from a Lagrangian because the original definition of IIB is
solely based on supersymmetry and the EOM. The Lagrangian in question is precisely the
one we have been studying given by Eq. (1).
The supersymmetry transformations of the dilatino and the gravitino M are given by
= iPM M ,

i
M = M QM .
2
(41)
226
From the supersymmetric variation of the dilatino it is easy to see that a supersymmetric
solution must have holomorphic or antiholomorphic . Namely,
=

e2i B 1
e2i B 1
1 + 1 2 2 .
1 + 2 2 =
22
22
(42)
Taking 1 2 = i one obtains that in order for the variation of the dilatino to be zero,
(1 + i2 ) = 0, which means that is a holomorphic function. The same analysis could
be carried out for antiholomorphic functions.
For the gravitino one has that the variation in the longitudinal directions is
= eAB m m A,
(43)
here we have assumed that all fields depend only on the transverse directions. One could
apply the same type of manipulation we applied to the dilatino transformation and conclude
that a solution to this equation is provided by any antiholomorphic function A, but A is a
metric entry and must, therefore, be real. The only choice we have is A = constant which,
by means of a trivial coordinate redefinition, is equivalent to A = 0. So far, everything
is compatible with the type of seven-branes considered in Section 2.1. The gravitino
transformation in the transverse directions is

1
1
i
i
m 1 .
(44)
m = m + n B m n m B m +
2
2
2
42
Substituting B = B0 +B1 ln r + 12 ln 2 , taking into account that is a holomorphic function
and considering a more general spinor = eif/2 0 where 1 2 0 = i0 , we obtain that
f + iB1 ln r must be a holomorphic function or in other words, f = B1 which
can always be satisfied for any given = ( ). In the case of the D7-brane of [1,3] one
has B1 = 0 which is a solution for = 0 (the preferred gauge fixing condition) and f = 0.
The solutions of [5] and [12] also enter in this scheme. To make contact with [5] we
need to drop the condition that B = B(r) and assume only that B must be real. In this
case the solution is specified up to a holomorphic function Eq. (13). The variation of the
gravitino in the directions perpendicular to the seven-brane demand that f + i(F + F )
be a holomorphic function. In [5] ( ) = Im ln( + i) up to an additive constant, this
combines with the fact that F = F ( ) is dictated by modular invariance, to provide a
supersymmetric solution for f = 0.
4. Instantonic solutions to IIB

In this section we will consider the following Euclidean action

Z
1
1 2
1
10
2
2
d x g R () + e (a)
S=
16G
2
2
and its EOM
RMN =

1
M N e2 FM FN ,
2
(45)

1
M gg MN N = e2 g MN FM FN ,
g
M gg MN e2 FM = 0.
227
(46)
One way to generalize the solution of [5] is to consider the following ansatz for the metric
ds 2 = e2B(r)mn dy m dy n .
(47)
As familiar by now, the Einstein equations have two tensor structures. The mn part is
17 0
(48)
B + 8(B 0 )2 = 0,
r
which becomes a linear differential equation after substituting B 0 = 1/u. The general
solution is

Q0
1
(49)
B B0 = ln 1 + 16 .
8
r
B 00 +
Note that for Q0 = 0 and fixing the constant B0 = 0 we recover the flat metric of the
solution obtained in [5].
The dilaton and axion equation take the following form
9
00 + 0 + e2 (a 0 )2 + 8B 0 0 = 0,
r
9
a 00 + a 0 + 2a 0 0 + 8B 0 a 0 = 0.
r
(50)
Using the axion equation one obtains that y m y n /r 2 Einstein equation can be rewritten as

288 16 0 2
00
2
0 9
0
+ 8B =
r (B ) .
(51)
+ () +
r
Q0
In terms of a new function D such that 0 = (D 4B 0 r 9 )/r 9 and also changing to
coordinate t: dt/dr = 1/r 9 the dilaton equation becomes
D + D 2 =
512Q0
.
(1 + 64Q0 t 2 )2
(52)
The solutions of this equation take different forms depending on the sign of Q0 . For
negative Q0 , substituting Q0 Q20 in all formulas we obtain the following solution:

Q20
2
ds = 1 16 dr 2 + r 2 d92 ,
r

3 1 + Q0 /r 8
e = e0 sinh x0 + ln
,
2 1 Q0 /r 8

3 1 + Q0 /r 8
0
coth x0 + ln
.
(53)
a a0 = e
2 1 Q0 /r 8
The limit in which this solution coincides with [5] is the following: Q0 , x0 0 and
exp 0 with the following relations going to constants: Q0 /x0 = C/3, e0 x0 = e0 .

(Here C and 0 are the parameters of the solution in [5].) This solution has a singularity at
228
r08 = Q0 . It is a nontrivial task to classify this singularity being a singularity in Euclidean

space. Let us simply note that it is a curvature singularity since for this metric the scalar
curvature equals

Q20 9/4
288Q20
.
1
R=
r 18
r 16
There is also a solution for positive values of the parameter Q0 , for this case we will
consider Q0 Q20 . The solution we obtained is

Q2
ds 2 = 1 + 160 dr 2 + r 2 d92 ,
r

3Q20 CQ20
Q20 3/2
C
0
,
1 + 8 16 24
1 + 16
e =e
r
r
3r
r
q

C 2 + 9Q20 C 3r 16 Q2
C 3r 16 Q20
0
0
1
1+ 8
.
(54)
a=e
3r 8
3r r 16 3Q20
r 16 3Q20
To recover the instanton of [5] from this solution one needs simply send Q0 0.
The Einstein equations of SII,1 (both for Minkowskian and Euclidean versions) imply
that the action is zero. This naive approach would render an instanton with zero action.
First of all one would turn to the surface term. The surface term originates from allowing
the surface but whose normal derivatives do not. The
variations of g MN
at
R which vanish
0
9
boundary term is (K K ) h d y where the surface may be taken to be the 9-sphere of
radius r. For the dilaton metric we consider here one obtains a contribution proportional
to O(1/r 8 ) which therefore vanishes for asymptotically large distances. A way out of this
situation which was successfully used in [5] consists in realizing that the same solution can
be considered as a solution of the supergravity but instead of considering an axion field one
needs to consider the C8 formulation with field strength F9 dual to the field strength of the
axion da. In this picture Einstein equation no longer implies that the solution must have
zero action. There is a more rigorous argument as to why the Euclidean version for the
axion field has to include the surface term, it is based on an analysis of the dual formulation
[5,24,25]. A subtlety here arises because of the need to perform Poincar duality to objects
that live in spaces of different signatures. Modulo these subtleties our result for the value of
the action is exactly that of [5]. We checked that our solution becomes that of [5] in certain
limit specified by the charges. It is straightforward to notice that this limit corresponds
precisely to the large r limit of the solution in the positive Q0 case and for negative Q0 we
need to assume that x0 = 0.
5. Conclusions
In this paper we have constructed new electric and magnetic seven-branes. We have
shown that the EOM for the seven-brane ansatz admit a one-parameter family of solutions
some particular cases of which were known in the literature as distinct objects. We
229
have also shown explicitly that the supersymmetry equations are flexible enough to
admit a solution for such one-parametric family. The electric and magnetic seven-branes
constructed here have a kink-like behavior similar to some solitons in gauge theories.
We have constructed instantonic solutions for IIB that generalize in a very clear way
the previously known D-instanton solution of [5]. We have not, however, analyzed the
supersymmetric structure of the instanton.
Acknowledgements
We wish to thank M. Duff for pointing out a solution we overlooked at the early stage.
We would also like to thank J.T. Liu and J.X. Lu for very interesting discussions and
several clarifying remarks. We would also like to acknowledge useful correspondence from
Y. Lozano and M. Gutperle. One of us (L.P.Z.) would like to thank the Office of the Provost
at the University of Michigan for support. Both of us wish to acknowledge the support of
the High Energy Physics Division of the Department of Energy.
Appendix
The calculations for the Ricci tensor for the p-brane metric have been presented in a
number of papers [1,3,4]. We present such results here for the sake of completeness. Let
us consider a metric in D spacetime dimensions, given by the following ansatz
ds 2 = e2A dx dx + e2B dy m dy n mn ,
(55)
where x ( = 0, . . . , d 1) are the coordinates of the (d 1)-brane world volume, and y m

are the coordinates of the (D d)-dimensional transverse space. Generally the functions
A and B depend only on r = y m y m . Here we will consider an arbitrary dependence in

the transverse dimensions. The Ricci tensor obtained from this metric is

R = e2(AB) mn m n A + d m An A + d m An B ,

Rmn = mn pq p q B + d pq p Aq B + d pq p Bq B
d m n B d m n A d m An A + d m Bn B
+ d(m An B + n Am B),
(56)
where d = D d 2. A convenient choice of vielbein basis for the metric is e =

eA dx and em = eB dy m , where underlined indices denote tangent space components. The
corresponding spin connection, which is needed for the supersymmetry transformation of
the gravitino, is
n = eB n A e ,
= 0,
mn = eB n B em eB m B en .
(57)
230
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
M.J. Duff, R.R. Khuri, J.X. Lu, Phys. Rept. 259 (1995) 213, hep-th/9412184.
D. Youm, Phys. Rept. 316 (1999) 1, hep-th/9710046.
H. Lu, C.N. Pope, E. Sezgin, K.S. Stelle, Nucl. Phys. B 456 (1995) 669, hep-th/9508042.
K.S. Stelle, BPS branes in supergravity, hep-th/9803116.
G.W. Gibbons, M.B. Green, M.J. Perry, Phys. Lett. B 370 (1996) 37, hep-th/9511080.
C. Vafa, Nucl. Phys. B 469 (1996) 403, hep-th/9602022.
O. DeWolfe, T. Hauer, A. Iqbal, B. Zwiebach, Nucl. Phys. B 534 (1998) 261, hep-th/9805220.
O. DeWolfe, A. Hanany, A. Iqbal, E. Katz, JHEP 03 (1999) 006, hep-th/9902179.
P. Meessen, T. Ortin, Nucl. Phys. B 541 (1999) 195, hep-th/9806120.
E. Eyras, Y. Lozano, Exotic branes and nonperturbative seven branes, hep-th/9908094.
Polchinski, String Theory, Cambridge Univ. Press, 1998.
E. Bergshoeff, M. de Roo, M.B. Green, G. Papadopoulos, P.K. Townsend, Nucl. Phys. B 470
(1996) 113, hep-th/9601150.
A. Dabholkar, Lectures on orientifolds and duality, hep-th/9804208.
B.R. Greene, A. Shapere, C. Vafa, S. Yau, Nucl. Phys. B 337 (1990) 1.
R. Rajaraman, Solitons and Instantons, North-Holland, 1982.
C.M. Bender, S.A. Orszag, Advanced Mathematical Methods for Scientist and Engineers,
McGraw-Hill, 1978.
J. Schwarz, Nucl. Phys. B 226 (1983) 269.
P.S. Howe, P.C. West, Nucl. Phys. B 238 (1984) 181.
P.C. West, Supergravity, brane dynamics and string duality, hep-th/9811101.
J.H. Schwarz, P.C. West, Phys. Lett. B 126 (1983) 301.
M.B. Green, S. Sethi, Phys. Rev. D 59 (1999) 046006, hep-th/9808061.
M.R. Gaberdiel, M.B. Green, JHEP 11 (1998) 026, hep-th/9810153.
M.B. Green, hep-th/9903124.
M.B. Green, M. Gutperle, Nucl. Phys. B 498 (1997) 195, hep-th/9701093.
K. Behrndt, I. Gaida, D. Lust, S. Mahapatra, T. Mohaupt, Nucl. Phys. B 508 (1997) 659, hepth/9706096.

On the effective N = 1 supergravity of M-theory

J.-P. Derendinger, R. Sauser
Institute of Physics, University of Neuchtel, CH-2000 Neuchtel, Switzerland
Abstract
In the low-energy limit, M-theory compactified on S 1 /Z 2 is formulated in terms of Bianchi
identities with sources localized at orbifold singularities and anomaly-cancelling counterterms to
the Wilson effective Lagrangian. Compactifying to four dimensions on a CalabiYau space leads to
N = 1 local supersymmetry. We derive a formulation of the effective supergravity which explicitly
relates four-dimensional supergravity multiplets and field equations with these fundamental Mtheory aspects. This formulation proves convenient for the introduction in the effective supergravity
of non-perturbative M-theory contributions. It also applies to the universal sector of generic
compactifications with N = 1 supersymmetry. 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
M-theory compactified on
O7 = X6 S 1 /Z 2 ,
(1.1)
where X6 is a CalabiYau three-fold, leads to a four-dimensional theory with N = 1 local

supersymmetry. At present, our knowledge of the resulting effective supergravity is based
on the few aspects of M-theory which are quantitatively understood and on the smallorbifold limit 1 which is the perturbative heterotic E8 E8 string compactified on the
CalabiYau space X6 . In the low-energy limit, M-theory information can be organized as
an expansion in powers of the eleven-dimensional gravitational constant 11 [1,2]. The
2
is eleven-dimensional supergravity [3]. In the case of a compactification
lowest order 11
1
on S /Z 2 only, the next orders in 11 are known to include orbifold plane contributions
Research supported in part by the Swiss National Science Foundation.

E-mail addresses: jean-pierre.derendinger@iph.unine.ch (J.-P. Derendinger), roger.sauser@iph.unine.ch (R.
Sauser).
1 The limit in which the length of the S 1 /Z orbifold direction is small compared to the size of the CalabiYau
2
manifold.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 9 6 - 0
232
J.-P. Derendinger, R. Sauser / Nuclear Physics B 582 (2000) 231258
(super-YangMills terms) as well as gauge and gravitational anomaly-cancelling terms [1,

2,4].
Similarly, the effective four-dimensional supergravity 2 can be formulated as an
expansion in the four-dimensional gravitational constant , even if a more common
choice suggested by string theory is to use the dilaton as expansion parameter. The
two options coincide provided the appropriate supersymmetric description of the dilaton
is adopted in a Wilson effective Lagrangian formulation. The lowest order 2 is the
O7 truncation of eleven-dimensional supergravity. The next order includes super-Yang
Mills, charged matter kinetic and superpotential contributions. Then come sigma-model
anomaly-cancelling terms contributing in particular to gauge threshold corrections. These
first corrections to the low-energy limit of compactified M-theory are identical to those
obtained from heterotic compactifications on CalabiYau. This is certainly expected since
the information content is identical. The literature gives a detailed description of these
results, with particular attention paid to the strong-coupling heterotic limit in which the
size of the CalabiYau space is smaller than the orbifold length, supersymmetry breaking
by gaugino condensation and non-standard embeddings [519].
Our goal in this paper is to provide a derivation of the effective supergravity which
explicitly relates four-dimensional supergravity statements with M-theory aspects like
Bianchi identities modified at singularities and anomaly cancellation. We reformulate these
basic facts of M-theory on O7 directly in terms of four-dimensional supermultiplets and
equations. For instance, Bianchi identities from M-theory are promoted to field equations,
as constraints on multiplets which are massless modes of M-theory bulk fields. With this
formulation, we expect to obtain a clean, direct derivation of the effective supergravity
suitable to cases more subtle than the universal modes of M-theory on O7 . A first use of our
formalism will be the coupling of five-brane moduli supermultiplets in the O7 case [20].
The organization of the paper is as follows. In Section 2, we establish our basic
supergravity formulation, using the bosonic bulk dynamics as starting point. The resulting
Lagrangian is the lowest order in the -expansion. It is essentially defined by a dynamical
Lagrangian involving tensor fields supplemented by Bianchi identities which are field
equations of the theory. We discuss in detail the bosonic component expansion of
the supermultiplet action, the question of the gravity frame and the generation of a
superpotential. In Section 3, we introduce the next order corrections: gauge multiplets
and charged matter contributions. We show that their introduction is controlled by a
simple modification of the four-dimensional Bianchi identities, in analogy with the
appearance of Z 2 fixed planes contributions in the M-theory Bianchi identities. Section 4
discusses anomaly-cancelling terms. We begin by modifying the four-dimensional effective
supergravity by adding terms similar to those appearing for gauge threshold corrections
in (2, 2) compactifications of the heterotic string. These modifications can be formulated
in terms of our particular set of multiplets directly related to M-theory bulk degrees
of freedom. We then directly compute these anomaly-cancelling terms by KaluzaKlein
reduction of the ten-dimensional GreenSchwarz counterterms arising from M-theory on
2 Which is a low-energy description.
233
S 1 /Z 2 . Section 5 gives some final comments and conclusions, and an appendix contains
our notations and conventions.
2. Four-dimensional effective M-theory supergravities: bulk dynamics

Our concern is compactifications of M-theory to four dimensions preserving N = 1
supersymmetry. Or compactifications in which supersymmetry would break spontaneously
or dynamically at solutions of the effective field equations. As a consequence, the light
(massless) modes can be described by a local effective N = 1 supergravity Lagrangian, to
be understood in the sense of Wilson.
These solutions are not as well understood as, for instance, CalabiYau compactifications of heterotic strings and it will prove useful to give a precise description of the aspects
which are better known. We will then focus on a precise description of two aspects which
are of importance in compactifications of M-theory. Firstly, we will use a supersymmetric description of the Bianchi identities verified by antisymmetric tensors, as arising from
M-theory or higher-dimensional supergravities. This procedure allows to avoid multiplet
ambiguities arising when duality transformations are performed at the bosonic level only.
Secondly, we will use a formulation which leaves explicitly the choice of gravity frame
open. This can be a relevant issue since an expansion, perturbative or not, is performed
around a gravitational background which selects a gravity frame. Standard Poincar supergravity is usually written in the Einstein frame, in which the gravitational Lagrangian is
21 2 eR. Corrections to the lowest order effective action, which includes this gravitational
term, induce in general (but not always) corrections to the gravitational Lagrangian which
affect the Einstein frame condition.
Keeping open the choice of gravity frame suggests to use superconformal supergravity.
And displaying Bianchi identities explicitly in an effective Lagrangian requires using
chiral, linear or vector supermultiplets with constraints.
The purpose of this first section is to establish our procedure by considering the wellknown bulk dynamics, which follows from O7 compactification of eleven-dimensional
supergravity.
2.1. Superconformal formalism
We use the superconformal formulation of N = 1 supergravity with a chiral compensating multiplet S0 to generate Poincar theories by gauge fixing. 3 Its conformal and chiral
weights are taken as w = 1 and n = 1. This formalism is particularly convenient to keep
control of a change of frame 4 which corresponds to a different Poincar gauge condition
applied on the modulus of the scalar compensator z0 , which fixes dilatation symmetry. Up
to two derivatives, a supergravity Lagrangian is written as 5
3 This is old minimal Poincar supergravity [21].
4 Mostly the so-called Einstein or string frames.
5 Except otherwise mentioned, our notation is as in Ref. [22] where also reference to the original literature can
be found.
234

1
L = S0 S 0 D + S03 W F + fab W a W b F .
(2.1)
4
The symbols [. . .]D and [. . .]F denote respectively the invariant D- and F -density formulae
given by (all fermion contributions are omitted)

1
(2.2)
[S]F = e(f + f ),
[V]D = e d + cR ,
3
where V is a vector multiplet with components (c, , m, n, b , , d) and S a chiral
multiplet with components (z, , f ). The real vector multiplet with zero weights is a
function (in the sense of tensor calculus) of the multiplets present in the theory, including in
general the compensating multiplet. The holomorphic function W of the chiral multiplets
is the superpotential. The last term contributes to gauge kinetic terms (the chiral multiplet
W is the gauge field strength for the gauge multiplets) and involves a holomorphic gauge
kinetic function fab of the chiral multiplets. 6 Expression (2.1) provides the most general
supergravity Lagrangian up to terms with more than two derivatives and up to terms which
would contribute to kinetic terms in a fermionic background only [23,24]. Besides S0 and
W, we will use chiral multiplets with zero weights and neither W nor fab will depend on
the compensator.
The chiral U (1) symmetry of the superconformal algebra can be extended to

(2.3)
S0 , W, S0 , 3 W, ( )1 S S ,
0
with an arbitrary chiral multiplet . This symmetry is at the origin of Khler invariance
of Poincar supergravity. The last transformation suggests that log transforms as the
corresponding gauge connection. Choosing = W 1/3 eliminates the superpotential except
if it vanishes. One can then use a U (1)/Khler gauge fixing in which the supergravity
Lagrangian (2.1) reads

1
(2.4)
L = S0 S 0 D + c S03 F + fab W a W b F ,
4
with an arbitrary constant c as superpotential and two arbitrary functions and fab .
The real function depends on matter multiplets, which will be either chiral multiplets
like the CalabiYau universal modulus T , or real linear multiplets (with weights w = 2,
n = 0) like the dilaton multiplet in the version of the theory with an antisymmetric tensor,
or real vector multiplets (n = 0, w arbitrary) like the multiplet of gauge potentials (w =
0). Vector multiplets will appear as essential ingredients in the effective description of
(N = 1)-preserving M-theory five-branes [20].
2.2. Bulk Lagrangian
The lowest order (in the expansion) effective four-dimensional supergravity of Mtheory compactified on O7 describes KaluzaKlein massless modes of eleven-dimensional
supergravity. It is the S 1 /Z 2 truncation of eleven-dimensional supergravity on a Calabi
Yau three-fold.
6 Gauge kinetic terms may also arise from the first term.
235
In the version given by Cremmer, Julia and Scherk (CJS) [3], the Lagrangian of elevendimensional supergravity can be written as 7

1
1
GM1 M2 M3 M4 GM1 M2 M3 M4
e1 LCJS = 2 R
2 4!
211

1 1
1 M1 ...M11
e
GM1 M2 M3 M4 GM5 M6 M7 M8 CM9 M10 M11

6 4!4!3!
+ fermionic terms.
(2.5)
Omitting all fields related to the detailed geometry of the CalabiYau manifold, the particle
content of the four-dimensional theory is the N = 1 supergravity multiplet, with metric
tensor g , and matter multiplets including (on-shell) four bosons and four fermions.
Two bosons are scalars and correspond to the dilaton and the universal modulus of the
CalabiYau space, the massless volume mode. Two bosons are KaluzaKlein modes of the
four-form field G, with Bianchi identity dG = 0. Explicitly, these fields and their Bianchi
identities can be written as
G4 ,
Gj k4 = iT j k ,
[ G 4] = 0,
[ T] = 0.
(2.6)
It will prove useful to identify these fields with the vector components of two real vector
multiplets V and VT , and to impose the Bianchi identities as field equations using the
appropriate multiplets as Lagrange multipliers. The bulk supergravity Lagrangian takes
then the form

3/2 1/2
1
V
S + S V + LT VT .
(2.7)
LB = S0 S 0 VT
2
D
The vector superfield V (weights w = 2, n = 0) includes in its components the vector
field v G 4 for which the Bianchi identity is v = 0. This condition is a
component of the (super)field equation of the chiral S (w = n = 0) which imposes V =
L, a real linear multiplet (w = 2, n = 0). Secondly, the vector multiplet VT (w = n =
0) includes in its components T . The supersymmetric extension of its Bianchi identity
[ T] = 0 is enforced by the real linear multiplet LT , which implies VT = T + T , with
a chiral weightless multiplet T . The usefulness of obtaining Bianchi identities via field
equations will become apparent with the introduction of higher orders in the expansion.
At this stage of the discussion however, it gives a formulation of the familiar duality relating
scalars and antisymmetric tensors or, for superfields, chiral and linear multiplets.
Solving in Eq. (2.7) for the Lagrange multipliers S and LT leads to the standard form
of the bulk four-dimensional Lagrangian [25,26]
1
b 3/2 1/2
L
,
LB,l = S0 S 0 eK/3
D
2
(2.8)
or, as defined in Eq. (2.1),

7 We use the flat spacetime metric (, +, +, , +). The gravitational constant has dimension (mass)9/2
11
and the three-index tensor field CMNP is dimensionless. Notations and conventions are defined in Appendix A.
236

=
2L
S0 S 0
1/2
eK/2 .
(2.9)
The Khler potential for the volume modulus T is

b = 3 log(T + T ).
K
(2.10)
We will see again below that this standard form is naturally obtained by direct reduction
of the CJS version of eleven-dimensional supergravity on O7 . Clearly, theory (2.8) is also
b
the CalabiYau truncation of ten-dimensional N = 1 pure supergravity [25]. Notice that K
can be regarded as the Khler connection for symmetry (2.3), with transformation
b
K
b + 3 log + 3 log ,
K
(2.11)
such that S0 S 0 eK/3 is chiral/Khler invariant.

A theory with a linear multiplet is in principle dual to an equivalent Lagrangian with
the linear multiplet replaced by a chiral one. In our case, solving for V and LT in
expression (2.7) leads to

3
S0 S 0 eK/3 D ,
2
b = log(S + S ) 3 log(T + T ).
K = log(S + S ) + K
LB,c =
(2.12)
This familiar chiral form [27] is not the most useful as long as one insists on the fourdimensional translation of eleven-dimensional Bianchi identities.
Notice that one can obtain another equivalent form of the Lagrangian (2.7) by choosing
to solve for S and VT . In this case, the CalabiYau modulus is described by a linear
multiplet LT . This form will not be useful since it is known that one-loop string corrections
in general break the chiral-linear duality for this modulus: they involve holomorphic
functions of T in a F -density which are intrinsically chiral [28]. Finally, there is an
obstruction when trying to solve for V and VT and one cannot write an expression in
terms of the chiral S and the linear LT .
Before turning to explicit component expressions, we should discuss the choice of
Poincar frame, and introduce the expansion in the four-dimensional gravity coupling ,
which effectively corresponds to the low-energy expansion of M-theory in powers of the
eleven-dimensional gravitational constant 11 .
2.2.1. Einstein frame
To gauge-fix dilatations, we impose as usual a condition on the Einstein term appearing
in the superconformal supergravity Lagrangian. According to the component expression
for the D-density and the tensor calculus of superconformal multiplets [22], the Einstein
term included in [S0 S 0 ]D is [21,26]

2
1
1X
wi Ci
,
(2.13)
LE = eR z0 z0
2
3
2
Ci
i
237
where wi is a Weyl weight, the sum is taken over the linear (wi = 2) and vector (wi
arbitrary) multiplets, and z0 , and Ci are the lowest, scalar components of, respectively,
S0 , , and of the vector or linear multiplets. 8
Applied to the bulk Lagrangian (2.7), expression (2.13) leads to

1
(2.14)
LE = eR (z0 z0 CT )3/2 (2C)1/2 .
2
As they should, the terms introduced to impose Bianchi identities do not contribute. The
Einstein frame is then selected by the dilatation gauge condition
2 = (z0 z0 CT )3/2 (2C)1/2.
(2.15)
It will be convenient to introduce the (composite) real vector multiplet

= (S0 S 0 VT )3/2(2V )1/2 ,
(2.16)
with conformal weight two. In the Poincar theory and in the Einstein frame, its lowest
component is precisely equal to 2 . With this definition, the bulk Lagrangian becomes
simply

(2.17)
LB = (S + S)V + LT VT D ,
and the equation of motion for V (the chiral-linear duality equation)
2V (S + S ) =
(2.18)
indicates that the Einstein Lagrangian also reads

LE = (2C Re s)eR.
In the Einstein frame (Planck units), Re s = (4 2C)1 .
Eq. (2.18) is compatible with the standard relation of heterotic strings 9 2 2hRe si = 0
if one identifies 2hCi = 1/ 0 . This equation defines string units, in which
LE =
e2
eR,
0
(2.19)
with a dilaton given by e2 = Re s.

2.2.2. Modified Bianchi identities and -expansion
Compactification of M-theory on S 1 /Z 2 is commonly discussed in an expansion in
powers of 11 . Compactification on O7 can similarly be formulated with as expansion
parameter. In the upstairs version, Bianchi identities are modified at the ten-dimensional
planes fixed by S 1 /Z 2 . Suppose now that we modify the four-dimensional supersymmetric
Bianchi identities of the bulk Lagrangian in the following way:

(2.20)
LB L = (S + S )(V + V ) + LT (VT + T ) D ,
8 We use in general the same notation for the lowest component of and the multiplet itself. Also, in expression
(2.13), we use the component expansion of vector multiplets indicated below [Eq. (2.23)], which differs in its
highest component from Ref. [22].
9 We use h i for a background value.
238
with two composite vector multiplets V (w = 2, n = 0) and T (w = n = 0). Solving for

the Lagrange multipliers now leads to
V = L V ,
VT = T + T T .
The Lagrangian to first order in these modifications reads then

3
V
T
L = LB
2V
2 VT
D

3
= LB (S + S )V
( T ) ,
2VT
D
(2.21)
with V and VT respectively replaced by L and T + T in these expressions. The multiplets

V and T , with canonical dimension w = 2, appear at order 0 0 , in comparison
with bulk terms of order 2 . This is the relation with the expansion in powers of 11
of M-theory in the low-energy limit. In M-theory compactification, the multiplets V and
T can thus be obtained either by considering the modified Bianchi identities on O7 ,
formulated as in Eq. (2.20), or from corrections to the Lagrangian of eleven-dimensional
supergravity on O7 , as in expression (2.21).
2.2.3. Component expressions
To analyze the Lagrangian (2.7), we will need to define some notations. Since we
will only explicitly consider the bosonic sector of the theory, all fermions in the N = 1
supermultiplets will be omitted. Since also we are concerned with Poincar supergravity,
we will immediately gauge-fix the superconformal symmetries not contained in N = 1
Poincar supersymmetry, with one exception, dilatation symmetry: we want to keep the
freedom of a frame choice as explicit as possible. These assumptions imply in particular
that superconformal covariant derivatives reduce in general to Dc = D 2i nA for
a complex field with chiral weight n and to
1
c = + w R,
6
(2.22)
if w is the Weyl weight and real. The gauge boson A of chiral U (1) symmetry is
auxiliary and D and would be covariantized with respect to Poincar symmetries.
We will use the following components for the various superconformal multiplets
appearing in Lagrangian (2.7):

V = C, 0, H, K, v , 0, d C 13 CR ,

VT = CT , 0, HT , KT , T , 0, dT CT ,

S = s, 0, f, if, i s, 0, 0 ,

e
LT = `T , 0, 0, 0, t , 0, `T 13 `T R , t = t ,
2

(2.23)
S0 = z0 , 0, f0 , if0 , iDc z0 , 0, 0 .
The role of the Lagrange multipliers S and LT follows from
239

e1 (S + S )V D = 2( Im s)v 2 (Re s C) + 2d Re s
f (H iK) f (H + iK)
= 2 Im s v + 2d Re s f (H iK) f (H + iK)
+ derivative,
e
[LT VT ]D = `T (dT CT ) ( T )t + derivative.
(2.24)
2
Solving for the components of S leads to v = d = H = K = 0, and V is a linear
multiplet. Solving for the components of LT leads to dT CT = [ T] = 0, and VT can
be written as T + T , with a chiral multiplet T (zero weights). 10
Since one can always write v = 6e v , we have generated with Im s and t the
Bianchi identities
e
[ v ] = [ T] = 0.
A modification of these Bianchi identities, as induced by S 1 /Z 2 compactification or
by five-brane couplings will then be phrased as a modification of the supermultiplets
appearing multiplied by S + S or LT in Eqs. (2.24).
To complete the identification of the four-dimensional supergravity (2.7) with the modes
of eleven-dimensional supergravity we need its complete bosonic expansion, which after
solving for the chiral U (1) auxiliary field A reads:
1
1
3
e1 LB = R + C 2 v v CT2 T T t T + 2 Im s v
2
4
4
1
3
C 2 ( C)( C) CT2 ( CT )( CT )
4
4

3
1
1
1
+ d C 2 Re s + (dT CT ) `T CT
2
2
1
+ ( )[ log C + log ]
2
+ e1 LAUX. + derivative,
(2.25)
where = (z0 z0 CT )3/2 (2C)1/2 , and

1
3
e1 LAUX. = C 2 (H + iK)(H iK) + CT2 (HT + iKT )(HT iKT )
4
4
(2.26)
+ f (H iK) + f (H + iK).
The last equality is obtained after solving for the f0 component of S0 . 11
The above component expansion of the bosonic Lagrangian is useful because it explicitly
displays the dependence on the gauge choice for dilatation symmetry. The gravitational
constant is the field-dependent quantity . Choosing a Poincar frame amounts to impose
the value of this quantity, and to use this condition to eliminate z0 . Notice also that the
choice of the phase of z0 is a gauge condition for the chiral internal U (1) superconformal
10 The components are: C = 2 Re T , T = 2 Im T , H = 2 Re f , K = 2 Im f .
T
T
T
T
T
11 Note that the result would be different with a superpotential.
240
symmetry. With the exception of z0 , all bosons are U (1)-neutral. As a consequence, the
bosonic Lagrangian only depends on the modulus of z0 .
Taking the Einstein frame, = 2 , and solving for the components of S and LT
leads to

1
1
e1 LB = 2 R 2 C 2 ( C)( C) v v
2
4

3 2
(2.27)
2 CT ( CT )( CT ) + T T ,
4
with v = 2e b since V is a linear multiplet, CT = 2 Re T and T = 2 Im T
since VT = T +T . This Lagrangian is to be compared with the reduction of the CJS version
of eleven-dimensional supergravity (2.5). The Z 2 orbifold projection eliminates all states
which are odd under x 4 x 4 , and since we disregard massless modes related to the
detailed CalabiYau geometry, the reduction of the D = 11 spacetime metric is
2
e e
g
0
0
(2.28)
gMN =
0 .
0
e2 e2
0
0
e ij
The SU(3)-invariant tensor ij refers to complex coordinates on the CalabiYau space.
The surviving components of the four-index tensor GMNP Q are only G4 and Gij 4 ,
with
G4 = 3[ C]4 ,
Gij 4 = Cij 4 ,
Cij 4 = ia(x)ij ,
and the four-dimensional Lagrangian for these fields reads

i
1
1 h
e1 LCJS = 2 R 2 9( )( ) + 16 e6 G4 G4
2
4

3
2 ( )( ) + e2 ( a)( a) .
4
In this expression, is the four-dimensional gravitational coupling
(2.29)
2
/V7 ,
2 = 11
V7 = V1 V6 being the volume of the compact space S 1 X6 .

At this stage, the identification of the bosonic components C and b of V , CT and T
of VT with the bulk fields , , C4 and a can only be determined up to a proportionality
constant for each multiplet. We will define these constants later on from the couplings of
C and CT = 2 Re T to charged matter and gauge fields, to obtain:
2 3
2
e
,
4 2b = C4 ,
V6
V6
2
2
T = 2 a = 2 Im T .
CT = 2 e = 2 Re T ,
V6
V6
4 2 C =
(2.30)
The quantity is the gauge coupling constant on the Z 2 fixed planes and 2 /V6 is a
dimensionless number which will actually never appear in the four-dimensional effective
theory.
241
2.2.4. Bianchi identities and symmetries

In the bulk Lagrangian (2.7), the terms [(S + S)V + LT VT ]D impose in particular the
Bianchi identities (2.6). They are certainly invariant under
V
VT
V + L,
VT + T + T ,
L linear,
T chiral.
These symmetries are the supersymmetric extensions of the gauge invariances of Bianchi
identities, G4 = 3[ ] and Gij 4 = i ij . Solving for S and LT implies then
that V and VT are pure gauge, V = L and VT = T + T . The last equation defines VT up
to a holomorphic redefinition of T , T f (T ). This redefinition is a symmetry of the bulk
Lagrangian if the function simultaneously transforms as in (2.3), with a holomorphic
function of T . The equation for the invariance of S0 S 0 VT is
f (T ) + f (T ) =
T +T
,
(T )(T )
and its solution is clearly Sl(2, R) symmetry,
aT ib
, ad bc = 1,
(2.31)
T f (T ) =
icT + d
the modular invariance of T (T-duality), extended to a continuous symmetry at the lowest
order.
This chiral symmetry is generically anomalous: in the presence of a N = 1 super-Yang
Mills sector, with or without chiral matter, mixed anomalies arise in the triangle diagram
for two gauge bosons and one connection 3 log(T + T ) for Sl(2, R) symmetry. This
anomaly is cancelled in particular by a GreenSchwarz mechanism as was demonstrated
in the effective Lagrangian description of gauge thresholds [24] calculated at one-loop
for (2, 2) compactifications of the heterotic strings [28]. We will see below that this
phenomenon is also a useful tool in the construction of effective supergravities of M-theory
compactifications.
2.2.5. Superpotential
The standard reduction of eleven-dimensional supergravity with unbroken N = 1
supersymmetry does not generate a superpotential. This fact is however not a direct
consequence of the eleven-dimensional Bianchi identity or of the CalabiYau and S 1 /Z 2
symmetries. In principle, the Bianchi identity [M GNP QR] = 0 allows a solution
Gij k4 = 2i 1 hij k ,
Gij k4 = 2i 1 hij k .
(2.32)
In these equations, h is a constant chosen real and ij k is the SU(3)-invariant CalabiYau

1
GMNP Q GMNP Q leads then to a contribution
tensor. The Lagrangian term 2e2 48
11
e
4 CT3 (2 2 C)h2
in the four-dimensional effective supergravity. This contribution corresponds to the

addition of a superpotential term
[ihS03 ]F
242
to the bulk Lagrangian, a contribution which however breaks supersymmetry [29]. Since
we have insisted in writing Lagrangians in which all Bianchi identities are field equations,
we prefer instead to use

(2.33)
U (W + W ) D + S03 W F .
The field equation of the vector multiplet U (weights w = 2, n = 0) implies that the
chiral multiplet W (w = n = 0) is an arbitrary imaginary constant, which can be zero
and supersymmetry stays unbroken, or non-zero.
With the addition (2.33) of a superpotential, the bulk Lagrangian takes its final offshell form

(2.34)
LB = (S + S )V + LT VT + U (W + W ) D + S03 W F ,
in which the Bianchi identities of eleven-dimensional supergravity are translated into field
equations of the Lagrange multipliers S, LT and U . At this stage, the introduction of
these multiplets is not fascinating. This approach encodes simply the (Poincar) dualities
relating antisymmetric tensors (in linear multiplets) and scalars (in chiral multiplets), and
the Bianchi identity for the superpotential is trivial. But this procedure will prove useful
and informative in the forthcoming sections.
3. Fixed planes: gauge and matter contributions

In this section, we start with the well-known effective N = 1 four-dimensional
supergravity for symmetric (2, 2) compactifications of heterotic strings. We then rewrite
this theory in a form where explicit Bianchi identities allow a direct comparison with O7
compactification of M-theory, in the so-called upstairs formulation [1,2].
The dependence on charged matter (in chiral multiplets collectively denoted by M,
with w = n = 0) and gauge multiplets (vector multiplet A, in the adjoint representation,
with w = n = 0) of the effective supergravity theory for CalabiYau compactifications of
heterotic strings is well-known [27,30,31], at least for the universal matter multiplets
arising from the simplest CalabiYau modes of the ten-dimensional super-YangMills
fields. Information on the non-trivial harmonic modes is more subtle [32], as for generic
CalabiYau moduli. In the chiral formulation, Eq. (2.12) becomes

3
1
Lc = S0 S 0 eK/3 D + SWW + S03 W ,
2
4
F
K = log(S + S ) 3 log(T + T 2MeA M),
W = M 3 .
(3.1)
For notational simplicity, we omit traces over the gauge group representation and their
normalization factors. The chiral multiplet W is the gauge field-strength for A (w = n =
3/2). Since the gauge group is in general not simple,
X
ca W a W a ,
(3.2)
WW =
a
243
with a (real) coefficient ca for each simple or abelian factor. 12 The superpotential should
be understood as a gauge invariant trilinear interaction with coupling constant defined
as an integral over the CalabiYau space. In the linear multiplet version, the equivalent
expression is [25,26]

1

Ll = (S0 S 0 )3/2 L 1/2 eK/2 D + S03 M 3 F .
2
(3.3)
With respect to Eq. (2.8), gauge and matter dependence arises in modifications of the linear
and of K:
b the new modulus and matter Khler potential is
multiplet L (to L)
b = 3 log(T + T 2MeA M),
K
(3.4)
instead of Eq. (2.10) and

L = L 2,
(3.5)
where (A) is the ChernSimons vector multiplet (w = 2, n = 0), defined by

=
X
a
ca a ,
( a ) =
1 a a
W W .
16
13
(3.6)
Insisting as before on Bianchi identities, both forms (3.1) and (3.3) are equivalent to
h
L = (S0 S 0 VT )3/2 (2V )1/2 (S + S )(V + 2)

i

+ S03 W F
+ LT (VT + 2MeA M) + U (W M 3 ) + c.c.
D
h
i

= (S + S )(V + 2) + LT (VT + 2MeA M) + S03 (ih + M 3 ) F .
(3.7)
D
Supersymmetric vacua have h = 0. As before, solving in the last expression for S and LT
imposes respectively V = L 2 = L and VT = T + T 2MeA M, leading to Eq. (3.3).
Alternatively, solving for V and LT leads back to the chiral form (3.1), with the tensor
calculus identity

1X a
c [SW a W a ]F + derivative,
2 (S + S ) D =
4 a
(3.8)
which follows from Eq. (3.6) and the definition of the F -density, [(. . .)]F = [. . .]D .
This reformulation of the gauge invariant Lagrangian suggests some remarks. Firstly,
it enhances the importance of ChernSimons multiplets in superstring effective actions:
gauge fields and matter fields couple to the bulk Lagrangian using a ChernSimons
multiplet. The gauge ChernSimons multiplet is defined by Eq. (3.6), which indicates
that its chiral projection () is the chiral multiplet for the kinetic super-YangMills
12 Corresponding to KacMoody levels in superstrings. All coefficients can be equal to one, as with the standard
embedding, but our discussion is not affected by their presence.
13 In global Poincar supersymmetry, () = 1 DD. A linear multiplet is defined by the condition (L) =
4
0.
244
Lagrangian. Similarly for chiral matter, the kinetic WessZumino Lagrangian can be
written as [S0 S 0 MeA M]D = [(S0 S 0 MeA M)]F , defining
M = S0 S 0 MeA M
(3.9)
as a matter ChernSimons multiplet (w = 2, n = 0) which then couples to S0 S 0 VT as

couples to V .
Secondly, the ChernSimons vector multiplet (A) is not gauge invariant: its variation
is a linear multiplet. Then, the variation of [(S + S)]D is a derivative and V remains
gauge invariant. When solving for S, it simply follows that L is gauge invariant and that
the linear multiplet transforms as
L = 2.
(3.10)
Finally, expression (3.7) shows that all gauge and chiral matter contributions can be
viewed as the supersymmetrization of modified Bianchi identities imposed by S, LT
and U . This is equally true in the ten-dimensional supergravityYangMills system: the
curl of the antisymmetric tensor field is modified by ChernSimons contributions which
are supersymmetry partners of the super-YangMills Lagrangian [33]. This observation
provides the link to the approach based on M-theory on O7 , in which the Z 2 -fixed planes
carrying the YangMills fields induce because of supersymmetry modifications to the
Bianchi identity of the four-form field strength of eleven-dimensional supergravity.
In the effective supergravity of M-theory on O7 (upstairs formulation), the various
components of the Lagrangian (3.7),
h
L = (S0 S 0 VT )3/2 (2V )1/2 (S + S)(V + 2)

i

+ S03 W F ,
+ LT (VT + 2MeA M) + U (W M 3 ) + c.c.
D
have the following origin. As already discussed at length, the first term is the bulk
supergravity contribution. Then [(S + S)(V + 2)]D is the supersymmetrization of the
Bianchi identity verified by the component G4 of the four-form field, modified by gauge
contributions on the fixed planes. Similarly, [LT (VT + 2MeA M)]D and [U (W M 3 ) +
c.c.]D are respectively the supersymmetric extensions of the Bianchi identities of Gj k4
and Gij k4 , when fixed plane contributions are included. Thus, all fixed plane contributions
are given at this order by the supersymmetrization of Bianchi identities, as obtained by
direct O7 truncation of the eleven-dimensional identities [1,2].
At this point, the gauge coupling constant for each simple or abelian factor a in the
gauge group appears to be
ca
1
a
,
=
c
Re
s
=
4C
ga2
(3.11)
s and C being respectively the lowest scalar component of the chiral S and the vector
V (or the linear L). At this order, ga is the tree-level wilsonnian and physical 14 gauge
14 The coefficient of 1 F a F a in the generating functional of one-particle irreducible Greens functions.
4
245
coupling. The second equality is the lowest component of the equation of motion of the
vector multiplet V , Eq. (2.18). In the Einstein frame, Re s = (4 2C)1 .
It is clear, as already observed [517], that as far as the structure of the fourdimensional effective supergravity is concerned, the same information follows from O7
compactification of M-theory at the next to lowest order in the expansion and from
CalabiYau compactifications of the heterotic strings, at zero string loop order.
Notice that Eq. (3.11) defines ca Re s as the coefficient of gauge kinetic terms. It defines
then this field in terms of the gauge kinetic action on the ten-dimensional Z 2 fixed planes,
Z
1
d 10 xe10 tr FAB F AB ,
(3.12)
Sgauge = 2
4
M10
reduced on X6 . This action is also at the origin of charged matter kinetic terms, which in
the effective supergravity read
b
6e
e 2K
(D M )(D M) = 2 (D M )(D M) + .
2
MM
CT
The CalabiYau reduction of the action (3.12) provides then the identification of C and
CT in terms of the bulk fields and appearing in the metric tensor (2.28). These results
have already been displayed in Eq. (2.30).
4. Anomaly-cancelling terms
In the ten-dimensional heterotic string, cancellation of gauge and gravitational anomalies is a one-loop effect in string or effective supergravity perturbation theory. In the
low-energy effective action description, we should then distinguish the Wilson effective
supergravity from the standard effective action S0 , defined as the generating functional
of one-particle irreducible Greens functions. The latter action can be obtained in a diagrammatic expansion built from the Wilson Lagrangian L, itself obtained from string
perturbation theory as an expansion
L = L(0) + L(1) + ,
the subscript being the string-loop order. The expressions given in the previous sections
were for L(0) , or for the tree-level S0 . At the string one-loop level, S0 includes tree and
one-loop diagrams generated by the Feynman rules of the tree-level Wilson Lagrangian
L(0) . These include anomalous loop diagrams. It also includes tree diagrams generated by
the GreenSchwarz counterterm [34] introduced in L(1) to cancel the anomalies generated
by L(0) . The mechanism for symmetry restoration implies that L(1) is not invariant under
the restored symmetry.
In four spacetime dimensions, the nature of the cancelled anomalies is known from
studies of (2, 2) compactifications of heterotic strings in the YangMills sector [24,28,35]:
target-space duality of the modulus T has a one-loop anomaly which is cancelled by a
counterterm in L(1) , in a generalization to sigma-model anomalies of the GreenSchwarz
mechanism [36]. The derivation of the complete counterterm requires a calculation to all
246
orders in the modulus T [28]. However, at the present stage of understanding, the Mtheory approach should be regarded as a large-T limit in which T-duality reduces to a shift
symmetry in the imaginary part of T .
Our goal in this section is to obtain some or all counterterms in L(1) associated with
anomaly-cancellation in the low-energy description. We are particularly interested in
contributions to gauge kinetic terms, the so-called threshold corrections. And we want
to formulate these terms using the M-theory multiplets V , VT and W corresponding to
the surviving components of G, in contrast to the heterotic multiplets S (or L) and T .
We begin by obtaining the relevant information from the case of heterotic (2, 2) symmetric
orbifolds.
4.1. Information from symmetric (2,2) orbifolds
Retaining only the universal modulus T and a linear dilaton multiplet L, the Wilson
one-loop Lagrangian for heterotic symmetric (2, 2) orbifolds includes a term 15 [24]
X

ba log (iT )W a W a F ,
(4.1)
L(1) = 2GS L log(T + T ) D +
a
ba
are numbers depending on the orbifold and (iT ) is the Dedekind

where GS and
function. Under Sl(2, Z) T-duality (2.31), the variation of L(1) is
h
i

1
+ ba log (T )W a W a F
L(1) = 2GS L log (T ) + log (T )
D
2

1
a
a
a
= (GS + b ) log (T )W W F ,
2
with (T ) = icT + d. On the other hand, the triangle one-loop diagram for two gauge
fields and one Khler connection 3 log(T + T ) is anomalous. Its variation is

1
= Aa log (T )W a W a F ,
2
where Aa is the chiral-anomaly coefficient, as obtained from the expression of the diagram.
The anomaly cancels since one finds that ba + GS + Aa = 0 for all factors in the gauge
group (the index a). The one-loop correction to gauge kinetic terms obtained from the
component expansion of L(1) and from the triangle diagram reads

1 a a
F
(GS + Aa ) log(T + T ) + ba log |(iT )|4 .
(4.2)
F
4
It is modular invariant since the anomaly is cancelled, and its value is controlled by the
F -density contribution to L(1) , with coefficients ba .
The Wilson Lagrangian depends on the coefficients GS and ba . But the information on
gauge thresholds is in the numbers b a . In general, the parameters of the Wilson Lagrangian
computed at a non-trivial loop order are not of direct physical significance and this is here
the case of GS in the sector of gauge kinetic terms.
P
15 In this paragraph, we use ca = 1 and L
a
=L2
a .
247
In the large-T limit, T-duality reduces to Im T Im T +constant, the Khler connection

3 log(T + T ) is invariant and, strictly speaking, no anomaly survives to be cancelled.
In addition, log |(iT )|4 3 (T + T ) dominates the logarithmic contributions. The
threshold correction is then of the simple form

1 X ba
a
(T + T ) F
F a ,
4 a
3
invariant under the imaginary shift symmetry of T . Its supersymmetrization is

2 X a
X a
b T WaWa F =
b (T + T ) a D .
12 a
3 a
As long as T only is considered, there seems to be no way to identify the D-density

contribution to L(1) in the large-T limit. The introduction of the matter multiplet M
changes the picture. In the lowest order Wilson Lagrangian L(0) , the Khler connection
3 log(T + T ) is modified to 3 log(T + T 2MeA M). As a consequence, the
contribution with coefficient Aa to the gauge threshold (4.2) involves the quantity
log(T + T 2MM). Then, either the one-loop term L(1) is accordingly modified
to 2GS [L log(T + T 2MeA M)]D and the parameter GS disappears from gauge
thresholds, or, less plausibly, this is not the case and GS acquires a physical significance in
M-dependent thresholds. In any case, since the holomorphic Dedekind function cannot
depend on MeA M, a calculation of the correction L(1) to the Wilson Lagrangian to
first order in MeA M will give access to the D-density GreenSchwarz term and to the
parameter GS .
From the point of view of M-theory on O7 or heterotic strings on CalabiYau, MeA M
arises from the ten-dimensional ChernSimons terms, at the same order as gauge kinetic
terms. This indicates that the D-density contribution to L(1) should be visible in a reduction
of the ten-dimensional GreenSchwarz terms.
4.2. The case of M-theory on O7
Before embarking in a derivation of the anomaly-cancelling terms from the low-energy
limit of M-theory on O7 , we consider the problem at the level of four-dimensional
supergravity only.
In the large-T limit, as discussed in the previous paragraph, the T -dependent corrections
to gauge kinetic terms are of the form

1 X a
T WaWa F ,
4 a
(4.3)
with coefficients which are in principle calculable in heterotic strings. To rewrite them in
terms of the M-theory multiplets, we first note that

X

1 X a
T W a W a F = 2 (T + T )
aa .
(4.4)
4 a
D
a
248
Since the field equation of the Lagrange multiplier LT implies that VT = T +T 2MeA M,
this expression can also be written

X a a
A

.
(4.5)
2 VT + 2Me M
a
The right-hand side of Eq. (4.4) is gauge invariant because a is a linear multiplet and
therefore [(T + T ) a ]D is a derivative. To ensure gauge invariance of expression (4.5),
we add the term [LT (VT + 2MeA M)]D included in Lagrangian (3.7). We obtain

X

a a
A

,
(4.6)
VT + 2Me M
LT 2
D
which is gauge invariant if we postulate that

X
a a .
LT = 2
(4.7)
The correction (4.4) to the super-YangMills Lagrangian is independent of the matter

fields and has a holomorphic character (it can be seen as a correction to the holomorphic
gauge kinetic function fab ). To enumerate possible matter-dependent contributions to
gauge kinetic terms, we consider for simplicity a single matter multiplet M transforming
in some unspecified representation of the gauge group. The first candidate counterterm is
a real density:
X

a MeA M a D .
a
Gauge invariance requires however its appearance in the combination

X
X
A
a a
A
a a
c
+ 2 Me M LT 2

,
2 Me M L 2
a
or

X
A
a a

,
2 Me M V D + 2 Me M LT 2
using M-theory multiplet V . In the first counterterm, each gauge group factor contributes
with weight ca , as in lowest order terms. The second contribution can be combined with
expression (4.6) into

X

a a VT + 2(1 + )MeA M
.
(4.8)
LT 2
a
We will also see below that the M-theory anomaly-cancelling terms generate a contribution
of the form

(4.9)
V |M 3 |2 D ,
involving the matter superpotential. Since the factor 1 + in expression (4.8) can be
eliminated by a rescaling of M, of and of the superpotential coupling constant , we
may take = 0 at our level of approximation, and the Wilson Lagrangian up to string
one-loop order is expected to become
249
h

L = (S + S )(V + 2) + U (W M 3 ) + c.c.
X
a a )(VT + 2MeA M)
+ (LT 2
a
+ V |M 3 |2 2MeA M
i
D
+ [S03 W ]F .
(4.10)
Notice that the contributions which correspond to string one-loop effects do not include
any correction to the Einstein Lagrangian, which remains simply
1
eR,
2
= (z0 z0 CT )3/2 (2C)1/2 ,
and the Einstein frame condition remains = 2 . This is expected since the gravitational
constant in the heterotic string is not corrected at string one-loop order.
From the general expression (4.10) in which Bianchi identities are field equations for S,
LT and U , we can derive various equivalent forms. For instance, solving for S, LT and
U leads to the version of the effective supergravity in which the dilaton is described by a
linear multiplet:
h
3/2

3/2
1/2 i
+ S03 (ih + M 3 ) F
T + T 2MeA M
2L
Ll = S0 S 0
D

X

1
3 2
A
a
a
a
+ L |M | 2Me M D + T
W W
.
(4.11)
4 a
F
The second line is a one-loop correction in the perturbative expansion of the heterotic
string in which L is the string loop-counting field [25]. Its T -dependent part corresponds
to the GreenSchwarz counterterm found in [24] for symmetric heterotic orbifolds. Each
of these one-loop corrections, with coefficients , and a , is related to a well-defined
counterterm which can be easily identified in, for instance, the low-energy limit of Mtheory on O7 . The GreenSchwarz counterterms controlled by and are intrinsically real
D-densities. They will appear as corrections to the Khler potential, as matter-dependent
wave-function renormalizations, in the dual version with a chiral dilaton multiplet. On
the other hand, the holomorphic T -dependent terms are true threshold corrections.
Alternatively, solving for LT , V and U leads to the version with a chiral dilaton
multiplet:

3
1 X a
(c S + a T )W a W a
Lc = S0 S 0 eK/3 D +
2
4 a
F
3

3
+ S0 (ih + M ) F ,
(4.12)
with the Khler potential 16
K = log S + S + 2MeA M |M 3 |2

3 log T + T 2MeA M ,

(4.13)
16 The superfield Khler potential includes covariantization contributions eA which disappear in the bosonic
expression used in component expansions.
250
and the gauge kinetic functions f a = ca S + a T . An ambiguity exists however because

one can perform a holomorphic redefinition of the two chiral multiplets. For instance,
S = S T ,
(4.14)
leads to the equivalent Khler potential

K = log S + S T + T 2MeA M |M 3 |2

(4.15)
with gauge kinetic functions f a = ca S + ( a ca )T . The origin of this ambiguity at the

level of M-theory multiplets is interesting. Suppose that we add the counterterm

L = A (V + 2) VT + 2MeA M D
to the fundamental Lagrangian (4.10), with an arbitrary constant A. The theory becomes
then

LA = (S + S )(V + 2) + U (W M 3 ) + c.c.

X

( a Aca ) a VT + 2MeA M
(4.16)
+ LT 2
a
+ V |M 3 |2 + AVT + 2(A )MeA M

D

+ S03 W F .
It is gauge invariant provided the appropriate transformation of LT is postulated. We

have apparently obtained a family of four-dimensional supergravities, depending on a
new parameter A. This is however only true before solving for the Lagrange multiplier
multiplets. Firstly, solving for S and LT leads to the spacetime derivative L = A[L(T +
T )]D . The counterterm L is then irrelevant in the version of the theory with a linear
dilaton. Secondly, if we instead solve for V and LT , we obtain the Khler potential

K = log S + S A(T + T ) + 2MeA M |M 3 |2

and the gauge kinetic functions f a = ca S + ( a Aca )T . This theory is clearly related
to Eq. (4.13) by the holomorphic redefinition S S AT , and the choice A = leads to
theory (4.15).
This discussion shows that the counterterm L is irrelevant in the four-dimensional
effective supergravity, that all values of A lead to equivalent Lagrangians, with the same
dynamical equations. Further information due, for instance, to compactification of extra
dimensions could however appear more natural with a specific value of A, if one insists to
use the version of the effective supergravity with a chiral dilaton multiplet. For instance,
all corrections linear in T appear as gauge thresholds with the choice A = 0. But one could
as well use the version with a linear dilaton which is free of ambiguities.
In addition, the holomorphic redefinition (4.14) mixes terms of different orders in
the string loop expansion. As a consequence, the distinction between terms in L(0) and
corrections in L(1) becomes ambiguous in general in the large T limit.
251
The values of the coefficients and can be inferred from a direct calculation of the
M-dependent anomaly-cancelling terms in M-theory on O7 . This is the subject of the next
paragraph. Notice however that such a calculation only provides the terms of first order in
the matter multiplets MeA M and |M 3 |2 . To this order, expression (4.13) becomes

K = log(S + S ) 3 log T + T 2MeA M

1
(4.17)
2MeA M |M 3 |2 .
S+S
The term with coefficient has been obtained in direct CalabiYau reductions of M-theory
on S 1 /Z 2 (see, for instance, [16,17] 17 ). The charged matter contribution with coefficient
was not included in these analyses.
The gauge contributions appearing in Eq. (4.10) read
X

ca (S + S ) + a VT + 2MeA M a D ,
2
a
so that the gauge coupling constants are given by

1
1
= ca Re s + a CT + 2MM .
ga2
2
(4.18)
This expression becomes harmonic once the Bianchi identity imposing CT + 2MM =
2 Re T has been used. Similarly, one obtains from Eq. (4.11)
ca
ca
1
+ a Re T ca MM +
|M 3 |2 ,
=
2
4C
2
ga
(4.19)
with = 2(z0 z0 )3/2 (Re T MM)3/2 C 1/2 . This second form of the gauge couplings is
never harmonic since it is obtained from a theory with a linear dilaton multiplet. Both
expressions do however agree since the chiral-linear duality relation between Re s and C
is
MM + |M 3 |2 .
Re s =
4C
2
4.3. On the eleven-dimensional origin of the anomaly-cancelling terms
In ten dimensions, anomaly-cancelling terms for the E8 E8 heterotic string are wellb8
known. There are two terms. The first couples a gauge and Lorentz
invariant eight-form X
R
b The second one is proportional to (3,1 + 3,2 3L ) X
b7 ,
to the two-form field B.
b8 .
b7 = X
with dX
The anomaly-cancelling terms arising from M-theory on S 1 /Z 2 [1,2] have recently
been precisely computed [38]. They arise from the following action terms in eleven
dimensions [1,2,3941]:
Z
Z
1
2
X
C G G,
(4.20)
7
2
2
(4)2 11
1211
17 It has also been obtained, in a quite different context, by Itoyama and Leon [37].
252
where
X7 =

1
1
1
2
(tr
R
)
7L
3L ,
12(4)3 2
8
(4.21)
11 is the eleven-dimensional gravitational constant and is the gauge coupling constant on

both ten-dimensional Z 2 -fixed planes, as defined by the gauge action (3.12). The Chern
Simons forms are defined by
d7L = tr R 4 ,
d3L = tr R 2 ,
d3,i = tr Fi2 ,
i = 1, 2.
Cancellation of gauge and gravitational anomalies and coherence of the reduction to ten
4 /12 [1,2,38]. This condition relates the gauge coupling
dimensions impose 6 = (4)5 11
1
and the S radius. Solving the Bianchi identity verified by the four-form field G on S 1 /Z 2
and extracting the zero modes leads to the following GreenSchwarz terms

Z
1
1
1
4
2 2
b (I4,1 )2 + (I4,2 )2 I4,1 I4,2 + 1
tr
R
(tr
R
B
)
SGS =
48
(4)4 2
8
M10
Z
1
(3,1 + 3,2 3L ) X7 .
(4.22)
(4)2
M10
In this expression,
I4,i =

1
1
2
2
tr
R
tr
F
,
i
2
(4)2
and 18
2
bAB =
B
2
11
i = 1, 2,
(4.23)
Z
dx 4 CAB4 .
(4.24)
S1
b couples to X
b8 . But the second term has a particular
As usual, the two-form field B
b
structure: X7 is replaced by the purely gravitational seven-form (4.21): the anomalycancelling terms derived from M-theory differ from the standard expression of the heterotic
string by a well-defined local counterterm, as permitted by the descent equations [38].
bAB leads to two zero modes of the bulk fields,
The CalabiYau compactification of B
b =
2B
since
2
2
11
2
C4 ,
V6
1 2 1
.
2 V6 V1
b =
2B
ij
2
C ,
V6 ij 4
By Eqs. (2.30), these states are related to our four-dimensional bulk
multiplets L (or S) and T by

b = 4b ,
B
b = i Im T .
2B
ij
ij
(4.25)
The number 2 /V6 disappears in this identification: it does not play any role in the fourdimensional effective theory.
18 Recall that x 4 is the S 1 coordinate.
253
Our task is then to compute the reduction of SGS on M4 (CalabiYau). Since we restrict
ourselves to contributions with at most two derivatives, we only need the reduction of
Z

1
b (I4,1 )2 + (I4,2 )2 I4,1 I4,2 .
B
=
48
M10
The global definition of the four-form field G (cohomology condition) implies [4]
hI4,1 i = hI4,2 i
for the CalabiYau background, which is a (2, 2) form. The counterterm becomes then
Z

1
b hI4,1 i tr F12 tr F22
B
=
3
4(4)
M10
Z

1
b tr F12 2 + tr F22 2 tr F12 tr F22 + ,
B
5
12(4)
M10
where gravitational contributions with more than two derivatives are omitted. Notice that
the two E8 factors contribute with opposite signs in the background-dependent term.
The standard embedding is defined by htr F22 i = 0, while htr F12 i is in the SU(3) direction
of the maximal embedding E6 SU(3) E8 . As a consequence,
htr F12 i = htr R 2 i = 2(4)2 hI4,1 i,
which in turn leads to
Z

1
b htr R 2 i tr F12 tr F22
B
=
5
8(4)
M10
Z
h

i
1
b tr F12 2 + tr F22 2 tr F12 tr F22 + .
B
12(4)5
(4.26)
M10
To derive the zero modes of tr F12 and tr F22 , it is simpler to consider the ChernSimons
forms, using the relation

tr Fi2 ABCD = 4 [A (3,i )BCD] , i = 1, 2.
In the standard embedding, the unbroken E8 group generates a N = 1 super-YangMills
multiplet only. The only massless mode is then

tr F22 ABCD tr FE28 , (3,2 )ABC (E8 ) .
Clearly, these massless modes of tr F22 and 3,2 are respectively components of the fourdimensional N = 1 multiplets W 2 W 2 and 2 used earlier. More precisely:
1
i
[W 2 W 2 ]f-component = tr FE8 FE8 tr FE8 FE8 + ,

2
4e
1
2
tr FE8 FE8 + ,
[ ]d-component =
16
1
[ 2 ]vector component = (E8 ) + .
8e
(4.27)
254
The gauge fields of the E8 group broken into E6 generate E6 gauge fields and the chiral
matter multiplet M, transforming in representation 27. Accordingly, the massless modes
of tr F12 are:

tr F12 ABCD tr FE26 ,

tr F12 ij = 2[ (3,1 )]ij ,

tr F12 ij k = (3,1 )ij k ,

tr F12 ij k = (3,1 )ij k .
While as before (3,1)ABC (E6 ) , the other components of 3,1 involve the scalar
component of the matter multiplet M and require more care since we have already precisely
defined the four-dimensional field M by its coupling to the bulk fields. To obtain the correct
relations, a detour is helpful.
As already explained, the gauge kinetic action (3.12) also generates the four-dimensional kinetic terms for the matter multiplet M. In the four-dimensional effective Lagrangian,
these contributions arise as the highest component of the matter ChernSimons multiplet
MeA M, which includes 2(D M)(D M ). This multiplet also contains in its vector
component the matter ChernSimons form
M = iM(D M) i(D M )M.
This is completely similar to the gauge ChernSimons multiplet which includes gauge
kinetic terms in its d-component and (3 ) in its vector component, as indicated by
expressions (4.27). A direct computation of the relation between kinetic terms due to the
action (3.12), and the highest component of MeA M delivers then the relation between
(3,1 )ij and this multiplet, by four-dimensional supersymmetry. A similar operation
gives the relation between (3,1 )ij k and the superpotential multiplet M 3 . The relations
are

1
i
(3,1 )ij = 2 (D M )M M(D M) ij = 2 ij MeA M vector component
6
6

1
1
(3,1 )ij k = 3 M 3 ij k = 3 ij k M 3 scalar component
(4.28)
3
3
together with the last equation (4.27) which applies to all gauge ChernSimons forms.
With the complete identification of the massless modes of the ChernSimons threeforms, we are equipped for translating the CalabiYau reduction of the GreenSchwarz
counterterm (4.26) in a four-dimensional supergravity density formula. After straightforward manipulations of Eq. (4.26), we obtain
2
1

I
( a)(E6 E8 )
192(4)5
V6
2

i

+
I
( C4 ) MD M MD M
5
2
384(4)
V6

V6 2
i 2
( C4 ) M 3 ( M 3 ) M 3 M 3 +
+
5
6
2
216(4)
V6
LGS =
1
I T (E6 E8 )
384(4)5

i
I ( b ) MD M MD M
96(4)5

i 2 V6
( b ) M 3 ( M 3 ) M 3 M 3 + ,
54(4)5 6
255
(4.29)
where the dots indicate the terms required by N = 1 supersymmetry and I is the
dimensionless integral
Z
2
dV6 ii ij k ij k htr R 2 ij kj k ,
I =
CY
in terms of the (1, 1) (the metric tensor), (3, 0) and (0, 3) CalabiYau tensors and the
background htr R 2 i.
Using Eqs. (4.28), we can write the GreenSchwarz counterterm in superfield form as:

I
I
VT + 2MeA M ( 1 2 ) D
V MeA M D
5
5
48(4)
48(4)

1
V6
+
V |M 3 |2 D .
27(4)5 6
LGS =
(4.30)
Comparing the three terms of LGS with the corresponding parts of (4.10), we can express
the coefficients a , and in terms of the CalabiYau dependent integral I :
= 1 = 2 =
I
,
96(4)5
1
V6
.
5
27(4) 6
(4.31)
The calculation predicts then 1 = 2 = , a result already obtained in references [16,

17], for instance. Once again, we stress that we have obtained next-order corrections to the
effective Wilson Lagrangian. As argued earlier in Section 4.1, while we expect 1 and 2 to
have physical significance as coefficients of the modulus-dependent threshold corrections,
the parameter is not necessarily a physical quantity. To decide of its relevance, a
calculation at the same order of threshold corrections in the effective action should be
performed, but this computation requires a detailed knowledge of the charged matter
spectrum and couplings.
5. Conclusions
In this paper, we have deduced the structure of the four-dimensional N = 1 effective
(wilsonnian) supergravity describing the universal massless sector of M-theory compactified on (CY)3 S 1 /Z 2 . The theory depends on three categories of multiplets: M-theory
multiplets V , VT and W describe the degrees of freedom of the M-theory four-index tensor, source multiplets a , MeA M and M 3 are related to the source terms in M-theory
Bianchi identities and Lagrange multiplets S, LT and U impose by their field equations
the Bianchi identities. In addition, the multipliers S and LT generate the four-dimensional
256
axion-tensor duality which is known to be an important ingredient of the formulation of

the string dilaton beyond the lowest order.
An effective supergravity similar to expression (4.10) is in principle valid for generic
compactifications of M-theory with unbroken N = 1 supersymmetry in four dimensions.
One needs to identify the appropriate supermultiplets appearing as sources in the Bianchi
identities generated by S, LT and U . Only the values of coefficients like a , and depend
on the detailed geometry of the compact space. This method is especially useful in deriving
contributions to the effective Lagrangian due to non-perturbative states like M-theory fivebrane degrees of freedom or condensates. Some of these more general N = 1 vacua will
be studied in a forthcoming publication [20].
Acknowledgements
The authors have benefited from discussions with A. Bilal, C. Kounnas, A. Lukas,
D. Lst and B. Ovrut. This research was supported in part by the European Union under
the TMR contract ERBFMRX-CT96-0045, the Swiss National Science Foundation and the
Swiss Office for Education and Science.
Appendix A. Notations and conventions

Metrics and coordinates
The spacetime metric has signature (, +, +, , +).
For coordinates, our notation is:
M = 0, . . . , 10
D = 11 curved spacetime:
xM
xA
x
= 0, 1, 2, 3
x4
S1 /Z 2 direction:
CalabiYau directions:
xa
a = 5, . . . , 10
CalabiYau complex (Khler) coordinates: zi , zi i = 1, 2, 3
For reduction purposes, we simply use
1
1
zl = (x l ix l+3 ),
l = 1, 2, 3.
zl = (x l + ix l+3 ),
2
2
ij k is the SU(3)-invariant CalabiYau tensor such that 123 = 123 = 1.
Antisymmetric tensors
Antisymmetrization of n indices has unit weight:

1
AM1 Mn (n! 1) permutations .
A[M1 Mn ] =
n!
Differential forms
For a p-index antisymmetric tensor, we define
1
A(p) = AM1 Mp dx M1 dx Mp .
p!
257
Then,
1
AM1 Mp BMp+1 Mp+q dx M1 dx Mp+q = C (p+q) ,
p!q!
(p + q)!
A[M1 Mp BMp+1 Mp+q ] .
=
p!q!
A(p) B (q) =
CM1 Mp+q
The exterior derivative is d = M dx M . The curl F (p+1) = dA(p) of a p-form reads then
1
(M AN1 ...Np ) dx M dx N1 dx Np
p!
1
FM ...M dx M1 dx Mp+1 ,
=
(p + 1)! 1 p+1
FM1 Mp+1 = (p + 1)[M1 AM2 Mp+1 ]
dA(p) =
= M1 AM2 Mp+1 p cyclic permutations.

The volume form in D spacetime dimensions is dx M1 dx MD = M1 MD dD x.
We use analogous conventions for forms in four spacetime dimensions.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
P. Horava, E. Witten, Nucl. Phys. B 460 (1996) 506, hep-th/9510209.

P. Horava, E. Witten, Nucl. Phys. B 475 (1996) 94, hep-th/9603142.
E. Cremmer, B. Julia, J. Scherk, Phys. Lett. B 76 (1978) 409.
E. Witten, Nucl. Phys. B 471 (1996) 135, hep-th/9602070.
T. Banks, M. Dine, Nucl. Phys. B 479 (1996) 173, hep-th/9605136.
E. Caceres, V.S. Kaplunovsky, I.M. Mandelberg, Nucl. Phys. B 493 (1997) 73, hep-th/9606036.
P. Horava, Phys. Rev. D 54 (1996) 7561, hep-th/9608019.
I. Antoniadis, M. Quirs, Phys. Lett. B 392 (1997) 61, hep-th/9609209; Nucl. Phys. B 505
(1997) 109, hep-th/9705037; Phys. Lett. B 416 (1998) 327, hep-th/9707208.
H.-P. Nilles, S. Stieberger, Nucl. Phys. B 499 (1997) 3, hep-th/9702110.
E. Dudas, C. Grojean, Nucl. Phys. B 507 (1997) 553, hep-th/9704177.
Z. Lalak, S. Thomas, Nucl. Phys. B 515 (1998) 55, hep-th/9707223.
A. Lukas, B.A. Ovrut, D. Waldram, Phys. Rev. D 57 (1998) 7529, hep-th/9711197.
E.A. Mirabelli, M.E. Peskin, Phys Rev. D 58 (1998) 065002, hep-th/9712214.
J. Ellis, Z. Lalak, S. Pokorski, W. Pokorski, Nucl. Phys. B 540 (1999) 149, hep-ph/9805377.
J. Ellis, Z. Lalak, S. Pokorski, S. Thomas, hep-th/9906148.
H.-P. Nilles, M. Olechowski, M. Yamaguchi, Phys. Lett. B 415 (1997) 24, hep-th/9707143;
Nucl. Phys. B 530 (1998) 43, hep-th/9801030.
A. Lukas, B.A. Ovrut, D. Waldram, Nucl. Phys. B 532 (1998) 43, hep-th/9710208.
Z. Lalak, S. Pokorski, S. Thomas, Nucl. Phys. B 549 (1999) 63, hep-ph/9807503.
J.-P. Derendinger, R. Sauser, in preparation.
S. Ferrara, L. Girardello, T. Kugo, A. Van Proeyen, Nucl. Phys. B 223 (1983) 191.
T. Kugo, S. Uehara, Nucl. Phys. B 226 (1983) 49; Nucl. Phys. B 222 (1983) 125.
E. Cremmer, S. Ferrara, L. Girardello, A. Van Proeyen, Phys. Lett. B 116 (1982) 231.
J.-P. Derendinger, S. Ferrara, C. Kounnas, F. Zwirner, Nucl. Phys. B 372 (1992) 145.
S. Cecotti, S. Ferrara, M. Villasante, Int. J. Mod. Phys. A 2 (1987) 1839.
J.-P. Derendinger, F. Quevedo, M. Quirs, Nucl. Phys. B 428 (1994) 282, hep-th/9402007.
258
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
E. Witten, Phys. Lett. B 155 (1985) 151.

L. Dixon, V. Kaplunovsky, J. Louis, Nucl. Phys. B 355 (1991) 649.
J.-P. Derendinger, L.E. Ibez, H.-P. Nilles, Phys. Lett. B 155 (1985) 65.
J.-P. Derendinger, L.E. Ibez, H.-P. Nilles, Nucl. Phys. B 267 (1986) 365.
C.P. Burgess, A. Font, F. Quevedo, Nucl. Phys. B 272 (1986) 661.
L. Dixon, V. Kaplunovsky, J. Louis, Nucl. Phys. B 329 (1990) 27.
G.F. Chapline, N.S. Manton, Phys. Lett. B 120 (1983) 105.
M.B. Green, J.H. Schwarz, Phys. Lett. B 149 (1984) 117.
J. Louis, in: P. Nath, S. Reucroft (Eds.), Particles, Strings and Cosmology, World Scientific,
Singapore, 1992, pp. 751765.
G.L. Cardoso, B.A. Ovrut, Nucl. Phys. B 369 (1992) 351; Nucl. Phys. B 392 (1993) 315, hepth/9205009.
H. Itoyama, J. Leon, Phys. Rev. Lett. 56 (1986) 2352.
A. Bilal, J.-P. Derendinger, R. Sauser, hep-th/9912150, to appear in Nucl. Phys. B.
C. Vafa, E. Witten, Nucl. Phys. B 447 (1995) 261, hep-th/9505053.
M.J. Duff, J.T. Liu, R. Minasian, Nucl. Phys. B 452 (1995) 261, hep-th/9506126.
S.P. de Alwis, Phys. Lett. B 392 (1997) 332, hep-th/9609211.

Gravity in the brane-world for two-branes model

with stabilized modulus
Takahiro Tanaka a,b , Xavier Montes a
a IFAE, Departament de Fisica, Universitat Autonoma de Barcelona, 08193 Bellaterra (Barcelona), Spain
b Department of Earth and Space Science, Graduate School of Science, Osaka University, Toyonaka 560-0043,
Japan
Received 2 February 2000; accepted 24 May 2000
Abstract
We present a complete scheme to discuss linear perturbations in the two-branes model of the
Randall and Sundrum scenario with the stabilization mechanism proposed by Goldberger and Wise.
We confirm that under the approximation of zero-mode truncation the induced metric on the branes
reproduces that of the usual 4-dimensional Einstein gravity. We also present formulas to evaluate the
mass spectrum and the contribution to the metric perturbations from all the KaluzaKlein modes.
We also conjecture that the model has tachyonic modes unless the background configuration for the
bulk scalar field introduced to stabilize the distance between the two branes is monotonic in the fifth
dimension. 2000 Elsevier Science B.V. All rights reserved.
PACS: 04.50.+h; 98.80.Cq
1. Introduction
Recently, there has been a growing interest in considering extra dimensions in nontrivial
form [14]. One important suggestion was done by Randall and Sundrum [4]. They realized
that if one considers 5-dimensional gravity with a negative cosmological constant bounded
by two positive and negative tension branes, the induced gravity on the negative tension
brane can possess a very small gravitational constant as compared with the energy scale
introduced in the original Lagrangian. This fact potentially gives a solution to the hierarchy
problem of explaining the extraordinary weakness of the gravitational coupling.
Later, the same authors pointed out that the model with the positive tension brane alone is
also possible [4]. In this case, the extension of the extra dimension is infinite. Nevertheless,
the induced gravity on the brane can be expected to mimic Einstein gravity.
There are many discussions about the cosmology based on these scenarios [510]. The
behavior of gravity on these models has been also investigated by many authors [1117]. In
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 2 8 - X
260
T. Tanaka, X. Montes / Nuclear Physics B 582 (2000) 259276
this direction, an explicit method to deal with the linear order metric perturbations induced
by the matter fields confined on the branes was developed in the paper by Garriga and
Tanaka [11] (Paper I). In that paper, considering the zero mode truncation approximation,
it was shown that the induced gravity on the branes becomes of the BransDicke type in
the two-branes model. On the other hand, Einstein gravity is recovered in the single-brane
model.
However, the analysis of the two-branes model in Paper I was not complete. The
stabilization mechanism of the distance between the two branes [18,19] by means of a bulk
scalar field was not taken into account. Moreover, the KaluzaKlein contribution was not
estimated for the two-branes model. In this paper we consider a model with stabilization
mechanism, and we confirm that the 4-dimensional Einstein gravity is recovered on both
branes under the approximation of zero-mode truncation. An approximate formula for
the mass of the lowest massive mode in the scalar-type perturbations is also obtained.
Furthermore, we present a method to evaluate the metric perturbations taking into account
all the KK contributions under a contact interaction approximation. This approximation is
valid if the source which excites the KK modes is smoothly distributed compared with the
scale corresponding to the mass of the lowest massive mode.
This paper is organized as follows. In Section 2 we describe the background model
that we consider, and derive the basic equations for the linear perturbations on it. In
Section 3 we discuss the mass spectrum of the perturbations. We identify the mode
functions for the massless degrees of freedom, and also derive the formula for the mass
and the mode function of the lowest massive mode. In Section 4 the mechanism to recover
Einstein gravity is explained. This result is expected from the notion of the mass spectrum
but its derivation is not so trivial when we consider the explicit construction of metric
perturbations. We shall find that the contribution from the massive modes of the scalartype perturbations must be also taken into account in some sense. For this purpose, we
introduce a contact interaction approximation. In Section 5 we apply the same technique
to the KK modes of the tensor-type perturbations to complete the description of the linear
perturbations of this system. Section 6 is devoted to summary.
2. Model and basic equations

We consider linear perturbations of the model proposed by Randall and Sundrum [4]
with the stabilization mechanism by Goldberger and Wise [18]. The fields existing on
the 5-dimensional spacetime (bulk) are the usual 5-dimensional gravity with a negative
cosmological constant 3, and a 5-dimensional scalar field introduced to stabilize the
distance between the two branes. We denote the 5-dimensional gravitational constant
by G5 . The unperturbed metric is supposed to take the form,
ds 2 = a 2 (y) dx dx + dy 2 ,
(2.1)
where is the 4-dimensional Minkowski metric with (+++) signature. We use

= a 2 (y) to raise 4-dimensional tensor indices. The y-direction is bounded by
261
two branes located at y = y () . On these two branes, reflection boundary conditions are
imposed. The Lagrangian for the bulk scalar field is
X

V ( ) () y y ( ) .
(2.2)
L = 12 g ab ,a ,b VB ()
=
For most of the present analysis, we do not need to specify the explicit form of the
potentials VB () and V () (). In the bulk, the background scale factor and the scalar field,
(a, 0), must satisfy

1 2
H 2 (y) =
0 (y) VB 0 (y) 1 3 ,
H (y) = 02 (y),
2
3
6
(2.3)
0 (y) + 4H (y) 0(y) VB0 0 (y) = 0,
where H (y) := a(y)/a(y)
3/6 and = 8G5 . Ordinary matter fields reside

on both branes, and the values of the 4-dimensional vacuum energy on both branes are
adjusted to realize a static background configuration. This way of model construction is
different from the standard approach [18] which assumes the model potential and the values
of the vacuum energy on the branes from the very beginning. In the present approach, we
obtain a constraint on the model potential by fixing the distance between the two branes.
Conversely, in the standard approach, the stabilized distance is determined for each given
model (although some tuning of one of the model parameters is necessary to realize a static
configuration). We call the brane at y = y (+) (y () ) the positive (negative) tension brane,
and adopt the convention y (+) < y () .
We first consider the metric perturbation (ds 2 ) = hab dx a dx b and the scalar field
perturbation in the bulk. We are using the convention that Latin indices run through
0, 1, 2, 3, 5, while Greek indices run through 0, 1, 2, 3. In the bulk, we can always impose
the Newton gauge condition:
h5 = 0,
h55 = 2 N ,
3
[y + 2H ] N,
=
2 0
N
h = h(TT)
,
(2.4)
(TT)
where h satisfies the transverse-traceless condition, and N is the 5-dimensional

Newtonian potential.
The equation for h(TT)
in the bulk is given by

1 (4) b (TT) (TT)
2 +L
(2.5)
h = 0,
a2
where 2(4) is the 4-dimensional dAlembertian operator with respect to , and we have
introduced the operator
1
1
b
L (TT) := 2 y a 4 y 2 .
a
a
For the scalar-type perturbations, we obtain
(4)
N
L ( ) N = 0,
2 +b
(2.6)
(2.7)
262
b (
where L
N)
is the operator defined by
2 2 2
1
N
b
a 0 .
y a 2
L ( ) := a 2 02 y
2
2
3
a 0
(2.8)
As discussed in Paper I, the y = constant hypersurface does not correspond to the

location of the branes in this gauge. Following it, we introduce two Gaussian normal
coordinate systems near both branes, respectively. We denote quantities in these coordinate
systems by associating a bar, such as y.
at
Then, the junction condition for the bulk scalar field is given by 2 = V 0() ()
y = y () 1 . Hence, its perturbation become

(2.9)
2 = V 00() (0 ) y = y () .
The junction condition for the metric perturbation is

()
1
2
0
(y 2H )h = T T
3
3

y = y () .
(2.10)
Here, we have introduced the energy momentum tensor of the matter fields confined on
()
()
()
, and T () := T
. T
satisfies the 4-dimensional conservation law
the branes, T
;
T = 0. Notice that there appears a contribution from the perturbation of the scalar field
which was not present in the models discussed in Paper I.
Whereas the junction conditions have been easily derived using Gaussian normal
coordinates, the equations of motion for the perturbations are simpler in the Newton
gauge. So we need to consider the gauge transformation which relates normal coordinates
and the Newton gauge. This is given by hab = h ab + a;b + b;a with
Zy
5
()
Zy
5
N (y 0 ) dy 0 + ()
,
(y ) dy =
N
Zy
=
()
y ()
0
Zy
(y 0 ) dy 0
Zy
=
(y ) dy
N
(y 00 ) dy 00
,
Zy
(2.11)
(y 00 ) dy 00 + ()
,
,
(2.12)
y ()
5
are independent of y. Then, we have
and ()
where ()
" Zy
(y)
= (y) 0 (y)
#
N
(y ) dy
5
+ ()
y ()
h (y) = h (y) + 2a 2(y)
Zy
dy 0
a 2 (y 0 )
Zy
N
(y 00 ) dy 00
,
1 The last means that we have to add and then take the limit 0.
Zy
2H (y)
N (y 0 ) dy 0.
263
(2.13)
Here we should stress that the arguments in the l.h.s. are not y but y.
Combining Eqs. (2.10) and (2.13), the junction condition in the Newton gauge for the
TT part is obtained as

()
(2.14)
y = y () ,
(y 2H )h(TT)
=
where we have defined
()
:= T 14 T
()

2 5
5 ,
(), 14 ()
, .
(2.15)
(TT)
Combining Eq. (2.14) with Eq. (2.5), we obtain the equation for h :

X

2(4) b (TT) (TT)
( )
+
L
y y ( ) .
h = 2
2
a
=
(2.16)
(TT)
The resulting h is automatically consistent with the traceless condition, but the
5
:
transverse condition gives us the equation which determines ()
1
2
a()
5
2(4) ()
= T () ,
6
(2.17)
where we have defined a() := a(y ()). The set of Eqs. (2.16) and (2.17) is exactly the
same that was obtained in Paper I once we specialize the background bulk geometry to
pure anti-de Sitter.
Using Eq. (2.17), it can be easily seen that the trace part of the metric junction condition
is trivially satisfied.
The remaining junction condition is the one for the scalar field, Eq. (2.9). After some
computations, it reduces in the Newton gauge to
() (4) N
2
5
0 ()
= 2 2
3
a 0

y = y () ,
(2.18)

() :=
2
.
V 00() 2(0 / 0 )
(2.19)
Combining Eq. (2.18) with Eq. (2.7), we obtain the equation for N :
X

4a 2 02 5
( ) y y ( )
3
=

X

(4) N
( )
( )
2 y y
.
= 2 1 +
N
b
L ( ) N
(2.20)
264
3. Mass spectrum
3.1. Zero modes
()
= 0. We
Let us first consider the solution of the source free equations by setting T
(4)
will first consider the zero eigenmodes of 2 , which are the most important because they
correspond to the 4-dimensional massless field responsible for the propagation of a long
range force.
If we set 2(4) = 0 in Eq. (2.16), the solution for h(TT)
is
h(TT)
2
(2) 2
= h (1)
(x )a (y) + h (x )a (y)
Zy
dy 0
,
a 4 (y 0 )
(3.1)
(1)
(2)
(i)
where h and h are 4-dimensional TT tensors independent of y satisfying 2(4) h = 0.
Furthermore, the junction condition (2.14) gives
2 5
h (2)
= 2a() (), .
(3.2)
(2)
Hence, there is no extra constraint on h (1)
, but h
must be written as the second derivative
of a scalar function. The former mode corresponds to 4-dimensional gravitational wave
perturbations. Although there seem to exist five independent degrees of freedom in this
mode, three of them are pure gauge. On the other hand, the latter mode should be classified
as scalar-type perturbations. If we forget about the stabilization mechanism, this mode
corresponds to the mode called radion in Ref. [20].
5 and 5 :
Notice that Eq. (3.2) gives a relation between (+)
()
2 5
2 5
(+) = a()
() .
a(+)
(3.3)
5
also appears in the junction condition for the
In the present case with stabilization, ()
scalar-type perturbations. Hence, the scalar field perturbation must also be chosen to be
compatible with this condition. Here we should note that this relation holds only when we
restrict our considerations to zero modes. In the general case, which will be considered
later, there is no reason for such a relation to be satisfied.
Let us now consider the solution of the zero eigenvalue scalar-type perturbation. One
solution is
N =
H (1)
f (x ),
a2
0 (1)
f (x ).
2a 2
(3.4)
This mode satisfies the scalar field junction condition (2.18) with (3.3) if f (1) =
2 5 . However, this solution can be transformed to nothing by a gauge transfor2a()
()
(1) R y 2 0 0
a (y ) (y ) dy 0 .
mation with parameters 5 = a 2 f (1) /2, = f,
The other solution is

Zy
2H
2 0
0
a (y ) dy f (2) (x ),
= 1 2
a
N
= f
(2)
0
(x ) 2
a
Zy
a 2 (y 0 ) dy 0 .
(3.5)
265
However, this mode is not compatible with condition (3.3). Hence, if we include
the stabilization mechanism, no physical massless mode is present in the scalar-type
perturbation spectrum, in contrast with the the case discussed in Paper I.
3.2. Massive modes
Let us now consider the lowest massive mode. They are also important because they
give the leading-order correction to the zero-mode truncation approximation. Furthermore,
if tachyonic modes are present, such a model must be rejected.
To estimate the lowest mass eigenvalue in the general case, one needs numerical
calculations. In order to obtain analytical approximations, here we assume that the effect
of the bulk scalar field back reaction to the background geometry is small. Namely, we
assume
02
|H |
=
1.
(3.6)
H2
3H 2
For the metric, we can use the pure anti-de Sitter form
a(y) = ey/` .
(3.7)
For simplicity, here we set

y (+) = 0,
y () = d.
(3.8)
First we consider the tensor-type perturbations. The expression for the mode functions
L (TT) satisfying the junction condition (y 2H )ui (y) = 0 on the
ui (y) for the operator b
positive tension brane is found in Refs. [4,11]. Denoting the eigenvalue corresponding
to ui (y) by m2i , the mode function is given in terms of Bessel functions: ui (y)
{J1 (mi `)Y2 (mi èy/` ) Y1 (mi `)J2 (mi èy/`)}. The boundary condition on the negative
tension brane reduces to

(3.9)
J1 (mi `)Y1 mi èd/` Y1 (mi `)J1 mi èd/` = 0,
and determines the discrete eigenvalues mi . For small mi `, this condition becomes
J1 (mi èd/`) 0, and hence the eigenvalues are given by mi ed/`ji `1 , where ji is
the ith zero-point of J1 . Hence the physical mass of the lowest massive KK mode on the
positive tension brane is given by 3.8ed/``1 while that on the negative tension brane
by 3.8`1 .
Next, let us consider the scalar-type perturbations. When we discuss the scalar-type
perturbations, we should not assume an scale factor of the form (3.7) from the beginning,
because the nonvanishing second or higher derivative of a(y) can be important. Hence we
will derive formula (3.22) for the lowest mass eigenvalue without assuming (3.7). Only to
obtain an estimate of this expression, we will use the scale factor (3.7).
To discuss scalar-type perturbations, it is convenient to introduce a new variable defined
by
q :=
3a 2 N
,
2A
(3.10)
266
where a prime 0 denotes derivative with respect to the conformal coordinate z defined by
dz = a(y)1 dy, and we have introduced A = a 1/200 . The signature in the definition of
A is chosen so that A is continuous on the branes. The junction condition for q is obtained
from (2.18):

5
(3.11)
y = y () .
z + (A0 /A) q = A()
Here, for simplicity, we have taken the () 0 limit. The perturbation equation in terms
of q becomes
00

X

1
2 0 2
5
0 q =
y y () .
2A()
(3.12)
2(4) + z2 A
A
3
Integrating the above equation once at the vicinity of the branes, we correctly reproduce
the junction condition (3.11).
To obtain the solution of Eq. (3.12), we will construct the Green function for the
differential operator appearing in it. It is given by
Z
d4 k ik 1x X qi (z)qi (z0 )
e
,
(3.13)
Gq =
(2)4
m2i + k 2
i
where mi is the mass eigenvalue. The mode functions satisfy
00

1
2 0 2
2
qi = m2i qi
z A
A
3 0
with the boundary condition

z + (A0 /A) qi = 0 y = y () .
(3.14)
(3.15)
In the weak back reaction case, the term U := 200 2 /3 in Eq. (3.14) is assumed to
be small. If we perturbatively expand the mode function and the mass eigenvalue with
(0)
(1)
(0)
(1)
respect to powers of U like qi = qi + qi + and mi = mi + mi + , the lowest
eigenmode for the unperturbed system is trivially given by
(0)
q0 =
N
,
A
(3.16)
2
with (m(0)
0 ) = 0, where N is a normalization constant.
It is straightforward to obtain the next-order correction. We find
N
(1)
q0 =
A
Zz
2
A dz
Zz0

dz00 2 0 2
(1) 2
m0
.
A2 3 0
(3.17)
It is convenient to express the above results in terms of , which is related to q through

=
1 d
Aq.
a 3/2A dz
Thus, the junction condition for the eigenmode i reads

i = 0 y = y () .
(3.18)
(3.19)
The above first-order correction is now given by

2

Zy
m(1)
(1)
0
0 2
.
dy
0 = N 0 (y)
3a 2 a 4 (y 0 )02 (y 0 )
267
(3.20)
y (+)
The junction condition (3.19) on the positive tension brane is already imposed due to the
choice of the integration constant.
If 0 vanishes at some point, the integrand in (3.20) becomes singular there. Even in this
case it can be shown that the expression for 0(1) is analytic. Suppose that 0 vanishes at
y = y0 . Then one can show that a 4 02 can be expanded around y0 as

2

(3.21)
a 4 02 = a 4 VB0 y=y (y y0 )2 + 2 (y y0 )4 + ,
0
VB00 )|y=y0 .
:=
+
Notice that the cubic term is absent in the above
where
(1)
expansion. Substituting this expression in Eq. (3.20), one finds that 0 is analytic at
y = y0 .
If we impose the junction condition (3.19) at y = y () on the first-order correction
(3.20), we obtain the formula for the lowest mass eigenvalue,
() ()
, Zy
y
Z
2
dy
dy
(3.22)
m20

.
3
a2
a 4 02
2
1
2
3 (4H
y (+)
y (+)
This expression is positive definite if there is no point at which 0 vanishes. However, when
0 vanishes, the integral in the denominator changes its signature at that point. Hence,
m20 can be negative. Due to the absence of the cubic term in Eq. (3.21), the integrand
is independent of the way of modification of the integration path. It is expected that the
dominant contribution to this integral comes from the region near the point y0 , where
vanishes. By substituting equation (3.21), we can approximately evaluate the denominator
in Eq. (3.22) as
4 0 2 1

Z
a (VB ) y=y
0
=
,
dy
(y y0 )2 + 2 (y y0 )4
a 4 (VB0 )2 y=y0
which becomes negative. Although we cannot give a proof here, this seems to be the
case in general. Here we simply conjecture it. If tachyonic modes appear, they mean the
breakdown of the model. Hence, we do not consider the case in which 0 vanishes at some
point, and hereafter we assume that 0 has a definite sign.
Without explicitly specifying any model, we can make a crude estimate of the lowest
mass eigenvalue. To evaluate the numerator in the r.h.s of Eq. (3.22), we can use the
form (3.7) for the scale factor a. To evaluate the denominator, we use the following fact.
Besides some exceptional cases, a 4 02 will have a minimum (cases with no minima will be
discussed later). There 0 + 2H 0 = VB0 2H 0 = 0, and accordingly the integrand has a
rather sharp peak. We denote this point by yc . At this point, the second derivative of a 4 02
is evaluated as (8H 2 + 2VB00 4H )a 4 02 . Hence, we can approximate the integral like
268
Z
1/a 4 02
dy 4H /a
2
(VB0 )2 y=yc

exp 4H 2 + VB00 y=yc (y yc )2 dy.
Under these approximations, the formula for the mass of the lowest eigenmode is obtained
as
s
"
#
2d/` `2
VB00 `2
e
2
4y/`
0 2
e
(VB ) 1 +
.
(3.23)
m0
4
6
y=yc
To proceed further, we consider the specific model for the bulk potential VB () discussed
in Ref. [18]:
M 22
.
(3.24)
2
For this model, the background solution for the bulk scalar field in the weak back reaction
case is already given by
VB () =
0 = B1 e1 y + B2 e2 y ,
where 1 = 2`1 + 4`2 + M 2 , 2 = 2`1 4`2 + M 2 and

B2 (+) () e1 d ,
B1 e1 d () (+) e2 d ,
(3.25)
(3.26)
(+) and () are the values of 0 on the positive and the negative tension branes,
respectively.
Let us briefly elaborate on the stabilization distance. For simplicity, we will assume that
the ratio (+) /() is not extremely large, and also that all the input scales are similar, i.e.
`3 . The condition for weak back reaction is
1
(3.27)
02 2 .
`
Since 02 does not have its maximum in the bulk, it is sufficient to consider the conditions
for weak back reaction on the boundaries. Then, we can evaluate the values of 0 on
boundaries,
q

0 y () 1 () 4`1 1 + (M 2 `2 /4) e2 d (+) . (3.28)
0 y (+) 2 (+) ,
The conditions for weak back reaction on both branes become
s
!

1 () 2
`3
M 2 `2
`3
4e2 d (+)
2 (+) 2
,
1+
.
1
5/2
5/2
4`
4`
1 ` ()
4
(3.29)
The condition on the positive tension brane is satisfied by choosing (case A) () `3/2
with M 2 . `2 or (case B) M 2 `2 with () & `3/2 .
In case A, the condition on the negative tension brane is automatically satisfied. In this
case, the stabilization distance is sensitive to the change of the values of the vacuum energy
on the branes. Hence, without specifying the complete model, we cannot estimate the
stabilization distance. If we consider the case in which M 2 `2 is not small, we find that
it is natural to assume that the signature of (+) is different from that of () . If (+) and
269
() have the same signature, the ratio between them must be chosen to be extremely large
to find a node-less solution of 0 with sufficiently large brane separation d. In the case that
(+) and () have different signature, the solution of 0 becomes node-less for any choice
of parameters.
Case B is the case discussed in Ref. [18] (see also Ref. [7]). In this case, to satisfy the
condition for weak back reaction on the negative tension brane,
!
p
1
4`1 (+) 1 + (M 2 `2 /4)
d
ln
(3.30)
` 2 `
1 ()
is required. Hence, (+) and () must have the same signature. To realize a sufficiently
large value of d/`, it is necessary that |(+) | is slightly larger than |() |. If we take the
small M 2 limit, the above expression for the stabilization distance coincides with that
obtained in Ref. [18].
Let us now return to the computation of m20 . First we consider the case in which a 4 02 has
a minimum and the estimate (3.23) is valid. As we shall see below, case B is exceptional
in this sense. From the condition that VB0 2H 0 = 0 at y = yc , we have e(2 1 )yc =
1 B1 /2 B2 . Substituting this estimate of yc into (3.23), we obtain
p
8M 2e2d/` 1 + (M 2 `2 /4)
2
(B1 B2 ).
(3.31)
m0
3
To solve the hierarchy problem on the negative tension brane, we need to set ed/`
1019(GeV)/103 (GeV) = 1016 . Hence, d/` 37. Therefore, it is rather natural to suppose
that M 2 `d is larger than unity. If |(+)| is not much larger than |() |, as assumed,
we can approximate B1 B2 |(+) () |e1 d . In this case, the mass eigenvalue m20
1
2 2
becomes O(|(+)() |M 2 e2` d 1+(M ` /4) ). Hence thephysical squared mass on the
1
negative tension brane becomes O(|(+) () |M 2 e2` d( 1+(M ` /4)1) ) while that on
the positive tension brane is the same as m20 . As for |()|, they must be smaller than
`3/2 for the
approximation of weak back reaction to be valid. Also, there is a factor
1
2 2
M 2 e2` d( 1+(M ` /4)1) , which takes its maximum value (0.2/`)2 at M 0.33`1 .
Hence, the mass scale on the negative tension brane tends to be smaller than the typical
background energy scale `1 .
Finally, we consider the special case in which a 4 02 in Eq. (3.22) has no minima. This
happens when one of the terms in 0 = 1 B1 e1 y + 2 B2 e2 y can be totally neglected. This
condition becomes

() (+) e2 d 2 |()|,
or
(3.32)
1

(+) () e1 d 1 |(+) |.
(3.33)
2
In the former case we can set B1 0, and in the latter case we can set B2 0. Then, for
the former case, formula (3.22) gives
" p
#
4 2 2 2d/`
1 + (M 2 `2 /4)
2
e
.
(3.34)
m0 =
2 2
3 2 ()
1 e4 1+(M ` /4)d/`
2 2
270
It is easy to see that case B corresponds to this case. The model discussed in Section 4.1
of Ref. [19] with negative small value of b also corresponds to this case (for the definition
of b, see Ref. [19]). The factor in the square brackets is almost unity. Substituting the
approximation 2 M 2 `/4, we recover the result obtained in [18] (see also [7]). Note
that their M 3 and k are our 1/(4) and `1 , respectively.
For the latter case, we have
#
" p
4 2 2 2d/`
1 + (M 2 `2 /4)
2
e
.
(3.35)
m0 =
2 2
3 1 ()
e4 1+(M ` /4)d/` 1
The model discussed in Section 4.1 of Ref. [19] with positive small value of b corresponds
to this case. To keep the mass sufficiently large, we have to choose |() | extremely large,
which is not compatible with the weak back reaction condition.
4. Recovery of Einstein gravity

As we have shown in the preceding section, in two-brane models with stabilization
mechanism, a physical massless mode is absent in the scalar-type perturbation spectrum,
which is different from what happens in the case without stabilization mechanism
discussed in Paper I. This fact indicates that the resulting 4-dimensional effective gravity
can resemble Einstein gravity at linear order. Let us show how it can be recovered.
To compute the induced metric on the branes, it is convenient to transform back
to Gaussian coordinates. Thus, applying minus the gauge transformation (2.12), the
induced metric on each brane is given by

(0) ()
5
()
()
+ ,
+ h(KK)
y () N y () + 2H ()
,
. (4.1)
h ()
= h y
into two parts, the zero mode contribution h(0)

Here we have decomposed h(TT)
(KK)
and the KK contribution h . The last term represents the residual 4-dimensional
gauge transformation. As we have already noted, Eq. (2.16) and Eq. (2.17) determining
(TT)
5
are essentially the same as the ones obtained in Paper I for the
h (y () ) and ()
case without stabilization mechanism. There, it was found that the induced gravity
approximated by the zero mode truncation becomes of the BransDicke type [11]. Here
(0)
we only quote the result for h (y () ) up to 4-dimensional gauge transformation:

(4) 1 X

2
T ( )
(0) ()
( )
16G
,
(4.2)
=
T
h y
3
a2
=
2 and
where 8G() := Na()
1
()
Zy
a 2 dy .
N := 2
(4.3)
y (+)
(KK)
Hence, the only possible source of an additional contribution is that from h or

N (y () ). Although we have just shown that there are no massless degrees of freedom
271
in the scalar-type perturbations, we will find that the contribution from N (y () ) gives the
correct long range force required to recover Einstein gravity.
First we consider the weak back reaction case discussed
in Section 3.2. From the
Rd
(0)
2
normalization condition of q0 we obtain N = [2 0 dz(1/A)2]1 . Comparing it with
the formula for the mass (3.22), we find a simple relation
m20
4N 2
Zd

dy 2`N 2 2d/`
e
1 .
2
a
3
(4.4)
Then, taking into account the contribution from the modes with the lowest mass
eigenvalue alone, the Green function for q, Eq. (3.13), is approximated by
Z
m20 A1 (y)A1 (y 0 )
d4 k
3eik 1x
Gq
(2)4 2`(e2d/` 1)
m20 + k 2

Z

3A1 (y)A1 (y 0 ) 4
d4 k k 2 eik 1x
=
(1x
)
.
(4.5)
2`(e2d/` 1)
(2)4 m20 + k 2
Using this Green function, and with the aid of (3.10) and (3.12), we find that N is given
by

2
1 2 (+)

2`1
5
5
2
(+)
+ ()
+ a()
T () . (4.6)
m0 2(4)
a(+) T
2
2d/`
a (e
1)
6
Only the first term inside the square brackets gives a long range contribution to the induced
5
. The source for the second
metric. The propagation of this force is essentially due to ()
term exactly becomes the trace of the energy momentum tensor of the ordinary matter field.
Hence, the force due to this term becomes short-ranged with typical length scale am1
0 .
5 , we finally obtain the contribution to the long
For the first term, using Eq. (2.17) for ()
()
range component of h coming from N as

8G() 2(4) 1 2

N ()
2
T () ,
(4.7)
a(+) T (+) + a()

y
2
3
a()
N
2 /(1 e2d/` ) in the weak back

where we have used the fact that 8G() = `1 a()
reaction case. Substituting this into Eq. (4.1), and using Eq. (4.2) and Eq. (2.17), the zero
mode truncation reproduces the formula for the linearized Einstein gravity:

T ( )
2(4) () X
( )
=
16G
.
(4.8)
T
h
2
2
a()
=
In the above derivation of Eq. (4.7), we have used several approximations. This
derivation has the merit of having a result with a rather easy intuitive interpretation. The
(5)
gravitational field is propagated through the massless field () . At a point far from the
(5)
source T , () generates a cloud of metric perturbations through the interaction with
the massive KK modes. However, the above approximate derivation is not completely
satisfactory because some aspects of general relativity are already tested with very high
precision. Hence, we present an alternative and more complete treatment below.
272
First we consider to apply the same trick used in the second line of Eq. (4.5) to the
complete Green function containing all massive modes. Then the long-range part of
P
the Green function reduces to Gq 4 (1x ) i qi (z)qi (z0 )/m2i . This replacement is
exactly valid when we focus on the long range force. Now one can notice that to use this
Green function neglecting the terms corresponding to the short range force is equivalent to
solve the equation for N (2.20) by setting 2(4) = 0 from the beginning.
For the case with 2(4) = 0, we have already obtained the general solutions to Eq. (2.20),
i.e., Eqs. (3.4) and (3.5). For convenience, here we quote the previous results in a slightly
different notation, namely:
X
u f ( ) (x ), with
(4.9)
N =
=
Zy
2H
u := 1 2
a
a 2 (y 0 ) dy 0 .
(4.10)
y ()
2 5 .
Then the junction condition (2.18) with 2(4) = 0 determines f () as f () = 2Na()
()
N
Substituting these back into the expression for , we obtain
X

5
2 5
2N
N y () = 2H ()
a(
(4.11)
) ( ) .
=
5
Adding this contribution to the contributions coming from the zero mode TT part and ()
in Eq. (4.1) for the induced metric on the branes, Einstein gravity at the linear order is
recovered.
It will be illustrative to give the next order correction. The source term for the next-order
P
correction is 2(4)0N (1 + = 2 ( ) (y y ( ) )), where we have denoted the solution
of the lowest order approximation (4.11) by 0N . By using Eq. (2.17), 2(4) 0N is evaluated
as
N X
4
( )
u (y)a(
.
(4.12)
2(4) 0N =
)T
3 =
Then, by using the standard Green function method, we obtain

" Zy
1N (y) = u+
y (+)
#

u 3N (4) N
0
2 0 dy + K+
02
" Zy
u
y ()
#

u+ 3N (4) N
0
2 0 dy K ,
02
(4.13)
where K+ and K are integration constants determined from the junction condition as
2 ()
()

N
H
()
4
4
()

+ a() T
a T
1 2
.
(4.14)
K =
2
a N
273
For simplicity, we take again the () 0 limit. Then, we have K = 0 and obtain
X
4
( )
I, a(
,
(4.15)
1N() = N 2
)T
=

()
Zy
ui (y) uj (y)
Ii,j :=
02
y (+)
dy.
(4.16)
In the weak back reaction case, we can approximately evaluate this integral by using
the following facts. First, we note that the combination a 2u can be shown to be a slowly
changing function for y & yc in the weak back reaction case. As before, yc is the peak
location of (a 4 02 )1 . To show this, we use the fact that 0 is approximated by 1 B1 e1 y for
y & yc . This constancy of a 2 u indicates that the integrand of Eq. (4.16) is approximately
proportional to (a 402 )1 . Then by using the background geometry defined by Eqs. (3.7)
and (3.8), I+,+ is evaluated like
I+,+ a u+
2
2
Zd
y=d
0
dy
a 4 02
`
4m20
e2d/`,
where we have used (3.22). The other components are also evaluated in a similar way
to obtain the relations I+,+ e2d/Ì+, e4d/` I, . The correction obtained by
substituting these estimates for Ii,j into (4.15) is consistent with the contribution from
the second term in the square brackets in Eq. (4.6). If we set 2(4) = 0 there, we recover the
same result.
It is easy to check that the treatment of taking 2(4) to be small is consistent if the
distribution of the energy momentum tensor is sufficiently smooth. In the treatment
presented here, we have taken into account the contribution from all the massive modes
simultaneously. The approximation is completely valid as long as we consider smooth
matter distribution as compared with the mass scale of the lowest massive mode. However,
the lowest mass on the positive tension brane becomes very small if we consider the case
in which we are living on the negative tension brane. In this sense, the treatment presented
here is rather restrictive when we discuss perturbations caused by the matter fields on the
positive tension brane.
5. TT part revisited
Applying the same technique that we have used in the preceding section for the scalar(TT)
type perturbations, we can also deal with the KK mode contribution for the TT part h .
By using the Green function method, we can easily calculate the zero mode contribution
like
X 2
2
(4) 1
( )
=
2Na
(y)
2
a( )
.
(5.1)
h(0)
274
(0)
(KK)
After using Eq. (2.17), we can recover Eq. (4.2). Substituting h(TT)
into
= h + h
(2.16), we obtain the equation for the KK contribution:
X

2 (4)
( ) 2
L (TT) h(KK)
a( )N y y ( ) .
(5.2)
a 2 +b
= 2
=
As before, neglecting the 2(4) -term for the massive KK contribution, we can solve this
equation like
h(KK)
= 2N
X
=
Zy
2
( ) 2
a(
) a (y)
dy 0
4
a (y 0 )
y ( )
Zy
00 2
dy a y
00
!
C( ) ,
(5.3)
y ( )
where C(+) and C() are constants. We should recall that we have already subtracted the
(KK)
zero mode contribution. Hence h must be orthogonal to the zero mode. This requires
()
y
Z
()
dy
u0 (y) h(KK)
(y)
a2
y (+)
Zy
dy h(KK)
(y) = 0.
(5.4)
y (+)
The constants C(+) and C() in the solution (5.3) are determined by imposing this
condition.
For simplicity we again adopt (3.7) with (3.8) as the background geometry. Then, from
condition (5.4), C() are explicitly calculated, and the resulting KK contribution becomes
h(KK)
=
X
`
( )
2d/`
4(1 e
) =

2
4a( )d`1
(2y2d)/`
2
2y/`
e
2a( ) + e
1 .
1 e2d/`
(5.5)
Hence, on the respective branes, the contributions from the KK modes become

`
`
(+)
()
(+)
()
+
(3 4d/`)
,
h(KK)
() 4 + , (5.6)
4
where we have assumed d/` 1. Here we should recall the remaining degrees of freedom
for the 4-dimensional gauge transformation. Using these degrees of freedom, can be
replaced with T T /3 with the aid of Eq. (2.17). Hence, the KK modes, of course,
(KK)
do not give any long range force contribution. If one takes the d/` limit, h(+)
seems to diverge. But this is just due to the breakdown of the approximation. In this limit,
the mass difference of the KK modes becomes zero.
h(KK)
(+)
6. Summary
In this paper we have developed a systematic procedure to evaluate the perturbations in
the 5-dimensional brane world model proposed by Randall and Sundrum [4] supplemented
with the moduli stabilization mechanism by Goldberger and Wise [18].
275
We have first investigated in detail the mass spectrum of this model. In the case without
stabilization mechanism, there was a scalar-type massless mode, which was called radion
in Ref. [20]. We have clarified how this massless mode disappears once we switch on the
stabilization mechanism.
We have also estimated the mass eigenvalue and the mode function corresponding to
the lowest mass eigenmodes for both tensor-type and scalar-type perturbations, assuming
that the back reaction to the background geometry due to the bulk scalar field introduced
for the moduli stabilization is not large. The physical mass of the lowest tensor-type mode
becomes `1 on the negative tension brane and ed/``1 on the positive tension
brane. Here ` is the curvature scale of the background geometry, and d is the proper
distance between the two branes. In the original model, `1 is supposed to be TeV scale.
For the physical mass of the lowest scalar-type mode, we have obtained rather
general formulas, (3.22) and (3.23). To proceed further in estimating the mass, we
have specified a model for the bulk potential of the scalar field. We have considered
a simple quadratic potential whose mass is given by M, and we have assumed that
the 5-dimensional gravitational constant is also `3 . We pointed out that in this
model there are two regimes in which the weak back reaction condition holds. The
first case is the one in which the vacuum expectation values of the scalar field on the
branes, () , are sufficiently small compared with the background energy scale, i.e.,
3/2 . In this case, we found that the mass of the lowest scalar-type mode becomes
()
q `
2 2
|()(+) |`3 Me( 1+(M ` /4)1)d/` . (For a more precise formula, see Eq. (3.31).) If
we take into account the fact that we need to set d/` 37 to
q solve the hierarchy problem
on the negative tension brane, this factor is at most 0.2`1 |() (+) |`3 for M` 0.33.
The second case is the one in which M is small compared with `1 , which is the situation
discussed in Refs. [7,18]. In this case, we reproduced the same result found there. Namely,
the mass is approximately given by (|() |`3/2)M 2 `. As pointed out in Ref. [18], the mass
scale on the negative tension brane tends to be smaller than the typical background energy
scale `1 . We confirmed that this is a general feature within the context of weak back
reaction. However, looking at these formulas, it seems that we can raise the mass by taking
a slightly larger vacuum expectation values of the bulk scalar field on both branes, although
a large value of |() | is inconsistent with our approximation. Hence, if we remove the
technical limitation of weak back reaction, it is not clear whether the physical mass of
the lowest scalar-type mode is always smaller than that of the lowest tensor-type mode.
We also mention that the mass scale on the positive tension brane is smaller by a factor
of ed/` .
Next, we have developed a method to evaluate the explicit form of the perturbations
caused by the matter fields confined on the branes. This is a generalization of the results
presented in the previous paper [11]. We found that, as is expected, Einstein gravity is
exactly recovered for the long range force at the order of linear perturbations when we
impose the stabilization mechanism. The formulas for the leading correction from the
scalar-type perturbations (4.15) and that from the KK modes (5.6) are also derived without
specifying the model for the potential of the bulk scalar field. From these formulas, we can
276
read the coupling of the metric perturbations induced on the branes to the matter fields on
both branes.
As we have confirmed in Section 5, there is no pathological behavior in all perturbation
modes at the level of linear perturbations. This is a rather expected result because almost all
important information at the level of linear perturbations is contained in the mass spectrum
except for the strength of coupling to the matter fields. However, it is still unclear what
happens once we take into account the nonlinearity of gravity. To investigate this issue, the
formulas obtained in this paper will be useful.
Acknowledgements
We thank J. Garriga for useful comments and discussions. We also thank O. Pujolas and
M. Sasaki for valuable suggestions. T.T. acknowledges support from Monbusho System
to Send Japanese Researchers Overseas. X.M. acknowledges support from CICYT under
contract AEN99-0766.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
P. Horava, E. Witten, Nucl. Phys. B 460 (1996) 506; Nucl. Phys. B 475 (1996) 94.
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263.
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257.
P. Bintruy, C. Deffayet, U. Ellwanger, D. Langlois, Phys. Lett. B 477 (2000) 285.
E.E. Flanagan, S.-H.H. Tve, I. Wasserman, hep-ph/9909373.
C. Cski, M. Graesser, L. Randall, J. Terning, hep-ph/9911406.
J. Garriga, M. Sasaki, hep-th/9912118.
S. Mukohyama, T. Shiromizu, K. Maeda, hep-th/9912287.
R. Maartens, D. Wands, B.A. Bassett, I. Heard, hep-ph/9912464.
J. Garriga, T. Tanaka, hep-th/9911055 (paper I).
T. Shiromizu, K. Maeda, M. Sasaki, gr-qc/9910076.
M. Sasaki, T. Shiromizu, K. Maeda, hep-th/9912233.
A. Chamblin, G.W. Gibbons, Phys. Rev. Lett. 84 (2000) 1090.
A. Chamblin, S.W. Hawking, H.S. Reall, Phys. Rev. D 6 (2000) 065007.
R. Emparan, G.T. Horowitz, R.C. Myers, JHEP 0001 (2000) 007.
T. Chiba, gr-qc/0001029.
W.D. Goldberger, M.B. Wise, Phys. Rev. Lett. 83 (1999) 4922; Phys. Lett. B 475 (2000) 275.
O. DeWolfe, D.Z. Freedman, S.S. Gubser, A. Karch, hep-th/9909134.
C. Charmousis, R. Gregory, V. Rubakov, hep-th/9912160.

Two-loop dimensional reduction and effective

potential without temperature expansions
M. Laine a,b,1 , M. Losada c,2
a Theory Division, CERN, CH-1211 Geneva 23, Switzerland
b Department of Physics, P.O. Box 9, FIN-00014, University of Helsinki, Helsinki, Finland
c Centro de Investigaciones, Universidad Antonio Nario, Cll. 59 No. 37-71, Santa Fe de Bogot, Colombia
Abstract
In many extensions of the Standard Model, finite temperature computations are complicated by a
hierarchy of zero temperature mass scales, in addition to the usual thermal mass scales. We extend
the standard thermal resummations to such a situation, and discuss the 2-loop computations of the
Higgs effective potential, and an effective 3d field theory for the electroweak phase transition, without
carrying out high or low temperature expansions for the heavy masses. We also estimate the accuracy
of the temperature expansions previously used for the MSSM electroweak phase transition in the
presence of a heavy left-handed stop. We find that the low temperature limit of dealing with the lefthanded stop is accurate up to surprisingly high temperatures. 2000 Elsevier Science B.V. All rights
reserved.
PACS: 11.10.Wx; 11.15.Ex; 11.30.Fs; 12.60.Jv; 98.80.Cq
Keywords: Finite temperature; Symmetry breaking; Baryogenesis; Supersymmetry; Cosmology
1. Introduction
Electroweak baryogenesis in the Minimal Supersymmetric Standard Model (MSSM) is
a viable option for explaining the matterantimatter asymmetry observed in the present
Universe, provided that there is a mild hierarchy between the right and left-handed stop
masses [18]. The dominantly right-handed stop should be lighter than the top in order to
make a strong transition, yet the left-handed stop should be rather heavy, 1 TeV, in order
to raise the Higgs mass upper bound to 110 GeV. Various details of the electroweak
phase transition in this regime are constantly being investigated [916].
1 mikko.laine@cern.ch
2 malosada@venus.uanarino.edu.co
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 9 8 - 4
278
M. Laine, M. Losada / Nuclear Physics B 582 (2000) 277295
Here we will be concerned with the thermodynamics of the phase transition. In the
perturbative approach, this problem is approached by computing the effective potential
for the Higgs field to some order in the loop expansion. In general, such a computation in
a weakly coupled gauge theory faces two problems:
(i) The system has a hierarchy of mass scales (2T , gT , g 2 T ), which spoils a
straightforward perturbative computation. Historically, this was observed by finding
large linear terms at 2-loop level [17], which were then shown to be absent after
an appropriate resummation [18];
(ii) At momenta of the order of the lowest of the mass scales (g 2 T ), the system is also
inherently non-perturbative [19].
The resummations needed at 2-loop level for dealing with the heavy scale 2T were
discussed in detail by Arnold and Espinosa [20]. However, the problem can also be
dealt with in another way, namely by constructing a sequence of effective field theories
by integrating out, to a given order in perturbation theory, the scales 2T , gT [2123].
This construction is highly accurate in the Standard Model [24,25]. The final theory
is three-dimensional (3d), purely bosonic, and contains only the momentum scale g 2 T .
A perturbative analysis of the 3d theory automatically reproduces the results of the
resummed 4d effective potential, but the theory can also be studied efficiently with
relatively simple lattice simulations [2631], to account for the non-perturbative part.
The problem we consider here is the observation that the hierarchy of mass scales can
be even more severe in extensions of the Standard Model such as the MSSM. Indeed, there
one tends to have new mass parameters that are not related to the temperature in the same
way as mH is in the Standard Model, where mH gTc . In particular, as mentioned above,
one prefers rather large left-handed squark mass parameters, say mQ 1 TeV. Previously,
the effects of mQ have been considered (on the 2-loop or non-perturbative level) only
in the high temperature expansion, or in the extreme limit mQ 2T where the finite
temperature effects decouple completely.
Our objective here is to treat in some detail the general situation mQ 2T . First of all,
we discuss how the resummations used previously need to be changed in such a situation
(Section 3). We then show with a simple example how the full resummed 2-loop effective
potential could be computed without any temperature expansions related to mQ , and how
the result can be used for a 2-loop computation of the mass parameter of an effective 3d
field theory (Section 4). Finally we consider a particular observable sensitive to mQ , the
critical temperature of the electroweak phase transition, and estimate the accuracy of the
high and low temperature expansions employed earlier on (Section 5). We conclude in
Section 6 and discuss several possible extensions of the computations presented in this
paper. The expressions used for the 1-loop tadpole and bubble, as well as 2-loop sunset
graphs are discussed in the appendices.
279
2. Parametric conventions
In order to be explicit yet concise, we illustrate the situation with a simple model
reminiscent of the scalar sector of the MSSM. We take
L = m2H H H + m2U U U + m2Q Q Q

+ h21 H H U U + h22 H Q Q H + h3 AH Q U + H.c. + .
(2.1)
Here H is an SU(2) doublet, U is an SU(3) (anti-)triplet, while Q changes under both

groups. We ignore gauge interactions for the moment. We assume that h1 h2 h3 g
are small couplings, and m2H , m2U (gT )2 . It is also important to specify the order of
magnitude of the dimensionful parameter A in Eq. (2.1). In this paper we work under the
assumption that
2 |A|2
A
g2 ,
m2Q
(2.2)
which simplifies the procedure considerably.

In the imaginary time formalism, the fields in Eq. (2.1) can be divided into Matsubara
modes. We assume that the only light modes are the zero Matsubara modes of H, U . The
non-zero Matsubara modes of H, U have effectively a mass parameter > (2T )2 . For the
field Q, we assume that mQ itself is large, mQ 2T , so that even the zero Matsubara
mode is heavy. If mQ gT , then the zero Matsubara mode of Q is light as well and the
procedure is the one described in [3234]. If mQ 2T /g, on the other hand, Q can be
integrated out at T = 0 with exponentially small corrections.
The issue of resummation can now be formulated as follows. Due to the presence of the
heavy mass scales, the n = 0 modes of H, U can receive radiative corrections as large as
the tree-level terms, m2H,U g 2 m2Q + g 2 T 2 . Such corrections have to be resummed. In
fact, close to the phase transition point, the effective mass parameters m2H eff,U eff can be
even smaller, of the non-perturbative magnitude (g 2 T )2 . Then resummation has to be
extended to the 2-loop level. Non-zero Matsubara modes, or the field Q, on the other hand,
do not require resummation [20], since the mass corrections g 2 T 2 , g 2 m2Q are according to
our convention small compared with the tree-level terms.
This statement can be formulated more precisely as follows. Let us write down the
effective Lagrangian obtained after integrating out all the heavy modes. The light fields
being the n = 0 modes of H , U , the form of the Lagrangian is
Leff = m2H eff H H + m2U eff U U + h21eff H H U U + .
(2.3)
Our aim is now to compute expressions of the form 3

m2H eff = m2H + c1 g 2 m2H + c2 g 2 m2U + c3 g 2 m2Q 1 + c4 g 2 + c5 g 2 T 2 1 + c6 g 2 , (2.4)
where the ci s are numerical coefficients.
3 In a gauge theory there are also corrections of order g 3 T 2 .
280
3. Leading order resummation

In order to carry out the resummation explicitly, let us consider the effective potential of
the theory. To illustrate the procedure, it is enough to consider
only the field H , keeping the
T
expectation value of U at zero. Introducing hH i = (0, ) / 2 changes the mass spectrum
of the system, m m = m + m. We are interested in a certain range of 0 . . . T .
Then, in the case of heavy modes, m m, and we can expand in m, while in the case
of light modes we cannot. For the purpose of illustration, let us suppress m for the light
modes here.
Then, in the standard thermal case, the 1-loop and 2-loop contributions to the effective
potential behave at small as
V1-loop T 2 (m )2 + T (m )3 + ,
(3.1)
V2-loop g T m + g T (m ) + .
(3.2)
The statement of resummation is now that the dominant 2-loop terms, the linear ones
g 2 T 3 m , arise from a badly convergent series which can be resummed into a better
convergent one [18]. The way the resummation proceeds is obvious from Eqs. (3.1), (3.2):
the non-analytic 1-loop and 2-loop terms combine to
3/2
.
(3.3)
T (m )3 + g 2 T 3 m T g 2 T 2 + m2
This corresponds simply to the corrections of order g 2 T 2 in Eq. (2.4). The extension we
make here is that when mQ 2T , the contribution to be resummed goes from g 2 T 2 to a
2 ).
non-trivial function g 2 T 2 f (mQ /T , |A|
In order to proceed systematically, we write the mass parameters related to the light
modes as
m2H = m2H eff r m2H ,
m2U = m2U eff r m2U ,
(3.4)
where m2H eff , m2U eff appear in the propagators, and r m2H , r m2U are treated as interactions.
Denoting the heavy modes by solid lines and the light modes by dashed lines, the graphs
suggest that
2

r m2H = 3h21 In6=0 (mU ) + 3h22 I (mQ ) + 3h23 A I (mQ ) In6=0 (0) ,
2

r m2U = 2h21 In6=0 (mH ) + 2h23 A I (mQ ) In6=0 (0) .
(3.5)
(3.6)
Here I, In6=0 are tadpole integrals defined in Eqs. (A.1), (A.11), and we have made use of
2 is small, Eq. (2.2), implies that wave function
m2H , m2U m2Q . Note that the fact that |A|
2 m2
corrections need not be considered, since their effect would be of order h2 |A|
3
g 4 m2H , beyond Eq. (2.4). For the same reason, we have dropped any m2H , m2U dependence
2.
in the terms proportional to |A|
281
In addition to the mass parameters of the scalar fields, resummation of course also
affects the zero components of the gauge fields. In fact, as is well known [17,18], in the
Standard Model the latter effect is more important for physical observables such as the
strength of the phase transition, while the former is important particularly for the critical
temperature. We do not discuss infrared dominated observables such as the strength of the
phase transition, nor gauge fields, to any length in this paper, but let us nevertheless note
that the contributions of H , U , Q to the Debye masses of the SU(2) and SU(3) fields A0 ,
C0 are, in the presence of mQ 2T ,

d
(3.7)
In6=0 (mH ) + 3IT (mQ ) ,
dT

d
(3.8)
In6=0 (mU ) + 2IT (mQ ) ,
r m2C0 = gS2 T
dT
where gS is the SU(3) gauge coupling and IT is defined in Eq. (A.6). In addition to these
terms, the Debye masses of course contain the usual gauge and fermion contributions.
In order to now show that the procedure introduced in Eqs. (3.4), (3.5), (3.6) is a
consistent one, we need to demonstrate that all linear terms at 2-loop level cancel, and the
remainder is quadratic in m . Recalling that we have set the quartic Higgs self-coupling to
zero (at tree-level) for the purpose of simplicity, we get for the shifts in the mass parameters
(Q1(2) denote the upper (lower) SU(2) component of Q)
r m2A0 = g 2 T
1
m2U eff = h21 2 ,
(3.9)
2
1
m2Q2 = h22 2 .
(3.10)
m2Q1 = 0,
2
2 g 2 we can ignore all corrections involving A here,
Note that due to the assumption |A|
2 g 6 . We
since the corresponding 2-loop contributions are at most of order h2 h2 |A|
m2H eff = 0,
i 3
will denote (mH eff )2 = m2H eff + m2H eff , etc.

Linear terms in the effective potential arise from graphs of the types (H ), (U ), (H U ),
(H Q), (H U Q) in the notation of Fig. 1. Denoting by I3d the 3d tadpole in Eq. (A.12) and
by H the bosonic sunset integral in Eq. (C.1), we obtain

(3.11)
(H ) + (U ) = 2r m2H I3d mH eff 3r m2U I3d mU eff ,

(H U ) + (H Q) = 6h21 I mH eff I mU eff + 3h22 I mH eff I mQ1 + I mQ2 ,(3.12)

(3.13)
(H U Q) = 3h23 |A|2 H mQ1 , mH eff , mU eff + H mQ2 , mH eff , mU eff .
Fig. 1. The 2-loop graphs considered. The blob means a counterterm. Different particles are denoted
by Pi = H, U, Q.
282
We then expand these contributions in m. Employing the expansions

I mQ = I (mQ ) m2Q D(mQ ) + O( mQ )4 ,

I meff = In6=0 (meff ) + I3d meff m2eff Dn6=0 (meff ) + O( meff )4 ,
(3.14)
(3.15)

1

H mQ , mH eff , mU eff = 2 I3d mH eff + I3d mU eff I (mQ ) + In6=0 (0)
mQ

(3.16)
+ I3d mH eff I3d mU eff + O( m)2 ,
where D, Dn6=0 are from Eqs. (B.5), (B.10) and we have used Eq. (C.17), we find that:
There are linear terms I3d (mH eff ), I3d (mU eff ) which however all get cancelled,
when the choice in Eqs. (3.5), (3.6) is made for r m2H , r m2U ;
There is an infrared sensitive contribution, quadratic in the masses, from the
Matsubara zero modes in the graphs (H U ), (H U Q):
2

(3.17)
(H U ) + (H U Q)|IR = 6 h21 h23 A I3d mH eff I3d mU eff .
2 corresponds to coupling constant resummation which
The appearance of h21 h23 |A|
we however do not discuss in any detail here, since the corresponding effects are in
principle beyond the accuracy of Eq. (2.4). Similarly, the graph (H U Q) also produces
2 2 , again beyond Eq. (2.4);
terms of order h4 |A|
Finally, there are ultraviolet sensitive (not from the zero modes) quadratic terms from
the graphs (H U ), (H Q):
(H U ) + (H Q)|UV = 6h21 m2U eff In6=0 (mH )Dn6=0 (mU )
3h22 m2Q2 In6=0 (mH )D(mQ ) + O( m)4 .
(3.18)
To summarize, we have observed that all linear terms get cancelled when the thermal
counterterms are chosen according to Eq. (3.11). The remainder involves quadratic terms,
which can either come from the ultraviolet or the infrared.
4. Next-to-leading order
We next evaluate the 2-loop contributions from the remaining graphs, and expand them
again in m; however, these graphs do not involve contributions linear in m. The graphs
left are the sunsets (H QQ), (H U U ), as well as the 1-loop graphs (H ), (U ), (Q), where
the blobs are now the bilinears obtained from the coupling constant counterterms after the
shift of H 4 .
After an expansion in m, we obtain
4 Mass counterterms do not contribute at the present order; terms proportional to m2 in them would, had we
Q
included self-interactions of the type (H H )2 , (U U )2 in Eq. (2.1).
(H ) + (U ) + (Q) =

2 1 1 4
9 h1 + h42 In6=0 (mH ) + 6h41 In6=0 (mU )
2
2 (4)

+ 12h42 I (mQ ) + O( m)3 ,
283
(4.1)
(H QQ) = 3h42 2 H (mQ , mQ , 0) + O( m)3 ,
(4.2)

3
(4.3)
(H U U ) = h41 2 H mH eff , mU eff , mU eff .
2
The numerical expression of H (mQ , mQ , 0) is discussed in Appendix C. As to the graph
(H U U ), on the other hand, we recall that it arises completely from the zero Matsubara
modes (H = H3d + O(m/T )), and is thus a purely IR quantity [39].
Adding all terms together from Eqs. (3.18), (4.1), (4.2) and using Eqs. (A.11), (B.5),
(B.10), (C.20), we obtain the 2-loop ultraviolet contribution to the 3d mass parameter,

1
UV
2
4
Ivac (mQ )
2-loopmH eff = h2 6Hvac (mQ , mQ , 0) + 12
(4)2

5 2
3T
T2
4 31
ln
+
c
3
ln
h
+
(4)2 1 4 4 2T

2
2
4 3
h2 ln 2 + 6I1 (mQ /T ) ln 2 + 2
4 mQ
mQ

1
(4.4)
+ D(mQ /T ) + 6H(mQ /T ) .
4
Here the first line is a 2-loop vacuum renormalization correction of order g 4 m2Q ,
T = 4eE T 7.0555 T ,

8 0 (2)
1
+
2E 0.34872274,
c = ln
2
9
(2)
(4.5)
and I1 , D, H are functions defined in Eqs. (A.7), (B.8), (C.21).

On the other hand, the IR sensitive part of the effective potential is, from Eqs. (3.17),
(4.3),
2

2IR-loopV = 6 h21 h23 A I3d mH eff I3d mU eff
3

(4.6)
h41 2 H3d mH eff , mU eff , mU eff .
2
The divergence in H3d (Eq. (C.2)) cancels against that from Eq. (4.4), m2H eff 2 /2.
Including also the 1-loop terms in Eq. (3.5), we can now write down the complete mass
parameter m2H eff with accuracy g 4 m2Q , g 4 T 2 . In order to do so, let us first note that 1-loop
radiative corrections generate couplings other than those in Eq. (2.1), viz.
L = h24 H H Q Q + (H H )2 + ,
(4.7)
which we have to include in the discussion for a moment. The corresponding contribution
in Eq. (3.5) is r m2H = 6h24 I (mQ ) + 6In6=0 (m2H ). Furthermore, in order to cancel spurious
-dependences,
we should express the MS parameters in terms of physical observables as
284
in [22]. In this paper we will not consider actual physical pole masses, etc., but simply some
finite physical scale independent parameters ( )phys which, dropping all terms beyond the
accuracy of Eq. (2.4), we define through the following relations:

2
3 2 2
2
2
2
2
2 2
= mH phys +
h m + h2phys + h3 A mQ ln 2 + 1
mH ()
(4)2 1 U
mQ

4
(4.8)
+ h2 6Hvac (mQ , mQ , 0) 12Dvac (mQ )Ivac (mQ ) finite part ,
= h21phys + h41
h21 ()
2
2
ln
,
(4)2 m2Q
h22 ()
= h22phys + h42
= h24phys + h42
h24 ()
2
1
ln
,
(4)2 m2Q
()
= phys +
2
2
ln
,
(4)2 m2Q
(4.9)
1
3 4
2
h1 + h42
ln
,
2
(4)2 m2Q
(4.10)
where Hvac (mQ , mQ , 0), Dvac (mQ ), Ivac (mQ ), are defined in Eqs. (C.19), (B.6), (A.5).
Moreover, let us now declare h24phys, phys 0. We then obtain the final expression for the
2-loop effective (bare) mass parameter m2H eff in the theory of Eq. (2.1):

m2Q
3
2 2
h
m
1
ln
(4)2 1 U
2T

2 3
2
1 2
h1phys h23 A + h22phys + h23 A I1 (mQ /T )
+T2
4
2

2
3T
31
5
1
4
T
ln
3
ln
+
c
+
h
+
(4)2 1 4 m2Q 4

1
4
h2 12I1 (mQ /T ) + D(mQ /T ) + 6H(mQ /T )
.
4
m2H eff = m2H phys
(4.11)
5. High-T and low-T expansions

We now wish to employ Eq. (4.11) to estimate in a non-trivial physical context the
accuracy of the high and low temperature expansions in mQ /T . We can do this by
inspecting the critical temperature Tc of the phase transition. Let us recall that the leading
(and next-to-leading in a gauge theory) terms in Tc are perturbative [35], thus ultraviolet
dominated and particularly sensitive to mQ /T . Most of the physical characteristics of the
phase transition, on the contrary, are infrared dominated and less sensitive to mQ /T . We
may also remind that in the MSSM the determination of Tc is physically more important
than in the Standard Model, since one has to address the question of whether other phase
transitions could take place before the electroweak one, in particular a transition to the
dangerous U -direction [18].
The transition will take place when m2H eff ( = g 2 T ) = #g 4 T 2 , where # is some nonperturbative coefficient, to be determined with lattice simulations. We shall keep the
285
physical parameters ( )phys fixed and vary mQ /T . It is then clear that the perturbative
contribution to Tc can equivalently be inspected by considering the finite part of the
2 0.
coefficient of T 2 in Eq. (4.11). We choose for simplicity h1phys, h2phys 1, h23 |A|
The high and low temperature limits of I1 , D, H are given in Eqs. (A.9), (B.9), (C.22).
To be in accordance with the limiting procedures usually applied in the literature, we
keep in the high temperature expansions terms up to logarithmic order, whereas in the
low temperature expansions we simply replace the exponentially small corrections in
Eqs. (A.9), (B.9), (C.22) with zero.
The numerically evaluated full expression for the coefficient of T 2 , as well as a
comparison of the high and low temperature expanded versions thereof with the full result,
are plotted in Fig. 2. We observe that the high temperature expansion gives typically too
large a coefficient of the T 2 -term, leading to too small a Tc . 5 With the low-temperature
expansion, on the other hand, Tc is slightly too large. Furthermore, we observe that while
naively one might have expected the crossover between the high and low temperature
regimes to be close to the first non-zero Matsubara frequency at mQ 2T , the low
temperature expansion is in fact perfectly sufficient already at mQ & 3T , which is the case
for realistic values mQ & 300 GeV. The fact that the high temperature expansion converges
relatively poorly as early as at mQ /T 2 is due particularly to the 2-loop function H,
whose behaviour is shown in Fig. 4 in Appendix C.
Fig. 2. Left: The coefficient of T 2 from Eq. (4.11), along with the high-T and low-T limits. The
parameters have been chosen as explained in Section 5. Right: The ratio of the high-T and low-T
limits to the non-expanded result.
5 Numerically the relative effect is larger here than in the realistic MSSM, since in that case there are other
fields contributing to the coefficient of T 2 (such as gauge bosons) for which the high temperature expansion
should work perfectly.
286
6. Conclusions
In this paper, we have pointed out that standard thermal resummations should be extended in two different ways, when one goes from the Standard Model to a general MSSM.
First of all, the left-handed stop mQ is typically of the order of magnitude 2T . Then it
cannot, a priori, be treated either in the high or in the low temperature expansion, but a more
general function appears. Second, the presence of dimensionful trilinear couplings leads to
the emergence of new linear terms coming from the scalar sunset diagrams. The results
for the scalar thermal counterterms including both of these effects in the model of Eq. (2.1)
are shown in Eqs. (3.5), (3.6), while the scalar contributions to the Debye masses are shown
in Eqs. (3.7), (3.8). In an effective theory approach such as the one followed in [36] (in [36]
it was assumed that mQ 2T , but the procedure can be extended to mQ 2T in a
straightforward way), all these effects of course arise automatically, whereas in a direct
computation of the 2-loop effective potential they should be explicitly taken into account.
In the framework of effective field theories, we have also extended the resummation
for the Higgs mass parameter to the next order beyond the effects described above. The
mass parameter thus determined, including corrections of order g 4 T 2 , could be used
for a precise estimation of the critical temperature of the corresponding electroweak phase
transition using 3d lattice Monte Carlo simulations. Let us stress that the only change with
respect to previous effective 3d theories is in the expressions for the effective parameters,
not in the functional form of the theory.
Using these results, we have estimated the accuracy of the high and low temperature
expansions used previously in the literature. Inspection of the critical temperature suggests
that the low temperature expansion, whereby all finite temperature contributions from
heavy particles are simply left out, works well already at m & 3T for bosonic particles.
Thus it should be completely clear that for the values mQ 1 TeV of interest for obtaining
a strong phase transition with experimentally allowed Higgs masses in the MSSM, the
Q-field can simply be left out in all finite temperature contributions.
The present results could clearly be extended in many directions. First of all, the
restricted model we have employed here can be extended to the full MSSM with gauge
fields and fermions in a straightforward way. Second, we have shown that the evaluation
of the integrals appearing in the perturbative 2-loop effective potential is numerically
feasible without any further temperature expansions thus the complete 2-loop potential
of the MSSM could in principle be computed, extending thus the results of [1,2,48,37,
38]. Third, we have here considered explicitly only the effects of a heavy mQ , while
in the MSSM many other mass parameters could be heavy as well. In particular, M2 ,
related to the gaugino and higgsino mass matrices and also relevant for providing sources
of CP violation can have values for which neither high nor low temperature expansions are
applicable. In the effective theory approach M2 , can be easily included at 1-loop level
without any temperature expansions [36], but this could now be extended to the 2-loop
level. The accuracy of previous approximations with respect to the contributions from the
second Higgs doublet with mA & 100 GeV could also be explicitly checked. Finally, we
287
2 . g 2 , an assumption
assumed that the trilinear couplings are not exceedingly large, |A|
which could be relaxed.
We believe that as long as the existence of a Higgs particle with MSSM type couplings
lighter than about 110 GeV is not experimentally excluded, these are worthwhile questions
to consider.
Acknowledgements
The work of M. Laine was partly supported by the TMR network Finite Temperature
Phase Transitions in Particle Physics, EU contract no. FMRX-CT97-0122. M. Losada
thanks A. Nieto for useful discussions during early stages of this work.
Appendix A. The tadpole

For completeness, let us review here some properties of the tadpole integral,
X Z d32 p
1
,
I (m) = T
32 p2 + p2 + m2
(2)
0
n
(A.1)
where p0 = pb 2nT for bosons, p0 = pf T (2n + 1) for fermions. We will need

two types of subdivisions of I (m). In the first case, relevant for all fermions and heavy
bosons, we write I (m) = Ivac (m) + IT (m), where IT vanishes at T = 0. In the second case,
relevant for light bosons (m2 (gT )2 ), we separate the contribution from the Matsubara
zero mode into I3d (m), writing I (m) = I3d (m) + In6=0 (m). We denote
1
1
,
nf () =
,
e 1
e +1
1/2
1/2
,
p = p2 + (m/T )2
,
p,i = p2 + m2i
Z
32
d
p nb(f ) (p,i )
.
IT ,b(f ) (mi ) =
32
p,i
(2)
nb () =
(A.2)
(A.3)
(A.4)
When the subscript ( )b,f is left out from I , we assume the bosonic case.
A.1. A heavy mass in the loop
Writing Ib (m) = Ivac (m) + IT (m), we get

2
m2
1
+
ln
+
1
,
Ivac (m) = 2
(4)2
m2

1 2 2
2
1 + 2 2 ln 2 + ln 2 I1 (m/T ) I2 (m/T ) ,
IT (m) = T
2
T
Z
nb ( p )
1
,
I1 (m/T ) = 2 dp p2
p
0
(A.5)
(A.6)
(A.7)
288
1
I2 (m/T ) = 2
Z
dp p2 (ln p2 )
0
nb ( p )
.
p
(A.8)
The limiting values are

y
y2
1
4
+
2E ,
1 + 2 ln
I1 (y) =
6 2
8 2
y
r
y ey
y1
=
,
2
y1

1
0 (2)
y
1 E +
ln y,
3
(2)
r
y ey
y1
(ln y + 2 E ln 2),
=
2
(A.9)
y1
I2 (y) =
(A.10)
where E = 0.57721566, 0 (2)/(2) = 0.56996099.

A.2. A light mass in the loop
Writing I (m) = In6=0 (m) + I3d (m), we obtain

2
m2
1
(1 + )
+ ln 2 + O 2 , m2 , m4 ,(A.11)
In6=0 (m) =
12
(4)2
T

2
mT
1 + ln 2 + 2 2 ln 2 + O 2 .
(A.12)
I3d (m) = 2
4
m
2 T
2
Here T is from Eq. (4.5) and [20]

= ln
2
0 (2)
.
+
2
2
ln
2
2
E
T2
(2)
(A.13)
A.3. The fermionic tadpole

Using the standard trick, the fermionic tadpole can be expressed in terms of the bosonic
tadpole:
If (m) = Ivac (m) + IT ,f (m),
IT ,f (m) = 2IT /2,b (m) IT ,b (m) = 2
(A.14)
1+2
IT ,b (2m) IT ,b (m).
(A.15)
Defining I1,f , I2,f as in Eqs. (A.7), (A.8) but with nf instead of nb , we obtain

1 2 2
2
IT ,f (m) = T 1 + 2 2 ln 2 + ln 2 I1,f (m/T ) + I2,f (m/T ) .
2
T
(A.16)
The high and low temperature expansions of I1,f , I2,f can be obtained from those of
I1 , I2 by noting that
289
1
I1,f (m/T ) = I1 (m/T ) I1 (2m/T ),
2
1
I2,f (m/T ) = I2 (m/T ) I2 (2m/T ) + (ln 2) I1 (2m/T ).
2
(A.17)
(A.18)
Appendix B. The bubble

Let us then consider the 1-loop bubble diagram with two propagators,
Z
X

1
1
1
=
(m
)
I
(m
)
, (B.1)
I
Db(f ) (m1 , m2 ) =
2
1
b(f
)
b(f
)
P 2 + m21 P 2 + m22 m21 m22
Pb(f )
where Pb(f ) = (pb(f ) , p) and we have taken the external momentum to zero. We can again
write
Db(f ) (m1 , m2 ) = Dvac (m1 , m2 ) + DT ,b(f ) (m1 , m2 ),

m21 + m22 m1
2
2 1
+
ln
+
1
ln
,
Dvac (m1 , m2 ) =
m1 m2
(4)2
m21 m22 m2
+ 2
DT ,b(f ) (m1 , m2 ) =()
(B.2)
(B.3)

1
T2
I1,b(f ) (m2 /T ) I1,b(f ) (m1 /T )
2
2
2 m1 m2
+ O().
The special case m1 = m2 gives the derivative of I (m) with respect to

dI (m)
= Dvac (m) + DT (m),
dm2

2
2 1
+ ln 2 ,
Dvac (m) =
(4)2
m
D(m)
2
D(m/T ) + O(),
(4)2
Z
nb ( p )
,
D(m/T ) = 4 dp
p
DT (m) =
(B.4)
m2 :
(B.5)
(B.6)
(B.7)
(B.8)
with
y
2
+ 2 ln
+ 2E ,
y
4
s
2 y
y1
e .
= 2
y
y1
D(y) =
We also need the derivative of In6=0 (m) with respect to m2 :

dIn6=0 (m)
2
2 1
+
ln
=
+ O , m2 .
Dn6=0 (m)
2
2
2
dm
(4)
T
(B.9)
(B.10)
290
Appendix C. The sunset

Let us then consider the bosonic and fermionic 2-loop sunset diagrams
Z X
Z
X
1
1
1
.
Hb(f ) (m1 , m2 , m3 ) =
2
2
2
2
P + m1 Q + m2 (P + Q)2 + m23
(C.1)
Pb(f ) Qb(f )
In the limit mi /T 1, it is known that [20,39]

1
T2
1
+ O(mi T ),
+
ln
+
Hb (m1 , m2 , m3 ) = 4
m1 + m2 + m3 2
(4)2 4
Hf (m1 , m2 , m3 ) = O(mi T ).
(C.2)
(C.3)
Our objective here is to compute these diagrams in the case of general mi . We are aware
of previous results in this direction in [40,41].
C.1. General case
The method we employ for evaluating Hb , Hf follows the standard procedure (see, e.g.,
[4042]). The twofold sum over the Matsubara modes is first written
R as a threefold sum
with a Kronecker delta, and the delta is then written as (p0 ) = T 0 dx exp(ip0 x). The
sums can now be performed,
T
X
pb(f )

nb(f ) (i ) (x)i
eipb(f ) x
+ xi .
=
e
() e
2
2
2
pb(f ) + i
i
(C.4)
The integral over x is then very simple. The outcome can be organized in a transparent
form, when different types of contributions are identified with known expressions; the
same result could also have been obtained from the rules of the real time formalism, as
noted for the 3-loop bosonic basketball diagram in [43]. In the remaining integral over the
spatial vectors p, q, we can perform at least the integration over z = p q/(|p||q|), leaving
for numerics at most a rapidly convergent 2d integral over p |p|, q |q|.
Let us denote
Z

d42 P
1

,
(C.5)
Q2 ; m21 , m22 =
42
2
2
(2)
P + m1 (P + Q)2 + m22

4 p2 + m2 q 2 + m2 m2 + m2 m2 2pq 2

1
2
1
2
3
fp,q (m1 , m2 ; m3 ) = ln
. (C.6)
4 p2 + m2 q 2 + m2 m2 + m2 m2 + 2pq 2
1
The explicit expression for
(Q2 ; m21, m22 ),
often denoted by B0 , is well known:

2 1

m21 + m22 m1
2
2
2
2
+
ln
+
1
ln
+
F
;
m
,
m
Q
Q2 ; m21, m22 =
E
1
2 ,
m1 m2
(4)2
m21 m22 m2
(C.7)
FE Q2 ; m21 , m22
= 1+
291
m21 + m22
ln
m21 m22
m1 m21 m22 m1
+
ln
m2
m2
Q2
q
2
2
1 m2 ) +Q
1 (m
p
2 +Q2
1 p
(m
+m
)
1
2
q
+ 2 (m1 + m2 )2 + Q2 (m1 m2 )2 + Q2 ln
.
2
2
Q
1 m2 ) +Q
1 + (m
(m +m )2 +Q2
1
(C.8)
The absolute value inside the logarithm in fp,q in Eq. (C.6) means that we take the real
part of the expression; the imaginary part would anyway cancel against Im .
With this notation, we obtain
Hb (m1 , m2 , m3 )
= Hvac (m1 , m2 , m3 ) +
X 4
32 4
i6=j 6=k
IT ,b (mi ) Re m2i ; m2j , m2k
i6=j 6=k
Z
dp p
0
dq q
nb (p,i ) nb (q,j )
fp,q (mi , mj ; mk ),
p,i
q,j
(C.9)
Hf (m1 , m2 , m3 )
= Hvac (m1 , m2 , m3 ) + IT ,b (m3 ) Re m23 ; m21 , m22
X

IT ,f (mi ) Re m2i ; m2j , m23
i6=j
4
+
32 4
Z
dp p
X 4
i6=j
where
i6=j 6=k
X
i6=j
dq q
0
Z
dp p
32 4
0
nb (p,3 ) nf (q,i )
fp,q (m3 , mi ; mj ),
p,3
q,i
dq q
(C.10)
nf (p,1 ) nf (q,2 )
fp,q (m1 , m2 ; m3)
p,1
q,2
(i,j,k)=(1,2,3),(2,3,1),(3,1,2)
(i,j )=(1,2),(2,1)
and the zero temperature contribution is

Z
Z
d42 P
1
1
1
d42 Q
.
Hvac (m1 , m2 , m3 ) =
42
42
2
2
2
2
(2)
(2)
P + m1 Q + m2 (P + Q)2 + m23
(C.11)
The only ultraviolet divergences are in Hvac (m1 , m2 , m3 ), which has 1/ 2 , 1/ poles, and
in the s, which have the pole 2 /(16 2 ), cf. Eq. (C.7).
292
Let us also mention a few words about the numerical evaluation of the 2d integrals
involving fp,q (m1 , m2 ; m3), left to be carried out in Eqs. (C.9), (C.10). These integrals
are of course well-defined and finite. However, if m3 < |m1 m2 | or m3 > m1 + m2 ,
they involve integration over logarithmic singularities. In our numerics, we found that the
integration is more effective if we factorize out the singularities explicitly. Suppose we, for
instance, first perform the integral over q. If m1 6= 0, we write

p
m23 m21 m22 ,
(C.12)
a=
2
2m1
1/2 4

2 1/2
1
,
(C.13)
p2 + m21
m3 2m23 m21 + m22 + m21 m22
b=
2
2m1

(q a b)(q + b a)
,
(C.14)
fp,q (m1 , m2 ; m3 ) = ln
(q + a + b)(q + a b)
and there are then singularities at q = |a + b|, |a b|. If m1 = 0, we write
2

q q0
4m22 p2 m23 m22

,

,
fp,q (0, m2 ; m3 ) = ln
q0 =
q + q0
4p m23 m22
(C.15)
and there is a singularity at q = |q0 |. If m2 = m3 , fp,q (0, m; m) = 0.

C.2. One heavy, one light mass
Let us now consider in more detail some special cases of Hb (m1 , m2 , m3 ) needed in
the main part of this paper. For the consideration in Section 3, we need to know how
Hb (mH , mU , mQ ) behaves for small mH , mU . We claim that there is a linear term
mH , mU (modulo logarithms). Since the result is symmetric in mH , mU , it is enough to
consider mH mQ .
Non-analytic terms can only arise from a zero Matsubara mode. Thus,
Z
Z
X
mH mQ
d32 p
1
1
1
. (C.16)
Hb
(2)32 p2 + m2H
q02 + (p + q)2 + m2U q02 + q 2 + m2Q
Q
Let us denote the latter integral by (p). It is then obvious that the leading behaviour must
be
Z

mH mQ
d32 p
1
1
(0) = I3d (mH ) 2
I (mU ) I (mQ )
Hb
32
2
2
2
(2)
p + mH
mQ mU

2

mU m2U
1
I
(m
)
I
(m
)
+
I
(m
)
+
I
(0)
1
+
O
,
.
=
3d
H
Q
3d
U
n6=0
m2Q
m2Q T 2
(C.17)
Indeed, the remainder,
Z

1
d32 p
(p) (0) ,
32
2
(2)
p 2 + mH
(C.18)
293
Fig. 3. The finite part of Hb (mH , mQ , 0) Hb (0, mQ , 0) (the zero temperature contribution has
been subtracted) with solid lines, compared with the linear term in Eq. (C.17) with dashed lines.
The results have been divided by T 2 /(4)2 .
R
behaves at small p as dp p4 /(p2 + m2H ). This is IR finite even after expanding in m2H ,
therefore there cannot be any further linear contributions.
In order to verify this behaviour explicitly, we set mU = 0 and compare the finite part of
Eq. (C.17) with a numerical evaluation of the finite part of Hb (mH , mQ , 0) in Eq. (C.9).
We fix = T , and choose mQ /T = 1.0, 2.0, 4.0. The result is shown in Fig. 3. Note
that in Eq. (C.17), one must include a contribution mH ln mH , arising when the O()
part of I3d (mH ) (cf. Eq. (A.12)) combines with the 1/ pole in I (mQ ). From the perfect
agreement at small mH /T in Fig. 3, we conclude that for mH /T 1 the behaviour is
indeed according to Eq. (C.17).
C.3. Two equal heavy masses
Finally, let us consider the special case needed in Section 4, Hb (m, m, 0) (cf. Eq. (4.2)).
is
The T = 0 part in Eq. (C.11), related to 2-loop vacuum renormalization of m2H (),

2
2
3
2 2 1
4 m
+ O() .
(C.19)
+ +7+
Hvac (m, m, 0) =
(4)4 m2
2
6
Using Eq. (A.6) as well as the simple expressions for (0; m2, m2 ), (m2 ; m2 , 0)
obtained from Eqs. (C.7), (C.8), we then get from Eq. (C.9)
Hb (m, m, 0)

2
1
2
1
+ I1 (m/T )
+ ln 2 + ln 2
= Hvac (m, m, 0) +
12

T
m

1
0 (2)
E ln 2
+ H(m/T ) ,
+ (4 2 ln 2)I1 (m/T ) I2 (m/T ) +
6
(2)
(C.20)
4
T2
(4)2

294
Fig. 4. Left: The function H Hfinite in Eq. (C.21), along with the high-T and low-T limits in
Eq. (C.22). Right: The ratio of the high-T and low-T limits to the non-expanded result.
where I1 , I2 are from Eq. (A.8), and

2
H(m/T ) = 2
Zp
dp p
dq q
0
nb ( p ) nb ( q ) p + q
.
ln
p
q
pq
(C.21)
The function H(m/T ), numerically very easily evaluated, is plotted in Fig. 4, together with
a comparison with the limiting values

1
0 (2)
2m 1
mT
+ E
,
H(m/T ) = ln
2
T
3
(2)
mT 1 2m/T
e
=
.
(C.22)
2
References
[1] D. Bdeker, P. John, M. Laine, M.G. Schmidt, Nucl. Phys. B 497 (1997) 387, hep-ph/9612364.
[2] M. Carena, M. Quirs, C.E.M. Wagner, Nucl. Phys. B 524 (1998) 3, hep-ph/9710401.
[3] M. Laine, K. Rummukainen, Phys. Rev. Lett. 80 (1998) 5259, hep-ph/9804255; Nucl. Phys.
B 535 (1998) 423, hep-lat/9804019.
[4] J.M. Cline, G.D. Moore, Phys. Rev. Lett. 81 (1998) 3315, hep-ph/9806354.
[5] M. Losada, Nucl. Phys. B 537 (1999) 3, hep-ph/9806519.
[6] M. Losada, Nucl. Phys. B 569 (2000) 125, hep-ph/9905441.
[7] S. Davidson, T. Falk, M. Losada, Phys. Lett. B 463 (1999) 214, hep-ph/9907365.
[8] J.M. Cline, G.D. Moore, G. Servant, Phys. Rev. D 60 (1999) 105035, hep-ph/9902220.
[9] S.J. Huber et al., hep-ph/9912278.
[10] F. Csikor et al., hep-ph/0001087.
[11] G.D. Moore, hep-ph/0001274.
[12] S. Davidson, M. Losada, A. Riotto, hep-ph/0001301.
[13] P. John, M.G. Schmidt, hep-ph/0002050.
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
295
T. Prokopec, hep-ph/0002181.
M. Joyce, K. Kainulainen, T. Prokopec, hep-ph/0002239.
J.M. Cline, K. Kainulainen, hep-ph/0002272.
M.E. Shaposhnikov, Phys. Lett. B 277 (1992) 324; Phys. Lett. B 282 (1992) 483 (Erratum).
M. Dine, R.G. Leigh, P. Huet, A. Linde, D. Linde, Phys. Rev. D 46 (1992) 550, hep-ph/9203203.
A.D. Linde, Phys. Lett. B 96 (1980) 289.
P. Arnold, O. Espinosa, Phys. Rev. D 47 (1993) 3546, hep-ph/9212235; Phys. Rev. D 50 (1994)
6662 (Erratum).
K. Farakos, K. Kajantie, K. Rummukainen, M. Shaposhnikov, Nucl. Phys. B 425 (1994) 67,
hep-ph/9404201.
K. Kajantie, M. Laine, K. Rummukainen, M. Shaposhnikov, Nucl. Phys. B 458 (1996) 90, hepph/9508379; Phys. Lett. B 423 (1998) 137, hep-ph/9710538.
A. Jakovc, A. Patks, Nucl. Phys. B 494 (1997) 54, hep-ph/9609364.
F. Csikor, Z. Fodor, J. Heitger, Phys. Rev. Lett. 82 (1999) 21, hep-ph/9809291.
M. Laine, JHEP 06 (1999) 020, hep-ph/9903513.
K. Kajantie et al., Nucl. Phys. B 466 (1996) 189, hep-lat/9510020; Phys. Rev. Lett. 77 (1996)
2887, hep-ph/9605288; Nucl. Phys. B 493 (1997) 413, hep-lat/9612006.
M. Grtler et al., Nucl. Phys. B 483 (1997) 383, hep-lat/9605042; Phys. Rev. D 56 (1997) 3888,
hep-lat/9704013.
F. Karsch et al., Nucl. Phys. B (Proc. Suppl.) 53 (1997) 623, hep-lat/9608087.
G.D. Moore, N. Turok, Phys. Rev. D 55 (1997) 6538, hep-ph/9608350.
O. Philipsen et al., Nucl. Phys. B 528 (1998) 379, hep-lat/9709145.
K. Rummukainen et al., Nucl. Phys. B 532 (1998) 283, hep-lat/9805013.
J.M. Cline, K. Kainulainen, Nucl. Phys. B 482 (1996) 73, hep-ph/9605235.
M. Losada, Phys. Rev. D 56 (1997) 2893, hep-ph/9605266.
M. Laine, Nucl. Phys. B 481 (1996) 43, hep-ph/9605283; Nucl. Phys. B 548 (1999) 637
(Erratum).
P. Arnold, Phys. Rev. D 46 (1992) 2628, hep-ph/9204228.
M. Laine, K. Rummukainen, Nucl. Phys. B 545 (1999) 141, hep-ph/9811369.
J.R. Espinosa, Nucl. Phys. B 475 (1996) 273, hep-ph/9604320.
B. de Carlos, J.R. Espinosa, Nucl. Phys. B 503 (1997) 24, hep-ph/9703212.
P. Arnold, C. Zhai, Phys. Rev. D 50 (1994) 7603, hep-ph/9408276.
R.R. Parwani, Phys. Rev. D 45 (1992) 4695, hep-ph/9204216; Phys. Rev. D 48 (1993) 5965
(Erratum).
A. Jakovc, Phys. Rev. D 53 (1996) 4538, hep-ph/9502313.
A.I. Bugrii, V.N. Shadura, hep-th/9510232.
J.O. Andersen, E. Braaten, M. Strickland, hep-ph/0002048.

Various modified solutions of the RandallSundrum

model with the GaussBonnet interaction
Jihn E. Kim 1 , Bumseok Kyae, Hyun Min Lee
Department of Physics and Center for Theoretical Physics, Seoul National University,
Seoul 151-742, South Korea
Abstract
The GaussBonnet interaction is the only consistent quadratic interaction below the Planck scale in
the RandallSundrum compactification. We study various static and inflationary solutions including
this GaussBonnet interaction. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Mj; 12.10.Dm; 98.80.Cq
Keywords: RandallSundrum compactification; GaussBonnet interaction; Brane cosmology
1. Introduction
Recently, Randall and Sundrum (RS) proposed a compactification scheme with
nonvanishing cosmological constant in the bulk [1] which has immediately attracted a
great deal of attention [25]. The most simple compactification studied in superstring
models before the RS proposal has been the orbifold compactification [6] in which the
compactified space is flat. On the other hand, the RandallSundrum compactification
allows a nonflat compactified space, but the analysis is relatively simple. Because of the
nonflat nature of the bulk between two branes, there exists an exponential warp factor
for metrics going from one brane to the other [1]. This exponential warp factor has been
suggested for a large hierarchy between the Planck scale MP = 2.44 1018 GeV and the
electroweak scale v ' 250 GeV.
Among two branes, let Brane 1 (B1) be the hidden-sector brane and Brane 2 (B2) the
visible sector brane. An exponential warp factor suppresses the soft mass in the visible
brane B2, and it is possible to obtain this small ratio because the Higgs mass term at B2 is a
dimension two operator. Thus, in the RS world, one changes the traditional gauge hierarchy
problem to a problem in geometry. Using the same argument, the nonrenormalizable
1 jekim@phyp.snu.ac.kr
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 1 8 - 7
J.E. Kim et al. / Nuclear Physics B 582 (2000) 296312
297
operators are suppressed not by MP but by v. Thus one has to make sure that the theory
has a high degree of symmetry to suppress sufficiently the unwanted operators.
Another problem is the problem of inflation. Generally, inflation occurs unless one finetunes the bulk cosmological constant and the brane tensions [24]. For the fine-tuned
relations [35], there exist static solutions. So far it has not been shown that any of the
static solutions is the t limit, not allowing a graceful exit from the inflationary period.
In addition, there is a possibility that the separation between the branes is expanding or
shrinking exponentially. However, this last problem may be understood by introducing a
scalar field in the bulk [7].
The most interesting point of the RS world is the interplay of the bulk and the brane
world. In particular, the bulk cosmological constant, k, and the brane tensions, ki (i = 1, 2),
must be related, k1 = k = k2 and k1 > 0. But the expansion rate of the observable
universe is measured by the Hubble parameter which is a function of k and k2 . These
ks are the appropriately defined from the original bulk cosmological constant b and the
brane tensions 1 , 2 at B1 and B2. 2 One interesting point of the RS compactification
is that there may exists a possibility of understanding the old cosmological constant
problem.
Below the Planck scale, the higher order effective interaction in the RS model is known
to be the GaussBonnet interaction [5]. In contrast to the models without the GaussBonnet
interaction, this model allows solutions with a positive 2 at the visible brane, which is
suitable for a proper expansion in the standard big bang cosmology after the inflationary
period.
In Section 2, the RS compactification with the GaussBonnet term is explored. In
Section 3, the static background solutions with the GaussBonnet interaction are presented.
In Section 4, simple inflationary solutions are given. In Section 5, we present other
possible inflationary solutions. In Section 6, metric perturbation near the static background
geometry is discussed.
2. GaussBonnet interaction
We will neglect the matter interaction, and consider only the gravitational interaction
with cosmological constants in the bulk and at the branes. The spacetime dimension is
D = 5. The fifth dimension x 5 y is compactified with an S1 /Z2 orbifold. The fivedimensional index is M, N = 0, 1, 2, 3, 5 and the four-dimensional brane world index is
, = 0, 1, . . . , 3. The fifth dimension variable y ranges in the region [0, 1/2]. The S1 /Z2
orbifold is used to locate the two branes at y = 0 and y = 1/2. The periodicity of y is 1.
Below the Planck scale, the higher order gravity effects can be added as effective
interaction terms. Since we are neglecting the matter interactions, the possible terms in
the Lagrangian is, up to O(R 2 /M 2 ),
2 Notations for ks are borrowed from Refs. [3,4].
298
Z
S=
3

M
1
1
1
R b + MR 2 + MRMN R MN + MRMNP Q R MNP Q
g
2
2
2
2
q
X Z
(1)
d4 x g (i) (Li i ),
d5 x
+
i=1,2 branes
where g, g (i) are the determinants of the metrics in the bulk and the branes, M is the
five-dimensional gravitational constant, b and i are the bulk and brane cosmological
constants, and , , are the effective couplings. We assume that the three-dimensional
space is homogeneous and isotropic, and hence the metric is parametrized by n, a, and b
ds 2 = n2 (, y) d 2 + a 2 (, y)ij dx i dx j + b2 (, y) dy 2,
(2)
where the Roman characters i, j denote the space indices 1, 2, and 3.

It has been found that there exist solutions consistent with the RandallSundrum setup
if the additional interaction is of the GaussBonnet type [5], namely, = 4 and =
are satisfied. In this case, the higher y derivative terms, n0000 , n000 n0 , n00 n00 , a 0000 , a 000 a 0 ,
a 00 a 00 , . . ., 3 are absent in the left-hand side of the Einstein equation. These conditions for
vanishing higher y derivative terms are necessary since the right-hand side of the Einstein
equation contains only one power of the Dirac delta function and the higher derivative
terms diverge more rapidly than the delta function at the branes. We find that this result is
highly nontrivial.
However, it can be anticipated from the fact that the GaussBonnet term can be rewritten
as a pseudoscalar quantity, MNOP QST U RMN QS ROP T U . From the antisymmetric
property of the Riemann tensor, we observe that there are no (a 00 )2 , (n00 )2 and a 00 n00
terms 4 in the action which would have given the unwanted fourth order derivatives in
the equations of motion. Thus, the GaussBonnet effective interaction does not contain
higher y derivatives.
The GaussBonnet term E = R 2 4RMN R MN + RMNP Q R MNP Q is a total derivative
in D = 4 spacetime, in which case it does not change the Einstein gravity. On the other
hand, for D 6= 4 it is not a topological quantity any more. Still, it does not contribute to
the massive poles of the spin-2 propagator [8,9]. It means that the metric variations near
the flat space do not give rise to a ghost graviton even with the GaussBonnet term. In
general, a combination of the quadratic curvature terms without the GaussBonnet ratio
leads to ghosts. But it may not be meaningful if the location of the ghost pole in the
graviton propagator is above the Planck scale where the derivative expansion breaks down.
However, the GaussBonnet term possibly excites ghost particles near anti-de Sitter space
in the sense that the sign of the propagator can be flipped [9].
The general curvature squared terms in any spacetime dimension D can be rewritten
as [10],
3 Here, 0 denotes the derivative with respect to y.
4 Under the metric assumption Eq. (2), the terms including b00 are absent in the action. We can see it from the
antisymmetric property of the Riemann tensor, R M NST = S M T N and unique non-vanishing Christoffel
symbol containing the first derivative of b with respect to y is 5 55 = b0 /b.
R 2 + RMN R MN + RMNP Q R MNP Q

D2
(D 2) + 4
E+
+ C2
=
4(D 3)
D3
4

4(D 1) + D + 4
R2 ,
+
4(D 1)
299
(3)
(4)
where E is the GaussBonnet term, C 2 is the square of the Weyl tensor as follows,
E = R 2 4RMN R MN + RMNP Q R MNP Q ,
2
C 2 = RMNP
Q
4
2
2
RMN
R2 .
+
D2
(D 1)(D 2)
(5)
(6)
Note that if the metric is conformally flat and 16 + 5 + 4 = 0 in D = 5, the

curvature squared terms appear necessarily in the GaussBonnet combination because the
Weyl tensor vanishes for a conformally flat metric, n(, y) = a(, y) = b(, y). Then the
coefficient of the resultant GaussBonnet term becomes (8 + )/4.
For n(, y) = a(, y) and 16 + 5 + 4 = 0, there still exist higher time derivatives
2 in the action from the curvature squared terms. Thus, to eliminate
(4 + )(a/a
b/b)
higher time derivatives too, in addition to the condition 16 + 5 + 4 = 0, we should
choose the GaussBonnet form in curvature squared terms or conformally flat metric,
n(, y) = a(, y) = b(, y).
Variations of the above action with the GaussBonnet term gives, apart from those for
the brane Lagrangian,

g RMN gMN R
gMN R 2 4RP Q R P Q + RST P Q R ST P Q
2
2
2M
2

2
+ 2 RRMN 4RMP RN P + RMQSP RN QSP + 2 gMN R;P ;P R;M;N
M
M

4
P
;P
PQ
2 gMN R ;P ;Q + RMN;P RM ;N;P RN P ;M;P
M

2
P Q
P Q
+ 2 RM N ;P ;Q + RM N ;Q;P
M

q
3
(1)
M N (y)
= M
b ggMN + 1 g (1) g

q
1
(2)
(2)
,
(7)
+ 2 g g M N y
2
where 1 refers to the brane of the hidden world B1 and 2 refers to the visible brane B2.
The left-hand side of the above equation contains the extra term due to the GaussBonnet
term, g XMN , in addition to the familiar Einstein tensor g GMN . With the metric
given in Eq. (2), the GMN and XMN are

a a b
3n2 a 00 a 0 a 0 b0
+
+
2
,
(8)
G00 = 3
a a b
b
a
a a
b
300

n a
b n
a
a 2 a b a
+
2
2
b n
a
n2 a b a n a
0

0

00
2
00
0
0
0
a
a
a
a0
n
b n
a n
+2 +
+2
2 +
,
+ 2
b n
a
a
n
a
b n
a

3b2 a a n a
a 0 n0 a 0
+
+3
,
G55 = 2
n a a n a
a n
a

0
b a 0 a 0
a n
+
,
G05 = 3
a n
ba
a
Gii =

0 2 a 0 2 a 00 n2 a 0 3 b0 n2
12 a 3 b
a 2 a 00 a 2 a 0 b0 a ba
+
+
3 5 ,
M 2 a 3 n2 b a 3 b2
a 3b3
a 3 b3
a 3 b4
a b

02
a 2n b
aa
00
4
aa
0b0
a 2 n00
a a b
ba
Xii = 2 2 4 + 3 5 + 2 2 2 2 2 3 + 2 3 + 3 2
M
n b
n b
n b
n b
n b
n b
2
a 2n0
a 2 b00
a 2 n0 b0
a na
00
a na
0b0
b 2a 0 2
4 2 4 2 3 3 2 3 2 +2 3 3 2 2 4
n b
n b
n b
n b
n b
n b
0 n0 b na
0
a ba
02
(a 0 )2
a 0an
a 0 ba
a 00 a 0 n0
0
2 3 3 3 3 2 2 2 + 4 3 2 + 4 2 3 2
n b
n b
n b
n b
n b
nb4

2
2
n00 a 0
a 0 n0 b0
+3
,
4
nb
nb5

12 a 3 nb
aa
02
a 2 a 0 n0 a na
02
a 0 3 n0
2 a a 2 b2
3 4 + 3 2+ 3 3 3 3 3 2 ,
X55 = 2
M
a 3 n5
a n
a n
a n
a n
a nb
3 0

2
3
0
0
aa
0 n0
12 a n
a 2 ba
ba
a 2 a 0
a 0 a 0 2
+
,
X05 = 2
M
a 3 n3 a 3 n2 b a 3 nb2
a 3 b3 a 3 n2 a 3 b2
X00 =
(9)
(10)
(11)
(12)
(13)
(14)
(15)
where 0 denotes the derivative with respect to y and denote the derivatives with respect
to .
The equation of motions Eq. (7) are
GMN + XMN = TMN ,
(16)
where
"
#
y 12
n2
(y)
1 +
2 ,
T00 = 3 b +
b
b
M
"
#
y 12
a2
(y)
1 +
2 ,
Tii = 3 b +
M
b
b
T55 =
b2
b ,
M3
T05 = 0.
(17)
301
3. Static solutions
To find static solutions, let us assume that the metric takes the following form,
ds 2 = e2 (y) dx dx + b02 dy 2 ,
(18)
where the length parameter b0 is a constant. Note that the modified Einstein equation for
the (00) component is identical to the (ii) component in Eq. (16). Thus, the (00), (ii), and
(55) components of Eq. (16) lead to two equations,

2
1
4
1
3 00
0 2
( ) = 3 (y) + 3 y
,
(19)
1
M b0
M b0
2
b02
M 2 b02
6( 0 )2
b02

2
b
0 2
( ) =
.
1
2
2
M3
M b0
(20)
There exist two solutions of Eq. (20), consistent with the orbifold symmetry y y,
2

M
4b 1/2 1/2
k b0 |y|.
(21)
1 1+
= b0 |y|
4
3M 5
Let us call + (or k+ ) and (or k ) solutions + and solutions, respectively. Note
that both positive and negative bulk cosmological constants are possible for the +
solution + . If b = 0, we have an AdS space for the + solution irrespective of ,
and a Minkowski space for the solution [9]. The RS solution is obtained by taking
0 in the solution.
These solutions exist for:
(i) < 0 and b < 0 allows only , and
(ii) > 0 allows both solutions. The + solution is possible for both b > 0 and
b < 0. The solution is possible only for b < 0. In any case, there exists the
lower limit of b , b > 3M 5 /4.
Comparing our results with that of Randall and Sundrum k = (b /6M 3 )1/2 , the
effective bulk cosmological constant by the GaussBonnet interaction can be defined as
!
r
3M 5
4b
(b)
2
.
(22)
1 1+
= 6M 3 k
eff
2
3M 5
This is because our geometry of the AdS space guarantees a negative bulk cosmological
constant effectively. For to be a real number it should have the negative sign, which is
the same as in the RS case.
Considering the discontinuities at the branes,

1
0
1,
(23)
|y| = 2 (y) y
2
which comes from the periodicity in y direction and the orbifold symmetry, we obtain two
solutions if the following relations among the brane cosmological constants are satisfied
302
p
Fig. 1. Possible solutions for 2 2 / 6M 3 |b | as a function of 4b /(3M 5 ). The star
point is the RS solution. The four quadrants have different sets of signs of and b , denoted as
(sign of , sign of b ).
=
2
4b
= 6k M 1 +
3M 5

2

4b 1/2
4b 1/2
3 M
1 1+
= 6M
,
1+
4
3M 5
3M 5
3
(24)
where k > 0. The RS solution is obtained by taking 0 in the solution. Note that
the visible brane can take a positive cosmological constant. It will be important in the later
stage of the evolution of the universe, which will be considered in the following section.
Possible solutions are depicted in Fig. 1 as a function of the GaussBonnet p
coupling .
The vertical axis ( 2 ) is the solution for 2 in the visible brane in units of 6M 3 |b |,
and the horizontal axis ( ) is defined as 4b /(3M 5 ).
Similarly, we can define the effective brane cosmological constants as
eff(i) p
i
1 + 4b /3M 5
6M 3 ki, .
Thus we see the RS-like fine tuning conditions again,

s
(b)
eff
eff(1)
eff(2)
=
.
6M 3
6M 3
6M 3
The Planck constant at B2 (visible sector) is given by [1]
Z1/2
MP = M b0
2
1/2
dy e2k b0 |y| =

M3
1 ek b0
k
(25)
(26)

=M

4b 1/2 1/2
1
1 1+
1 ek b0 .
4
3M 5
303
(27)
The Higgs boson mass parameter at the visible sector is obtained by redefining the Higgs
field such that the kinetic energy term of the Higgs boson takes a standard form [1]. Thus
the Higgs mass parameter is given by

b
b0 M 2
4b 1/2 1/2
20 k
m0 = m0 exp
1 1+
(28)
me
2 4
3M 5
where m0 is the mass given in the fundamental Lagrangian, before redefining the Higgs
field.
For k+ b0 ' 74, the + solution gives a needed large mass hierarchy through the warp
b0
factor e 2 k+ from the input mass parameter (M) of order 1019 GeV, leading to a TeV
scale observable mass. To achieve a sufficient hierarchy, Randall and Sundrum set k 2 =
b /(6M 3 ) M 2 MP2 and b0 /2 37/M. These results can be reproduced for our
+ solution with = O(1) and b = O(1) M 5 . For example, k+ = M MP for
b /M 5 = 12 6 and > 1/4. In this case, the brane cosmological constants are given by
+
4
1 = 2 = 6M |4 1|. Thus, we have to set b0 /2 = 37/M to explain the hierarchy

between the Planck and TeV scale. Note that in this case k+ is not zero even for b = 0 or
= 1/2.
The value of k+ can be smaller or larger using the parameter and the bulk cosmological
constant b . Smaller k+ require longer interval length to explain the hierarchy between
those scales, which results in lighter KaluzaKlein (KK) modes of the graviton. As the KK
modes must interact with the standard model particles through the gravitational interaction,
the lighter KK mode has the longer lifetime. And sufficiently longer life time of the KK
modes could have an effect on nucleosynthesis. According to Ref. [11], the masses of the
KK modes should be larger than about a few GeV, which corresponds to k+ b0 /2 . 40 to
be consistent with the nucleosynthesis scenario. Therefore, smaller k+ than M cannot be
consistent with the current cosmology.
On the other hand, a larger k+ corresponds to a larger curvature and then it would
locate our theory out of perturbative regime. Thus, it isnt desirable. We will show,
however, in Section 6 that if quadratic curvature terms have the GaussBonnet ratio, at
least the quadratic corrections do not affect linearized 4-dimensional Einstein gravity or
non-relativistic Newtonian gravity regardless of the curvatures magnitude.
For k b0 ' 74, the solution has the same behavior. The case of b /M 5 = 12 6
4
and < 1/4 corresponds to k = M MP and +
1 = 2 = +6M |1 4|. Then we
have only to set b0 /2 = 37/M to explain the hierarchy between two scales too. Of course,
we also have the freedom to make the k smaller or larger than M depending on the
parameters and b .
The small warp factor [1] makes it possible to generate a TeV scale mass from
the fundamental parameter of O(M). But these TeV scale masses also appear in the
other mass parameters of the effective operators. In particular, the operators leading to
proton decay are also parametrized by a TeV scale mass. Therefore, one has to suppress
304
sufficiently the low dimensional proton decay operators such that it is sufficiently longlived (p > 1032 years), allowing operators with D > 14 only.
In non-GaussBonnet cases satisfying the condition 16 + 5 + 4 = 0, the RS type
solution is still valid except for the substitution 4 8 + in Eq. (21) because the RS
metric can be redefined to be conformally flat.
4. Inflationary solutions
For inflationary solutions we impose an ansatz,
n = f (y),
a = g( )f (y),
b = b0 ,
(29)
where b0 is a constant. Now adding the (00) and (ii) equations in Eq. (16), we obtain

4 f 00
g
= 0.
(30)
1 2 2
2
g
M b0 f
0. So we define
Since f 00 necessarily gives rise to a delta function we should take (g/g)=
(g/g)
H0 = constant. Then the (55) equation gives

2 0 2

2 0 2
f
f
b
1
1 2
2
H0
H0
=
.
(31)
+ 2
3
2
2
f
f
M
f
f
6M
b0
b0
After little algebra, we obtain
0 2
f
2 2
= H02 + k
f ,
b0
(32)
2 is defined in Eq. (22). For the case with = 0, we arrive at the solution
where the k
2 = /6M 3 . Note that k 2 contains the cases of both positive and
given in Ref. [4], k
b
+
2 can take both positive and negative
negative cosmological constants in the bulk and the k
signs. Inflationary solutions were obtained for a flat bulk geometry and for an AdS bulk
geometry [2,3,17], which can solve the hierarchy problem in the static limit. In our case,
as one can see below, inflationary solutions exist also for a positive bulk cosmological
constant.
2 > 0, the solution consistent with the orbifold symmetry is
For k
f=

H0
sinh k b0 |y| + c0 .
k
(33)
The (00) or (ii) equations of Eq. (16) just give a boundary condition for the solution. Using
the relation given in Eq. (23), one can find that the imposed conditions determine the extra
dimension scale b0 and the integration constant c0 as follows,

1
k2, = k coth kb0 + c0
(34)
k1, = k coth(c0 ),
2
where the ki are defined in Eq. (25). Here, the solutions are valid only for k < |k1, | <
|k2, | in case c0 > 14 kb0 and k < |k2, | < |k1, | in case c0 < 14 kb0 . One can also check
305
easily that the ki tends to those of Ref. [4] in the limit of 0 in the lower case (i.e., the
solution). In general, inflation occurs if parameters , b , 1 , and 2 do not satisfy
the two relations implied by Eq. (24). From the above relations the k1, (k2, ) diverges as
k2, (k1, ) k coth( 12 kb0 ).
Then the metric is
2

H0
sinh2 k b0 |y| + c0 d 2 + e2H0 ij dx i dx j + b02 dy 2.
(35)
ds 2 =
k
To obtain the RS static solution with the warp factor in the visible brane, we should take
H0 0 and c0 + while keeping the ratio (H0 ec0 )/(2k ) 1 fixed. Then we obtain
the fine tuning condition k1, = k2, = k from Eq. (25), which is the same result
as Eq. (24). Here one can see the possibility of the warp factored brane with the positive
cosmological constant again.
After 4-dimensional coordinate transformation at a given y to make the 4-dimensional
metric be in the form ds42 = dt 2 + e2H (y)t ij dx i dx j [4], we get the Hubble parameter
expressed in terms of the cosmological constant and the energy density, Hvis, =
q
2 . Here k 2
2
(kvis, )2 k
vis, = k for the static solutions and the two parameters
corresponding to the + and solutions at the visible brane, kvis, , are given by
2 + vis
p
,
(36)
kvis, =
6M 3 1 + (4b /3M 5 )
where
2 vis . Thus the Hubble parameter at B2 is given by [5]

vis vis + 2
vis
vis
2
2
p
=
Hvis, =
1+
.
36M 6(1 + 4b /3M 5 ) 3M 2 1 + 4b /3M 5
2
2
Pl
(37)
The second equation above is derived with the use of Eqs. (24) and (27). With vis = 0, we
obtain the previous static solution. But with
2 = b < 0 where the original RS solution
sits, there exists a possibility that vis (22 + vis ) < 0 at a sufficiently low temperature,
and hence it is difficult to obtain a real Hubble parameter [1216]. But with a positive
+
2 , there does not exist such a problem. This is possible for our + solution for > 0.
Therefore, with the + solution we can obtain a plausible FriedmannRobertsonWalker
universe after inflation ends.
2 term gives a correction to the conventional Friedmann equation but
In Eq. (37), the vis
+
in the limit 2 and b /M 5 0 we recover the standard 4-D general relativistic
result. The modified Friedmann equation leads to the modified inflation condition,

1
2vis
vis
vis
3pvis
a
p
=
1+ + +
1+ +
>0
(38)
a 3M 2 1 + 4b /3M 5 2
2
2
2
Pl
or

vis +
2 + 2vis
,
pvis <
3
+
2 + vis
where the vis and pvis satisfy the fluid equation, vis + 3H (vis + pvis ) = 0.
(39)
306
Now let us consider the case that the only matter in the 4-D universe is a self interacting
scalar field, inflaton . Then the vis and pvis are given by vis = 12 2 + V () and pvis =
1 2
2 V (), respectively, and the Eq. (39) becomes [19]

2

+ 2V ()
2
2
(40)
5 2V () < 0,
V () +
8+
2
which reduces to 2 < V when 2 + 2V +
2 . Assuming that the inflaton field rolls down
to a true vacuum very slowly, the energy density is dominated by the potential V and the
inflaton field evolution is strongly damped, which implies

V
V
p
1+
,
(41)
H2 '
2+
3MP2 l 1 + 4b /3M 5
2
V0
,
'
3H
(42)
where we use ' to denote equality within the slow-roll approximation [18]. Our brane
physics modifies also the e-folding number as follows;
Ztf
dt H '
N=
ti
MP2 l 1 + 4b /3M 5
Zf
d
i

V
V
1
+
.
V0
2+
2
(43)
5
For a large +
2 and small b /M we obtain the standard e-folding formula again.
Next, let us consider the corrections to the scalar and tensor density perturbations.
The scalar density perturbation can be related to the curvature perturbation on uniform
density hypersurfaces when modes re-enter the Hubble radius during the matter dominated
era [18,19],
A2S =
4
2
,
25
H
,
(44)
where the scalar field fluctuation at Hubble crossing (k = aH ) are given by h 2 i '
(H /2)2 . Thus, using the slow-roll conditions Eqs. (41) and (42), the amplitude of scalar
perturbations becomes

3 3

1
V
1
V 3
p
.
(45)
1
+
A2S '

75 2 M 2 1 + 4b /3M 5 V 02
2+
k=aH
2
Pl
So the amplitude of scalar perturbations is increased relative to the standard result. Of
5
course, we recover the standard one for a large +
2 and small b /M .
The amplitude of tensor (gravitational wave) perturbation at Hubble crossing is given
by [18,19]

1
H 2
2
.
(46)
AT =
50 2 MP l k=aH
In the slow-roll approximation, this yields
A2T '

1
1
V
p
V
1
+

150 2 M 4 1 + 4b /3M 5
2+
k=aH
2
Pl
307
(47)
which is increased by brane effects, but with a smaller factor than in the case the scalar
5
perturbation. This tends to the standard form as +
2 and b /M 0.
In the previous section, we discussed the phenomenologically favored values of b /M 5 ,
+
4
2 and . The result was k+ = M MP , b /M 5 = 12 6, +
2 = 6M |4 1| and
> 1/4. If we take b = 0 or = 1/2, we could recover the existing results in cosmology.
5. Other solutions
If we take a non-separable ansatz for the metric tensor,
n(, y) = a(, y) =
1
,
f (y) + g0
b(, y) = k b0 a(, y),
(48)
the solution is
ds 2 =
d 2 + ij dx i dx j + (k b0 )2 dy 2
,
[k sinh(k b0 |y| + c0 ) + g0 ]2
where b0 and c0 are constants and determined by the boundary walls conditions.

k1,
,
c0 = cosh1
k

k2,
k1,
cosh1
.
k b0 = 2 cosh1
k
k
(49)
(50)
Actually this has the same form as in Ref. [4] except that the cosmological constants are,
as before, given by Eq. (22) and Eq. (25). The constant g0 remains as a free parameter and
its physical role is discussed in [4]. Setting g0 = 0, we get the solution given in Ref. [3]
and b0 becomes independent of .
If 16 + 5 + 4 = 0 is satisfied but the GaussBonnet conditions are not, the
inflationary solutions in given Eq. (35) (with a separable metric) and Eq. (49) (with a
nonseparable metric) are still valid except the substitution 4 8 + in our solutions
Eq. (22), etc., even though there exist higher time derivatives in the equations of motion.
Taking a different ansatz, n(, y) = a(, y) = b(, y), which is conformally flat, the
solution is given by
ds 2 =
d 2 + ij dx i dx j + dy 2

2 ,
2 k 2 1/2 + k
k1,
1, |y| + c0
(51)
where c0 is a constant. This metric describes inflation in both the spatial dimensions and the
extra dimension. The ks are given by Eqs. (22) and (25) as before. For non-GaussBonnet
case with 16 + 5 + 4 = 0, the solution is still valid also except for the substitution
4 8 + because it is conformally flat.
308
6. Metric perturbation near the RS background geometry

It is also of interest to study the gravitational interaction with the RS background. In
fact, Randall and Sundrum demonstrated that the Newtons force law does not imply
only four non-compact dimensions in the presence of a non-factorizable background
geometry [2026]. The example they studied is the case of a single 3-brane embedded
in non-compact five dimension. In this section, let us reconsider the case with the Gauss
Bonnet interaction.
The graviton is a linearized tensor fluctuation near the background geometry,
g = e2k |y| + h (x, y),
(52)
where the x indicates the coordinate for the 4-dimensional space embedded in the
5-dimensional bulk. Since we are interested in the 4-dimensional graviton only, which
is the longitudinal component of the metric fluctuation, we set h5 = h5 = h55 = 0.
Inserting Eq. (52) into Eq. (7) and taking only the linear terms in h , we obtain

2
2
8k
4k
1
2
+
(y) sgn(y)y
1
G + X =
y
2
M2
M2

2
4k
24 2k |y|
8k
e
+
(y)
1
2
M2
M2

2
2
6k
5k
2
(53)
+ 4k 2
h (x, y)
8k (y) 1
M2
M2
and

1
T = 3 b +
1 (y) h (x, y)
M

2
2
4k
2k
2
(y)
1
6k
h (x, y),
= 6k
M2
M2
(54)
where the underlined quantities denote the linear part in h in the full expressions and 24
is . Here we set b0 = 1 for simplicity. Eq. (54) is obtained by the use of Eq. (22)
and Eq. (24). Here we choose the traceless transverse gauge conditions, h = h = 0
[2026]. Under this gauge condition all components of h satisfy the same equation of
motion, and hence we will omit the indices below. Here we note again that in the Gauss
Bonnet case the unwanted higher derivative terms disappear in the linear approximation as
in the background case.
To perform a KaluzaKlein reduction down to 4-dimension and get an understanding of all modes that appear in the assumed 4D effective theory, we separate the variables; h(x, y) = (y) eipx , where the p is a 4-dimensional momentum. Since the
4-dimensional mass m2 of the KK excitation is p2 = m2 , Eq. (53) = Eq. (54) leads to

2
2
4k
4k
1
2
2
+
2k
1
y
2
M2
M2

2
12k
4k
sgn(y)y + 1
(y)
2k (y)
M2
M2

2
4k
m2 2k |y|
8k
+
(y)
(y).
e
1
2
M2
M2
309
(55)
Note that the above equation remains the same regardless of the sign of the branes
cosmological constant.
In the bulk, we can easily check that the GaussBonnet interaction does not modify the
2 /M 2 )
equation of motion because all terms have the exactly same common factor (1 4k
neglecting the Dirac delta functions. Therefore, we obtain the same eigenfunctions and
eigenvalues as in the RSs solutions [2023] except for the definition of k . On the
other hand, the Dirac delta functions gives the boundary condition, so the GaussBonnet
2 /M 2 .
interaction modifies only the boundary condition in the order of magnitude of k
Thus, one can imagine that there exists a possibility that the massless mode has not only
an exponentially decaying component but also an exponentially growing one at order
2 /M 2 . Note that the coefficient of the growing mode is exactly zero in the absence
k
of the higher curvature terms in the action [2023]. We have found, however, that the
exponentially growing mode does not appear even in the presence of the GaussBonnet
interaction.
To follow the RS process, let us make change of variables; z sgn(y)(ek |y| 1)/k ,
z) h(x, y)ek |y|/2 . Then, Eq. (55) reads
(z)
(y)ek |y|/2 and h(x,

2
15k
2B
k
3C
1
(z)
(z)
+
sgn(z)
z2 +
z
2
8(k |z| + 1)2
2
Ak
A

B
m2
1+
(z) (z),
(56)
=
2
Ak
2 /M 2 , B = 8k 2 /M 2 and C = 1 12k 2 /M 2 . Note that A = B + C.
where A = 1 4k
2
For m = 0, the eigenfunctions in the bulk satisfying the orbifold symmetry are
(k |z| + 1)3/2 /k (= exp( 32 k |y|)/k ) and (k |z| + 1)5/2/k (= exp( 52 k |y|)/k ).
Therefore, the solution is a linear combination of them
(k |z| + 1)5/2
(k |z| + 1)3/2
+b
.
k
k
(57)
To satisfy the boundary condition at z = 0, let us insert Eq. (57) into Eq. (56) and assemble
the coefficients of the Dirac delta functions. Then, we obtain
a(A + B + C) + b(A B + C) = 0,
(58)
where A = B + C. Thus, b = 0, i.e., the massless graviton is confined on the brane and the
Newtons force law on the brane holds good even under the non-compact extra dimension.
Particularly, we note that the result is not changed even though the branes cosmological
constant is negative.
As
h (x, y) e
k
2 |y|
ipx
e2k |y| eipx ,

(z)e
the fluctuation near the background metric can be written down as
310

0
g
= e2k |y| + eipx ,
(59)
where the superscript 0 denotes massless fluctuation and the is a polarization tensor of
the graviton wave function and from which we can see that the massless mode fluctuates
only in the longitudinal direction to the brane. Besides, from Eq. (55) we can get the
4-dimensional linearized Einstein equation in the Minkowski space,
24
eipx = 0,
2
(60)
which is of course true also after 4-dimensional coordinate transformation at a given y to

make the 4-dimensional background metric be in the form dS42 = dx dx . Note that
the above result is not affected by the GaussBonnet correction.
We usually worry about the instability of anti-de Sitter space due to excitations of ghost
particles [9]. In our case, we still have such a problem since the sign of the kinetic term for
the k+ background is opposite to that of the case without the GaussBonnet term, viz.
r

2

4k
1
4b 2k |y|
1
2k |y|
2
2
e
2
+
1
+
2
+
h
e
h . (61)
1
4
4
y
y
2
M2
2
3M 5
However, the equation of motion itself describes the same behavior of gravity localization
on the hidden sector brane (B1) because the brane cosmological constant contributing to
energy momentum tensor changes its sign as well. Thus, we have no ghost problem as far
as the brane cosmological constant is concerned as energy density. But we cannot regard
the brane with negative cosmological constant at z = 0 as our universe due to the later
cosmological problem that was discussed in Section 4.
For m2 > 0, the solutions for the above equation of motion in the bulk are

1/2
1/2
J2 m(|z| + 1/k ) + b |z| + 1/k
Y2 m(|z| + 1/k ) , (62)
a |z| + 1/k
which is the same solution as in the RS case except for the definition of k . The imposed
boundary condition at z = 0 fixes the ratio of a and b,
2 /M 2 + 2m2 /M 2
4k 2 1 4k
a
= 2
.
2 /M 2 + m2 /M 2
b m 1 12k
(63)
In this case, the GaussBonnet interaction modifies Newtons non-relativistic gravitational

potential through the KK states as follows,
m1 m2
+
V GN
r
Z
0
dm m
m1 m2 emr
GN
k k
r
2 /M 2 + m2 /M 2
1 12k
2
2 /M 2 + 2m2 /M 2
1 4k

2 /M 2 2
1 12k
m1 m2
GN
1+
.
2 /M 2
r
(k r)2 1 4k
(64)
The above result is obtained through the RS technique given in Ref. [2023]. Of course,
the potential given above is not ruled out yet [16].
For the case of two branes and bulk with S 1 /Z2 symmetry, the Eq. (55) is modified into

2
2
4k
4k
1
2
2
+
2k
1
2
M2
M2

2
12k
4k
sgn(y)y + 1
2k (y)
M2
M2

2
12k
4k
1
sgn(y)
+
1
(y)
+ 2k y
y
2
M2
M2

2
4k
8k
8k
1
m2 2k |y|
e
+
(y)
y
(y),
1
=
2
M2
M2
M2
2
311
(65)
where the sgn(y) |y|0 = 2( (y) (y 12 )) 1 and we can check the solution Eq. (57)
with b = 0 satisfies the above equation (65) regardless of the length scale by use of the
k
and Eq. (23). Therefore, from Eq. (60) the non-relativistic

relation h = e 2 |y| (z)
Newtonian gravity could be restored at the visible sector brane for sufficiently small
interval length b0 r.
7. Conclusion
We studied various static and inflationary solutions in the RandallSundrum framework
with the GaussBonnet term added to the standard Hilbert action. It has been argued that
in this RS framework the GaussBonnet term is the only acceptable curvature square
term. Then there exist additional coupling , the coefficient of the GaussBonnet term.
Depending on various values of , there exist static solutions and also the inflationary
solutions. In particular, there exist solutions for a positive visible sector tension 2 for
> 0, which makes it possible to transit to a standard Big Bang cosmology after inflation.
Acknowledgements
We thank K. Choi, S. Moon and J.D. Park for useful discussions. This work is supported
in part by the BK21 program of Ministry of Education, and by the Korea Science and
Engineering Foundation (KOSEF 1999 G 0100).
References
[1]
[2]
[3]
[4]
[5]
[6]
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-ph/9905221.

N. Kaloper, Phys. Rev. D 60 (1999) 123506, hep-th/9905210.
T. Nihei, Phys. Lett. B 465 (1999) 81, hep-ph/9905487.
H.B. Kim, H.D. Kim, hep-ph/9909053.
J.E. Kim, B. Kyae, H.M. Lee, hep-ph/9912344.
L. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 261 (1985) 678; Nucl. Phys. B 274
(1986) 285.
[7] W.D. Goldberger, M.B. Wise, Phys. Rev. Lett. 83 (1999) 4922, hep-ph/9907447.
[8] B. Zwiebach, Phys. Lett. B 156 (1985) 315.
312
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
D.G. Boulware, S. Deser, Phys. Rev. Lett. 55 (1985) 2656.

A.L. Maroto, I.L. Shapiro, Phys. Lett. B 414 (1997) 34.
S. Chang, M. Yamaguchi, hep-ph/9909523.
P. Binetruy, C. Deffayet, D. Langlois, hep-th/9905012.
C. Cski, M. Graesser, C. Kolda, J. Terning, hep-ph/9906513.
J.M. Cline, C. Grojean, G. Servant, Phys. Rev. Lett. 83 (1999) 4245, hep-ph/9906523.
C. Cski, M. Graesser, L. Randall, J. Terning, Phys. Lett. B 462 (1999) 34, hep-ph/9911406.
J. Lykken, L. Randall, hep-th/9908076.
J.E. Lidsey, A.R. Liddle, E.W. Kolb, E.J. Copeland, T. Barreiro, M. Abney, Rev. Mod. Phys. 69
(1997) 373, astro-ph/9508078.
R. Maartens, D. Wands, B.A. Bassett, I. Heard, hep-ph/9912464.
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690, hep-th/9906064.
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, N. Kaloper, hep-th/9907209.
C. Cski, J. Erlich, T.J. Hollowood, Y. Shirman, hep-th/0001033.
S.B. Giddings, E. Katz, L. Randall, hep-th/0002091.
M.G. Ivanov, I.V. Volovich, hep-th/9912242.
Y.S. Myung, G. Kang, hep-th/0001003.
Y.S. Myung, G. Kang, H.W. Lee, hep-th/0001107.

Black hole pair creation in de Sitter space:

a complete one-loop analysis
Mikhail S. Volkov 1 , Andreas Wipf
Institute for Theoretical Physics, Friedrich Schiller University of Jena, Max-Wien Platz 1, D-07743, Jena,
Germany
Abstract
We present an exact one-loop calculation of the tunneling process in Euclidean quantum gravity
describing creation of black hole pairs in a de Sitter universe. Such processes are mediated by S 2 S 2
gravitational instantons giving an imaginary contribution to the partition function. The required
energy is provided by the expansion of the universe. We utilize the thermal properties of de Sitter
space to describe the process as the decay of a metastable thermal state. Within the Euclidean path
integral approach to gravity, we explicitly determine the spectra of the fluctuation operators, exactly
calculate the one-loop fluctuation determinants in the -function regularization scheme, and check
the agreement with the expected scaling behaviour. Our results show a constant volume density of
created black holes at late times, and a very strong suppression of the nucleation rate for small values
of . 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
Instantons play an important role in flat space gauge field theory [45]. Being stationary
points of the Euclidean action, they give the dominant contribution to the Euclidean
path integral thus accounting for a variety of important phenomena in QCD-type
theories. In addition, self-dual instantons admit supersymmetric extensions, which makes
them an important tool for verifying various duality conjectures like the AdS/CFT
correspondence [4]. More generally, the Euclidean approach has become the standard
method of quantum field theories in flat space.
Since the theory of gravity and YangMills theory are somewhat similar, it is natural
to study also gravitational instantons. An impressive amount of work has been done in
this direction, leading to a number of important discoveries. A thorough study of instanton
E-mail addresses: vol@tpi.uni-jena.de (M.S. Volkov), wipf@tpi.uni-jena.de (A. Wipf).
1 Supported by the DFG.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 8 7 - X
314
M.S. Volkov, A. Wipf / Nuclear Physics B 582 (2000) 313362
solutions of the vacuum Einstein equations and also those with a -term has been carried
out [17,20,24]. These solutions dominate the path integral of Euclidean quantum gravity,
leading to interesting phenomena like black hole nucleation and quantum creation of
universes. Perhaps one of the most spectacular achievements of the Euclidean approach is
the derivation of black hole entropy from the action of the Schwarzschild instanton [22]. In
addition, gravitational instantons are used in the KaluzaKlein reductions of string theory.
Along with these very suggestive results, the difficulties of Euclidean quantum gravity
have been revealed. Apart from the usual problem of the non-renormalizability of gravity,
which can probably be resolved only at the level of a more fundamental theory like string
theory, the Euclidean approach presents other challenging problems. In field theories in flat
space the correlation functions of field operators are holomorphic functions of the global
coordinates in a domain that includes negative imaginary values of the time coordinate,
t = i , where is real and positive [48]. This allows one to perform the analysis in
the Euclidean section and then analytically continue the functions back to the Lorentzian
sector to obtain the physical predictions. In curved space the theorems that would ensure
the analyticity of any quantities arising in quantum gravity are not known. As a result, even
if Euclidean calculations make sense, it is not in general clear how to relate their result to
the Lorentzian physics.
This difficulty is most strikingly illustrated by the famous problem of the conformal
sector in Euclidean quantum gravity. If one tries evaluating the path integral over
Riemannian metrics, then one discovers that it diverges because the Euclidean gravitational
action is not bounded from below and can be made arbitrarily large and negative by
conformal rescaling of the metric [25]. Such a result is actually expected, for if the integral
did converge (with some regularization), then one could give a well-defined meaning to the
canonical ensemble of the quantum gravitational field. However, the possibility of having
a black hole causes the canonical ensemble to break down since the degeneracy of
black hole states grows faster than the Boltzmann factor decreases. One can, improve the
Euclidean gravitational action by analytically continuing the conformal modes, let us call
them h, via h ih, and this improves the convergence of the integral [25]. This shows
that if there is a well-defined Euclidean path integral for the gravitational field, then the
relation to the Lorentzian sector is more complicated than just via t i .
Unfortunately, it is unknown at present whether one can in the general case find a
physically well-defined and convergent path integral for the gravitational field. At the
same time, the idea of constructing it is conceptually simple [46]: one should start from
the Hamiltonian path integral over the physical degrees of freedom of the gravitational
field. Such an integral certainly makes sense physically and is well-convergent, since
the Hamiltonian is positive at least in the asymptotically flat case. The Hamiltonian
approach is not covariant, but one can covariantize it by changing the integration variables,
which leads to a manifestly covariant and convergent path integral for gravity. The main
problem with this program is that in the general case it is unclear how to isolate the physical
degrees of freedom of the gravitational field. For this reason, so far the program has been
carried out only for weak fields in the asymptotically flat case [46]. Remarkably, the result
has been shown to exactly correspond to the the standard Euclidean path integral with the
315
conformal modes being complex-rotated via h ih. This lends support to the Euclidean
approach in gravity and allows one to hope that the difficulties of the method can be
consistently resolved (see, for example, [6,7] for the recent new developments within the
lattice approach).
One can adopt the viewpoint that Euclidean quantum gravity is a meaningful theory
within its range of applicability, at least at one-loop level, by assuming that a consistent
resolution of its difficulties exists. Then already in its present status the theory can be
used for calculating certain processes, most notably for describing tunneling phenomena,
in which case the Euclidean amplitude directly determines the probability. The analytic
continuation to the Lorentzian sector in this case is not necessary, apart from when the
issue of the interpretation of the corresponding gravitational instanton is considered. The
important example of a tunneling process in quantum gravity is the creation of black holes
in external fields. Black holes are created whenever the energy pumped into the system
is enough in order to make a pair of virtual black holes real [33]. The energy can be
provided by the heat bath [5,30,38], by the background magnetic field [15,16,19,21], by
the expansion of the universe [10,28,41], by cosmic strings [37], domain walls [11], etc.,
(see also [35,36,43]). Besides, one can consider pair creation of extended multidimensional
objects like p-branes due to interaction with the background supergravity fields [14]. In all
these examples the process is mediated by the corresponding gravitational instanton, and
the semiclassical nucleation rate for a pair of objects on a given background is given by

(1.1)
= A exp (Iobj Ibg ) .
Here Iobj is the classical action of the gravitational instanton mediating creation of the
objects, Ibg is the action of the background fields alone, and the prefactor A includes
quantum corrections.
In most cases the existing calculations of black hole pair creation processes consider
only the classical term in (1.1). This is easily understood, since loop calculations
in quantum gravity for non-trivial backgrounds are extremely complicated. To our
knowledge, there is only one example of a next-to-leading-order computation, which was
undertaken in [30] by Gross, Perry, and Yaffe for the Schwarzschild instanton background.
The aim of the present paper is to consider one more example of a complete one-loop
computation in quantum gravity.
The problem we are interested in is the quantum creation of black holes in de Sitter
space. This problem was considered by Ginsparg and Perry [28], who identified the
instanton responsible for this process, which is the S 2 S 2 solution of the Euclidean
Einstein equations R = g for > 0. Ginsparg and Perry noticed that this solution
has one negative mode in the physical sector, which renders the partition function complex,
thus indicating the quasi-classical instability of the system. This instability leads to
spontaneous nucleation of black holes in the rapidly inflating universe. This is the dominant
instability of de Sitter space, since classically the space is stable [28]. The energy necessary
for the nucleation is provided by the -term, which drives different parts of the universe
apart thereby drugging the members of a virtual black
hole pair away from each other.
The typical radius of the created black holes is 1/ , while the the nucleation rate is
316
of the order of exp(/G), where G is Newtons constant. As a result, for G 1

when inflation is fast, the black holes are produced in abundance but they are small and
presumably almost immediately evaporate. Large black holes emerge for G 1 when
inflation slows down, and these can probably exist for a long time, but the probability of
their creation is exponentially small. This scenario was further studied in Refs. [9,10,18]
(see also references in [9]), where the generalization to the charged case was considered
and also the subsequent evolution of the created black holes was analyzed. However, the
one-loop contribution so far has not been computed.
A remarkable feature of the S 2 S 2 instanton is its high symmetry. In what follows,
we shall utilize this symmetry in order to explicitly determine spectra of all relevant
fluctuation operators in the problem. We shall use the -function regularization scheme
in order to compute the one-loop determinants, which will give us the partition function
Z[S 2 S 2 ] for the small fluctuations around the S 2 S 2 instanton. We shall then need
to normalize this result. The normalization coefficients is Z[S 4 ], the partition function for
small fluctuations around the S 4 instanton, which is the Euclidean version of the de Sitter
space. The one-loop quantization around the S 4 instanton was considered by Gibbons and
Perry [27], and by Christensen and Duff [13], but unfortunately in none of these papers
the analysis was completed. We shall therefore reconsider the problem by rederiving the
spectra of fluctuations around S 4 and computing the determinants within the -function
scheme, thereby obtaining a closed one-loop expression for Z[S 4 ].
In our treatment of the path integral we follow the approach of Gibbons and Perry [27]
(see also [42]). In order to have control over the results, we work in a one-parameter family
of covariant gauges and perform the Hodge decomposition of the fluctuations. These are
then expanded with respect to the complete sets of basis harmonics, and the perturbative
path integration measure is defined as the square root of the determinant of the metric
on the function space of fluctuations. To insure the convergence of the integral over the
conformal modes, which enter the action with the wrong sign, we essentially follow the
standard recipe h ih [25]; (see also Ref. [42], where a slightly disguised form of the
0=
same prescription was advocated). The subtle issue is that the conformal operator 1
3 4 has a finite number, N , of negative modes, and these enter the action
with the correct sign from the very beginning. Our treatment of these special modes is
different from that by Hawking [32], who suggests that such modes should be complexrotated twice, the partition function then acquiring the overall factor of i N . However, the
presence of this factor in the partition function would lead to unsatisfactory results, and on
these grounds we are led to not rotating the special conformal modes at all.
The path integral is computed by integrating over the Fourier expansion coefficients,
which leads to infinite products over the eigenvalues. The only conformal modes giving
contribution to the result are the special negative modes discussed above. We carefully
analyze the resulting products to make sure that all modes are taken into account and that
the dependence of the gauge-fixing parameter cancels thereby indicating the correctness of
the procedure. We give a detailed consideration to the zero modes of the FaddeevPopov
operator, which arise due to the background isometries. The integration over these modes
requires a non-perturbative extension of the path-integration measure, and we find such a
317
non-perturbative measure in the zero mode sector to be proportional to the Haar measure of
the isometry group. Collecting all terms yields the partition function for small fluctuations
around a background instanton configuration in terms of infinite products over eigenvalues
of the gauge-invariant operators. We then use the explicitly known spectra of fluctuations
around the S 2 S 2 and S 4 backgrounds in order to calculate the partition functions.
The rest of the paper is organized as follows. In Section 2 we present our derivation of
the black hole nucleation rate within the finite temperature approach. In Section 3 the path
integration procedure is considered. The spectra of small fluctuations around the S 2 S 2
instanton are computed in Section 4 via a direct solving of the differential equations
in the eigenvalue problems. The spectra of the fluctuations around the S 4 instanton are
rederived in Section 5 with the use of group theoretic arguments. The partition functions
are computed in Section 6, and Section 7 contains the final expression for the black
hole nucleation rate together with some remarks. We present a detailed analysis of the
-functions in the appendix. We use units where c = h = kB = 1.
2. Black hole nucleation rate

In this section we shall derive the basic formula for the black hole nucleation rate in
de Sitter space, whose different parts will be evaluated in the next sections. The existing
derivations of the nucleation rate [10,28] recover only the classical factor in (1.1). In
addition, it is not always clear to which volume the rate refers. We argue that our formula
(2.14) gives the nucleation probability per Hubble volume and unit time as measured by a
freely falling observer. The basic idea of our approach is to utilize the relation between the
inflation and thermal properties of de Sitter space. This will allow us to use the standard
theory of decay of metastable thermal states [3,39,40].
Let us consider the partition function for the gravitational field
Z
(2.1)
Z = D[g ] eI ,
where the integral is taken over Riemannian metrics, and I = I [g ] is the Euclidean action
for gravity with a positive terms; see Eq. (3.1) below. The path integration procedure
will be considered in detail in the next section. At present let us only recall that in the
semiclassical approximation the integral is approximated by the sum over the classical
extrema of the action I , that is
X
Zl .
(2.2)
Z
l
Here Zl = Z[Ml ] is the partition function for the small gravitational fluctuations around
l subject to the Euclidean Einstein equations
a background manifold Ml with a metric g
R = g . Schematically one has
Z
X exp(Il )
,
Det 1l
l
(2.3)
318
where Il is the classical action for the lth extremum, and 1l is the operator for the small
fluctuations around this background.
The dominant contribution to the sum in (2.3) is given by the S 4 instanton, which is
the four-dimensional sphere with the radius 3/ and the standard metric. Since this is a
maximally symmetry space, its action I = 3/G is less than that of any other instanton.
Hence,
exp(3/G)
.
(2.4)
Z Z[S 4 ] =
Det 1
On the other hand, the S 4 instanton describes the thermal properties of de Sitter space [22,
23], since it can be obtained by an analytic continuation via t = it of the region of the
de Sitter solution

2
dr 2
2
+ r 2 d 2 + sin2 d 2
(2.5)
ds = 1 r dt 2 +
2
3
1 3r
contained inside the event horizon, r < 3/. Let us call this region a Hubble region.
Its boundary, the horizon,
has the area A = 12/. The temperature associated with this
q
horizon is T =
1
2
the entropy S = A/4G = 3/G and the free energy F = T S.
3,
The same values can be obtained by writing the partition function for the S 4 instanton as

(2.6)
Z S 4 = eF
with = 1/T . Indeed, since S 4 is periodic in all four coordinates,qany of them can be
chosen to be the imaginary time. The corresponding period, = 2 3 , can be identified
with the proper length of a geodesic on S 4 , all of which are circles with the same length.
This gives the correct de Sitter temperature. Comparing (2.6) and (2.4) one obtains F =
3/G + , the dots denoting the quantum corrections, and this again agrees with the
result for the de Sitter space. To recapitulate, the partition function of quantum gravity with
> 0 is approximately
Z eF ,
(2.7)
where 1/ is the de Sitter temperature and F is the free energy in the Hubble region.
Let us now consider the contribution of the other instantons. One has

X0 Z[Ml ]
F
,
(2.8)
1+
Ze
Z[S 4 ]
l
where the prime indicates that Ml 6= S 4 . Now, for G 1 all terms in the sum are
exponentially small and can safely be neglected as compared to the unity, if only they
are real. If there are complex terms, then they will give an exponentially small imaginary
contribution. The S 2 S 2 instanton is distinguished by the fact that its partition function
is purely imaginary due to the negative mode in the physical sector [28]. This is the only
solution for > 0 which is not a local minimum of the action in the class of metrics with
constant scalar curvature [20]. Hence (see Fig. 1),

Z[S 2 S 2 ]
Z[S 2 S 2 ]
F
exp F
,
(2.9)
1+
Ze
Z[S 4 ]
Z[S 4 ]
319
Fig. 1. The leading contribution to the partition function comes from the S 4 and S 2 S 2 gravitational
bubbles, the effect of the latter being purely imaginary.
where Z[S 2 S 2 ] is purely imaginary. As a result, the partition function can still be
represented as Z eF , where the real part of F is the free energy of the Hubble region,
and the exponentially small imaginary part is given by
=(F ) =
Z[S 2 S 2 ]
.
Z[S 4 ]
(2.10)
It is natural to relate this imaginary quantity also to the free energy. We are therefore led to
the conclusion that the free energy of the Hubble region has a small imaginary part, thus
indicating that the system is metastable. The decay of this metastable state will lead to a
spontaneous nucleation of a black hole in the Hubble region, which can be inferred from
the geometrical properties of the S 2 S 2 instanton.
The S 2 S 2 instanton can be obtained via the analytic continuation of the Schwarzschildde Sitter solution [10,26,28]
ds 2 = N dt 2 +

dr 2
+ r 2 d 2 + sin2 d 2 .
N
(2.11)
2
2
Here N = 1 2M
r 3 r , and for 9M < 1 this function has roots at r = r+ > 0
(black hole horizon) and at r = r++ > r+ (cosmological horizon). One has N > 0 for
r+ < r < r++ , and only this portion of the solution can be analytically continued to the
Euclidean sector via t = it. The conical singularity at either of the horizons can be
removed by a suitable identification of the imaginary time. However, since the two horizons
have different surface gravities, the second conical singularity will survive. The situation
improves in the extreme limit, r+ r++ 1 , since the surface gravities are then the
same and both conical singularities can be removed at the same time. Although one might
think that the Euclidean region shrinks to zero when the two horizons merge, this is not so.
The limit r+ r++ implies that 9M2 = 1 3 2 with 0. One
can introduce new
coordinates 1 and 1 via cos 1 = ( r 1)/ + /6 and 1 = . Passing to the
new coordinates and taking the limit 0, the result is
ds 2 =

1
d1 2 + sin2 1 d1 2 + d 2 + sin2 d 2 ,
(2.12)
and this S 2 S 2 metric fulfills the Einstein equations. Since the instanton field determines
the initial value for the created real time configuration, one concludes that the S 2 S 2
320
Fig. 2. Left: The conformal diagram of de Sitter space in coordinates (2.13). The trajectories
= const are timelike geodesics. The diamond-shaped region in the center is the Hubble region
of the geodesic observer at = /2. Although this region completely covers the hypersurface 0 , at
later times one needs more observers to cover the hypersurface 1 with the interiors of their horizons
the Hubble regions proliferate. Right: The de Sitter hyperboloid in the embedding Minkowski
space (with two dimensions suppressed). The Hubble region of the inertial observer moving along
the hyperbola x = 0, y > 0 is the portion of the hyperboloid lying to the right from the two shaded
strips. This corresponds to the interior region of the observers static coordinate system.
instanton is responsible for the creation of a black hole in the Hubble region. This black
hole fills the whole region, since its size is equal to the radius of the cosmological horizon.
It is well known that the region r < 3/ of the static coordinate system in (2.5)
covers only a small portion of the de Sitter hyperboloid [47] (see Fig. 2). In order to
cover the whole space, one can introduce an infinite number of freely falling observers
and associate the interior of the static coordinate system with each of them. Hence, the
spacetime contains infinitely many Hubble regions. It is also instructive to use global
coordinates covering the whole de Sitter space,

3
2
2
2
2
2
2
d
+
d
+
sin
d
+
sin
d
,
(2.13)
ds 2 =
cos2
where [/2, /2] and [0, ]. The trajectory of a freely falling observer is
= 0 (and also = 0 , = 0 ), and the domain of the associated static coordinate
system, the Hubble region, is the intersection of the interiors of the observers past and
future horizons [34]. Let be a spacelike hypersurface, say = 0 . If 0 = 0 then
is completely contained inside the Hubble region of a single observer with = /2
(see Fig. 2). However, for late moments of time, /2, one needs more and more
independent observers in order to completely cover by the union of their Hubble regions.
One can say that the Hubble regions proliferate with the expansion of the universe.
Since de Sitter space consists of infinitely many Hubble regions, the black hole
nucleation will lead to some of the regions being completely filled by a black hole, but
most of the regions will be empty. The number of the filled regions divided by the number
of those without a black hole is the probability for a black hole nucleation in one region.
This is proportional to =(F ) in (2.10).
321
Fig. 3. The conformal diagram for the extreme Schwarzschildde Sitter solution.
One can argue that the black holes are actually created in pairs [33,36], where the two
members of the pair are located at the antipodal points of the de Sitter hyperboloid. This
can be inferred from the conformal diagram of the Schwarzschildde Sitter solution, which
contains an infinite sequence of black hole singularities and spacelike infinities; see Fig. 3.
One can identify the asymptotically de Sitter regions in the diagram related by a horizontal
shift, and the spacetime will then consist of two black holes at antipodal points of the
closed universe. This agrees with the standard picture of particles in external fields being
created in pairs.
The surface gravity of the extreme Schwarzschildde Sitter solution is finite when
defined with respect to the suitably normalized Killing vector [10]. This gives a nonzero
value for the temperature of the nucleated black holes, which can be read off also from
S 2 metric: it is the inverse proper length of the equator of any of the two spheres,
the S 2
TBH = 2 . How can it be that this is different from the temperature of the heat bath, which
1
/3? For example, in the hot Minkowski space the
is the de Sitter space with TdS = 2
nucleated black holes have the same temperature as the heat bath [30]. However, the global
structure of de Sitter space is different from that of Minkowski space. The fluctuations
cannot absorb energy from and emit energy into the whole of de Sitter space, but can only
exchange energy with the Hubble region. Thus the energy exchange is restricted. As a
result, the local temperature in the vicinity of a created defect may be different from that of
the heat bath, but reduces to the latter in the asymptotic region far beyond the cosmological
horizon.
The relation of the imaginary part of the free energy to the rate of decay of a metastable
thermal state was considered in [3,39,40]. If the decay is only due to tunneling then =
2=(F ). Suppose that there is an additional possibility to classically jump over the potential
barrier. In this case on top of the barrier there is a classical saddle point configuration whose
real time decay rate is determined by the saddle negative mode . At low temperatures
|
the tunneling formula is then still correct, while for T > |2 | one has = |
T =(F ). In
2
2
our problem the saddle point configuration also exists, the S S instanton, but its real
time analog, the Schwarzschildde Sitter black hole, is stable. It seems therefore that there
is no classical contribution to the process and the black hole nucleation is a purely quantum
phenomenon. 2 (One can imagine that the effective potential barrier is infinitely high, such
2 We do not understand the classical interpretation of the Euclidean saddle point solution suggested in [30].
The argument uses a family of non-normalizable deformations of the instanton, and the action is finite as long as
they are static. However, if one considers a time evolution along such a family then the action will be infinite,
322
that a classical transition is forbidden, but at the same time so narrow that the tunneling
rate is non-zero.) As a result, the rate of quasiclassical decay of the de Sitter space is given
by = 2=(F ). Using Eq. (2.10),
= 2T
Z[S 2 S 2 ]
.
Z[S 4 ]
(2.14)
q
1
Here T = 2
3 is the temperature of the de Sitter heat bath, which was originally defined
with respect to the analytically continued Killing vector /t. Since t is the proper time of
the geodesic observer resting at the origin of the static coordinate system (2.5), we conclude
that the formula gives the probability of a black hole nucleation per Hubble volume and
unit time of a freely falling observer.
In order to use the formula (2.14), we should be able to compute the one-loop partition
functions Z[S 2 S 2 ] and Z[S 4 ]. Now we shall calculate them within the path integral
approach.
3. The path integration procedure

In this section we shall consider the path integral for fluctuations around an instanton
solution of the Einstein equations R = g in the stationary phase approximation. We
shall largely follow the approach of Gibbons and Perry [27].
3.1. The second variation of the action
Our starting point is the action for the gravitational field on a compact Riemannian
manifold M,
Z
1
(R 2) g d 4 x,
(3.1)
I [g ] =
16G
M
whose extrema, I = 0, are determined by the equations

R = g .
(3.2)
Let g be an arbitrary solution, and consider small fluctuations around it, g g +

h . The action expands as
I [g + h ] = I [g ] + 2 I + ,
(3.3)
where 2 I is quadratic in h and dots denote the higher order terms. One can express 2 I
directly in terms if h . However, it is convenient to use first the standard decomposition
of h ,
h = +
1
1
h g + + g .
4
2
(3.4)
which shows that the classical picture does not apply. Even if one uses the classical formula for in this case,
one arrives at the quantum result, since | |/T = const 1.
323
Here is the transverse tracefree part, = = 0, h is the trace, and the piece
due to is the longitudinal tracefree part. Under the gauge transformations (general
diffeomorphisms) generated by one has h h + + . The TT-tensor
is gauge-invariant, while the trace h changes as h h + 2 . It follows that
h = h 2
(3.5)
is gauge-invariant. For further references we note that can in turn be decomposed into
its coexact part , for which = 0, the exact part , and the harmonic piece H ,
= + + H .
(3.6)
The number of square-integrable harmonic vectors is a topological invariant, which is

equal to the first Betti number of the manifold M. Since the latter is zero if M is simplyconnected, which is the case for > 0, we may safely ignore the harmonic contribution in
what follows.
With the decomposition (3.4) the second variation of the action in (3.3) is expressed in
terms of the gauge-invariant quantities and h alone,

1

1

, 12
h, 10 h .
(3.7)
2
16
Here and below we consider the following second order differential operators: the operator
for the TT-tensor fluctuations
2 I =
12 = 2R ,
(3.8)
the vector operator acting on coexact vectors

11 = ,
(3.9)
and
and the scalar operators for h, h,
10 = ,
0 = 310 4,
1
0 10 ,
=1
1
0
(3.10)
with being a real parameter. Since for > 0 the manifold M is compact, these operators
are (formally) self-adjoint with respect to the scalar product
Z

g d 4 x;
(3.11)
, =
32G
M
similarly for vectors h , i and scalars h, i.

while
The action 2 I in (3.7) contains only the gauge-invariant amplitudes and h,
the dependence on the gauge degrees of freedom cancels. Pure gauge modes are thus
zero modes of the action. Fixing of the gauge is therefore necessary in order to carry out
the path integration. To fix the gauge we pass from the action 2 I to the gauge-fixed action
2 Igf = 2 I + 2 Ig ,
(3.12)
324
where, following [27], we choose the gauge-fixing terms as

1
1
2

Ig = h h, h h ,
(3.13)
with and being real parameters. We shall shortly see that it is convenient to choose
[27]
4
.
(3.14)
=
+1
This choice, however, implies that 2 Ig does not vanish for 0. It is often convenient
to set = 1, in which case = 2. However, we shall not fix the value of , since this will
provide us with a check of the gauge-invariance of our results.
Using the decompositions (3.4), (3.6) the gauge-fixing term reads

1

, 10 h + 21
.
h + 21
(3.15)
2 Ig = , 121 +
0
0
16
Adding this up with 2 I in (3.7) one obtains the gauge-fixed action 2 Igf . It is now
convenient to pass from the gauge-invariant variable h defined in (3.5) back to the trace h,
since with the choice in (3.14) the resulting action then becomes diagonal:
2 Igf =

1

, 12 + , 121
2

1
01

0h .
h, 1
+ , 10 1
0
4
16
(3.16)
This action generically has no zero modes, but it depends on the arbitrary parameter ,
which reflects the freedom of choice of gauge-fixing. In order to cancel this dependency,
the compensating ghost term is needed.
3.2. The mode decomposition of the action
We wish to calculate the path integral
Z

Z[g ] = eI D[h ] DFP exp 2 Igf ,
(3.17)
where I = I [g ] is the classical action, and the FaddeevPopov factor is obtained from
Z

(3.18)
1 = DFP D[ ] exp 2 Ig .
In order to perform the path integration, we introduce the eigenmodes associated with the
operators 12 , 11 and 10 :
(k)
(k)
= k
,
12
11 (s) = s (s) ,
10 (p) = p (p) .
(3.19)
Throughout this paper we shall denote the eigenvalues and eigenfunctions of the tensor
(k)
, and those for the vector operator 11 by s and (s) ,
operator 12 by k and
325
respectively. (Later we shall use the symbol s also for the argument of the -functions,
and this will not lead to any confusion.) Eigenvalues of the scalar operator will be denoted
by p , and it will be convenient to split the set {p } into three subsets, {p } = {0 , i , n },
where 0 = 0, i = 43 , and n > 43 ; see Eqs. (3.25)(3.27) below. Accordingly, the set
of the scalar eigenfunctions will be split as { (p) } = { (0) , (i) , (n) }.
Since the manifold is compact, we choose the modes to be orthonormal. This allows us
to expand all fields in the problem as
X
X
(k)
Ck
,
=
Cs (s),
(3.20)
=
s
and
=
Cp (p) ,
h=
Cph (p) ,
h =
Cph (p) .
(3.21)
As a result, the action decomposes into the sum over modes, and the path integral reduces
to integrals over the Fourier coefficients.
(a) Vector and tensor modes
Let us consider the mode decomposition for the gauge-fixed action in (3.16). This action
is the sum of four terms. For the first two terms we obtain
1 X
1

2
k Ck ,
(3.22)
, 12 =
2
2
k
X

2
(s )2 Cs .
(3.23)
, 121 =
s
These quadratic forms should be positive definite, since otherwise the integrals over the
coefficients C would be ill-defined. We can see that the quadratic form in (3.23) for the
vector modes is indeed non-negative definite. Next, the expression in (3.22) for the gaugeinvariant tensor modes is positive-definite if all eigenvalues k are positive. If there is a
negative eigenvalue, < 0, as in the case of the S 2 S 2 instanton background, then it
is physically significant. The integration over C is performed with the complex contour
rotation, which renders the partition function imaginary thus indicating the quasiclassical
instability of the system.
Let us consider now the contribution of the longitudinal vector piece to the action (3.16).
We obtain
1 X
2
1
01
=
, 10 1
p p p Cp ,
(3.24)
0
4
4 p
0 and 1
, respectively.
where p = 3p 4 and p = p p are the eigenvalues of 1
0
We note that while p > 0, the p and p can be negative and should therefore be treated
carefully. Let us split the scalar modes into three groups according to the sign of p :
(3.25)
0 = 43 ,
i = 0,
n > 0.
(3.27)
10 (0) = 0
10 (i) = 43 (i)
10
(n)
= n
(n)
(3.26)
326
First we consider the constant mode (0) in (3.25). This exists for any background, and for
compact manifolds without boundary this is the only normalizable zero mode of 10 . Since
this mode is annihilated by 10 , it does not contribute to the sum in (3.24).
Consider now the scalar modes with the eigenvalue 4/3 in (3.26). In view of the
LichnerowiczObata theorem [49], the lowest non-trivial eigenvalue of 10 for > 0 is
bounded from below by 4/3, and the equality is attained if only the background is S 4 .
Hence the modes in (3.26) exist only for the S 4 instanton, and there can be no modes in
between (3.25) and (3.26). In the S 4 case there are five scalar modes with the eigenvalue
4/3, and their gradients are the five conformal Killing vectors that do not correspond
to infinitesimal isometries. If R = g , a theorem of Yano an Nagano [49] states that
such vectors exist only in the S 4 case. Let us call these five scalar modes conformal Killing
modes. Notice that these also do not contribute to the sum in (3.24).
To recapitulate, the lowest lying modes in the scalar spectrum are the constant conformal
mode in (3.25), which exists for any background, and also 5 conformal Killing modes in
(3.26) which exist only for the S 4 instanton and generate the conformal isometries. As we
shall see, these 1 + 5 lowest lying modes are physically distinguished, since they are the
only scalar modes contributing to the partition function. However, they do not enter the
sum in (3.24).
For the remaining infinite number of scalar modes in (3.27) (these are labeled by n) the
eigenvalues n and n are positive, and it is not difficult to see that all the n s are also
positive, provided that the gauge parameter is positive and large enough. To recapitulate,
the contribution of the longitudinal vector modes to the action is given by
1 X
2
1
01
=
, 10 1
n n n Cn ,
(3.28)
0
4
4 n
which is positive definite. We shall see that all modes contributing to this sum are
unphysical and cancel from the path integral.
(b) Conformal modes
We now turn to the last term in the gauge-fixed action (3.16). Using (3.25)(3.27) we
obtain
h 2
X h 2
1 X h 2
1
0h =
h, 1
C0 +
(3.29)
Ci
n Cn .
16
4
12
16 n
i
The expression on the right has a finite number of positive terms, corresponding to the
distinguished lowest lying modes, and infinitely many negative ones. As a result, an
increase in the coefficients Cnh makes it arbitrarily large and negative, thus rendering the
path integral divergent. This represents the well-known problem of conformal modes in
Euclidean quantum gravity [25]. A complete solution of this problem is lacking at present,
but the origin of the trouble seems to be understood [46]. In brief, the problem is not
related to any defects of the theory itself, but arises as a result of the bad choice of the path
integral. If one starts from the fundamental Hamiltonian path integral over the physical
degrees of freedom of the gravitational field, then one does not encounter this problem.
The Hamiltonian path integral, however, is non-covariant and difficult to work with. One
327
can covariantize it by adding gauge degrees of freedom, and this leads to the Euclidean
path integral described above, up to the important replacement [25]
h ih.
(3.30)
The effect of this is to change the overall sign in (3.29), such that the infinite number of
negative modes become positive. Unfortunately, such a consistent derivation of the path
integral has only been carried out for weak gravitational fields [46] (and for = 0), since
otherwise it is unclear how to choose the physical degrees of freedom. Nevertheless, the
rule (3.30) is often used also in the general case [25], and it leads to the cancellation of the
unphysical conformal modes. However, some subtle issues can arise.
For > 0 the expression in (3.29) contains, apart from infinitely many negative terms,
also a finite number of positive ones, which are due to the distinguished lowest lying
scalar modes. If we apply the rule (3.30) and change the overall sign of the scalar mode
action, then the negative modes will become positive, but the positive ones will become
negative. As a result, the path integral will still be divergent. It was therefore suggested
by Hawking that the contour for these extra negative modes should be rotated back, the
partition function then acquiring the factor i N , where N is the number of such modes [32].
As we know, N = 6 for the S 4 instanton, and N = 1 for any other solution of R = g
with > 0.
Unfortunately, this prescription to rotate the contour twice leads in some cases to
physically meaningless results; the examples will be given in a moment. We suggest
therefore a slightly different scheme: not to touch the positive modes in (3.29) at all and to
change the sign only for the negative modes. The whole expression then becomes
h 2
X h 2
1 X h 2
1
0h =
h, 1
C0 +
n Cn .
(3.31)
Ci +
16
4
12
16 n
i
We make no attempt to rigorously justify such a rule. We note, however, that it is essentially
equivalent to the standard recipe (3.30) up to a finite number of modes which we
handle differently as compared to Hawkings prescription. We shall now comment on this
difference.
When compared to Hawkings recipe [32], the ultimate effect of our prescription is to
remove the factor i N from the partition function. We are unaware of any examples where it
would be necessary to insist on this factor being present in the final result. On the contrary,
the examples are in favour of the factor being absent. For the S 4 instanton one has N = 6,
such that i N = 1, and this would render the partition function for hot gravitons in a de
Sitter universe negative, which would be physically meaningless. Next, for the S 2 S 2
instanton, which already has one negative mode in the spin-2 sector, one has N = 1. As
a result, the factor i N would make the partition function real instead of being imaginary,
and there would be no black hole pair creation!
These arguments suggest that Hawkings rule should be somehow modified, and we
therefore put forward the prescription (3.31). Let us also note that our rule leads to gauge
invariant results the dependence on the gauge parameter cancels after the integration.
Finally we note that the lowest lying scalar modes are physically distinguished, and since
they are positive, they should be treated similarly to the physical tensor modes.
328
To recapitulate, the mode expansion of the gauge-fixed action 2 Igf is given by the sum
of (3.22), (3.23), (3.28), and (3.31):
X
2 1 X
2
1 X
2
k Ck +
(s )2 Cs +
n n n Cn
2 Igf =
2
4
s
n
k
X
X

2
h 2
1
2
n Cnh .
+
(3.32)
C0 +
Cih +
4
12
16 n
i
In a similar way we obtain the following mode expansion for the gauge-fixing term 2 Ig
in (3.15):

X

16 3 X
3 h 2
2
2
2
C
(s ) Cs +
Ci
Ig =
27
8 i
s
i
1 X
2
+
n 2 n Cn + Cnh .
(3.33)
16 n
This expression is non-negative definite.
3.3. The path integration measure
In order to compute the path integrals in (3.17), (3.18) we still need to define the
path-integration measure. The perturbative measure is defined as the square root of the
determinant of the metric on the function space of fluctuations:
p
p
D[ ] Det(hd , d i).
(3.34)
D[h ] Det(hdh , dh i),
Here it is assumed that the fluctuations are Fourier-expanded and the differentials refer to
the Fourier coefficients, while the meaning of the proportionality sign will become clear
shortly. Let us first consider D[ ]. It follows from (3.4), (3.6) that

0 + 1 hh, hi,
h , h = , + 2 , 11 + , 10 1
4

, = , + h, 10 i.
(3.35)
Expanding the fields on the right according to (3.20), (3.21) and differentiating with respect
to the Fourier coefficients we obtain the metric for the vector fluctuations
X

2 X0
2
p dCp ,
(3.36)
dCs +
d , d = d , d + hd, 10 di =
s
which yields
p
Det(hd , d )i =
!
dCs
Y0 p
!
p dCp .
(3.37)
Here the prime indicates that terms with p = 0 do not contribute to the sum in (3.36),
and should therefore be omitted in the product in (3.37). To obtain the measure D[ ] we
endow each term in the products in (3.37) with the weight factor 2o / :
!
!
!
r
Y 2
Y 2
Y 2 p
4
o
o
o
dCi
dCs
n dCn .
(3.38)
D[ ] =
3
s
n
i
Such a normalization implies that

Z

1 = D[ ] exp 4o , .
329
(3.39)
Here o is a parameter with the dimension of an inverse length. In a similar way we obtain
the measure D[h ], which is normalized as

Z

2
(3.40)
1 = D[h ] exp o h , h ;
2
we shall shortly comment on the relative normalization of D[h ] and D[ ]. The result
is
!
!
!
Y o
Y0 o p
Y o q
dCk
2s dCs
n n dCn
D[h ] =
2
2
2
n
s
k
!
!
Y

Y o 1
o 1
o 1
h
h
h
dC0
dCi
dCn .
(3.41)

2 2
2 2
2 2
n
i
Here the prime indicates that the zero modes of the vector fluctuation operator do not
contribute to the product. Notice, however, that these modes do contribute to the measure
D[ ].
The following remarks are in order. We use units where all fields and parameters have
dimensions of different powers of a length scale l. One has [1/G] = [] = [2o ] = [l 2 ].
Eigenvalues of all fluctuation operators have the dimension [l 2 ]. The coordinates x
are dimensionless, while [g ] = [h ] = [l 2 ]. For the vectors, [ ] = [ ] = [l 2 ], and
for the scalars [h] = [l 0 ] and [] = [l 2 ]. We assume that the scalar, vector and tensor
eigenfunctions in (3.19) are orthonormal with respect to the scalar product in (3.11). As a
(k)
] = [l], [(s)] = [l 0 ], [ (p) ] = [l 1 ],
result, the dimensions of the eigenfunctions are [
which gives for the Fourier coefficients in (3.20), (3.21) [C ] = [C h ] = [l], [C ] = [l 2 ],
and [C ] = [l 3 ].
The normalization of D[h ] can be arbitrary, which is reflected in the presence of
the arbitrary parameter o in the above formulas. However, the relative normalization of
D[h ] and D[ ], which is defined by Eqs. (3.39) and (3.40) is fixed by gauge invariance.
Had we chosen instead a different relative normalization, say dividing each mode in (3.38)
by 2, then the path integral would acquire a factor of 2N0 , where N 0 is the number of
. (The issue of relative normalization
eigenvalues of the non-gauge-invariant operator 1
0
of the fluctuation and FaddeevPopov determinants seldom arises, since in most cases the
absolute value of the path integral is irrelevant.)
3.4. Computation of the path integral
Now we are ready to compute the path integrals in (3.17), (3.18). Let us illustrate the
procedure on the example of Eq. (3.18), which determines the FaddeevPopov factor DFP .
Using 2 Ig from Eq. (3.33) and the measure D[ ] from (3.38) we obtain
330
(DFP )1 =
Y Z 2
2
o dCs exp (s )2 Cs
s
r

Z
Y
16 3
3 h 2
2o 4
dCi exp
Ci
C
3
27
8 i
i

Y Z 2 p
1 X
2
o n dCn exp
n 2 n Cn + Cnh
,
16 n
which gives
(DFP )1 = 1
Y0
s
2o
Y 3 2o
i
Y 2 2o
n
(3.42)
!
.
(3.43)
3.4.1. Zero modes

The factor 1 in (3.43) arises due to the gauge zero modes, for which s 0 = 0 and
the integral is non-Gaussian:
Z Y 2
o dC0j ,
(3.44)
1 =
with the product taken over all such modes. The existence of zero modes of the Faddeev
Popov operator indicates that the gauge is not completely fixed. This can be related to
the global aspects of gauge fixing procedure known as the Gribov ambiguity. However,
Gribovs problem is usually not the issue in the perturbative calculations, where zero
modes arise rather due to background symmetries. This will be the case in our analysis
below. Specifically, the isometries of the background manifold M form a subgroup H of
the full diffeomorphism group. Sometimes H is called the stability group; for the S 4 and
S 2 S 2 backgrounds H is SO(5) and SO(3) SO(3), respectively. Since the isometries do
not change h (in the linearized approximation), their generators, which are the Killing
vectors K , are zero modes of the FaddeevPopov operator.
We therefore conclude that the integration in (3.44) is actually performed over the
stability group H. Since the latter is compact in the cases under consideration, the integral
is finite. In order to actually compute the integral, some further analysis is necessary,
in which we shall adopt the approach of Osborn [44]. First of all, let us recall that all
eigenmodes in our analysis have unit norm. If we now rescale the zero modes such that the
Killing vectors Kj Kj x become dimensionless (remember that the coordinates x

are also dimensionless), then the expression in Eq. (3.44) reads
Z Y 2
o kKj k dCj ,
(3.45)
1 =
where now [kKj k] = [l 2 ] and [Cj ] = [l 0 ]. For small values of the parameters Cj they can
be regarded as coordinates on the group manifold H in the vicinity of the unit element.
x
Since H acts on M via x x (Cj ), one has Kj = C j C
. However, strictly
j x
speaking Cj are not coordinates on the group manifold H but rather on its tangent space
331
at the group unity, such that their range is infinite. We wish to restrict the range of Cj , and
for this we should integrate not over the tangent space but over H itself. In other words, to
render the integral in (3.45) convergent we must treat the zero modes non-perturbatively,
Q
and for this we should replace the perturbative measure j dCj by a non-perturbative one,
d(C).
In general it is a difficult issue to construct the non-perturbative path integration
measure. However, in the zero mode sector this can be done. We note that the measure
should be invariant under the group multiplications, d(CC 0 ) = d(C), and this uniquely
requires that d(C) should be the Haar measure for H. The normalization is fixed
by the requirement that for Cj 0 the perturbative result (3.45) is reproduced. This
unambiguously gives
!
Z Y 2

o
d(C),
(3.46)
1 =

Cj
j
where /Cj is computed at Cj = 0 and d(C) is the Haar measure of the isometry group
Q
H normalized such that d(C) j dCj as Cj 0.
3.4.2. The path integral
Now, using (3.32) and the measure (3.41), we compute the path integral in (3.17) first
without the FaddeevPopov factor DFP :
!
!
!
Z
Y0 o
Y0 o
Y 2 o

2
q
D[h ] exp Igf = 2
k
s
n
s
k
n
!
!
Y 3 o
Y 2 o
o
q
.
(3.47)
2 i
2
n
n
Here the primes indicate that zero and negative modes should be omitted from the products.
Zero vector modes do not contribute since they are not present in the path-integration
measure (3.41), and we assume that there are no negative vector modes, since otherwise
the metric on the space of fluctuations would not be positive definite. For tensor fluctuations
negative and zero modes are present in the measure (3.41), and their overall contribution
is collected in the factor 2 in (3.47). Let us further assume that there are no zero tensor
modes, which is the case for the manifolds of interest. If negative modes are also absent
then 2 = 1. If there is one negative tensor mode with eigenvalue < 0, then
Z
o
2
dC exp 12 C .
(3.48)
2 =
2
The integral is computed via the deformation of the contour to the complex plane, which
gives the purely imaginary result
o
,
(3.49)
2 =
2i | |
with the factor of 1/2 arising in the course of the analytic continuation [12].
332
Both the FaddeevPopov factor in (3.43) and the path integral in (3.47) depend on the
gauge parameter . However, the -dependence exactly cancels in their product, which
provides a very good consistency check. In particular, the relative normalization of the
integration measures fixed by Eqs. (3.39) and (3.40) is important. If we had divided each
factor in the mode products in (3.38) by a 6= 1, then the resulting path integral would be
Q
proportional to ( n a) a (0) with being the -function of the -dependent operator
. Thus, unless a = 1, the result would be gauge-dependent.
1
0
We therefore finally obtain the following expression for the path integral in (3.17):
!
!
!
r
Y 2 1
Y0 s
Y 0 o
2 o
(3.50)
eI .
Z[g ] =
1 2
3 o

o
k
s
i
k
Here 2 is the contribution of the negative tensor mode, and 1 is the isometry factor. As
we expected, the contribution of all unphysical scalar modes has canceled from the result.
The only scalar modes which do contribute are the several lowest lying modes which seem
to be physically
distinguished. These are the constant conformal mode giving rise to the
factor o / 2, and the 5 conformal Killing scalars which exist only in the S 4 case
and give rise to the product over i. The next two factors in (3.50) is the contribution of
the transversal vector modes and the TT-tensor modes. Finally, I = I [g ] is the classical
action.
In order to apply the above formula for Z[g ] we need the eigenvalues of the fluctuation
operators. Now we shall determine the latter for the manifolds S 2 S 2 and S 4 .
4. Spectra of fluctuation operators

In this section we derive the spectra of small fluctuations around the S 2 S 2 and S 4
instantons. In the S 2 S 2 case the problem is tackled via solving the differential equations.
It turns out that in a suitable basis the system of 10 coupled equations for the gravity
fluctuations splits into 10 independent equations. The latter are solved in terms of spinweighted spherical harmonics. In the S 4 case the equations do not decouple and the direct
approach is not so transparent. However, the problem can be conveniently analyzed with
group theoretic methods, which was done some time ago by Gibbons and Perry [27]. We
shall describe the group theory approach in some detail with the same principal result
as in [27].
4.1. Fluctuations around the S2 S2 instanton
Let us consider the metric of the S 2 S 2 instanton background,
ds 2 =

1
(d1 )2 + sin2 1 (d1 )2 + (d2 )2 + sin2 2 (d2 )2 .
(4.1)
It is convenient to set = 1 for the time being; at the end of calculations the -dependence
is restored by multiplying all eigenvalues with . We introduce the complex tetrad

1
i
e1 = d1 +
d1 ,
sin 1
2

1
i
d2 ,
e3 = d2 +
sin 2
2
The metric in (4.1) splits as g
1
ab = g ea eb =
0
0

1
i
d1 ,
e2 = d1
sin 1
2

1
i
d2 .
e4 = d2
sin 2
2
= ea eb ab , where the tetrad metric is
1 0 0
0 0 0
.
0 0 1
0 1
333
(4.2)
(4.3)
4.1.1. Tensor modes

First we consider the eigenvalue problem
2R = ,
(4.4)
where
= 0,
= 0.
(4.5)
We expand with respect to the complex basis (4.2),

= ea eb ab ,
(4.6)
insert this into (4.4) and project the resulting equation onto the basis (4.2) again.
Remarkably, the system of 10 coupled equations splits then into 10 independent equations
for the 10 tetrad projections ab . A partial explanation of this fact is the existence of two
different parity symmetries acting independently on the two spheres.
It is convenient to introduce the operator
2
2
, ] = + cot + 2is cot + 1
s2 cot ,
D[s,
2
sin
sin2 2
whose eigenfunctions are the spin-weighted spherical harmonics s Yj m [29],

, ]s Yj m (, ) = s2 j (j + 1) Yj m (, ).
D[s,
s
(4.7)
(4.8)
Here j and m are such that j > |s| and j 6 m 6 j . One has s Yj m = 0 for j < |s|.
(Notice that we use the bold-faced s for the spin weight.) The following relations between
harmonics with different values of the spin weight s are useful:
p
(4.9)
L [s, , ] s Yj m = (j s)(j s + 1)s1 Yj m ,
where
i
s cot .
L [s, , ] =
sin
The harmonics for a fixed s form an orthonormal set on S 2 .
Using the above definitions, Eqs. (4.4) can be represented as

2
2

D s1ab , 1 , 1 + D s2ab , 2 , 2 s1ab s2ab + 2 + ab = 0,
(4.10)
(4.11)
where 1 6 a, b 6 4 (no summation over a, b). Here the nonzero elements of the symmetric
matrices s1ab and s2ab are
334
s111 = s122 = 2,
s113 = s114 = s123 = s124 = 1,
s233 = s244 = 2,
s213 = s223 = s214 = s224 = 1.
(4.12)
The solution to Eqs. (4.11) reads

ab = Cab s1 Yj1 m1 (1 , 1 )s2 Yj2 m2 (2 , 2 ),
ab
ab
(4.13)
with Cab being integration constants. The eigenvalue, , is the same for all ab :
= j1 (j1 + 1) + j2 (j2 + 1) 2.
(4.14)
This is essentially the sum of squares of the two SO(3) angular momentum operators acting
independently on the two spheres.
Let us now count the degeneracy of the modes. For this one should take into account
the additional conditions (4.5), which gives algebraic constraints for the coefficients Cab .
We consider first the trace condition = 0. In the language of the tetrad projections this
reduces to 12 + 34 = 0, or equivalently to
C12 + C34 = 0.
(4.15)
Hence only 9 out of the 10 constants Cab are independent.

Next, we consider the Lorenz condition = 0. This implies

L s1a1 , 1 , 1 a1 + L + [s1a2 , 1 , 1 ] a2

+ L s2 , 2 , 2 a3 + L + s2 , 2 , 2 a4 = 0
a3
a4
(4.16)
(no summation over a). Inserting the solution (4.13) and using the recurrence relations in
(4.9), these conditions reduce to algebraic constraints
1 C11 1 C12 + 2 (C13 C14 ) = 0,
1 C12 1 C22 + 2 (C23 C24 ) = 0,
1 (C13 C23 ) + 2 C33 2 C34 = 0,
(4.17)
1 (C14 C24 ) + 2 C34 2 C44 = 0.
Here = j (j + 1) (with = 1, 2) and = (j + 2)(j 1) for j > 1 while = 0

for j = 0.
For j > 2 (which corresponds to quadrupole or higher deformations of each sphere)
none of the coefficients , vanish, and the algebraic constraints (4.17) reduce the
number of independent coefficients Cab to five. This gives the degeneracy d:
j1 > 2, j2 > 2,
d = 5 (2j1 + 1)(2j2 + 1).
(4.18)
The situation is different for small values of j . Consider, for example, the j1 = j2 = 0
sector. Since the harmonics s Yj m vanish for j < |s|, we must set in the solution (4.13) all
Cab s to zero, apart from C12 = C34 . The Lorenz condition (4.16) is then fulfilled. As a
result, there is only one independent integration constant, which yields
j1 = j2 = 0,
d = 1.
Notice that in this case = 2.
(4.19)
335
In a similar way one can consider the sector where j1 = 0 and j2 = 1 (or j1 = 1 and
j2 = 0), in which case = 0. One discovers then that the Lorenz constraints (4.16) require
that all non-trivial coefficients Cab mush vanish. As a result,
j1 = 0, j2 = 1,
or
j2 = 1, j1 = 0,
d = 0,
(4.20)
which shows that there are no zero modes.

Next,
j1 = j2 = 1,
d = (2j1 + 1)(2j2 + 1) = 9;
(4.21)
and finally
j1 > 2, j2 = 0,
d = 2j1 + 1;
j1 > 2, j2 = 1,
d = 9(2j1 + 1),
(4.22)
which conditions are symmetric under interchanging j1 and j2 .

To recapitulate, the spectrum of the tensor fluctuations contains one negative mode and
no zero modes.
4.1.2. Vector modes
Let us now consider the eigenvalue problem

=
(4.23)
subject to the condition

= 0
(4.24)
for the vector fluctuations on the S 2 S 2 background. We again expand the fluctuations
with respect to the basis (4.2),
= ea a ,
(4.25)
insert this into (4.23), and project back to the tetrad. Similarly as in the tensor case, the
equations decouple to give

(4.26)
D s1a , 1 , 1 + D s2a , 2 , 2 + 1 + a = 0,
where 1 6 a 6 4 (no summation over a), and nonzero coefficients read s11 = s12 = s23 =
s24 = 1. The solution is
a = Ca s1a Yj1 m1 (1 , 1 )s2a Yj2 m2 (2 , 2 ),
(4.27)
with Ca being integration constants, and the eigenvalue is the same for all a :
= j1 (j1 + 1) + j2 (j2 + 1) 2.
The Lorenz condition, = 0, reads

L s11 , 1 , 1 1 + L + s12 , 1 , 1 2

+ L s23 , 2 , 2 3 + L+ s24 , 2 , 2 4 = 0,
(4.28)
(4.29)
336
which reduces upon inserting (4.27) to the algebraic condition

p
p
j1 (j1 + 1)(C1 C2 ) + j2 (j2 + 1)(C3 C4 ) = 0.
(4.30)
This allows one to count the degeneracies:

j1 > 1, j2 > 1 ( > 0),
d = 3 (2j1 + 1)(2j2 + 1);
(4.31)
and also
j1 > 2, j2 = 0 ( > 0),
d = j1 (j1 + 1);
j1 = 1, j2 = 0 ( = 0),
j1 = 0, j2 = 0 ( = 2),
d = 3;
d = 0.
(4.32)
These results are symmetric under j1 j2 , hence there are no negative modes, there are
six zero modes corresponding to the six Killing vectors of S 2 S 2 , and the rest of the
spectrum is positive.
4.1.3. Scalar modes and the orthogonality conditions
The eigenvalue problem for the scalar modes,
h = h,
(4.33)
reduces to the equation

2 , 2 ] + h = 0,
1 , 1 ] + D[0,
D[0,
(4.34)
whose solutions are

h = Yj1 m1 (1 , 1 )Yj2 m2 (2 , 2 )
(4.35)
(for s = 0 the spin-weighted spherical harmonics coincide with the usual spherical
harmonics). The eigenvalues are just
= j1 (j1 + 1) + j2 (j2 + 1),
(4.36)
and the degeneracies are

j1 > 0, j2 > 0,
d = (2j1 + 1)(2j2 + 1).
(4.37)
We have obtained the spectra of all relevant fluctuation operators. Although the
eigenfunctions are complex, one can pick up their real part in a way that is consistent
with the orthogonality conditions. For example, for the scalar modes one considers
1+i
(4.38)
<(h) = Yj1 m1 Yj2 m2 + c.c.,
2
and one can easily see that the modes <(h) with different quantum numbers (j1 m1 j2 m2 )
are orthogonal with respect to the scalar product defined in Eq. (3.11).
For the vector modes a the procedure is slightly more complicated, since the tetrad
metric ab is not diagonal. In addition, harmonics s Yj m for different values of the spin
weight are not orthogonal. Consider, however, the real combinations
1+i
<( ) = ea a + c.c.,
2
(4.39)
337
Table 1
Spectra of fluctuations around the S 2 S 2 instanton
Operator
Eigenvalue
Degeneracy
12
2
2
(j (j + 1) 2)
j (j + 1)
(j1 (j1 + 1) + j2 (j2 + 1) 2)
1
9
2(2j + 1)
18(2j + 1)
5(2j1 + 1)(2j2 + 1)
j >2
j >2
j1 , j2 > 2
11
0
(j (j + 1) 2)
(j1 (j1 + 1) + j2 (j2 + 1) 2)
6
2(2j + 1)
3(2j1 + 1)(2j2 + 1)
j >2
j1 , j2 > 1
10
(j1 (j1 + 1) + j2 (j2 + 1))
(2j1 + 1)(2j2 + 1)
j1 , j2 > 0
where a has quantum numbers (j1 m1 j2 m2 ). Consider <((1)) and <((2) ) with different
quantum numbers. Their scalar product (defined in Eq. (3.11)) can be computed using the
relations

ab = ea eb g ,
(4.40)
ab = ea eb g ,
which gives

X
(1) (2)
(1) (2)
(1) (2)
a , a
+ i 1 , 2 + i 3 , 4 + c.c.
< (1) , < (2) =
a
(4.41)
P
(1)
(2)
i
a ha , a
is a bilinear combination of spin-weighted

Here each term in the sum
harmonics with the same value of the spin weight, such that the orthogonality relation
holds. Next, integrating by parts and using the recurrence relations (4.9) one can show that
(1)
(2)
(1)
(2)
the remaining term in the scalar product, ih1 , 2 i + ih3 , 4 i + c.c., vanishes.
This shows that vector modes <( ) with different quantum numbers are orthogonal.
A similar procedure can be carried out for the tensor modes. Hence for all eigenmodes
considered above one can choose the real part in such a way that the orthogonality
condition holds. This is a manifestation of the fact that the fluctuation operators are selfadjoint. We finally restore the dependence on and summarize the results of this section
in Table 1. There is one negative mode in the spectrum, and this plays a crucial role in
our analysis. The corresponding deformation increases the radius of one of the spheres,
shrinking at the same time the second one.
4.2. Fluctuations around the S4 instanton
The S 4 instanton can be viewed as the four-dimensional sphere with radius 3/ in

five-dimensional Euclidean space E 5 . Although the corresponding eigenvalue problem for
fluctuations was considered in [27], we have reanalyzed it for the sake of completeness
(with the same result) and shall present below the key steps of our analysis. The problem
338
essentially reduces to studying representations of SO(5) [1,2,8,31], whose Casimir operator

can be related to the invariant Laplacians on S 4 with the help of the projection formalism
[34]. We shall therefore first outline the group theory part by summarizing the relevant
facts about representations of SO(5). We shall work on the unit 4-sphere rescaling at the
end the eigenvalues by the factor /3.
4.2.1. Representations of SO(5)
The unit sphere S 4 in E 5 is defined in Cartesian coordinates by the equation
P5
a 2
1 = (x 1 ix 2 )/ 2,
a=1 (x ) = 1. Itis convenient to use the complex coordinates
2 = (x 3 ix 4 )/ 2, 0 = x 5 . We shall not distinguish between lower and upper indices,
P
i = i . In these new coordinates the defining quadratic form reads 2i=2 i i = 1,
which is annihilated by the generators of SO(5):
(4.42)
Yji = i j j i ,
whose commutation relations are

i k
j
pq
p
i
k
i
Yl Cij,kl Yq .
(4.43)
Yj , Yl = jk Yli li Yjk + jl Yk
j
Since Yji = Yi , the independent generators can be chosen to be those for i < j . Y11 and
j
Y22 generate the Cartan subalgebra, while Yji and Yi for i < j < i are the raising and
lowering operators, respectively. One has
i k
(4.44)
Yi , Yl = lk (i)Ylk ,
where
i
i
k
lk (i) = ki li + l
(4.45)
(lk (1), lk (2)). The roots corredetermine the root vectors with the components
2 = (1, 1), and
sponding to the four raising operators are 12 = (1, 1), 02 = (0, 1), 1
1
0 = (1, 0).
lk
Irreducible representations of SO(5) are characterized by two numbers denoted by m

(m1 , m2 ), where m2 > m1 and both m1 and m2 are either integer or half-integer. The
highest weight vector m is annihilated by all raising operators, Yji m = 0 for i > j >
i, and it is an eigenvector of the Cartan subalgebra generators, Yii m = mi m , i = 1, 2.
j
j
i
i
Yj
, one finds the eigenvalues
Using these properties and also [Yji , Yi ] = Yii Yj 2j
of the Casimir operator,
X
j
m1
Yji Yi m = Cm m ,
(4.46)
C
2
i,j
where
Cm = m1 (m1 + 1) + m2 (m2 + 3)
(4.47)
is the same for all vectors of the representation. The dimension of representations is given
by
Y
Y
h, ri.
(4.48)
dim(m) = h, r + mi
339
P
Here the product is over the four root vectors described above, and r = 12 = ( 12 , 32 ).
One has r + m = ( 12 + m1 , 32 + m2 ), and h , i is the scalar with respect to the Cartan metric
pq
kl
= 6ij (here i, j = 1, 2). As a result,
gij = Cii,kl Cjj,pq
1
dim(m) = (2m1 + 1)(2m2 + 3)(m2 m1 + 1)(m1 + m2 + 2).
6
(4.49)
4.2.2. Scalar modes

Using Eqs. (4.47), (4.49) one can find the spectra of the relevant fluctuation operators.
It is now convenient to pass back to the Cartesian coordinates x a = xa (a = 1, . . . , 5),
such that the sphere S 4 is determined by the condition r x a xa = 1. Let na

x a /r be the unit normal to the sphere. The (anti-hermitean) generators of SO(5) in
Cartesian coordinates are given by Mab = na b nb a , and the Casimir operator is C =
P
12 (Mab )2 12 ab (Mab )2 .
Let us define the projection operator Pab = ab na nb = P ab = Pba , which can be
thought of as the induced metric on the sphere. In what follows we shall use the projection
formalism [34] to describe geometrical 4-objects tangent to the sphere in terms of 5-objects
of the embedding space. For example, a 4-vector can be described as a 5-vector a
subject to the condition na a = 0. The covariant derivative of a tensor is obtained by taking
the partial derivative and then projecting all the indices down to the sphere. For example,
p q
a b = (p q )Pa Pb . One has na = na , while for objects tangent to the sphere 5-indices
can be raised and lowered either with Pab or with ab . The curvature tensor is given by
Rpsqt = Ppq Pst Ppt Psq .
Consider first scalar fluctuations. The covariant Laplacian for a scalar field h can be
expressed in terms of the angular momentum operator as
1
(4.50)
2h P ab a Pbc c h = (Mab )2 h = Ch.
2
Scalars transform according to the (0, j ) representations, which correspond to the Young
tableaux 1 2 j and can be represented in terms of homogeneous polynomials on S 4
as
h = h(a1 ...aj ) na1 . . . naj .
(4.51)
Hence, the eigenvalues of the Casimir operator in Eq. (4.50) are C(0,j ) = j (j + 3), which
gives the spectrum of the scalar eigenvalue problem, 10 h = h with 10 3 2:
=
j (j + 3),
3
d=
1
(2j + 3)(j + 2)(j + 1),
6
j > 0.
(4.52)
4.2.3. Vector modes

Consider a tangent vector field s = Psa a . The invariant Laplacian reads
1
p q
(4.53)
2s P ab a Pbc c p Pq Ps = (Mab )2 s + 2(a a )ns + s .
2
Here the last two terms on the right can be related to the contribution of the spin operator.
Under general SO(5) rotations a vector (x) transforms into (x)
= R (R 1 x), where
340
R = exp(ab Sab ) with ab = ba being the rotation parameters and Sab (Sab )pq =
p q
p q
a b b a . For |ab | 1 one obtains = ab (Mab + Sab ), such that Sab can be
p
identified with the spin operator: (Sab )s = (Sab )s p . As a result,

1
2
(Mab + Sab ) + 3 s (C + 3) s ,
(4.54)
2s =
2
where the Casimir operator is now the square of the total angular momentum. The general
vector harmonics on S 4 can be obtained by considering the product of a constant vector in
E 5 with scalar harmonics on S 4 . Such a product decomposes into three irreducible pieces,
(0, 1) (0, j ) = (1, j ) (0, j + 1) (0, j 1), which can be visualized as
1 2
1 2
0 1
1 2
j-1
(4.55)
The first term on the right here is the (1, j )-piece, and in the language of homogeneous
polynomials it reads
s = [sa](a1...aj1 ) na na1 . . . naj2 naj1 ,
(4.56)
where [sa](a1...aj1 ) is traceless with respect to any pair of indices. This is manifestly
tangential and coexact. As a result, the eigenvalues of the Casimir operator are C(1,j ) =
j (j + 3) + 2, and the spectrum of the vector eigenvalue problem 11 s ( 3 2 )s =
s in the sector where a a = na a = 0 is given by
=
j (j + 3) 4 ,
3
d=
1
j (j + 3)(2j + 3),
2
j > 1.
(4.57)
One can also directly verify that the harmonic s in Eq. (4.56) fulfills the condition
1
2
2 (Mab ) s = j (j + 3)s . It follows then from Eq. (4.53) that 2s = (j (j + 3) 1)s ,
and this again yields the spectrum in Eq. (4.57). The correct degeneracy can be obtained
by counting the independent components of [sa](a1...aj1 ) [31].
The remaining two pieces in (4.55), when represented in terms of the polynomials on
S 4 , can be related to the exact tangential and the normal components of the vector field.
4.2.4. Tensor modes
For a symmetric tensor hpq = Ppa Pqb hab a direct calculation gives
p q p q
2hpq + 2Rpsqt hst P ab a Pbc (c hp q )Pp Pq Pp Pq + 2(Ppq Pst Ppt Psq )hst

1
= (Mab )2 hpq + 2np a haq + 2nq a hap + 2pq haa
2

1
2
(Mab + ab ) + 6 hpq (C + 6)hpq .
=
(4.58)
2
Here the spin operator is defined in the same way as for vectors, which gives (ab h)pq =
(Sab )p s hsq + (Sab )q s hsp . The general tensor harmonics on S 4 are obtained by the direct
products (0, 2) (0, j ) = (0, j + 2) (1, j + 1) (0, j ) (2, j ) (0, j + 1) (1, j 1)
(0, j 2). Again this can be visualized by Youngs diagrams and represented in the
341
Table 2
Spectra of fluctuations around the S 4 instanton
Operator
Eigenvalue
Degeneracy
12
j (j + 3)
3
11
(j (j + 3) 4)
3
j (j + 3)
3
5 (j 1)(j + 4)(2j + 3)
6
1 j (j + 3)(2j + 3)
2
1 (j + 1)(j + 2)(2j + 3)
6
10
j >2
j >1
j >0
language of homogeneous polynomials. The transverse and tracefree harmonics tangent to

the sphere correspond to the (2, j ) piece, whose explicit representation is
hpq = h[pa][qb](a1...aj2 ) na nb na1 . . . naj2 .
(4.59)
Here h[pa][qb](a1 ...aj2 ) is traceless with respect to any pair of indices and is symmetric
under interchange of the [pa] and [qb] pairs. As a result, the eigenvalues of the Casimir
operator are C(2,j ) = j (j + 3) + 6. This gives the spectrum of the tensor eigenvalue
problem 12 hpq = hpq in the sector where a hab = na hab = haa = 0:
5
j (j + 3),
d = (j 1)(j + 4)(2j + 3), j > 2.
(4.60)
3
6
The same result can be obtained by directly verifying that hpq in Eq. (4.59) fulfills the
condition 12 (Mab )2 hpq = j (j + 3)hpq .
The other tensor harmonics in the expansion of (0, 2) (0, j ) correspond to the exact
and coexact pieces of the longitudinal vector part of the 4-metric, to those of the 4-vector
h5 , and to the trace [27].
We summarize the results of our analysis in this section in Table 2. Notice that the scalar
and tensor eigenvalues are the same (for j > 2), while the vector spectrum is shifted by a
constant.
=
5. Partition function
Now we are able to derive the explicit expressions for the one-loop partition functions
for fluctuations around the S 2 S 2 and S 4 instantons. The corresponding formula was
obtained in Eq. (3.50) above. It is convenient to pass from o to the dimensionless
normalization parameter 0 via the rescaling
(5.1)
o = 0 .
The one-loop partition function for gravity fluctuations around an Euclidean background
then reads
s
0
2 Det0 11 I
e .
(5.2)
Z[g ] = 0
1 Det0 12
2
Herethe first two factors on the right are the contributions of the scalar modes. The factor
0 / 2 is due to the constant conformal mode, which is always present, and 0 is the
342
contribution of the 5 scalar modes with eigenvalue 4/3 which exist only for the S 4
instanton (see Table 2):
r

2 1 5
.
(5.3)
0 =
3 0
For any background other than S 4 one has 0 = 1. As was discussed above, other scalar
modes do not contribute to the partition function.
The factor 2 in (5.2) is the contribution of the negative tensor mode,
0
.
(5.4)
2 =
2i | |
For the S 2 S 2 instanton there is one such mode with = 2, while in the S 4 case all
tensor modes are positive and 2 = 1. Next,
!
Y 2

0

Vol(H),
(5.5)
1 =

Cj
is the isometry factor. If the background has no isometries then 1 = 1.

The determinants in Eq. (5.2) are the contributions of the positive vector and tensor
modes. One has
!
s

p
Y0

1
s
1 0
2
0
ln
Det 11 =
(0)
(0)
,
(5.6)
=
exp
0 1
2 1
2
20
s
where the -function for the positive, transverse vector modes is
X0 z
.
1 (z) =
s
s
Similarly for the positive, transverse traceless tensor modes:
!
s

p
Y0

1
k
1 0
2
0
ln
Det 12 =
(0)
(0)
=
exp
2
0
2 2
2
20
(5.7)
(5.8)
with
2 (z) =
X0 z
k
k
(5.9)
The last factor in Eq. (5.2) is the classical contribution, with I being the action for the
background. Let us now apply these formulas.
5.1. The S2 S2 instanton
The classical action is I [S 2 S 2 ] = 2/G, and according to Table 1,
0
2 = .
(5.10)
0 = 1,
2i 2
Consider now the isometry factor 1 in (5.5), which is due to the background SO(3)
SO(3) symmetry. Each of the two SO(3) groups can be parameterized by matrices Uik =
343
exp(ikj Cj ). The invariant metric on the SO(3) space is gik = 12 tr(i U k U 1 ) ik
for Cj 0. RThe Haar measure is d(C) = det gik dC1 dC2 dC3 , and the volume
Vol(SO(3)) = d(C) = 8 2 . For later use, we reproduce this result in a different way.
The measure for a (compact, semi-simple) Lie group G can be represented as the product
of the measure for the maximal subgroup H and that for the coset G/H. This implies that
Vol(G) = Vol(H) Vol(G/H).
(5.11)
In particular, Vol(SO(3)) = Vol(SO(2)) Vol(S 2 ), where Vol(SO(2)) = 2 , and the

volume of the S 2 coset with unit (due to the normalization of the measure) radius is
Vol(S 2 ) = 4 . As a result, Vol(SO(3)) = 2 4 = 8 2 .
When acting on S 2 , the SO(3) generators /Cj generate rotations in the three
orthogonal planes of the embedding Euclidean 3-space. Let /C3 be the generator of
rotations in the XY-plane, such that the azimuthal angle of the spherical coordinate system
changes as + C3 . Then the norm k C 3 k is the square root of

Z
,
.
(5.12)
g g d 4 x =
=

32G
33 G
S 2 S 2
k
k C
1
and k C 2 k and those of the generators of the second SO(3)

Obviously, the norms
factor are the same. Hence,
2 6

2 64 4 (0 )12

=
.
(5.13)
Vol(SO(3))
1 = 0
27(G)3

Consider now the positive modes. The -function associated with the positive vector
modes is (see Table 1)
1 (s) =
X
j =2
X
X
2(2j + 1)
3(2j1 + 1)(2j2 + 1)
+
.
s
{j (j + 1) 2}
{j1 (j1 + 1) + j2 (j2 + 1) 2}s
(5.14)
j1 =2 j2 =2
This can be represented as

1 (s) = 4s 2 (2, 9|s) + 3 Z(1, 10|s) ,
(5.15)
where the following two functions have been introduced:

(k, |s) =
Z(k, |s) =
2j + 1
,
{(2j + 1)2 + }s
j =k
X
X
j1 =k j2 =k
(2j1 + 1) (2j2 + 1)
.
{(2j1 + 1)2 + (2j2 + 1)2 + }s
(5.16)
(5.17)
These functions are studied in detail in the appendix. Similarly, using the results of Table 1
one obtains the -function for the positive tensor modes

(5.18)
2 (s) = 9 2s + 4s 2 (2, 9|s) + 18 (2, 1|s) + 5 Z(2, 10|s) .
The following relation implied by the definitions in (5.16), (5.17), will be useful:
Z(1, 10|s) = Z(2, 10|s) + 6 (2, 1|s) + 9 8s .
344
5.1.1. The scaling behaviour

Before we proceed further, it is very instructive to pause and check whether the
expressions above agree with the general formulas for the scaling behaviour of effective
actions. We shall follow the approach of Christensen and Duff [13], who relate this scaling
behaviour to
Z

1
R R + 6362 g d 4 x,
N0 =
2
180 (4)
Z

1
N1 =
11R R + 9842 g d 4 x,
2
180 (4)
Z

1
(5.19)
189R R 7562 g d 4 x.
N2 =
180 (4)2
Here N0 is the number of eigenvalues of the scalar operator 10 2 acting on a manifold
with R = g . N1 is the number of eigenvalues of the vector operator 11 acting in the
space of all vectors, that is, including both transverse and longitudinal fluctuations. Finally,
N2 counts both transverse and longitudinal eigenstates of the tensor operator 12 , with the
only requirement that the fluctuations must be traceless.
Let us apply these formulas to the S 2 S 2 background. The volume of the manifold is
VS 2 S 2 = (4)2 /2 , while R R = 82 . As a result,
161
224
21
,
N1 =
,
N2 = .
(5.20)
45
45
5
Now let us obtain the same result via a direct evaluation of the -functions. First we
consider the scalar case. Using the results of Table 1, the operator 10 2 has one negative
mode, six zero modes, while the rest of the spectrum is positive and gives rise to the
-function

(5.21)
0 (s) = 4s 2 (2, 9|s) + Z(1, 10|s) .
N0 =
Hence the number of all eigenvalues is 7 + 0 (0). In order to compute 0 (0), we use the
results of the appendix, where the following values are obtained:
1
1
k2,
12 4

2 2
1
1
13
1 2

+ 2k 4 +
k +
.
Z(k, |0) =
32
24
2
3
360
(k, |0) =
(5.22)
(5.23)
This gives for the -functions in (5.15), (5.18), (5.21)

0 (0) =
154
,
45
1 (0) =
18
,
5
2 (0) =
38
.
9
(5.24)
161
Using these, the number of scalar eigenvalues is N0 = 7 154
45 = 45 , which agrees with
(5.20).
Next, the vector operator 11 has 6 zero modes, such that the number of its eigenvalues in
the transverse sector is 6 + 1 (0). Now, one should take into account also the longitudinal
vectors, which are gradients of scalars. It is not difficult to see that if is an eigenvector
of 11 , such that 11 = , then (10 2) = . We see that the eigenfunctions
345
of 10 2 are in one-to-one correspondence with the longitudinal vectors. The number

of the latter is therefore N0 1, where the one is subtracted because the ground state scalar
eigenfunction is constant, which vanishes upon differentiation. We therefore conclude that
161
224
N1 = 6 + 1 (0) + N0 1 = 6 18
5 + 45 1 = 45 , which also agrees with (5.20).
Finally, the number of traceless eigenvalues of 12 is 1 + 2 (0) (here the one is
the contribution of the negative mode) plus the number of longitudinal traceless tensor
L = + 1 g .
harmonics

2
L associated with one has 1 L = L . Hence,
Now, if 11 = then for
the number of longitudinal tensors is determined by the number of vectors, which gives
N2 = 1 + 2 (0) + (N1 6). Here six is subtracted because the six Killing vectors do not
L = 0. We therefore
contribute to the tensor spectrum, since for Killing vectors one has
224
21
obtain N2 = 1 + 38
9 + 45 6 = 5 , which again is in perfect agreement with (5.20).
The overall scale dependence of the partition function is given by the factor
(0 )N2 +N0 2N1 . For the S 2 S 2 instanton one has N2 + N0 2N1 = 98
45 , and we shall
shortly see that this agrees with our analysis.
5.1.2. The partition function Z[S2 S2 ]
It is now a simple task to insert the formulas above into the expression for the partition
function. We obtain
s

Det0 11
= exp 0 (0) + ln 20 (0) ,
(5.25)
0
Det 12
where
(s)

1
2 (s) 1 (s) = 9 2s + 4s Z(2, 10|s).
2
Using the values Z(2, 10|0) =

in the appendix) we find
s
352
1567
Det0 11
45 45 e .
=
2
0
Det0 12
581
45
(5.26)
and Z 0 (2, 10|0) = 18.3118 (see Eq. (A.50)
(5.27)
Finally, taking into account the contributions of the negative, zero, and scalar modes
computed in (5.10), together with the classical term, we obtain
s
3

2
27
(G)
Det0 11 I
e
(5.28)
Z S S 2 = i
Det0 12
256 410
0

2
98
.
= i 0.3667 (G)3 0 45 exp
G
This is our final result in the S 2 S 2 sector.
5.2. The S4 instanton
The classical action is I [S 4 ] = 3/G. Using the results in Table 2 we find
346
r
0 =
2 1
3 0
5
2 = 1.
(5.29)
Let us consider the symmetry factor 1 . The isometry group is now H = SO(5), and this
can be represented by matrices Uik = exp(Cik ), where Cik = Cki , i, k = 1, . . . , 5. The 10
generators /Cik generate rotations of S 4 in the 10 orthogonal planes of the embedding
Euclidean 5-space. Let /C12 be the generator of rotations in the XY-plane, such that the
standard azimuthal angle changes as + C12 . The norm k C12 k is the square root of

Z
9
1

,
,
(5.30)
g g d 4 x =
=

32G
103 G
S4
which applies also to the the norms of the remaining 9 generators.

The volume of SO(5) can be computed by directly constructing the invariant metric
and the Haar measure with the use of the matrix representation Uik = exp(Cik ). The
Q
measure should be normalized such that for Cik 0 it reduces to i<k dCik . However,
it is much simpler to use the coset reduction formula (5.11). One has SO(5)/SO(4) =
S 4 and SO(4)/SO(3) = S 3 , such that Vol(SO(5)) = Vol(S 4 ) Vol(S 3 ) Vol(SO(3)). We
know that Vol(SO(3)) = 8 2 , while the volumes of unit S 3 and S 4 are 2 2 and 8 2 /3,
respectively. As a result, Vol(SO(5)) = 128 6 /3. Summarizing,
2 10
5

128 6 (0 )20
Vol(SO(5)) = 9
.
(5.31)
1 = 0
10
3 (G)5

Let us consider the positive modes. The -function associated with the positive vector
modes is (see Table 2)
1 (s) =
1 s X j (j + 3)(2j + 3)
3
.
2
{j (j + 3) 4}s
(5.32)
j =2
This can be written as

1
1 (s) = 3s Q(1, 4, 0|s),
2
where the following function has been introduced
Q(k, , c|s) =
X
(2j + 3)(j (j + 3) + c)
j =k
{j (j + 3) + }s
(5.33)
(5.34)
Similarly, using the results of Table 2, one obtains the -function for the positive tensor
modes
5
(5.35)
2 (s) = 3s Q(2, 0, 4|s).
6
Finally, consider the scalar operator 10 2. According to Table 2, its eigenvalues,
measured in units of , are given by (j (j + 3) 6)/3, and the degeneracy is (j + 1)(j + 2)
(2j + 3)/6 with j > 0. Hence, the -function for the positive scalar modes is
0 (s) =
1 s
3 Q(2, 6, 4|s).
6
(5.36)
347
5.2.1. The scaling behaviour

Let us again check the consistency with the general formulas for the scaling behaviour
of quantum fields (for fluctuations around S 4 this was done by Christensen and Duff [13]).
Applying again the formulas in (5.19), where now the volume of the manifold is VS 4 =
24 2 /2 , while R R = 82 /3, one has
479
358
21
, N1 =
, N2 = .
(5.37)
90
45
10
On the other hand, using the result of the appendix,

11
1 2
3
1
1 4
3
k + (3 2c)k + 2 + c , (5.38)
Q(k, , c|0) = k 2k c +
2
2
2
3
15
N0 =
one obtains for the -functions in (5.32), (5.33), (5.35)

61
191
61
, 1 (0) =
, 2 (0) = .
(5.39)
90
30
90
Now, since the spectrum of 10 2 contains six non-positive modes, one has N0 =
479
6 + 0 (0) = 6 61
90 = 90 , which agrees with (5.37). Next, 11 has 10 zero modes, such
that there are 10 + 1(0) transverse vector eigenstates, plus (N0 1) longitudinal ones (the
479
358
constant scalar mode gives no contribution). As a result, N1 = 10 191
30 + 90 1 = 45 ,
which agrees with (5.37). Finally, there are N2 = 2 (0) + N1 15 traceless tensor modes,
where 15 is subtracted because 10 Killing vectors and 5 conformal Killing vectors of S 4 do
358
21
not contribute to the longitudinal tensor modes. One obtains N2 = 61
90 + 45 15 = 10 ,
which again agrees with (5.37).
The overall scale dependence of the partition functions is expected to be (0 )N2 +N0 2N1 ,
where N2 + N0 2N1 = 571
45 .
0 (0) =
5.2.2. The partition function Z[S4 ]

Let us now obtain the partition function. One finds
s

Det0 11
= exp 0 (0) + ln 0 (0) ,
0
Det 12
(5.40)
where

1
1
s 5
Q(2, 0, 4|s) Q(2, 4, 0|s) .
(s) 2 (s) 1 (s) = 3
2
12
4
(5.41)
0
One has (0) = 509
90 and (0) 1 = 6.1015 (see Eq. (A.36) in the appendix). This yields
s
509
Det0 11
45 1
=
(5.42)
0 e .
Det0 12
Finally, collecting the contributions of the negative, zero, and scalar modes computed in
(5.10), together with the classical term, we obtain
5 s

571
4
3
35
Det0 11 I
5 45
.
(5.43)
e
=
0.0047
(G)
exp
Z S =
0
Det0 12
G
312 6 24
0
348
To our knowledge, this formula has been obtained here for the first time, since in Refs. [13,
27] a closed expression for Z[S 4 ] was not achieved. In particular, the isometry factor 1
was not taken into account and the derivative of the -function was not computed.
6. Summary
Our last step is to use the expressions for Z[S 2 S 2 ] and Z[S 4 ] in (5.28) and (5.43) and
insert these into Eq. (2.14) to find the decay rate
r
1 =Z[S 2 S 2 ]
=
3
Z[S 4 ]

473
2
.
(6.1)
= 14.338 (G) (o ) 45 exp
G
This is the final result of our analysis. This formula gives the rate of semiclassical decay
of de Sitter space due to the spontaneous nucleation of black holes. This is the leading
mode of decay, since classically de Sitter space is stable [28]. The numerical coefficient
in the formula originates
from the fluctuation determinants evaluated in the -function
scheme. The factor comes from the heat bath temperature coefficient in (2.14) and
gives the correct dimension of an inverse time. The coefficient (G)2 arises due to
the combined effect of the background isometries. The power of o contains the effect
of rescalings, where we have passed again to the dimensionful renormalization parameter
o . Since quantum gravity is non-renormalizable, o remains undetermined, and we have
nothing to say about this problem. For numerical estimates it is reasonable to assume that
o G. The last factor in the formula is the classical term. The formula is obtained in
the one-loop approximation, which is good as long as the classical term is large compared
to the quantum corrections, that is for G 1. Under this condition the nucleation rate
is exponentially small. Notice that since the overall power of is positive, the quantum
corrections provide an additional suppression of the transition rate for small .
The formula gives the probability of black hole nucleation per unit proper time of a
freely falling observer in his Hubble region. The latter is the region enclosed insidethe
observers cosmological horizon. If a black hole is created, then it has the radius 1/
and fills the whole Hubble region. This does not mean that the whole space will be eaten
by a giant black hole, since de Sitter spacetime consists of many Hubble regions, whose
number grows as the universe expands. Some of these regions will contain a black hole
but most of them will be empty. The black holes are actually born in pairs, where the
two members of the pair are created at the opposite sides of the 3-space. The interesting
conclusion is that for G 1, when inflation is slow, the rate of black hole nucleation
is strongly suppressed, but the created black holes are large. This can be understood as a
consequence of the fact that the black holes are made of the energy contained inside the
Hubble region. As the size of the latter is large for small , the created black holes are also
large. On the other hand, if one is allowed to extrapolate the formula for G 1, when
inflation is fast, then the created black holes are small, but they are created in abundance.
349
One can see that for late times the number of black holes per unit physical volume will
be constant. Let us choose for de Sitter spacetime the global coordinates associated with
the freely falling observers:

r
3
2
2
2
d32.
(6.2)
ds = d + cosh
3
Here is the (dimensionful) proper time and d32 is the volume element of the
unit 3-sphere. The volume of the global hypersurface of constant is V () =
2
2 2 ( 3 )3/2 cosh3 ( /3 ) 4 ( 3 )3/2 exp( 3 ). The portion of contained inside
3 3/2
(for late .)
the future event horizon of any observer has the volume VH = 4
3 ()
This is the spatial Hubble volume. (This quantity slightly depends on the choice of the
hypersurface. Even though for any given observer one has = t, which is the time
associated with the observers coordinate system, one has 6= t , unless = t = 0,
in which case the spatial Hubble volume is VH = 2 ( 3 )3/2 ). As a result, the number of
Hubble volumes on the hypersurface is NH () = V ()/VH . (One has NH (0) = 2: the de
Sitter throat consists of two causally disconnected parts belonging to the Hubble regions
of two antipodal observers [47].) Multiplying NH () by gives the black hole nucleation
rate per ,

dNBH 3
=
exp 3 .
d
16
(6.3)
Integrating with respect to and dividing by V () yields the average volume density of
created black holes on ,
BH =
,
12
(6.4)
which does not depend on .

The subsequent real time evolution of these black holes is an interesting issue.
Presumably most of them will immediately evaporate, unless is very small and the black
holes are large, in which case however the nucleation rate is strongly suppressed. It was
argued in [9] that this process could dramatically change the global structure of de Sitter
space. For more information on this issue we refer to [9,10,18] and to the papers cited in
Ref. [9].
The following steps have been essential in our analysis. We have derived Eq. (2.14)
for the nucleation rate using the thermal properties of de Sitter space. For this we have
approximated the partition function for Euclidean quantum gravity with > 0 by the
semiclassical contributions of the S 4 and S 2 S 2 instantons, of which the first yields the
free energy F in the Hubble volume while the contribution of the second can be regarded
as a purely imaginary part of F . In a sense one can think of the created black holes as being
the bubbles of the new phase spontaneously created out of thermal fluctuations via quantum
tunneling. We have argued that these bubbles may have temperature different from that of
the heat bath, since they cannot thermalize via interactions with the whole reservoir and
only exchange energy inside the Hubble region.
350
To compute the one-loop contributions of the S 4 and S 2 S 2 instantons we have used

the standard FaddeevPopov approach to the path integral. We have worked with a oneparameter family of covariant background gauges and employed the Hodge decomposition
of the fluctuations with their subsequent spectral expansion. In our treatment of the
conformal modes we have followed the standard recipe of complex rotation, up to
several lowest lying modes for which a different prescription has been applied. In
order to integrate over zero modes of the FaddeevPopov operator arising due to the
background isometries, we have gone beyond the perturbation theory and showed that the
corresponding integration measure is the Haar measure on the isometry group. There are
no other zero modes in the problem for example, the standard rotational zero modes are
absent because rotations are isometries of the backgrounds under consideration.
We have explicitly determined the spectra of the fluctuation operators. For fluctuations
around the S 2 S 2 instanton the spectrum was obtained by directly solving the differential
equations, while in the S 4 case group theoretic methods have been applied, in which we
followed the approach of [27]. These spectra have been used in order to compute the
functional determinants within the -function regularization scheme, the corresponding
-functions being studied in detail in the appendix below. We have checked that our results
agree with the general formulas for the anomalous scaling behaviour. Finally, we have
obtained in (5.28), (5.43) the one-loop partition functions for fluctuations around the S 4 and
S 2 S 2 backgrounds. To our knowledge, in both cases such closed expressions have been
obtained for the first time. The last step has been to use the resulting partition functions in
order to calculate the nucleation rate . This describes a constant density of created black
holes per unit physical volume of the expanding 3-space.
After the work of Gross, Perry and Jaffe [30], our analysis presents the second example
of a complete one-loop computation on a non-trivial background. 3 One may hope that our
results can lend further support to the Euclidean approach to quantum gravity.
Note added in proof

We would like to thank Dima Vassilevich for bringing to our attention a number of
relatively recent papers considering one-loop Euclidean quantum gravity on S 4 . Although
in none of these papers a closed expression for the one-loop partition function Z[S 4 ] is
achieved, it is worth mentioning the work by Allen [50], by Polchinski [51], and by Taylor
and Veneziano [52]. We refer to the paper by Vassilevich [53] for more references. Not
all papers agree on the scaling behaviour of the partition function. The reason is that
some authors do not take into account the contribution of the 10 zero modes due to the
+ 329
background isometries, thereby obtaining Z[S 4 ] to be proportional to 0 45 instead of
571
45
[52]. However, since these zero modes are in the path integration measure, they
3 Note also that the analysis in [30] was not quite complete, since the spectrum is unknown and the -functions
have not been computed, even though the undetermined quantities can be absorbed into the renormalization
parameter. We also do not understand their treatment of the background isometries and that of the nonnormalizable deformations of the instanton.
351
do contribute to the anomalous scaling on equal footing with all other modes. In fact, the
example of flat space gauge theories [45] shows that the background symmetry zero modes,
when treated non-perturbatively as was done above, are of vital importance for obtaining
the correct running behaviour of the coupling constant. Our result for the scaling behaviour
agrees with that of Christensen and Duff [13] and with the general analysis of Fradkin and
Tseytlin [54].
Acknowledgements
We thank Michael Bordag for suggesting the idea to use the AbelPlan formula
in the analytic continuation of the -functions. M.S.V. would also like to thank Gary
Gibbons for discussing the role of the special conformal modes and Raphael Bousso
for interesting conversations. The work of M.S.V. was supported by the Deutsche
Forschungsgemeinschaft, grant Wi 777/4-2.
Appendix A. Calculation of -functions

In this appendix we shall study the -function
Z(k, |s) =
X
X
n=k m=k
(2n + 1) (2m + 1)
,
{(2n + 1)2 + (2m + 1)2 + }s
(A.1)
which is used in the main text for computing the one-loop fluctuation term on the
S 2 S 2 instanton background. Here is real while k is a positive integer such that
2(2k + 1)2 + > 0. It is assumed that <(s) is positive and large enough to ensure the
convergence of the series. Despite its apparent simplicity, the analysis of this -function is
lacking in the literature. This is probably due to the fact that the summation in (A.1) cannot
be extended to all integers and the standard Poisson resummation techniques do not apply.
For this reason we use other methods, which are unfortunately rather lengthy. However we
think that it is necessary to describe the basic steps, especially in view of other possible
applications of our results.
In what follows we shall perform the analytic continuation by finding the integral
representation for Z(k, |s) that is valid for any s. This will be used to compute the
d
Z(k, |s) at s = 0. As a first step, we shall consider the related
values of Z(k, |0) and ds
-function:
(k, |s) =
X
n=k
(2n + 1)
{(2n + 1)2 + }s
(A.2)
with (2k + 1)2 + > 0. The integral representation for this function will be useful. In
addition, we shall study the -function
Q(k, , c|s) =
X
(2j + 3)(j (j + 3) + c)
j =k
{j (j + 3) + }s
(A.3)
352
where k(k + 3) + > 0, and shall find its value and its s-derivative at s = 0. This function
is needed in the analysis of fluctuations around the S 4 instanton.
A.1. Computation of Z(k, |0) and (k, |0)
First we shall compute the values of these functions at s = 0 using the standard heat
kernel technique. These values determine the scaling properties of the system. Later we
shall rederive the same values by using the integral representations for Z(k, |s) and
(k, |s), and this will provide us with a good consistency check. For Q(k, , c|s) we
shall consider only the integral representation, since the values of Q(k, , c|0) have been
computed in [13].
A -function related to a second order elliptic operator with a positive spectrum can be
expressed as
1
(s) =
(s)
Z
t s1 (t) dt.
(A.4)
On compact spaces the heat kernel (t) vanishes exponentially fast for large t, while for
small t there is the asymptotic expansion
X
Cr t r ,
(A.5)
(t)
r
with r assuming in general both integer and half-integer values. It is not difficult to see that
(0) = C0 .
(A.6)
The problem therefore reduces to determining the asymptotic expansion of the heat kernel.
The heat kernels in our problem are given by
2
(A.7)
(k, |t) = (t) (k|t) et
for Z(k, |s) and

(k, |t) = (t) (k|t) et
(A.8)
for (k, |s), where

(t) =
(2n + 1) et (2n+1)
(A.9)
n=0
and
(k|t) =
k1
X
2
(2n + 1) et (2n+1) .
(A.10)
n=0
The only difficulty is to find the asymptotic expansions for small t for the function (t) in
(A.9). 4 (t) is a partition function for a two-dimensional rotator at temperature 1/t. We
4 We note that (t) cannot be expressed in terms of theta-functions in a simple way, and that the Poisson
resummation formula does not directly apply.
353
wish therefore to find its high-temperature expansion, and for this we shall construct the
integral representation for (t).
Let us consider the generating function
(t, ) =
et (2n+1)
2 +i(2n+1)
(A.11)
n=0
such that
(t, ).
(t, ) fulfills the differential equation

(t) = i lim
=
.
t
2
This has the special solution

( 0 )2
1
exp
(t,
) =
4t
4t
(A.12)
(A.13)
(A.14)
with the property (0,

) = ( 0 ), which allows us to represent the general solution
of (A.13) as
Z
(t,
0 ) (0, 0 ) d0 .
(A.15)
(t, ) =
The initial value (0, 0 ) is obtained directly from the definition (A.11):
(0, 0 ) =
ei0 (2n+1) =
n=0
i
,
2 sin 0
(A.16)
where we assume that 0 has a small positive imaginary part in order to ensure convergence
of the geometrical series. We can now insert this into (A.15) and the result into (A.12).
Introducing the new variable x = 02 /4 we obtain the sought for integral representation
1
(t) =
4t 3
ex/t
dx
.
sin(2 x)
(A.17)
Here we should remember that x has a small imaginary part, such that the integration is
actually performed along a contour parallel to the positive real axis and approaching it
from above.
It is now a straightforward task to find the asymptotic expansion of the integral in (A.17)
for small t, since the only non-trivial contribution comes from a small neighbourhood of
x = 0:

1
7 2
1
1+ t +
t + O t3 .
(A.18)
(t)
4t
3
30
Inserting this into (A.7) and (A.8) gives the asymptotic expansions for the heat kernels
(k, |t) and (k, |t), whose coefficients C0 determine the -functions at s = 0:
354
Z(k, |0) =
1 2
1
2
13
1

+ k 2 + 2k 4 k 2 +
,
32
24
2
3
360
(A.19)
(k, |0) =
1
1
k2.
12 4
(A.20)
and
To check these results we note that the definitions in (A.1) and (A.2) imply that
Z(k1 , |s) = Z(k2 , |s) + 2
kX
2 1

(2m + 1) k2 , + (2m + 1)2 s
m=k1
kX
2 1 kX
2 1
n=k1 m=k1
(2n + 1)(2m + 1)
,
{(2n + 1)2 + (2m + 1)2 + }s
(A.21)
with k2 > k1 . Setting here s = 0 we obtain a non-trivial relation for Z(k, |0) and (k, |0),
and this is fulfilled by the expressions in (A.19) and (A.20). Finally we use (A.19), (A.20)
to obtain the values used in the main text:
Z(2, 10|0) =
581
,
45
5
(2, 9|0) = ,
3
(2, 1|0) =
11
.
3
(A.22)
A.2. Computation of (k, |s), and Q(k, , c|s)

It is usually more difficult to determine the derivative of a -function at s = 0 than the
value of the function itself, since the knowledge of its behaviour in a neighbourhood of
s = 0 is required. We shall perform the analytic continuation of the -functions defined by
Eqs. (A.1), (A.2) and (A.3) to arbitrary values of s with the use of the relation sometimes
called
A.2.1. The AbelPlan formula
This can be derived using the obvious relation
Z
X
f (z)
dz,
f (n) =
e2iz 1
n=k
(A.23)
where the contour C encompasses the part of the real axis with Re(z) > k (see Fig. 4) and
f (z) is analytic for Re(z) > k. The idea is to split C into three parts, C1 + C2 + C3 , as
shown in Fig. 4. For the first part, C1 , the integral can be written as
Z
C1

Z
Z
f (z)
1
1 f (z) dz = f (t) dt +
dz,
1 e2iz
1 e2iz
k
(A.24)
C1
where in the integral over C1 on the right the contour is then rotated to the position C 1
as shown in Fig. 4. Such a rotation is possible if only f (z) tends to zero fast enough for
Re(z) > k and |z| .
355
Fig. 4. Left: Starting from the contour C1 + C2 + C3 and rotating we arrive at C 1 + C2 + C 3 . Right:
the same when a branching point at z = z+ ( ) is present. The contour C1 will then wrap around the
The point z = z+ ( ) is in the region of interest
cut leading to the additional contribution due to C.
for > .
The integral over the second portion of the contour, C2 , is equal to 12 f (k), while in the
integral over C3 the contour is rotated to the position C 3 as shown in Fig. 4. As a result,
we arrive at the AbelPlan formula
Z
Z
X
1
f (k + it) f (k it)
dt.
(A.25)
f (n) = f (k) + f (t) dt + i
2
e2t 1
n=k
This formula can be used for analytic continuation of -functions, in which case f (t)
depends also on s, f = f (t, s). The analytic continuation to small values of s is performed
in the first integral on the right in (A.25). This usually converges only for <(s) large and
positive, but can often be computed in a closed form, and then one can continue the result
to arbitrary s. The second integral on the right in (A.25) usually cannot be computed in
a closed form, but it converges for any s. Let us first apply the AbelPlan formula to the
-functions in (A.2) and (A.3).
A.2.2. Analytic continuation of (k, |s)
Applying (A.25) to the series for (k, |s) in (A.2), we have
f (z) =
2z + 1
,
{(2z + 1)2 + }s
(A.26)
which is analytic for <(z) > 1/2 and decays fast enough for |z| provided that <(s)
is large enough. As a result, we can use the AbelPlan formula, which gives

1
1
1
1
+
(k, |s) = k +
2 {(2k + 1)2 + }s 4(s 1) {(2k + 1)2 + }s1

Z
2k + 1 2it
2k + 1 + 2it
idt
+
.
e2t 1 {(2k + 1 + 2it)2 + }s {(2k + 1 2it)2 + }s
0
(A.27)
356
This representation is finite for all s, apart from s = 1, where the pole is located. The
remaining integral here converges uniformly for |s| < , which allows us to differentiate
with respect to s. If we set s = 0, then the integral can be easily computed. We find
1
14 k 2 , and this agrees with the value obtained above in (A.20).
(k, |0) = 12
Similarly, we can differentiate (A.27) with respect to s and then set s = 0. This gives

1
1
ln W
0 (k, |0) = W (ln W 1) k +
4
2
Z

dt
t ln A + (2k + 1) ,
(A.28)
+2
2t
e 1
0
where W
= (2k + 1)2
A = W 4t 2
+ and
2
+ 16(2k + 1)2 t 2 ,
= arctan
4(2k + 1)t
.
W 4t 2
(A.29)
For any k and the integral in (A.28) is convergent and can be evaluated numerically.
Notice that 0 (k, |0) is not needed in the main body of the paper, and for this reason we
do not quote the actual number here.
A.2.3. Analytic continuation of Q(k, , c|s)
The procedure is exactly the same as above. Denoting
f (z) =
(2z + 3)(z(z + 3) + c)
{z(z + 3) + }s
(A.30)
the direct application of the AbelPlan formula (A.25) yields

1
3
k(k + 3) + c W s +
W 2s
Q(k, , c|s) = k +
2
s 2
Z

c 1s
dt
W
i f (k + it) f (k it) ,
+
+
2t
s1
e 1
(A.31)
where W = k(k + 3) + . Setting s = 0 the integral can be easily computed leading to

1 2
1
(A.32)
k
Q(k, , c|0) = k 4 2k 3 c +
2
2

11
4
1
c .
+ (3 2c) k + 2 +
2
3
15
Next, differentiating (A.30) with respect to s and setting s = 0 gives

3
0
(W + c ) ln W
Q (k, , c|0) = k +
2

1
1
ln W
W 2 + (c )(ln W 1) W + G.
+
2
2
(A.33)
357
Here
Z
G=
dt
t (6k(k + 3) + 2c + 9 2t 2 ) ln A
1

+ 4k 3 + 18k 2 + (18 + 4c) k + 6c 6 (2k + 3) t 2
e2t
0
(A.34)
with
(2k + 3) t
.
W t2
Evaluating the integral numerically, the two values used in the main text are
A = t 4 + (2W 4 + 9) t 2 + W 2 ,
Q0 (2, 0, 4|0) = 3.72344,
= arctan
(A.35)
Q0 (2, 4, 0|0) = 6.65246.
(A.36)
5
Q(2, 0, 4|s) 14 Q(2, 4, 0|s)) used in Eq. (5.41)
Finally, for the function (s) = 3s ( 12
in the main text one obtains with the help of (A.32) and (A.36)
(0) =
509
,
90
0 (0) 1 = 6.10158.
(A.37)
A.3. Computation of Z(k, |s)

Let us not turn to our main task the evaluation of the double-sum function Z(k, |s),
which has been defined for large values of <(s) by (A.1). The idea is to express it in terms
of the single-sum function (k, |s).
It follows from the definitions (A.1) and (A.2) that
Z(k, |s) =

(2n + 1) k, + (2n + 1)2 s .
(A.38)
n=k
Here we can use the integral representation (A.27) for (k, + (2n + 1)2|s). Indeed, if is
real and (2k + 1)2 + > 0 then the same remains true upon replacement + (2n + 1)2,
and the formula (A.27) therefore applies. Now, replacing in (A.27) by + (2n + 1)2 and
assuming for a moment that <(s) is large and positive, the integral in (A.27) converges
uniformly with respect to n for n . This allows us, upon insertion of (A.27) into
(A.38), to interchange the orders of summation and integration. The result then can be
extended to any s by analytic continuation. This gives

2n + 1
1 X
Z(k, |s) = k +
2
2
{(2k + 1) + + (2n + 1)2 }s
+
n=k
1
4(s 1)
n=k
{(2k + 1)2
2n + 1
+ + (2n + 1)2 }s1
X
id
2n + 1
(2k + 1 + 2i )
2
e
1
{(2k + 1 + 2i )2 + + (2n + 1)2 }s
n=k
0
!
X
2n + 1
(2k + 1 2i )
.
(A.39)
{(2k + 1 2i )2 + + (2n + 1)2 }s
+
n=k
358
One can see that all sums here are exactly the same as in the definition of (k, |s) in (A.2)
up to the replacements + (2k + 1)2 and ( ) + (2k + 1 + 2i )2 . Since
the definition in (A.2) makes sense for arbitrary values of (the series always converges
for <(s) big enough), we can express the sums in (A.39) in terms of (k, + (2k + 1)2 |s)
and (k, ( )|s). This leads to the the following formula:

1
1
k, + (2k + 1)2 s +
k, + (2k + 1)2 s 1
Z(k, |s) = k +
2
4(s 1)
Z

i d
F ( ) F ( ) ,
(A.40)
+
e2 1
0
with F ( ) = (2k + 1 + 2i ) (k, ( )|s). In this formula the first two terms on the right are
determined by the integral representation (A.27) for arbitrary s. We are left with computing
the remaining integral over . The problem here is that the parameter ( ) is complex,
and for this reason we cannot directly apply the integral representation (A.27) to compute
(k, ( )|s).
Let us recall that the formula (A.27) was derived assuming that the function f (z) in
Eq. (A.26) had no poles for <(z) > k. This allowed us to rotate the integration contour as
shown in the left part of Fig. 4 without intersecting singularities. Let us now replace by
( ) + (2k + 1 + 2i )2 . As a result, f (z) in Eq. (A.26) is replaced by
2z + 1
2z + 1
=
,
(A.41)
2
s
{(2z + 1) + ( )}
{4(z z+ ( ))(z z ( ))}s
with z ( ) = 12 (1 i ( )). For = 0 one has <(z (0)) = 12 . As increases, the

point z+ ( ) moves to the right in the complex plane (while z ( ) moves to the left), but
as long as <(z+ ( )) < k one can still use the formula (A.27). However, for large enough
values of the pole at z = z+ ( ) enters the region of interest, that is the part of the complex
plane with <(z) > k, and we can no longer use the formula (A.27).
To tackle the problem we notice that the pole of f (z) at z = z+ ( ) is a branching point,
and one can choose the cut in the complex plane as shown in the right part of Fig. 4. We then
repeat the steps leading to the AbelPlan formula and the additional problem we encounter
is the following: when we rotate the integration contour as we did before, the contour will
wrap around the cut as shown in Fig. 4. The resulting contour will then consist of two
disconnected pieces. The first piece will be the same as the old contour C 1 + C2 + C 3 (see
Fig. 4). The second piece is the contour C wrapping around the cut. Integrating around
such a combined contour, the result will consist of two parts,
Z
f (z)
dz.
(A.42)
(k, ( )|s) = old (k, ( )|s) + ( )
2iz
e
1
f (z) =
Here the first term on the right, old (k, ( )|s), is the function given by the previous
expression in (A.27) with being replaced by ( ). The second term, with f (z) given
by (A.41) and the contour C as shown in the right part of Fig. 4, is the contribution of
the cut. The step function ( ) reflects the fact that the cut contributes only for large
359
enough when the pole enters the region <(z) > k. Here (x) = 0 for x < 0 and (x) = 1
for x > 0, and <(z+ ( )) = k.
The representation (A.42) applies for all values of s and for any > 0. Similarly, one
can obtain (k, ( )|s) (the cut then resides in the upper half-plane). As a result, the
function F ( ) F ( ) in the integrand in Eq. (A.40) is defined for any > 0, and the
integral converges due to the damping exponential factor. This finally gives Z(k, |s) for
any s.
Let us first check our result by computing Z(k, |0). For s = 0 the function f (z) has no
poles and the contribution of the cut vanishes. The remaining integrals then can be easily
computed, which gives for Z(k, |0) exactly the same expression as in Eq. (A.19).
Let us now compute Z 0 (k, |0). Since all integrals in (A.40), (A.41) converge uniformly
with respect to s (at least for |s| < ), we can differentiate the integrands with respect to
s and then set s = 0. The result can be represented in the following form:
Z
Z (k, |0) = H + 2
0
Z
+
0
dt
G(t)
1
e2t
d
e2 1
Z
0
dt
W(, t) + S.
e2t 1
(A.43)
Here

2
1
4 + 4 2 ln 2(2k + 1)2 +
k +k+
H = 2k 4 + 2
2
32

1
3
1
3 2 4 20 .
+ 3k 4 + 2k 3 + ( 2)k 2 + ( 6)k +
4
4
64
In addition,

t
G(t) = 4t 2 16k 2 12k 2 ln P
2

+ (2k + 1) 6t 2 2k (2k + 1) /2 ,
(A.44)
(A.45)
where we have used

P = 2 + 4(2k + 1)2 8t 2 + 4(2k + 1)4 + 16t 4 ,

= Phase (2k + 1)2 + /2 2t 2 + i 2(2k + 1)t ,
and < Phase[x + iy] 6 is the phase of the complex number. Next,

W(, t) = (2k + 1)2 4 t ln Q + 4(t )(2k + 1)

(t, ) (t, )
with

+ ( + 2)2 8 t 2 + 2 + 128 k 2 + k + 1/4 t

+ 16 k(k + 1) + 32 k(k + 1) 2k 2 + 2k + 1 ,

= Phase (2k + 1)2 + /2 2 t 2 + 2 + i 2(2k + 1)(t ) .
Q = 16 t 2 + 2
(A.46)
(A.47)
2
(A.48)
360
Finally, the contribution of the cut is

Z
S = 4
d
2
e
1
Z
=
0
(2k + 1 2i )(2z( ) + 1 + 2it)

dt,
e2(t iz( )) 1
(A.49)
p
where z( ) = 12 + 4 2 (2k + 1)2 + 4i(2k + 1) , and <(z( )) = k.
We now use the formulas above in order to evaluate Z 0 (2, 10|0), which value is
needed in the main text. Setting k = 2 and = 10 we obtain for the first term on the
right in (A.43) H = 1.9445. The second term, containing the integral over t, is evaluated
numerically to give 19.9469. The numerical value of the term containing the double
integral is 0.1294. As for the last term, S, it is exponentially small and is of the order
(A.49), the value of S is suppressed by the
of 1012 . This is because, as one can see from
factor of exp{2( + =(z( ))} = exp{4 5}.
Summing everything up, we obtain
Z 0 (2, 10|0) = 18.3118.
(A.50)
References
[1] S. Adler, Massless, Euclidean quantum electrodynamics on the 5-dimensional unit hypersphere,
Phys. Rev. 6 (1972) 34453461.
[2] S. Adler, Massless electrodynamics on the 5-dimensional unit hypersphere, an amplitudeintegral formulation, Phys. Rev. 8 (1973) 24002418.
[3] I. Affleck, Quantum-statistical metastability, Phys. Rev. Lett. 46 (1981) 388391.
[4] O. Aharony, S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory
and gravity, Phys. Rep. 323 (2000) 183386, hep-th/9905111.
[5] B. Allen, Euclidean Schwarzschild negative mode, Phys. Rev. D 30 (1984) 11531157.
[6] J. Ambjorn, J. Jurkiewicz, R. Loll, A non-perturbative Lorentzian path integral for gravity, hepth/0002050.
[7] J. Ambjorn, J. Jurkiewicz, R. Loll, Lorentzian and Euclidean quantum gravity analytical and
numerical results, hep-th/0001124.
[8] A.O. Barut, R. Raczka, Theory of Group Representations and Applications, PWN Polish
Scientific Publishers, Warszawa, 1977.
[9] R. Bousso, Proliferation of de Sitter space, Phys. Rev. D 58 (1998) 083511, hep-th/9805081.
[10] R. Bousso, S.W. Hawking, Pair creation of black holes during inflation, Phys. Rev. D 54 (1996)
63126322, hep-th/9606052.
[11] R.R. Caldwell, A. Chamblin, G.W. Gibbons, Pair creation of black holes by domain walls, Phys.
Rev. D 53 (1996) 71037114, hep-th/9602126.
[12] C.G. Callan, S. Coleman, Fate of the false vacuum. II. First quantum corrections, Phys. Rev. 16
(1977) 17621768.
[13] S.M. Christensen, M.J. Duff, Quantizing gravity with cosmological constant, Nucl. Phys. B 170
(1980) 480506.
[14] F. Dowker, J.P. Gauntlett, G.W. Gibbons, G.T. Horowitz, Nucleation of p-branes and fundamental strings, Phys. Rev. D 53 (1996) 71157128, hep-th/9512154.
[15] F. Dowker, J.P. Gauntlett, S.B. Giddings, G.T. Horowitz, On pair creation of extremal black
holes and KaluzaKlein monopoles, Phys. Rev. D 50 (1994) 26622679, hep-th/9312172.
[16] F. Dowker, J.P. Gauntlett, D.A. Kastor, J. Traschen, Pair creation of dilaton black holes, Phys.
Rev. D 49 (1994) 29092917, hep-th/9309075.
361
[17] T. Eguchi, P.B. Gilkey, A.J. Hanson, Gravitation, gauge theory and differential geometry, Phys.
Rep. 66 (1980) 213393.
[18] E. Elizalde, S. Nojiri, S.D. Odintsov, Possible quantum instability of primordial black holes,
Phys. Rev. D 59 (1999) 061501, hep-th/9901026.
[19] D. Garfinkle, S.B. Giddings, A. Strominger, Entropy in black hole pair production, Phys. Rev.
D 49 (1994) 958965, hep-th/9306023.
[20] G.W. Gibbons, Gravitational instantons, a survey, Lecture Notes in Physics 116 (1979).
[21] G.W. Gibbons, Quantum flux tubes in EinsteinMaxwell theory and noncompact internal
spaces, in: A. Jadczyk (Ed.), Proceedings of the 22nd Karpacz Winter School of Theoretical
Physics, Fields and Geometry 1986, World Scientific, Singapore, 1986.
[22] G.W. Gibbons, S.W. Hawking, Action integrals and partition functions in quantum gravity,
Phys. Rev. D 15 (1977) 27522756.
[23] G.W. Gibbons, S.W. Hawking, Cosmological event horizons, thermodynamics, and particle
creation, Phys. Rev. D 15 (1977) 27382751.
[24] G.W. Gibbons, S.W. Hawking, Euclidean Quantum Gravity, World Scientific, 1993.
[25] G.W. Gibbons, S.W. Hawking, M.J. Perry, Path integrals and the indefiniteness of the
gravitational action, Nucl. Phys. B 138 (1978) 141149.
[26] G.W. Gibbons, M.J. Perry, CP2 as a gravitational instanton, Commun. Math. Phys. 61 (1978)
239248.
[27] G.W. Gibbons, M.J. Perry, Quantizing gravitational instantons, Nucl. Phys. B 146 (1978) 90
108.
[28] P. Ginsparg, M.J. Perry, Semiclassical perdurance of de Sitter space, Nucl. Phys. B 222 (1983)
245268.
[29] J.N. Goldberg, A.J. Macfarlane, E.T. Newnam, F. Rohrlich, E.C.G. Sudarshan, Spin-s spherical
harmonics and , J. Math. Phys. 8 (1967) 21552161.
[30] D.J. Gross, M.J. Perry, L.G. Yaffe, Instability of flat space at finite temperature, Phys. Rev. D 25
(1982) 330355.
[31] M. Hamermesh, Group theory and its application to physical problems, Dover Publications,
New York, 1962.
[32] S.W. Hawking, The path-integral approach to quantum gravity, in: S.W. Hawking, W. Israel
(Eds.), General Relativity, an Einstein centenary survey, Cambridge University Press, 1979,
p. 746.
[33] S.W. Hawking, Virtual black holes, Phys. Rev. D 53 (1996) 30993107, hep-th/9510029.
[34] S.W. Hawking, G.F.R. Ellis, The Large Scale Structure of Space-Time, Cambridge University
Press, 1973.
[35] S.W. Hawking, G.T. Horowitz, S.F. Ross, Entropy, area, and black hole pairs, Phys. Rev. D 51
(1994) 43024314, hep-th/9409013.
[36] S.W. Hawking, S.F. Ross, Duality between electric and magnetic black holes, Phys. Rev. D 52
(1995) 58655876, hep-th/9504019.
[37] S.W. Hawking, S.F. Ross, Pair creation of black holes on cosmic strings, Phys. Rev. Lett. 75
(1995) 33823385, hep-th/9506020.
[38] J.I. Kapusta, Nucleation rate for black holes, Phys. Rev. D 30 (1984) 831832.
[39] J.S. Langer, Theory of the condensation point, Ann. Phys. 41 (1967) 108157.
[40] J.S. Langer, Statistical theory of the decay of metastable states, Ann. Phys. 54 (1969) 258275.
[41] R.B. Mann, S.F. Ross, Cosmological production of charged black holes, Phys. Rev. D 52 (1995)
22542265, hep-th/9504015.
[42] P.O. Mazur, E. Mottola, The gravitational measure, solution of the conformal factor problem
and stability of the ground state of quantum gravity, Nucl. Phys. B 341 (1990) 187212.
[43] F. Mellor, I. Moss, Black holes and gravitational instantons, Class. Quant. Grav. 6 (1989) 1379
1385.
362
[44] H. Osborn, Semiclassical functional integrals for self-dual gauge fields, Ann. Phys. 135 (1981)
373415.
[45] M. Schifman, Instantons in Gauge Theories, World Scientific, 1994.
[46] K. Schleich, Conformal rotation in perturbative gravity, Phys. Rev. 36 (1987) 23422363.
[47] E. Schrdinger, Expanding Universes, Cambridge University Press, 1956.
[48] R.F. Streater, A.S. Wightman, PCT, Spin and Statistics, and All That, Benjamin, New York,
1964.
[49] K. Yano, Integral Formulae in Riemannian Geometry, Interscience, 1970.
[50] B. Allen, Graviton propagator in de Sitter space, Phys. Rev. D 34 (1986) 36703675.
[51] J. Polchinski, The phase of the sum over spheres, Phys. Lett. B 219 (1989) 251257.
[52] T.R. Taylor, G. Veneziano, Quantum gravity at large distances and the cosmological constant,
Nucl. Phys. B 345 (1990) 210230.
[53] D.V. Vassilevich, One-loop quantum gravity on de Sitter space, Int. J. Mod. Phys. A 6 (1993)
16371652.
[54] E.S. Fradkin, A.A. Tseytlin, On the new definition of off-shell effective action, Nucl. Phys.
B 234 (1984) 509527.

Special properties of five-dimensional

BPS rotating black holes
C.A.R. Herdeiro
Department of Applied Mathematics and Theoretical Physics, Centre for Mathematical Sciences,
University of Cambridge, Wilberforce Road, Cambridge CB3 0WA, UK
Abstract
Supersymmetric, rotating, asymptotically flat black holes with a regular horizon are rare
configurations in string theory. One example is known in five spacetime dimensions, within the
toroidal compactification of type IIB string theory. The existence of such special solution is allowed
by the existence of a ChernSimons coupling in the supergravity theory and by the possibility
of imposing a self duality condition on the rotation 2-form. We further exemplify the use of
such duality condition by finding a new Brinkmann wave solution in D = 6 simple gravity,
possessing Killing spinors. We then explore three peculiar features of the aforementioned black
holes: (1) Oxidising to D = 10 the five-dimensional configuration may be interpreted as a system
of D1D5 branes with a Brinkmann wave propagating along their worldvolume. Unlike its fivedimensional KaluzaKlein compactification, the universal covering space of this manifold has no
causality violations. In other words, causal anomalies can be solved in higher dimensions. From
the dual SCFT viewpoint, the causality bound for the compactified spacetime arises as the unitarity
bound; (2) The vanishing of the scattering cross section for uncharged scalars and sufficiently high
angular momentum of the background is shown still to hold at the level of charged interactions. The
same is verified when a non-minimal coupling to the geometry is used. Therefore, the repulson
behaviour previously found is universal for non accelerated observers; (3) The solutions are shown
to have a non-standard gyromagnetic ratio of g = 3. In contrast, the superpartners of a static, BPS,
five-dimensional black hole have g = 1. At the semi-classical level, we find that a Dirac fermion
propagating in the rotating hole background has g = 2 1, depending on the spinor direction of the
fermion being parallel to Killing or anti-Killing spinors. 2000 Elsevier Science B.V. All rights
reserved.
PACS: 04.65.+e; 04.70.-s; 04.70.Dy; 04.50.+h
Keywords: Rotating black holes; Supersymmetry; Causality
car26@damtp.cam.ac.uk
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 3 5 - 7
364
C.A.R. Herdeiro / Nuclear Physics B 582 (2000) 363392
1. Introduction
One of the most appealing features of string theory is that it provides a finite quantum
theory of gravity which reduces to the well tested general relativity in the appropriate
limit. Quantum gravitational effects become particularly important in at least two physical
situations the early universe and black holes physics which are therefore laboratories
to test the novel stringy physics. But whereas some significative progress has been done in
understanding microscopics of black holes, stringy cosmology is still in a very early stage.
One important reason for this distinction is supersymmetry.
Supersymmetry has allowed quantitative mapping between classical gravitational
configurations and quantum states of strings. But one pays a price: preservation of
supersymmetry requires the initial classical configuration to be dynamically very simple. In
particular, it must contain a causal Killing vector field, in order to possess a Killing spinor.
Whereas this is not an obstacle in understanding some features of black holes, since we
may study extreme ReissnerNordstrm (RN) type solutions, it becomes a major obstacle
in studying cosmology, since there seems to be no easy way to reconcile supersymmetry
with an expanding universe.
Even within the domain of black holes one has the natural desire of understanding more
complex configurations than the aforementioned ones, in particular to include rotation,
which we expect will be present in the black holes we find in nature. Rotation makes the
dynamical properties of a black hole spacetime richer, and therefore less compatible with
supersymmetry. In fact, regular (on and outside a horizon), rotating, supersymmetric black
hole solutions are rare configurations in string theory. Let us consider the heterotic string
theory, compactified on a T 10D torus to 9 > D > 4 spacetime dimensions. For 9 > D > 6,
the low energy effective field theory will contain (36 2D) one-form gauge potentials,
giving rise to electric charges Qa , a = 1 . . . 36 2D. Of course, these charges have no
absolute meaning in themselves; they can be mixed together by the O(10 D, 26 D)
symmetry of the low energy effective action. In particular, the subgroup [SO(10 D)
SO(26 D)]/[SO(9 D) SO(25 D)] has dimension (34 2D). Therefore, a 9 >
D > 6 dimensional black hole solution classified by mass, the maximal number of angular
momentum parameters, li , i = 1 . . . [(D 1)/2] = rank of SO(D 1), and two electric
charges can be used to generate a solution with all (36 2D) charges, by the use of the
above subgroup. The solution with two charges is called the generating solution. Since the
O(10 D, 26 D) symmetry does not act on the D-dimensional metric, the geometry for
the most general electrically charged rotating black hole solution can be seen at the level
of the generating solution.
In D = 5, the (reduced) NeveuSchwarz field can be Hodge dualized to couple to
0-branes. It carries a magnetic charge for a black hole. Hence, the generating solution
will have 3 charges, and the most general electrically charged black hole 26 electric and 1
magnetic charge. In D = 4, the one form fields also carry magnetic charges for black holes.
Hence the generating solution has 4 charges and the most general black hole solution 56
365
charges, 28 electric and 28 magnetic. 1

The geometry of these generating solutions in the BPS limit has the following behaviour:
(i) In D = 4 [1]; the spacetime with non-vanishing angular momentum yields a naked
(timelike) singularity. Putting all charges equal it reduces to the extreme (i.e., Q2 =
M 2 and arbitrary J ) KerrNewman (KN) spacetime of EinsteinMaxwell theory;
(ii) In D = 5; a regular rotating black hole spacetime [2]. Putting all charges equal it
reduces to the BreckenridgeMyersPeetVafa (BMPV) spacetime [4], of N = 2,
D = 5 supergravity;
(iii) In D > 6; a naked singularity. When only one li is non-zero it reduces to a better
behaved null singularity [3].
Regularity, therefore, distinguishes five-dimensional, BPS rotating black holes. But why?
In Section 2 we perform a comparative analysis between the extreme KN black
hole and the BMPV black hole. In particular, we emphasize how a Hodge self duality
condition, which can be implemented in a four-dimensional transverse space to the world
volume, allows for the harmonic function in the metric and the Sagnac connection [12]
to be independent. This is essential for the properties of the solution. We then give a
simple example of using such duality condition. Specifically we find a supersymmetric
Brinkmann wave type solution to D = 6 pure gravity and mention the corresponding fivedimensional KaluzaKlein rotating black hole.
In Section 3 we exhibit the D = 10 description of five-dimensional, BPS, rotating
black holes. We also analyse the interesting causal structure of the D = 5 black holes
from the ten-dimensional perspective, finding that the unavoidable closed timelike curves
(CTCs) of the five-dimensional solution acquire a trivial character in ten dimensions (in
the language of [10]). As far as we are aware this is the first example of resolving causal
anomalies in higher dimensions. In order to give a complete description of the causal
structure of this black holes, we review the string theory dual picture of the compactified
spacetime in terms of an N = 4 super conformal field theory (SCFT), which has allowed
the microscopic computation of the entropy [4]. The point we wish to stress is the loss
of unitarity in this theory when the spacetime undergoes a loss of causality. This unitarity
bound can be seen by looking at general unitarity requirements for the conformal weights
of SCFT operators. Alternatively one may examine the representations of an N = 2
SCFT [35] and the condition for these to be unitary, that arises from analysing the sign
of the determinant of the Verma module.
Spacetimes where causality is violated may have odd effects related to geodesic motion.
One such example is the totally imprisoned incompleteness found for the Lorentzian
Taub-NUT metric [11]. The phenomenon consists in the existence of a family of null
geodesics which are imprisoned inside a compact region of spacetime whose boundary they
reach within finite affine length. Such geodesics are therefore incomplete. However, they
meet neither an s.p. (divergence of a curvature invariant) nor a p.p. singularity (divergence
of the components of the Riemann tensor in a parallelly propagated frame).
1 The four-dimensional case is actually more complex and the generating solution should have five charges;
a case which has not been dealt with in the literature [1].
366
The hypersurface from which they cannot be extended separates a causally well behaved
spacetime region (the region initially found by Taub) from the outer reaches where closed
timelike curves exist. We should remark, however, that there exists another family of null
geodesics that passes through these surfaces.
In Section 4.1 we explore an interesting effect arising for the BMPV spacetime whenever
CTCs arise outside the horizon the over-rotating case firstly pointed out in [12]. The
phenomenon a repulson effect is shown still to be present when we consider test
particles with charge, naturally following charged geodesics. For scalar waves we also try
a non-minimal coupling to the geometry, in the KleinGordon equation, by including the
Ricci scalar. Separation of variables can still be achieved, and a radial equation obtained.
We conclude the universality of the effect for non accelerated observers. As for the TaubNUT case we could find neither p.p. nor s.p. singularities, and nothing special seems to
occur for an accelerated observer trying to overcome this natural obstacle. We also note that
this repulson behaviour is distinct from singular repulsons recently studied and resolved in
the context of AdS/CFT [36].
The uniqueness or no-hair theorems for black holes tell us that a small set of
independent quantities completely determines black hole spacetimes. Within Einstein
Maxwell theory such set includes solely ADM mass (M), angular momentum (J i ), electric
(Q) and magnetic (P ) charges. In spite of such restriction, black holes may have other
multipole moments, albeit not independent. The magnetic dipole moment (i ) is the most
natural one, since we expect any electrically charged rotating object to have it. In D = 4 it
is related to the previous quantities by i QJ i /(2M), therefore defining a constant of
proportionality which is called the gyromagnetic ratio.
In ordinary electrodynamical systems, it follows from the definitions that g = 1 for any
rotating homogeneous distribution of charges. But for four-dimensional KerrNewman
black holes, g = 2 [10] in analogy with the quantum mechanical value for the electron
(up to loop quantum corrections). The same value for g is found for other heterotic black
holes in four dimensions [13], 2 and some p-brane solutions in other dimensions, namely
the membrane in D = 11 [15]. However, this is by no means universal. KaluzaKlein
black holes have a gyromagnetic ratio that depends on the several metric parameters [16].
In the limit of large electric charge and vanishing magnetic charge, g approaches one. This
is the natural value for massive KaluzaKlein modes in five-dimensional KaluzaKlein
theory [17]. The classical value arises because the charge is orbital motion in the compact
direction. Another KaluzaKlein black hole is the Dirichlet 0-brane supergravity solution
in D = 10. It is known to have g = 1 [18]. In fact, it is well known that the identification
of the spectrum of IIA D0-branes with the KaluzaKlein modes of the 11D graviton was a
major guideline in the discovery of M-theory [19].
In finding the value of g for the M2 and D0 branes, the superpartners technique was
used. This method is usually applied to static BPS solutions, yielding a new solution
2 The charges for the KerrNewmanSen [13] black hole and for the electron have different origins, however,
in heterotic string theory. Therefore there is a distinction in their gyromagnetic ratios [14].
367
with fermionic hair. 3 The interpretation of these superpartners is ambiguous due to the
existence of bilinears of odd Grassmann numbers in the bosonic fields. Nevertheless, the
method is useful in reading quantities like the gyromagnetic ratio. The technique was first
used in [20] and applied to the extreme RN background. The superpartners were found to
have g = 2.
In Section 4.2, we compute the gyromagnetic ratio for both the BMPV background and
its superpartners showing they are different. This is in contrast with the four-dimensional
case, where superpartners of the extreme RN background have the same g as the Kerr
Newman. We then analyse the problem of a Dirac fermion in the BMPV spacetime and
show it is isomorphic to the problem of a non-minimally coupled Dirac field in flat space.
This coupling alters the usual value for g, creating a matching between the behaviour of
the elementary particle and either the BMPV black hole or the superpartner of the static
D = 5 RN spacetime.
We close with a discussion.
2. BMPV versus extreme KN and the self-duality condition
In D = 4, 5 the simplest supergravity theories containing a Maxwell field are the N = 2
theories. The bosonic truncations of the general theories are, respectively, the Einstein
Maxwell theory and EinsteinMaxwellChernSimons theory, with actions given by

Z
1
1 2
(4)
4
d x g R F ,
S =
16G4
4

Z
1
2
(5)
5
2
(2.1)
d x g R F A F F .
S =
16G5
3 3
Finding supersymmetric solutions to these theories involves solving the gravitino
variation equations which are, respectively,

i
1
D + ea bc a 4eb c Fbc = 0,
(2.2)
D Fab ab = 0,
4
4 3
where we have denoted the covariant derivative acting on spinors by D = d +
1/4wab ab . For the four-dimensional theory all solutions possessing Killing spinors were
found by Tod [39]. They split into two families, according to the existence of a null or
a timelike Killing vector field. The former type corresponds to a family of gravitational
waves, carrying electromagnetic fields, with the standard plane-fronted waves with parallel
rays (pp-waves) arising as a special case. The latter case corresponds to the IsraelWilson
Perjs (IWP) metrics [40,41]. The IWP solutions can be written in the form:
2
ds 2 = |H |2 dt + wi dx i + |H |2 ij dx i dx j ,

1
F = i R H 1 dt dx i + |H |2 kl ij k l I H 1 + 2wi j R H 1 dx i dx j .
2
(2.3)
3 The new configuration is only a solution to the full supergravity equations of motion if the method is carried
out to all orders. Otherwise it solves the equations of motion only to some order in fermionic parameters.
368
The Latin indices denote spatial coordinates and take values 1, 2, 3. We are using cartesian
coordinates for the spatial metric. The complex function H (x i ) is required to solve
Laplaces equation in E3 , and |H | represents its modulus. It completely determines the
solution since the rotation vector, wi , is determined by
w = i(H H H H ),
(2.4)
where the curl is taken for E3 and * denotes complex conjugation. The symbols R, I
denote, respectively, real and imaginary part of a complex quantity.
There are two distinct families within the IWP solutions, according to how one
complexifies H (x i ) [42]. Complexifying the constants in the harmonic function leads to
a multi-object generalisation of the charged Taub-NUT solution, whereas complexifying
the coordinates in a specific way leads to the multi-object generalisation of the extreme
(i.e., Q2 = M 2 , where M, Q are the ADM mass and charge of the solution) KerrNewman
(KN) spacetime. The one object limit of the latter the usual KN solution is obtained
by choosing
M
,
(2.5)
H =1+ p
2
2
x + y + (z ia)2
and contact with the standard form in BoyerLinquist coordinates is made by changing
from cartesian to oblate spheroidal coordinates, (r, , ), with transformations
p
z = (r M) cos ,
(2.6)
x + iy = (r M)2 + a 2 sin ei ,
where the constant a will be the angular momentum parameter. 4
For the five-dimensional case, not all supersymmetric solutions are known, but at least
the equivalent to pp-waves and to the extreme KerrNewman are. The latter is known as
the BMPV solution and given by the fields

2
ds 2 = H 2 dt + ai dx i + H ij dx i dx j ,

3
i H 1 dx i dt + i H 1 aj dx i dx j .
(2.8)
F=
2
The Latin indices run now from 1 to 4. The solution is determined by two independent
quantities. The real function H (x i ) is required to be harmonic on E4 , and the E4 one
form a is required to have a Hodge self-dual field strength. In hyperspherical coordinates
(, , 1 , 2 ), obtained from cartesian coordinates as
x 1 + ix 2 = cos ei1 ,
x 3 + ix 4 = sin ei2 ,
(2.9)
4 Aside: Within the KerrNewman family, only the non-rotating case or the extreme Q2 = M 2 case can be
expressed in the IWP form (2.3). To see this notice that a necessary condition is that after completing the square,
(dt + w)2 , the remaining (i.e., transverse) metric must be conformally flat. Since the Weyl tensor is identically
zero in three dimensions, it cannot be used to check conformal flatness. But there is a conformal tensor in three
dimensions, firstly defined by Eisenhart [50] but often called the York tensor [47], which is defined as
Yabc 2[c Rb]a + 12 ga[c b] R.
(2.7)
The square parenthesis denote antisymmetrization with unit weight. For the KerrNewman spacetime the nonvanishing components for the York tensor applied to the transverse metric are proportional to a 2 (Q2 M 2 ) as
claimed.
369
the metric on E4 can be written as

dsE4 = d 2 + 2 d 2 + sin2 d12 + cos2 d22 ,
(2.10)
and the functions in (2.8) expressed as

J
(2.11)
H =1+ ,
a = 2 sin2 d1 cos2 d2 ,
2
where J , are constants.

Comparing the two solutions, the crucial difference is that the KN is determined by
one complex function, whereas the BMPV is determined by one real function and a
self dual form, independent of one another. For the latter spacetime, the gravitational
potential, determined by the harmonic function, H , and the dragging effects, determined
by the Sagnac connection, ai , will be independent, while for the former spacetime they are
interconnected. There are several consequences:
(i) The location of the null hypersurface defining the future event horizon, H+ (which,
for these coordinates, is determined by the poles of the harmonic function), will
depend on the angular momentum J for the KN case, but not for the BMPV case. In
particular we know that for the KN case only for J = 0 will the harmonic function
have real poles. In contrast the BMPV case will have a well defined horizon for
non-vanishing J . The horizon becomes, however, ill-defined for large J , since the
locus of H+ becomes a timelike hypersurface [12];
(ii) For the BMPV case, the norm of the timelike Killing vector field, does not depend
on the angular momentum. Therefore, there cannot be an ergoregion and the
angular velocity of the horizon must be zero. Since there is an overall angular
momentum the spacetime cannot be rigidly rotating.
(iii) This two cases are the only possibilities when trying to include rotation in a
supersymmetric black hole spacetime: either we destabilise the horizon and create
a naked singularity, or we keep a non-rotating horizon. A rotating horizon is
incompatible with supersymmetry since it will create an ergoregion [5].
Comparing the two theories, two major differences are apparent: the dimension and
the ChernSimons term. The role of the ChernSimons term was discussed in [5] where
it was conjectured that the stability of the solution (and the regularity of the horizon)
was associated with the presence of such term with the particular coefficient required
by supersymmetry. But the particular spacetime dimension, D = 5, was not sufficiently
appreciated. A simple comparison might help. The similarity between the bosonic sector
of N = 2, D = 5 and N = 8, D = 11 supergravity has been long known, particularly the
identical ChernSimons terms [6]. The major difference is in the rank of the gauge field;
a 1-form potential for the former versus a 3-form for the latter. They naturally couple to a
black hole or a membrane. However, no regular, supersymmetric, rotating M-brane solution
exists in the literature. In [7], the BPS limit of the non-extreme rotating M2-brane yields a
spacetime with vanishing angular momentum (and an irregular horizon). Attempting to find
such solution immediately reveals a major difference with the five-dimensional case. In the
derivation of [5] it was crucial to use the fact that the one form in the metric describing the
rotation has a field strength with a self-duality property in the transverse space to the world
370
volume (as will be review in Section 4.2.1). As it stands, this property cannot be used in
D = 11, and singles out D = 5 as a special dimension.
We now give a simple example of how a self-duality condition can be useful in finding
a solution with Killing spinors in a particular dimension. One can define a Brinkmann
wave as a geometry that admits a covariantly constant null vector field: N = /v obeying
N = 0. In general these waves can be described by a metric of the type

(2.12)
ds 2 = du dv + f du + Ai dx i + ij dx i dx j .
We are using light cone coordinates u = t + y, v = t + y. Both the scalar function f , the
spatial vector Ai and the spatial metric ij may admit u and x i dependence. Pp-waves are
the special case Ai = 0 (or a pure gauge Ai , i.e., Ai = i K for some function K(x i )) and
ij = ij . We will be interested in (2.12) with a flat spatial metric and no u dependence.
Then, demanding this geometry to be a solution to pure gravity (in arbitrary dimension)
yields the constraints [32]
1
f = F ij Fij ,
(2.13)
8
where Fij = i Aj j Ai . If in addition one is looking for a supersymmetric configuration,
the spacetime must possess a supercovariantly constant spinor, or equivalently for pure
gravity, a covariantly constant spinor. Such metrics have a restricted holonomy and obey
k Fki = 0,
v = 0,
u +
1
1
Fij ij + i f ui = 0,
16
4
1
i Fj i uj = 0. (2.14)
8
For the pp-wave case, constant spinors with u = 0 (or, in t, y coordinates, (1 + ty )

= 0) are solutions to (2.14) showing that these waves are one half supersymmetric. But for
nontrivial Ai we have to deal also with the term Fij ij . A way to make this term vanish
is to require the x i space to be four-dimensional (E4 ) and Fij to be the components of a
Hodge self-dual two form in this space. We then find

1
Fij ij = Fij ij 1 + ty 7 ,
2
(2.15)
where 7 ty1234 is the chirality operator for the six-dimensional spacetime. Thus,
in D = 6, (2.12) admits Killing spinors with one quarter of the degrees of freedom of a
general constant spinor, due to the conditions
7 = ,
u = 0.
(2.16)
The self-duality condition implies that the Maxwell equation in (2.13) is identically
obeyed. To solve the Poisson type equation we parametrise E4 with the hyperspherical
coordinates (2.9). Then,
A=

J
sin2 d1 cos2 d2 ,
2
f=
J2
Q
+
,
r 2 12r 6
(2.17)
with Q constant, is a solutions. Actually, the first term in f could be any harmonic function
on E4 . We therefore find the following geometry to be a Brinkmann wave solution to D = 6
pure gravity and admitting Killing spinors:
371

Q
J2
+
(dt dy)2
r 2 12r 6

J
+ 2 sin2 d1 cos2 d2 (dy dt) + dsE4 .
r
ds 2 = dt 2 + dy 2 +
(2.18)
Compactifying along the y direction and performing KaluzaKlein reduction, we find

a 5D black hole type solution to KaluzaKlein theory (in 5 dimensions). In the Einstein
frame

2
J
2
= 2/3 dt + 2 sin2 d1 cos2 d2
+ 1/3 ds 2 (E),
ds5,E
2

J
A = 1 1 dt + 2 sin2 d1 cos2 d2 ,
2
2
Q
J
.
(2.19)
e2 = 1 + 2 +
12 6
The solution has some unusual features, but we will postpone a detailed examination to
somewhere else.
In Section 3 we will need a 10D Brinkmann wave compatible with a system involving
D1 and D5 branes. Such geometry will be similar to (2.18) but without the J 2 /12r 6 term.
Some components of the Ricci tensor will then be non-zero, namely
Rt t = Ryy = Rty = Ryt =
J2
.
r8
(2.20)
Therefore it is not a solution to pure gravity. It cannot be a solution to D = 6, minimal

Weyl
sd
). The gravitino
supergravity either. In fact, the graviton multiplet is then (gMN , M , BMN
sd
derived from the potential two form
is a Weyl vector-spinor and the field strength HMNP
sd is required to be self-dual. 5 Hence, due to the Lorentzian ( + + + ) signature,
BMN
sd will have a vanishing energy momentum tensor, and any purely bosonic solution to
HMNP
simple D = 6 supergravity must be Ricci flat.
In order to solve nontrivially the Killing spinor conditions we required the space
transverse to the Brinkmann wave propagation to be E4 . However, we may add more flat
directions, z , without spoiling the supersymmetry, as long as the metric functions Ai , f
do not depend on z . In this way we build the D = 10 configuration:

ds 2 = du dv + f du + Ai dx i + ij dx i dx j + dz dz ,
B=
Ai
du dx i ,
2
(2.21)
with = 5, . . . , 8. Assuming self-duality of Fij in the aforementioned sense, (2.21) is a

supersymmetric configuration of D = 10, type II supergravities.
5 The self duality condition is crucial for the matching of on-shell degrees of freedom: g
MN has (D 2)
D
Weyl
(D 1)/2 1 = 9; M
has 2[ 2 ]1 (D 3) = 12; an usual two form potential BMN has (D 2)(D 3)/2
= 6, which can be halved by the self duality condition, obtaining the necessary matching for supersymmetry.
372
Notice that the essential piece in this construction was to maintain an SO(4) symmetry
in the transverse space to the direction of propagation of the wave. This is an isometry
of the space where the metric functions vary. The same isometry is present for a system
of D1-branes inside D5-branes. Therefore we will be able to superimpose the systems
and still have a supersymmetric solution. But the same is not true for the system of D2
D6NS5 branes in type IIA that describe 4-dimensional black holes [8]. From the higherdimensional viewpoint we can still see why we should expect different properties of 4- or
5-dimensional black holes.
For completeness we note that (2.21) with A given by (2.17) and f = C/r 2 indeed
solves the field equations for a gravitonaxion configuration of D = 10 type II supergravity,
which read [33]

1
1
1
(2.22)
H H g H H ,
D H = 0.
R g R =
2
4
6
Let us remark that the ten-dimensional Brinkmann wave we have just described is not a
particular case of the solutions derived in [9], due to the special dependence of the rotation
vector on some of the coordinates of the transverse space rather than only one of the light
cone directions.
3. Causality in supergravity and string theory

In spite of fairly recent proposals for infinitely large extra dimensions [44], the
most accepted way to match the observed number of dimensions with the theoretical
requirements of string theory is still compactification. One assumes, therefore, a special
topology for the universe, namely that it is a K 6 bundle over M 4 , where M 4 is a fourdimensional manifold and K 6 a compact six-dimensional space. Compactification induces
a lower-dimensional spectrum consisting of an infinite tower of states. Most often only
the massless ground states are of any relevance at low energies. These are related to
the higher-dimensional fields through the KaluzaKlein procedure. When spacetime is a
twisted K 6 bundle over M 4 , the four-dimensional metric obtained by the procedure will not
be conformal to the 10-dimensional one, implying a modification in the causal structure:
light cones for 10- and 4-dimensional gravity do not coincide anymore. 6
In this section we analyse the distinctions between higher- and lower-dimensional
viewpoints for the particular example of the BMPV black hole. To complete the picture we
then review the string theory dual description of this spacetime [4], and stress the relation
between microscopic unitarity and macroscopic causality (which was implicit in [4]).
6 As an aside, let us remark on theories with variable speed of light. In fact, some recent attempts to deal
with the cosmological problem of a (possibly) negative deacceleration parameter via gravityscalar theories [21]
show a close technical similarity. In such theories, matter couples to gravity via a combination of the metric plus
the derivatives of the scalar field. In our language it would be as if gravity was propagating in the whole ten
dimensions, whereas matter would couple to a four-dimensional KaluzaKlein metric.
373
3.1. Causality and dimensional reduction

3.1.1. The D1D5-Brinkmann wave system in D = 10
In 11 dimensions, a supersymmetric configuration was found describing the intersection
of a five brane and a two brane on a string, with momentum along the string direction
and with nontrivial angular momentum [7]. 7 Using the following set of 11-dimensional
coordinates we represent the configuration as:
t
y1
y2
y3
y4
y5
5
2
5
2
w
y6
The last row refers to the spatial direction in which the wave carrying the momentum
is propagating. Compactifying the y 6 direction to a circle and performing KaluzaKlein
reduction we obtain a solution of type II supergravity (IIA or IIB since no RR fields are
excited). Transforming to the Einstein frame and S-dualizing we then get a solution to
type IIB supergravity which can be interpreted as a D1-brane inside a D5 with a Brinkmann
wave propagating along the string, as described in the last section:

1/4 3/4
2
f1
dt 2 + dy12 + fK (dt dy1)2
dsE = f5

J
2
2
+ 2 sin d1 cos d2 (dy1 dt)
1/4

f1
3/4 1/4
ds 2 E4I + f5 f1 ds 2 E4E ,
+
f5
BRR = f11 dt dy1 P cos2 d1 d2

J
+ 2 f11 (dy1 dt) sin2 d1 cos2 d2 ,
2
f5
e2( ) = .
f1
(3.1)
Solutions with some similarities have been studied in [51]. Setting J = 0 we recover the
standard D1D5-pp wave system [22]. The Euclidean space E4I is parametrised by y2 , y3 ,
y4 , y5 , while the Euclidean space E4E is parametrised by , , 1 , 2 . The three functions
f1 , f5 , fK are given by:
f5 = 1 +
P
,
2
f1 = 1 +
Q
,
2
fK =
QKK
.
2
(3.2)
7 The reader should notice that in formula (82) of this reference the expressions for B (11) and B 11 should
t16
t2 6
read, respectively T and J cos2 T /2r 2 .
374
This solution is invariant under the transformations generated by four supercharges of the
type IIB supersymmetry algebra. This corresponds to 1/8 of the vacuum supersymmetry.
Recall that the supersymmetry invariance of this theory is generated by two Weyl
Majorana spinors, L , R . The chirality conditions are 11 L,R = L,R . The Killing
spinors obey three sets of conditions. The presence of the D-string and the D5-brane give
rise to, respectively:
ty1 L = R ,
ty1 y2 y3 y4 y5 L = R ,
(3.3)
whereas the Brinkmann wave requires

ty1 L = L ,
ty1 R = R .
(3.4)
Of course, these are exactly the same conditions as in the J = 0 case [8]. The presence of
angular momentum does not break any further supersymmetry, in the same way as it does
not cost any more energy. However, as mentioned before, in solving the equation for the
gravitino variation it becomes essential to use the fact that the one form representing the
rotation, i.e., Ai in Ai dx i dt, is self-dual in the transverse space E4E .
Compactifying further on T 4 S 1 , we obtain a five-dimensional solution describing
a rotating black hole with three different charges. Using U-duality arguments, one can
quantise the charges P , Q, QKK in terms of stringy quantities [8]. Denoting the radius of
the circle by R and the volume of T 4 by (2)4 V the result is:
Q1 03 g
g 2 04 NR
,
QKK =
.
(3.5)
V
R2 V
The quantities Q1 , Q5 and NR are integers and are counting, respectively, the number of
D1-branes, D5-branes and units of right moving momentum, g is the string coupling and
0 the Regge slope. Unlike these quantities, the angular momentum parameter J is still
continuous. We will address its quantisation below.
P = Q5 g 0 ,
Q=
3.1.2. The BMPV black hole in D = 5

Let us describe the five-dimensional configuration. The metric reads

2/3
2
J
2
2
2
dt + 2 sin d1 cos d2
dsE = f1 f5 (1 + fK )
2

1/3 2 4
+ f1 f5 (1 + fK )
ds EE ,
(3.6)
while for the gauge fields we get

Qi
J
sin2 d1 cos2 d2 ),
dt +
Ai = 2
+ Qi
2(Qi + 2 )

J
dt sin2 d1 cos2 d2 ,
(3.7)
B = P cos2 d1 d2
2
2(Q + )
where i = 1, 2, Q1 QKK and Q2 Q. Hence A1 is the KaluzaKlein gauge field
arising from compactifying y1 and A2 is the winding gauge field arising from the same
compactification. The moduli are
1/4
f5
1 + fK
f1
2( )
21
2s
= ,
e =
,
e =
,
(3.8)
e
1
3
f1
f5
(f5 f ) 4
1
375
where s = 2, 3, 4, 5. An equivalent solution was firstly obtained in [2]. The ADM mass and
the entropy of this black hole are

RV
1 RgQ1 RV gQ5 g 2 NR
,
+
+
MADM = 04 2 (Q + P + QKK ) = 2
0
R
g
g
03
s
2
J2
.
(3.9)
P QQKK
SSugra =
2G5
4
G5 = 04 g 2 /(4V R) is the five-dimensional Newtons constant which relates to the tendimensional one (G10 = 8 04 g 2 ) by the moduli of the compact manifold. Notice that
the ADM mass is unchanged from the static case as expected from supersymmetry. For the
special case when all the three charges coincide, i.e., QKK = P = Q , the configuration
is equivalent to the one in [4]. The RR two form potential is then encoded in the remaining
gauge fields through
dB = ?d A A d A,
(3.10)
where A A1 = A2 and ? denotes Hodge duality with respect to the metric (3.6). All the
scalars are constants. Therefore, in this special case, the solution is specified by the metric
In this form, the configuration can also be obtained as a solution
and the gauge field A.
to five-dimensional N = 2 supergravity [5]. Using a Schwarzchild type radial coordinate
r 2 = 2 + , gE and A can be expressed as
2

2
2
(d + cos d)
ds = (10 ) dt +
2(r 2 )

dr 2
r2 2
d + d 2 + d 2 + 2 cos d d ,
+
+
2
(10 )
4

3
3
w
(3.11)
A = 2 dt (d + cos d) .
A
2
2
2r
The angles (, , ) are Euler angles on SU(2) ' S 3 , we defined w = J /2 to make
contact with [12] and for future convenience we have introduced the notation
i 2j
.
(3.12)
ij = 1 2
r
r
The configuration (3.11) is the BMPV black hole, which in isotropic coordinates becomes
(2.8). We define the quantities rH , rL , rQ and rA as, respectively the zeros of 10 , 21 ,
11 , 01 . The first quantity defines the black hole horizon (r = rH ). The second quantity
defines the surface r = rL , which we call the velocity of light surface (VLS). Inside the
VLS there are closed timelike curves (CTCs). We distinguish two cases. For rL > rH
there are naked CTCs. This is the over-rotating case. For rL < rH the CTCs are hidden
behind the horizon, which we refer to as the under-rotating case. Notice that
J2
6 QKK QP ,
(3.13)
4
which is therefore the causality bound. We remark that the interpretation of r = rH as the
black hole horizon only makes sense in the under-rotating case, since the horizon should
rL 6 rH w2 6
376
be a null hypersurface which is no longer true in the over-rotating case. The interpretation
of rQ and rA will be given in Section 4.1.
At this point we should comment on some conventions. The standard
treatment of
3 |Q|/2 [5,23].
N = 2, D = 5 supergravity
leads
to
a
Bogomolnyi
bound
of
M
>
This leads to a factor of 3/2 in the gauge potential, so that the field A introduced in
(3.11) is consistent with this treatment. This factor is then essential for supersymmetry.
The
potential A coming from dimensional reduction is consistent with a bound without the
3/2 factor. Therefore, one must perform a rescaling of the field to make contact with the
five-dimensional theory. Although using A leads to a more awkward set of conventions,
we will use it when dealing with the five-dimensional viewpoint.
3.1.3. Causality
The geometry (3.11) contains strong causality violations. Even a freely falling particle
can move backwards in time (as seen from the observer at infinity) [12]. What happens in
D = 10?
The geometry (3.1) does not have any obvious closed timelike curves, in the sense
that there is no periodic direction whose metric coefficient changes sign in some spacetime
region. However if we choose to compactify the y1 direction, we create CTCs. The point
is that there are linear combinations like t = 2B/r(/ ) + A(/y1 ) that become
timelike (we have chosen to parametrise the ten-dimensional solution with Euler angles for
these remarks). But only when the y1 direction is made periodic, may the curves with such
tangents become closed. Rewriting the 10D configuration with all the charges equal and
changing to Schwarzchild coordinates as before, the condition for such curves to become
null is:
3
p
B
= rL 21 .
(3.14)
A
r
Only for r < rL two distinct roots arise. Values of |B/A| in between them correspond to
timelike curves. This happens only inside the VLS, which can therefore be seen from the
higher-dimensional perspective as well. The condition for the curves to be closed is
r
B
=
q,
A 2R
(3.15)
where q Q.
The universal covering space of the manifold with geometry (3.1) is therefore causally
well behaved, whereas its non-simply connected compactification has CTCs. Notice that
this is quite different from creating causal anomalies in flat space by identifying the
time coordinate or from the AdS causal problems. In both these cases, it is a timelike
direction that becomes compact, whereas in our case CTCs become possible through the
compactification of a spacelike direction. The similarity is, of course, they both can be
resolved by going to the universal covering space.
The KaluzaKlein reduction to five dimensions gives rise to the identifications
377
Fig. 1. Creation of CTCs for the BMPV spacetime from the D = 10 causal configuration.
(10,E)
gMN
(5,KK)
+ e21 A A
e21 A
0
e21 A
e21
0
0
0 .
(3.16)
e2s
The superscripts on the metric refer to the Einstein or KaluzaKlein frame. 8 The first
row (and column) refers to the directions t and E4E , the second to the direction y1 and the
last to the directions in E4I . The five-dimensional Einstein frame is then obtained as
2
g (5,E) = e 3
P5
i=1 i
g (5,KK) .
(3.17)
The procedure eliminates the y1 direction and effectively projects down the CTCs into the
5,KK (or g 5,E ) has unavoidable closed timelike curves, whereas
direction. So, the metric g
the combination
(5,KK)
+ e21 A A
g
(3.18)
does not. The latter describes the local geometry of a manifold were any existing CTCs
will not be homotopic to a point.
In Fig. 1 we illustrate the above procedure. The picture on the left describes the D = 10
universal covering manifold, i.e., y1 ' R, and we illustrate the curve with tangent t .
Step 1 is the compactification of y1 , i.e., y1 ' S 1 , so that the curve with tangent t
becomes closed. Step 2 is the KaluzaKlein reduction, upon which the curve is projected
onto the direction, corresponding to the causal anomalies seen in the BMPV black hole.
We should remark that the pictures are misleading in two senses. Firstly, the manifold
parametrised by y1 and E4E is a nontrivial R (or S 1 ) bundle over E4E , which is precisely
what allows the curve described by t to become timelike. Secondly, the direction
does not go around a nontrivial cycle of the manifold.
Can we associate other acausal spacetimes with better behaved higher-dimensional
configurations in a similar fashion? This would rely on performing a transformation of
8 By KaluzaKlein frame we mean the frame obtained from a KaluzaKlein compactification starting from the
higher-dimensional Einstein frame and without rescaling the lower-dimensional metric by using the moduli, as in
using ansatz (3.16)
378
the type (3.18). Thus, one cannot apply the method to any acausal spacetime configuration,
since not all examples have a gauge field. The Gdel manifold is the simplest example [24].
But it turns out that this procedure does not work for the general five-dimensional family
of black holes [2]. Moreover it is not something associated to supersymmetry either, since
the four-dimensional KerrNewman spacetime with Q2 = M 2 still exhibits the traditional
closed timelike curves in the negative r region after applying an analogous transformation
to (3.18). Therefore, we have to conclude that this spacetime has a peculiar property,
which, using Carters terminology [10], one can put as: the nontrivial CTCs of the fivedimensional spacetime arise as trivial CTCs from the ten-dimensional viewpoint.
3.2. Causality and unitarity
Consider the ten-dimensional supergravity D1D5-brane solution, i.e., (3.1) with
QKK = J = 0, and y1 , E4I compact. The D-brane tensions depend on the string coupling
constant g as 1/g (as can be seen from (3.9)) and the ten-dimensional Newtons as G10
g 2 . Hence, the gravitational potential for D-branes goes as V g. Decreasing the string
coupling we turn off gravity and find a flat space configuration: a gas of open strings with
boundary conditions determined by the presence of the D-branes.
The theory describing these open strings and effectively describing the D-branes is a
(1 + 1)-dimensional SCFT, which can be thought of as living on the worldvolume of the
D1-branes inside the D5-branes [34]. Within the AdS/CFT correspondence this is because
the near horizon geometry of the D = 10 configuration is AdS3 S 3 T 4 , and the 1 + 1
SCFT lives on the conformal boundary of the AdS piece. Since we compactified the
supergravity configuration to 5D, our SCFT will be on the cylinder R S 1 , were R is
the time direction. The amount of supersymmetry of this SCFT must be the same as in the
classical configuration, i.e., 1/4 of the vacuum supersymmetry. Thus, we have an N = 4
SCFT on R S 1 describing the compactified supergravity D-brane configuration.
Supersymmetric states belong to short multiplets. These should be stable at any value
of the string coupling. Therefore we expect our continuous variation of g to leave the
degeneracy of states unchanged, i.e., to be adiabatic. The SCFT has a central charge
determined by the massless degrees of freedom of the open strings, c = NB + NF /2. 9
These are fairly easy to count: the massless excitations have the same number of bosonic
(NB ) and fermionic (NF ) physical degrees of freedom, 4Q1 Q5 , effectively carried by the
(1, 5) and (5, 1) strings in the flat space D-brane configuration [22]. Hence c = 6Q1 Q5 .
Now we want to turn on momentum (QKK ) and angular momentum (J ) in the
supergravity side and see the correspondence in the SCFT side.
The supergravity angular momentum is classified by the SO(4) rotation group which is
an isometry of the geometry (3.1) or (3.6) acting on the space transverse to the branes. We
have to identify this symmetry on the SCFT side. It is known that an SO(4) symmetry arises
in the N = 4 superconformal algebra, corresponding to endomorphisms in the graded
9 In our setup the states describing the entropy are right moving. That is why the central charge has a tilde. But
this is the usual Virasoro central charge, related to the complex dimension dc of the target space for the sigma
model by c = 3dc .
379
algebra that rotate the fermionic generators Gim (i = 1, . . . , 4) amongst themselves, i.e.,
R-symmetry. Such symmetry is gauged in the sense that linear combinations of the Gim
(for negative m) create states that carry (FL , FR ) charges of the U (1)L U (1)R Cartan
subalgebra of SO(4). These two quantities are therefore to be identified with the two
linearly independent spacetime angular momentum parameters

VR
21
43
21
43
J = 0, 04 2 J .
(3.19)
= 0,
(JL , JR ) J J , J + J
4G5
g
The J ik is the angular momentum tensor that can be read off from the geometry (3.6) in
the usual way. In this way, the JL , JR become quantised in terms of (FL , FR ). Using the
above results for the quantisation of P , Q, QKK , we may now express the entropy of the
classical solution in terms of quantised charges:
s
s
F2
2
J2
= 2 Q1 Q5 NR R .
(3.20)
P QQKK
SSugra =
2G5
4
4
We have not identified NR in the SCFT side yet, i.e., established the correspondence
between SUGRA momentum and a quantity in the SCFT. In supergravity the integer NR
arises due to the compactification of the y1 direction into S 1 . It is therefore a U (1) charge.
The conformal algebra of the cylinder the Virasoro algebra has certainly a U (1)
subalgebra. States carry a U (1) charge given by the eigenvalue of the zero mode in the
theory, i.e., the Virasoro generator e
L0 . Hence, NR must correspond in the SCFT to the
right moving level nR of the states.
What is the degeneracy of states of the form
|nL , FL ; nR , FR i = |0, 0; NR , FR i
(3.21)
in an N = 4 SCFT of central charge c = 6Q1 Q5 ? For a 2D CFT on a cylinder,

the degeneracy of states at highest available level M 1 can be computed by using
the modular invariance of T 2 in the Euclidean section [37,38], yielding, up to power
corrections
q
d(M, c)
e
M c
6
(3.22)
The highest level available for the states (3.21) is not NR , after we have fixed the other
quantum numbers, namely FR . In a CFT, states associated with a primary operator give a
contribution to the total e
L0 eigenvalue equal to their conformal weight. In particular, an
operator creating states with charge FR has a conformal weight not smaller than 3FR2 /(2c)
if we assume that the CFT is unitary, which demands all conformal weights to be non plus the
negative [43]. The total e
L0 eigenvalue, NR , is therefore the sum of 3FR2 /(2c)
remain
, which is non-negative by unitarity:
conformal weight of other operators, NR
3FR2
F2
+ NRremain Q1 Q5 NR > R .
(3.23)
2c
4
This is the unitarity bound, which coincides with the causality bound (3.13). Moreover,
the highest available level is now M = NRremain . This is the total level available to find
NR =
380
as different combinations of operators. It therefore determines the degeneracy of states. It

follows that

(3.24)
SSCFT = ln d NRremain , c = SSugra .
The above bound can also be seen in the explicit constructions of unitary representations
of the SCFT. First notice that the states we are interested in are in the left-moving ground
state. So, we need to look only at the representation theory of an N = 2 SCFT, which
was worked out in [35]. The generators are Lm , the Virasoro generators, Tm , the modes
of a U (1) current, and Gim , i = 1, 2, the modes of the supercurrent. The U (1) symmetry
is the only surviving piece of the SO(4) symmetry of the N = 4 algebra, which we want
to think of as U (1)R . There are 3 possible N = 2 algebras, according to the moddings
one can choose for the generators, via the boundary conditions. We are interested in the
P algebra, where the fermions have periodic boundary conditions and therefore generalises
the usual Ramond sector of the open string theory. The reason is that the spacetime angular
momentum is carried by the fermionic modes of the (1, 5) and (5, 1) strings [8].
There are 3 possible classes of unitary representations for the P algebra. The ones with
a two-dimensional moduli space obey [35] 10

2

c
c
1 c
1 h
q2 +
+ 1 > 0,
2
(3.25)
3
24
4 3
which, for c = 6Q1 Q5 1, level h = NR and charge q = FR gives (3.23).
The upshot is that the charge (in CFT language) or angular momentum (in spacetime
language) we create must be bounded by the energy we create; otherwise we violate
unitarity or causality.
4. Other properties of the BMPV spacetime

4.1. The repulson behaviour
The first property we would like to address is related with the over-rotating case. Let
us start by computing the charged geodesics. We follow the HamiltonJacobi method with
the usual minimally coupled Hamiltonian
H=
where g
g
(p + qA )(p + qA )
2
and A are given by (3.11). The ansatz for the action function is
S = Et + H (, , ) + W (r),
H (, , ) = jL + jR + ().
(4.1)
(4.2)
This generalises the construction in [12]. The Killing tensor found there is reducible. Since
each of the Killing vector fields into which it decomposes is a symmetry of the Maxwell
field (in the sense that L3 A = Ri A = 0, where denotes the Lie derivative and Li (Ri )
10 Note: c in [35] is the complex dimension of the target space, hence 3 times our usual Virasoro central charge.
381
the left (right) invariant vector fields on SU(2)), we conclude that the Killing tensor still
commutes with the minimally coupled Hamiltonian. Therefore the quantity
j 2 (L1 H )2 + (L2 H )2 + (L3 H )2 ,
(4.3)
is still a constant of motion, as in the case of purely gravitational interactions. The

following set of equations
of motion is then obtained (for the BMPV black hole, the mass
to charge ratio is M/Q = 3/2, but we keep Q and M in order to make the interpretation
of the several terms clear):

2
4j 2
qQE
qQ 2
dr
= E 2 21 m2 + 2 (10 )2 2 11 +
01
d
r
r
2r 2

2jR
4
+ 4 qQ ME 10 ,
3
r

1
qQ
4MjR
dt
=
E
21
11
10 ,
d (10 )2
2r 2
3r 4

4 jR jL cos
4
d
=
ME qQ .
(4.4)
+ 2
d r 2
4r 10 3
sin2
The and equations of motion are the same as in the purely gravitational case [12]. The
quantity m2 is the mass squared of the test particle, which arises in the usual fashion as
the integral of motion associated with the metric Killing tensor. Notice that the RHS of
the r equation is even under CPT, whereas the RHS of the t and are odd, as expected
from the form of the LHS. Also notice that the affine parameter is not proper time for
the massive case, but = m, and m can be made to vanish in the equations by defining
energy, charge and angular momentum per unit mass.
We now discuss some features of these equations. There are four interesting surfaces:
the horizon (at r = rH ), the VLS (at r = rL ) and the timelike surfaces rQ , rA , defined as
the zeros of 11 and 01 respectively. The relative location of these surfaces is rA < rQ <
rL < rH (rA > rQ > rL > rH ) for the under-rotating (over-rotating) case. The VLS bounds
the region where light cones allow causal travelling into the past of the observer at infinity.
The energy term in the t equation changes sign corresponding to the possibility of geodesic
time travelling, or equivalently, to a change of character from particle to antiparticle. The
energy term in the radial equation also changes sign and becomes repulsive. The surface
r = rQ couples to the Coulomb term both in radial and time equations. Both of them change
signs when crossing this surface, i.e., the Coulomb interaction changes from attractive to
repulsive or vice-versa. We call this surface Coulomb conjugation surface (CCS). The
r = rA surface corresponds to a change in sign for the general relativistic correction to the
electromagnetic interaction (which is asymptotically subleading). Effectively q 2 changes
sign there.
On the surface r = rH , the right hand side of the r equation of motion becomes

6

rL
3qQ 2
1
.
(4.5)
E
4M
rH
382
This is always negative in the over-rotating case. We conclude, therefore, that the repulson
like behaviour found in [12] for the gravitational interactions is still true in general when
charge interactions are taken into account: for the over-rotating solution there are no freely
case singled out
falling orbits, with or without charge, entering the r = rH surface. The
by (4.5) corresponds to an energy to charge ratio for
the test object of 3/2. With our
conventions, within a supersymmetric theory E > m > 3/2|q|. Thus a particle for which
(4.5)
vanishes everywhere should have E = m, i.e., be at rest, so the mass to charge ratio is
also 3/2 a BPS particle. The behaviour is then better seen by rewriting the r equation
of motion:

2
2jR
4
2
4
dr
ME
= qQ ME
qQ
10
d
3
r4
4r 6
3

1
16
4
1
2
2
2
2
2
+ E m 2 qQE Mm + 4 (qQ) (mM)
3
9
r
4r
4j 2
(10 )2 .
(4.6)
r2
Since the BPS test particle must have for initial conditions the lowest possible quantum
numbers, |E| = m, j = 0, it will stay at rest in the spatial coordinates, i.e., a no-force
configuration.
The repulson result is readily confirmed at the semi-classical level for charged scalar
waves minimally coupled to the background electromagnetic field. We also introduce a
non-minimal coupling to the geometry. We are still able to separate variables in the latter
model, which follows from the Ricci scalar being a function of r only:

2
22 2 r 2 2 .
8
r
The wave equation then takes the form
R=
(4.7)

g (D + iqA)(D + iqA ) = m2 + R .
(4.8)
The non-minimal coupling might be thought of as a renormalisation of the mass.

j
Following [12] we use the ansatz (x ) = eit E DjL ,jR F (r) and change variables as
x=
,
r2
in which case (4.8) can be written as

B
C
D
x2
2 x 3
d 2F
= A+ + 2 + 3 2
F.
dx
x
x
x
x (1 + x)3 (1 + x)2
A, B, C, D are given by

6

qQ 2
rL
1
1 E
,
A=
4
rH
2

E qQ
E ,
C = D + j (j + 1) +
2
2
(4.9)
(4.10)
383

qQ 2
Qq 2 jR
1
1
2 2
E
(qQ 2E),
B=
E
4
2
2
2
2

2
D=
m E2 .
(4.11)
4
Near the horizon the scalar equation (4.10) is approximated by keeping only the A term on
the right hand side. Oscillating solutions arise only if rL < rH , i.e., in the under-rotating
case. Therefore the absorption cross section for scalar charged waves is zero in the overrotating case, confirming the repulson behaviour.
Now we ask if an accelerated observer can enter the horizon in the over-rotating case.
Obviously, there are timelike curves that can achieve that. The simplest example is a radial
orbit with tangent vector
t 10 r ,
(4.12)
t =
10
which has a unit norm. But it is well known in general relativity that a generic timelike
trajectory might not be realistic. The simplest example is given by the Reissner
Nrdstrom spacetime. The repulsive character of the timelike singularity precludes any
timelike trajectory with bounded proper acceleration to reach the physical singularity at
r = 0 (in Schwarzchild type coordinates) [25]. The non real character of such curves is
manifest in the fact one would need an infinite acceleration (and so an infinite payload for
the rockets) to perform such a deed (even assuming one could survive the tidal forces). In
the ReissnerNrdstrom case, the divergence in the proper acceleration is a consequence
of the existence of an s.p. singularity. In our case, however, the most obvious curvature
invariants (R, R R , R R ) show no such singular behaviour on or outside the
r = rH surface. Furthermore, we could not find any pp-singularity. 11
What happens to an accelerated observer? Along the trajectory described by (4.12), the
proper acceleration gives

82
w2

(4.13)
a a = t D t t D t = 6 1 + 2 .
r
r
There is no divergence here. Therefore it seems one would be able to travel into the r < rH
region if (and only if) one would have a rocket and a (very) robust spaceship. This is clearly
an odd behaviour.
Let us conclude this section by commenting on some recent work on repulsons. As
mentioned above, the RN curvature singularity is repulsive. In particular, the extreme
RN spacetime with negative mass has a repulsive naked singularity. This is the type of
repulsons dealt with in [36]. In order to make contact with the AdS/CFT conjecture, the
repulsons therein are brane configurations with near horizon geometries containing an AdS
piece. One example is (3.1) with J = QKK = 0 and Q < 0. Others may be obtained by
T-duality. These repulsons are therefore quite different from our case, since the repulsive
hypersurface is non-singular for the over-rotating BMPV spacetime.
11 Black holes with this kind of singular behaviour outside the horizon (naked black holes) were studied in [26].
384
4.2. The gyromagnetic ratio for the BMPV black hole

We also want to address the gyromagnetic ratio of these black holes. As seen before,
from the asymptotic behaviour of the metric (3.11), we can read off how the two parameters
and are related to the ADM mass and the two angular momenta parameters of the
SO(4) rotation group:

3
,
(JL , JR ) = 0,
.
(4.14)
MADM =
4G5
2G5
To get the latter expression we can use the relation between Euler angles and the cartesian
coordinates on E4 , Xi . Such relation is obtained by solving the embedding constraints for
the S 3 embedding in E4 . We can then write
d + cos d =
2(L3 )ki Xk dXi

,
2
where L3 is a Hodge self dual two form on E4 , and =

momentum can now be expressed as
3 ki
,
L
J ki =
4G5
(4.15)
p
Xi Xi . The spacetime angular

(4.16)
and our definitions of left and right angular momenta are J 21 J 43 , with the + sign for
JR .
The charge associated with the Maxwell field is given by 12
I
1
3
Q=
F dS =
,
(4.17)
8G5
2G5
S3
and so the mass to charge ratio is 3/2, i.e., it saturates the Bogomolnyi bound for the
D = 5, N = 2 supergravity theory [23]. We should remark that the normalization in (4.17)
and for the ADM mass are consistent with Newtonian and Coulomb force laws of the form
8G5 M1 M2
2G5 Q1 Q2
,
kFC k =
,
(4.18)
kFN k =
3 r 3
r3
which means that a gravitationalelectrostatic force balance is achieved when 4M1 M2 =
3Q1 Q2 . An object experimenting a no-force condition
on a BPS background is then
expected to possess the same mass to charge ratio of 3/2, as seen in the last section.
Rotation endows a charged black hole with a magnetic dipole moment, which can be
read off from the spatial components of the vector potential. Using (4.15) we get
3 3 ij
,
(4.19)
L
ij =
4G5
H
12 An alternative quite frequent charge normalization is Q = (2A

1
dS . Then, the
D2 GD )
S D2 F
charge, magnetic dipole moment and coefficient in Bogomolnyi bound will differ from ours by a factor of /2.
In this section we set G5 = /2. So our formulae are the same as the ones we would obtain with the alternative
charge definition and setting G5 = 1.
385
and from the usual relation we can read off the gyromagnetic ratio:
ij = g
Q ij
J
2M
g = 3.
(4.20)
Notice we would obtain the same value if we had used A in (3.11) instead, since an overall
factor gives the same contribution to charge and magnetic moment.
4.2.1. Black hole superpartners
The field configuration (3.11) can be expressed very simply using an isotropic type radial
coordinate, defined as 2 + = r 2 . Then we have (2.8). The harmonic form H and the
one-form a are given by (2.11) for a spacetime with a connected even horizon. The multi
black hole configuration is obtained by using a harmonic function with N poles:
N
X

H xi = 1 +
=1
|x i
,
xi |2
a=
k
J
L3 i k H dx i ,
4
(4.21)
where xi are constants. A quite natural set of frames is

e0 = H 1 (dt + a),
ei = H 1/2 dx i ,
(4.22)
for which the field strength F takes the form F = 3 d(e0 )/2. Using the Cartan structure
equation for the spin connection and expressing F and the exterior derivative in terms of
the frames we get [5]:
fij j
fij 0
i H
e ,
ij =
e + H 3/2k[i j ] H ek ,
0 i = 3/2 e0 +
2
2
H
2H
2H

fij i
3
1/2
i
0
j
e
H
e
e
+
H
,
F=
i
2
2H 2
d = e0 H t + H 1/2 ei (i ai t ).
(4.23)
The nontrivial supersymmetry variation of the D = 5 simple supergravity theory for our
bosonic background is the gravitino variation: 13

1
i
= d + ab ab + ea bc a 4eb c Fbc
4
4 3

ij

fij
ii H i
0
1

= e0 H t +
8H 2
2H 3/2

i 0 k H

+ ek H 1/2(k ak t ) +
2H 3/2

i H ik ifki i
i0
0
1
i
+
(f
?f
)
.

ik
ik
4H 3/2
2H 2
4H 2
(4.24)
13 We follow the conventions of [27]. However, the signature choice therein is (+ ), so that our
equations differ by a few factors of i. We choose for the flat Gamma matrices 01234 = i, and make use of
the four-dimensional Majorana representation plus 4 = i 0 1 2 3 . Therefore, 0 , 1 , 2 , 3 are real and
4 purely imaginary. Furthermore, 0 , 4 are antisymmetric and the remaining gamma matrices symmetric.
386
We have introduced the notation fij for the components of the two form f = da, and ?f
is the Hodge dual of f on E4 . One can check that the f that follows from (2.11) (or its
multi black hole generalisation) is a self dual form on E4 . Then, for
= H 1/2 0K ,
0K = i 0 0K ,
(4.25)
we get = 0, i.e., such is a Killing spinor. However, if one chooses the constant spinor
0 obeying
0AK = i 0 0AK ,
(4.26)
we get a nontrivial gravitino variation, which can be fed back into the vielbein and gauge
field variations to yield the first order superpartners of the BMPV spacetime. These spinors
are often called anti-Killing spinors. We will comment more on the choice of this specific
form for the anti-Killing spinors below.
We now generate the superpartners for the BMPV (multi)-black hole spacetimes. To first
nontrivial order the variations of the gravitino and gauge field are:

1
1
ifki i
ek
e0
,
= 2 iH 1/2i H i fij ij 3/2 k H + i H ik +
4
2
H
H
H 1/2
0 k

3i
3 i e fi
1/2 k
( ) =
H e i H ik ,
(4.27)
A =
2
2
2
2H
a a yielding
while for the vielbein they follow from ea =

i e0 k
fi H 1/2ek i H ik ,
e0 = 2
H
2

j
e = 2iH 2 fik ek e0 H 1/2i H ij .
In the holonomic basis the first order superpartner takes the following form:

j
ds 2 = H 2 1 + iH 2 fi ij dt 2

j
j
2H 2 ak + i H 2 ak fi 3i H k ij dt dx k

4i
H 1 i H ak + fki ij dx k dx j ,
+ H dx k dxk +
H

i
3
j
i
f

dt

1+
A=
j
2H
2H 2 i

1
3
j
j
i
ak + i
ak fi i H k j dx k .
+
2H
2H 2
(4.28)
(4.29)
(4.30)
Comparing the three terms of g0i with the three terms of Ai , we see that they have similar
form, but with different coefficients:
(i) The
term with purely bosonic angular momentum, ak , implies a relation J ki =
ki
/ 3, leading to a gyromagnetic ratio g = 3 as seen before;
(ii) The term with purely fermionic angular momentum, H 2 , gives J ki = 3 ki ,

implying g = 1. This should be regarded as the gyromagnetic ratio for the fermionic
superpartners of a static background;
387
(iii) The mixed term, af 2 , yields J ki = 2/ 3 ki and gives rise to g = 3/2.

As we move along the supermultiplet, the gyromagnetic ratio varies. By this we mean that
g for the BMPV black hole differs from the one for its superpartners. But notice that the
supermultiplets are labelled by some value of the bosonic angular momentum, so that the
superpartners of the D = 5 RN black hole are not in the same supermultiplet as the BMPV
black hole with a 6= 0.
The result to keep in mind is that the gyromagnetic ratio for the 5D black holes
with some kind of angular momentum seems to cover the range 1 6 g 6 3, with the
extreme values attained for purely fermionic and bosonic angular momentum, respectively.
A similar behaviour will be seen when we study the Dirac equation.
Let us now justify the use of the form (4.26) for the anti-Killing spinors. It is quite clear
that any choice of = f (x)0 , where f 6= H 1/2 or 6= K , will lead to a nontrivial
gravitino variation and hence some kind of superpartners. The new set of fields will
represent the same physical spacetime configuration as the initial one, just in a different
superframe. This is in a very direct analogy with the excitation of the magnetic field
by going to a frame moving with respect to a purely electric source. Then, we still
have the same physical setup, i.e., some configuration of electric sources; the fact that
we see a different set of fields (namely electric and magnetic) results from the fact that
only the components of the electric plus the components of the magnetic field fit into
a complete relativistic multiplet, while each of these fields would fill a Newtonian
multiplet. Of course, the fundamental point is that the Lorentz group is a more fundamental
symmetry than the Galilean group, which is a low energy approximation. Similarly, the
superdiffeomorphisms group of a supergravity theory allows the same physical setup to
be described by many different sets of fields. In many physical situations some of these
sets of fields contain fermionic excitations while others will be purely bosonic. We can
therefore gauge away the fermions, which can then be labelled as pure gauge. In many
situations, however, the fermions cannot be gauged away (in the context of N = 1, D = 4
supergravity, necessary and sufficient conditions for an excited gravitino to be pure gauge
were given in [28]).
One way to deal with such gauge arbitrariness is to impose the tracelessness condition
to the first order gravitino, i.e., = 0 [30]. Physically this means that the first
order gravitino is a pure spin 3/2 excitation, since the projection of the gravitino
transforms as a spin 1/2 representation of the Lorentz group universal covering. Moreover,
b , where D
b is the supercovariant derivative, we impose a condition on
since = D
the existence of superpartners, namely that the candidate background has regular solutions
b = 0. Of course, we are interested in obeying this
to a modified Dirac equation D
b 6= 0, in order to get nontrivial superpartners. Our choice for antiequation but with D
Killing spinors satisfies this criterion.
It has also been argued that a more fundamental criterion for suitable superpartners is to
require the first order gravitino to be normalisable [31]. For the superpartners of the static
background it is very easy to show this is the case, just as for the D = 4 RN spacetime [30]:
388
Z
| |2 =
d 4x

g(4) g = 2 2 MADM 0AK 0AK ,
(4.31)
where 6 is a spacelike surface.

4.2.2. The gyromagnetic ratio for an electron
A test Dirac fermion with charge q and mass m interacting with gravitational and
electromagnetic fields is described by a wave function obeying the minimally coupled
Dirac equation, which takes the standard form

1
(4.32)
+ ab ab + iqA m = 0.
4
Using (4.23), the spin connection term can be rewritten in terms of the Maxwell tensor:

1
0
ab
(4.33)
+ iqA Fab m = 0,
4 3
where 0 is the flat gamma matrix. Hence we see that the problem of studying a Dirac
fermion in the background (3.11) can be restated as the study of a non-minimally coupled
Dirac fermion interacting solely with an electromagnetic field. Of course this should be
expected for a supersymmetric background, since the vanishing of the supercovariant
derivative acting on some spinor means that there will be a cancellation of terms between
the Maxwell field and the spin connection.
In four-dimensional flat space, adding a non-minimal electromagnetic interaction of the
type
iq
g F ,
(4.34)
8m
to the minimally coupled Dirac equation yields in the non relativistic limit, a magnetic
dipole interaction in the Hamiltonian of the form
q
s B(2 + g).
(4.35)
Hdipole =
2m
The natural higher-dimensional generalisation of this result yields for the gyromagnetic
ratio associated with a Dirac fermion obeying (4.33) the non-standard value g = 2 + g
and
2m
(4.36)
g = i 0 .
3q
Therefore we arrive at the interesting conclusion that the gyromagnetic ratio of a spin 1/2
particle depends on its spinor direction being parallel to the anti-Killing or Killing spinors.
In particular we should get g = 3 for the latter case and g = 1 for the former, when the
Dirac fermion is BPS. It is clear that there is a conspiracy between the behaviour of
the elementary particle and the behaviour of the black holes seen in the last section. Its
meaning, however, is not quite clear. Moreover, this seems to be a particular property of
the five-dimensional family of black holes. For the D = 4 extreme KN, such conspiracy
does not arise: g = 2 both for the bosonic background [10] and for the superpartners of the
389
extreme RN background [20]. The Dirac equation in the extreme KN background cannot be
easily cast into the non-minimally coupled flat space form. But a result similar to the fivedimensional one would require g = 0, i.e., the vanishing of the non-minimal coupling.
This is not expected to occur. Indeed, even for the D = 4 extreme RN, the Dirac equation
can easily be put in a form similar to (4.33) with g = 1/2.
Let us close this section by discussing the possible string theory counterpart of these
results. Massive string states have a Schwarzchild radius greater than their Compton wave
length; in other words their mass is greater than the Planck mass (in ten dimensions). This
led to the suggestion that such states should be identified with some extremal black holes
in supergravity [45]. One possible check on this conjecture was made for gyromagnetic
ratios. On the string theory side, the gyromagnetic ratio for heterotic states in the presence
of a background gauge field were computed in [46]. A matching was verified with the g
value for some black holes of D = 4, N = 4 supergravity coupled to 22 vector multiplets
(i.e., the low energy field theory for heterotic on T 6 ) [14]. On the string theory side,
the computation follows from the knowledge of the action for the heterotic string in the
presence of the background field. In our case, the black hole does not correspond to string
states but rather to D-brane states. Since the open strings describing the D-branes do not
couple to the RamondRamond charge it is not clear how one could compute g for the
microscopic configuration.
5. Conclusions
In this paper we argued that within all known BPS, rotating, asymptotically flat stringy
black holes, the five-dimensional case is rather special. And that one may use these special
spacetimes to learn more about the connections between microscopic and macroscopic
gravity. Our framework was toroidally compactified string theory, but one may embed the
BMPV geometry in M-theory compactified on more general CalabiYau spaces [52]. In
Section 2 we performed a comparison with the typical irregular case: the four-dimensional
extreme KerrNewman. The special properties arise not only from the ChernSimons
term [5] but also from the possibility of having a Hodge self dual rotation two-form. This
is illustrated by studying a special class of gravitational waves.
The rich causal structure of these spacetimes also presents a novel feature: CTCs
homotopic to a point in the five-dimensional spacetime are resolved in the universal
covering of the ten-dimensional uplifted geometry. Although CTCs are clearly a nonperturbative effect in string theory (as the repulson effect of Section 4.1), they manifest
themselves at weak coupling by the loss of unitarity. The point here is that states violating
(3.23) would imply the existence of states with negative e
L0 eigenvalue. Consequently there
would be negative norm states, i.e., non-unitary or ghost states.
Much controversy has surrounded causality as a fundamental principle in theories
of gravity. Despite the doubts cast by the information paradox, unitarity is generally a
more solid principle in quantum theories. In the same way we still have to understand
quantitatively the microscopic description of the most fundamental black hole
390
Schwarzchild it will be an interesting problem to understand the microscopic states

associated with more fundamental acausal spacetimes, as the Gdel manifold, and
scrutinise the role of unitarity.
Another interesting property of these black holes concerns the gyromagnetic ratio. In
quantum field theory, requiring a good behaviour for the tree level scattering amplitudes
singles out g = 2 as the most natural value for elementary particles [48]. This is
implemented by a non-minimal electromagnetic coupling for fields with spin higher
than 1/2. For spin 1 charged matter, another argument in favour of g = 2 is that a
coupling consistent with such gyromagnetic ratio gives rise to a non electromagnetic gauge
symmetry [49]. One could argue that a similar statement applies to black holes, if one is
trying to interpret them as field theory realizations of fundamental states. It would therefore
be interesting to understand in the microscopic context the results of Section 4.2.
Acknowledgments
I would like to thank Miguel Costa, Malcolm Perry, Harvey Reall and Paul Townsend
for discussions. I am particularly grateful to Gary Gibbons for many discussions and
suggestions. The author is supported by FCT (Portugal) through grant no. PRAXIS
XXI/BD/13384/97. This work is also supported by the PPARC grant PPA/G/S/1998/00613.
References
[1] M. Cvetic, D. Youm, Entropy of non-extreme charged rotating black holes in string theory,
Phys. Rev. D 54 (1996) 2612, hep-th/9603147.
[2] M. Cvetic, D. Youm, General rotating five-dimensional black holes of toroidally compactified
heterotic string, hep-th/9603100, Nucl. Phys. B 476 (1996) 118.
[3] M. Cvetic, D. Youm, Near-BPS-saturated rotating electrically charged black holes as string
states, Nucl. Phys. B 477 (1996) 449, hep-th/9605051.
[4] J. Breckenridge, R. Myers, A. Peet, C. Vafa, D-branes and spinning black holes, hepth/9602065, Phys. Lett. B 391 (1997) 93.
[5] J. Gauntlett, R. Myers, P. Townsend, Black holes of D = 5 supergravity, hep-th/9810204, Class.
Quant. Grav. 16 (1999) 1.
[6] E. Cremmer, Supergravities in 5 dimensions, in: S.W. Hawking, M. Rocek (Eds.), Superspace
and Supergravity, Cambridge University Press, 1980, p. 267.
[7] M. Cvetic, D. Youm, Rotating intersecting M-branes, Nucl. Phys. B 499 (1997) 253, hepth/9612229.
[8] J. Maldacena, Black holes in string theory, Ph.D. thesis, hep-th/9607235.
[9] E. Berghoeff, R. Kallosh, T. Ortin, Supersymmetric string waves, hep-th/9212030, Phys. Rev.
D 47 (1993) 5444.
[10] B. Carter, Global structure of the Kerr family of gravitational fields, Phys. Rev. 174 (1968)
1559.
[11] S. Hawking, G. Ellis, The Large Scale Structure of the SpaceTime, Cambridge University
Press, 1973.
[12] G. Gibbons, C. Herdeiro, Supersymmetric rotating black holes and causality violation, hepth/9906098, Class. Quant. Grav. 16 (1999) 3619.
391
[13] A. Sen, Rotating charged black hole solution in heterotic string theory, Phys. Rev. Lett. 69
(1992) 1006.
[14] M. Duff, J. Liu, J. Rahmfeld, Dipole moments of black holes and string states, Nucl. Phys.
B 494 (1997) 161, hep-th/9612015.
[15] V. Balasubramanian, D. Kastor, J. Traschen, K. Win, The spin of the M2-brane and spinspin
interactions via probe techniques, hep-th/9811037, Phys. Rev. D 59 (1999) 084007.
[16] F. Larsen, Rotating KaluzaKlein black holes, hep-th/9909102.
[17] A. Hosoya, K. Ishikawa, Y. Ohkuwa, K. Yamagishi, Gyromagnetic ratio of heavy particles in
KaluzaKlein theory, Phys. Lett 134B (1984) 44.
[18] M. Duff, J. Liu, J. Rahmfeld, g = 1 for Dirichlet 0-branes, hep-th/9810072, Nucl. Phys. B 524
(1998) 129.
[19] E. Witten, String theory dynamics in various dimensions, Nucl. Phys. B 443 (1995) 85, hepth/9503124.
[20] P. Aichelburg, F. Embacher, Exact superpartners of N = 2 supergravity solitons, Phys. Rev.
D 34 (1986) 3006.
[21] M. Clayton, J. Moffat, Scalartensor gravity theory for dynamical light velocity, gr-qc/9910112.
[22] C. Callan, J. Maldacena, D-brane approach to black hole quantum mechanics, Nucl. Phys. B 472
(1996) 591, hep-th/9602043.
[23] G. Gibbons, D. Kastor, L. London, P. Townsend, J. Traschen, Supersymmetric self-gravitating
solitons, Nucl. Phys. B 416 (1994) 850.
[24] K. Gdel, An example of a new type of cosmological solutions of Einsteins field equations of
gravitation, Rev. Mod. Phys. 21 (1949) 447.
[25] S. Chakrabarti, R. Geroch, C. Liang, Timelike curves of limited acceleration in general
relativity, J. Math. Phys. 24 (1983) 597.
[26] G. Horowitz, S. Ross, Naked black holes, hep-th/9704058, Phys. Rev. D 56 (1997) 2180.
[27] A. Chamaseddine, H. Nicolai, Coupling the SO(2) supergravity theory through dimensional
reduction, Phys. Lett. 96B (1980) 89.
[28] P. Aichelburg, H. Urbantke, Necessary and sufficient conditions for trivial solutions in
supergravity, Gen. Relativ. Gravitation 13 (1981) 817.
[29] G. Gibbons, in: F. Aguila et al. (Eds.), Proceedings of the XVth GIFT seminar, Supersymmetry,
Supergravity and Related Topics, World Scientific, Singapore, 1985.
[30] R. Brooks, R. Kallosh, T. Ortn, Fermionic zero modes and black hole hypermultiplet with rigid
supersymmetry, hep-th/9505116.
[31] K. Win, Fermion zero modes and cosmological constant, hep-th/9811082.
[32] C. Duval, G. Gibbons, P. Horvthy, Celestial mechanics, conformal structures, and gravitational
waves, Phys. Rev. D 43 (1991) 3907.
[33] M. Duff, R. Khuri, J. Lu, String solitons, Phys. Rep. 259 (1995) 213, hep-th/9412184.
[34] C. Vafa, Gas of D-branes and the Hagedorn density of BPS states, Nucl. Phys. B 463 (1996)
415, hep-th/9511088.
[35] W. Bucher, D. Friedan, A. Kent, Determinant formulae and unitarity for the N = 2
superconformal algebras in two dimensions, or exact results in string compactification, Phys.
Lett. B 172 (1986) 316.
[36] C. Johnson, A. Peet, J. Polchinski, Gauge theory and the excision of repulson singularities,
hep-th/9911161.
[37] J. Cardy, Operator content of two-dimensional conformally invariant theories, Nucl. Phys. B 270
(1986) 186.
[38] J. Polchinski, String Theory, Cambridge University Press, 1998, Chapters 7 and 14.
[39] K. Tod, All metrics admitting super-covariantly constant spinors, Phys. Lett. 121B (1983) 241.
[40] W. Israel, G. Wilson, A class of stationary electromagnetic vacuum fields, J. Math. Phys. 13
(1972) 865.
392
[41] Z. Perjs, Solutions of the coupled EinsteinMaxwell equations representing the fields of
spinning sources, Phys. Rev. Lett. 27 (1971) 1668.
[42] J. Hartle, S. Hawking, Solutions of the EinsteinMaxwell equations with many black holes,
Commun. Math. Phys. 26 (1972) 87.
[43] J. Polchinski, String Theory, Cambridge University Press, 1998, Chapter 2.
[44] L. Randall, R. Sundrum, An alternative to compactification, Phys. Rev. Lett. 83 (1999) 4690.
[45] C. Callan, J. Maldacena, A. Peet, Extremal black holes as fundamental strings, Nucl. Phys.
B 475 (1996), hep-th/9510134.
[46] J. Russo, L. Susskind, Asymptotic level density in heterotic string theory and rotating black
holes, Nucl. Phys. B 437 (1995) 611.
[47] J. York, Gravitational degrees of freedom and the initial value problem, Phys. Rev. Lett. 26
(1971) 1656.
[48] S. Ferrara, M. Porrati, V. Telegdi, g = 2 as the natural value of the tree level gyromagnetic ratio
of elementary particles, Phys. Rev. D 46 (1992) 3529.
[49] R. Jackiw, g = 2 as a gauge condition, Phys. Rev. D 57 (1998) 2635, hep-th/9708097.
[50] L. Eisenhart, Riemannian Geometry, Princeton University Press, 1997, 8th printing, p. 92.
[51] G. Papadopoulos, Rotating rotated branes, JHEP 9904 (1999) 014, hep-th/9902166.
[52] A. Chamseddine, W. Sabra, Metrics admitting Killing spinors in five dimensions, Phys. Lett.
B 426 (1998) 36, hep-th/9801161.

Regular BPS black holes: macroscopic and

microscopic description of the generating solution
Matteo Bertolini a,1,2 , Mario Trigiante b,3
a NORDITA, Blegdamsvej 17, DK-2100 Copenhagen , Denmark
b Department of Physics, University of Wales Swansea, Singleton Park, Swansea SA2 8PP, United Kingdom
Received 3 March 2000; accepted 3 April 2000
Abstract
In this paper we construct the BPS black hole generating solution of toroidally compactified string
(and M) theory, giving for it both the macroscopic and microscopic description. Choosing a proper
U -duality gauge the latter will be given by a bound state made solely of D-branes. The axionic nature
of the supergravity solution will be directly related to non-trivial angles between the constituent Dbranes (type IIB configuration) or, in a T -dual gauge, to the presence of magnetic flux on constituent
D-brane worldvolumes (type IIA configuration). As expected, the four-dimensional axion fields arise
from the dimensional reduction of non-diagonal metric tensor components or KalbRamond B field
components for type IIB or type IIA cases, respectively. Thanks to this result it is then now possible
to fill the full 56-dimensional U -duality orbit of N = 8 BPS black holes and to have a macroscopic
and microscopic description of all of them. 2000 Elsevier Science B.V. All rights reserved.
PACS: 04.65.+e; 04.50.+h; 11.25.-w
Keywords: BPS black holes; D-branes; Generating solution, N = 8 supergravity; Solvable Lie algebra
1. Introduction and summary of the results

In the last 4 years the study of black hole solutions in supergravity and string theory
has acquired a renewed interest. This is due, essentially, to some successful microscopic
entropy countings which have given, for the first time [14], a statistical interpretation of
the BeckesteinHawking entropy formula in the context of a consistent quantum theory
of gravity, such as string theory. The general properties of both supersymmetric and near
supersymmetric black holes have been systematically studied and many new results have
been obtained.
1 Supported by INFN, Italy.
2 teobert@nordita.dk, teobert@nbi.dk
3 m.trigiante@swan.ac.uk
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 1 6 - 9
394
M. Bertolini, M. Trigiante / Nuclear Physics B 582 (2000) 393406
Despite this progress, it is still not possible nowadays to have a complete control on
the common microscopic properties underlying these black holes. Many efforts have been
made in order to find a way of making these microscopic entropy countings more based
on first principles rather than on case by case computations (see for example [512])
and to understand which are the actual microscopic degrees of freedom underlying the
microscopic/macroscopic matching. However, a definite answer has not been found, yet.
In this respect, it would be useful to find a precise description, both at macroscopic and
microscopic level, of the so-called generating solution of regular BPS black holes obtained
in the framework of superstring (or M) theory compactified on the torus T 6 (T 7 ).
There are at least three known macroscopic generating solutions in the literature,
[1315]. However, their structure is too involved to allow a direct construction of the
corresponding microscopic configuration. On the other hand, it was proposed in [16] one
possible stringy (i.e., microscopic) structure for the generating solution but it was not
possible to find its field theory description as a solution of the relevant supergravity theory.
In this paper we fill the above gap and finally find the N = 8 BPS black holes
generating solution, giving for it both a macroscopic and a microscopic description in a
clear and simple way. As expected, the solution will depend on five independent parameters
which will have a precise physical meaning both at macroscopic and microscopic level.
Despite the number of independent parameters is five, the number of independent
harmonic functions entering the solution will be four. This is somewhat expected and
is related to the well known fact that within the five invariants of the U -duality group
of toroidally compactified M-theory, i.e., E7(7) , four are moduli-dependent and one is
moduli-independent (and proportional to the entropy). Another feature of the generating
solution is to have non-trivial axion fields switched on in it (this is related, as we shall
see, to the intrinsic property of the central charge eigenvalues evaluated on the solution
of being intrinsically complex, i.e., not all imaginary). On the microscopic side this will
correspond to have a configuration of D-branes intersecting at non-trivial angles (IIB) or, in
a T -dual picture, a D-brane configuration with non-trivial magnetic fluxes on the D-brane
worldvolumes (IIA). As expected, the non-trivial four-dimensional axions will come from
off-diagonal components of the ten-dimensional metric GMN in the type IIB case and of
the antisymmetric tensor BMN in the type IIA case, along the worldvolume of the D-branes
(and constant with respect to the worldvolume coordinates). Switching off the magnetic
flux one would recover a four parameter solution of D-branes orthogonally intersecting,
which turns out to be a pure dilatonic one.
Once a macroscopic solution it is given, its specific microscopic counterpart it is
not uniquely defined: it depends on how the solution is embedded in the original tendimensional theory. Our choice is to use two suitable RR embeddings recently defined
in [17], in order to have, as anticipated, a stringy configuration made solely of D-branes.
Relying on the structure of the superalgebra central charges and thanks to the geometrical
control of the embedding of the solution in the full type II string theory, it will be quite easy
to figure out the microscopic systems corresponding to our solution. Moreover, using the
tools developed in previous papers ([15,18,19] and especially [17]) we are able to generate,
from the above simple pure D-brane description, any other configuration, even pure NS
395
NS ones, i.e., those made of solely NS states (as fundamental strings, NS5-branes and KKstates). 4 The property for the entropy of being an U -duality invariant ensures that all these
U -dual configurations share the same entropy. And, as already stressed, the possibility of
having a control both at macroscopic and microscopic level of all these configurations,
could give some help in unraveling the very conceptual basis of the microscopic entropy
counting.
2. The macroscopic generating solution

Let us start with the macroscopic description of the generating solution. It has been
shown in [18] that the BPS black holes generating solution of toroidally compactified
type II string is also solution of a consistent truncation of the relevant N = 8 supergravity
effective theory, the so-called STU model. 5 This is a N = 2 effective model characterized
by the graviton multiplet and just three vector multiplets, each of them containing two
scalar fields. Therefore its full bosonic field content is: a graviton, four vector fields (i.e.,
eight charges, (p , q ) where = 0, 1, 2, 3) and three complex scalars zi = ai + ibi
spanning the special Khler manifold MSTU = [SL(2, R)/SO(2)]3 (bi being the dilatonic
fields and ai the axions). We are not going to describe the detailed structure of the model
since it has been described in a complete way in [15,19]. In fact, we will use the same
conventions and notations adopted in those papers. Let us then briefly summarize the
procedure to follow in order to derive the generating solution in the framework of the STU
model, leaving to the next section the discussion of its embedding in the N = 8 theory.
The BPS condition is equivalent to imposing the vanishing of the fermion supersymmetry
variation along the Killing spinor a direction:
fermions = 0,
0 a =
Z
ab b ,
|Z|
if a, b = 1, 2,
(2.1)
Z(z, z , p, q) being the N = 2 supersymmetry central charge. We adopt the following

anstze for the metric, Killing spinor, scalars and vector field-strengths:
x 2,
ds 2 = e2U (r) dt 2 e2U (r) dE
r 2 = xE 2 ,
a (x) = f (r)a ,
zi (x) = zi (r),
p
l (r)
x,
F (r) = 3 krs x k dx r x s 3 e2U (r) dt xE dE
2r
r
= 0, 1, 2, 3,
(2.2)
4 As shown in [20], the generating solution of heterotic black holes is also a generating solution for the type II
ones. It is the U -duality group which changes in the two cases and which specifies the U -duality properties of
the solution. In the heterotic case is U = SL(2) SO(6, 22) while in the present case, i.e., type IIA, type IIB or
M-theory compactified on tori (T 6 or T 7 , respectively), we have U = E7(7) .
5 BPS solutions of N = 2 supergravity have been studied, for instance, in [2127].
396
l being the moduli-dependent electric charges defined in [19]. From the BPS conditions (2.1) we may derive an equivalent system of first order equations for the scalars
and metric function U :
U (r)
e
dzi
?
= 2
hij j ? |Z(z, z , p, q)|,
dr
r2
U (r)
e
dU
=
|Z(z, z , p, q)|.
(2.3)
dr
r2
The explicit expression of the right hand side of Eqs. (2.3) is quite involved and may be
derived from the equations in [19] (computed

using the
p too restrictive condition Z = Z)
p
as far as the equations for the

by multiplying their right hand side by Z/Z or by Z/Z
scalars or the one for U are concerned, respectively. In order to characterize the generating
solution we need to compute the skew-eigenvalues {Z } = {Zi , Z4 } ( = 1, . . . , 4) of the
N = 8 central charge ZAB in terms of the supersymmetry and matter central charges
{Z, Zi } of the N = 2 model, using the following relations:
Z = iZ4 ,
Z i = hij j ? Z = Pi Z i
(i, i = 1, 2, 3),
(2.4)
where Pi = 2bi (r) is the vielbein transforming the rigid indices i (the one characterizing
i
the eigenvalues of the N = 8 central charge in its normal form) to the curved indices i
of the STU scalar manifold (see [19], Section 3, for details). Following [19], the explicit
expression of the N = 8 central charge eigenvalues in terms of quantized charges and
moduli is given in the appendix. The five U -duality invariants characterizing a generic
P
solution are the four norms |Z | and the overall phase = Arg(Z ), [28]. These
invariants may be combined to give four moduli-dependent invariants and one moduliindependent invariant (the quartic invariant of the U -duality group). The generating
solution is defined [15] as the BPS black hole solution depending on the least number of
parameters such that, on the point of the moduli space defining the boundary condition
at radial infinity of its scalar fields, the five invariants can assume all 5-plets of values
(consistent with the positivity condition of the quartic invariant). A necessary condition
for the generating solution is thus to depend only on five quantized charges, obtained from
the original 8 by suitably fixing the SO(2)3 gauge. Our choice for the gauge fixing is p0 =
q2 = q3 = 0. As an a posteriori check that the solution is a generating one, it is necessary to
verify that the five invariants computed in the corresponding are independent functions
of the five remaining charges: q0 , q1 , p1 , p2 , p3 .
With the above gauge choice the system of first and second order differential equations
simplifies considerably and the fixed values for the scalar fields (namely the values the
scalars get at the horizon, [29]) turn out to be the following ones:
s

q
1 q1 p1 2
p
q0 p1
1 1
,
b1fix =
,
a1fix =
2p2 p3
p2 p3 4 p2 p3
s

q
1 q1 2
q0 p2
1
fix
fix
,
b2 =
,
a2 =
2p3
p1 p3 4 p3
s
a3fix
q1
=
,
2p2
b3fix

1 q1 2
q0 p3
.
p1 p2 4 p2
Let us now introduce the following harmonic functions:
2pi
i
with i = 1, 2, 3,
H (r) = 1 +
r
2q0
and
H0 (r) = 1 +
r
q1
2q1
, g
,
H1 (r) = g +
r
p1 + p2
397
(2.5)
(2.6)
where the above value for the parameter g is fixed by supersymmetry (first order equation
for U ).
One can now see that the following anstze for the ai (r), the bi (r) and the scalar function
U(r):
s

H1 H 1 + gH 2
1 H1 H 1 gH 2 2
H0 H 1
,
b1 =
,
a1 =
2H 2 H 3
H 2H 3 4
H 2H 3
s

H1 H 1 gH 2
1 H1 H 1 gH 2 2
H0 H 2
,
b
=
,
a2 =
2
2H 1 H 3
H 1H 3 4
H 1H 3
s

H1 H 1 + gH 2
1 H1 H 1 gH 2 2
H0 H 3
,
b
=
,
a3 =
3
2H 1 H 2
H 1H 2 4
H 1H 2

2
1
1
,
(2.7)
U = ln H0 H 1 H 2 H 3 H1 H 1 gH 2
4
4
satisfies both the first and second order differential equations and hence is the solution
we were looking for. 6 The values of the constants characterizing each harmonic function
have been chosen in such a way to have (1) asymptotic flat space and (2) asymptotic unitary
values for the dilatons bi , so to have unitary radii of compactification. The corresponding
point in the moduli space at infinity is thus:

a = a2 = 0; a3 = g,
(2.8)
1
bi = 1.
Since the boundary values of the scalar fields at infinity define a bosonic vacuum of the
theory, they characterize also the microscopic configuration realizing our solution in the
opposite string coupling regime. In next section we shall describe two particularly simple
microscopic configurations corresponding to the choice of in Eq. (2.8).
Notice that the number of truly independent harmonic functions in Eq. (2.7) is four, as
expected, although the number of independent charges is five: q0 , q1 , p1 , p2 , p3 . Indeed,
from the conditions (2.6), one sees that H1 (r) = g(H 1 (r) + H 2(r) 1). Finally, according
to the anstze (2.2), the metric has the following form:
6 This result is consistent with the analysis in [27].
398

1/2 2
1
1 2 3
1
2 2
ds = H0 H H H H1 H gH
dt
4

2 1/2 2
1
H0 H 1 H 2 H 3 H1 H 1 gH 2
dE
x
4
2
(2.9)
and the macroscopic entropy, according to BeckensteinHawking formula, reads:

r
1
(2.10)
Smacro = 2 q0 p1 p2 p3 (q1 p1 )2 ,
4
which is the expected expression for the entropy of a generating solution, [13,14].
As anticipated, in order to check that the above solution is indeed a five parameters one,
one has to work out the expression of the N = 8 central charge skew-eigenvalues on .
From the explicit expressions of the field dependent central charge in the appendix and
from (2.8) it follows that:

q1 p1
1
+ i(q0 + p1 p2 p3 ) ,
Z1 ( , p, q) = 2
p1 + p2
2 2

q
1
1 p1
+ i(q0 p1 + p2 p3 ) ,
Z2 ( , p, q) = 2
2 2 p1 + p2
1
Z3 ( , p, q) = [0 + i(q0 p1 p2 + p3 )],
2 2
1
(2.11)
Z4 ( , p, q) = [0 + i(q0 + p1 + p2 + p3 )].
2 2
From the above equations it is clear that the 5 invariant quantities |Z ( , p, q)|,
( , p, q) are independent functions of the five charges q0 , q1 , p1 , p2 , p3 .
Switching off the fifth parameter q1 our solution becomes exactly the four parameters
one studied in [17] (indeed H1 (r)|q1 =0 = 0 and g(q1 = 0) = 0). Indeed putting q1 to zero
the central charges Z become pure imaginary and the axion fields vanish uniformly.
3. The microscopic description of the generating solution

The microscopic counterpart of a four dimensional macroscopic solution is of course
not uniquely defined. Indeed it depends on the interpretation of the four-dimensional fields
describing the supergravity solution in terms of dimensionally reduced ten-dimensional
ones. In the light of the analysis put forward in [32] and then completed in [17] 7 , the
microscopic interpretation of our generating solution can be uniquely defined in terms of
the embedding of the STU model within the original N = 8 theory (in particular of the
embedding of the SLA generating the STU scalar manifold inside the SLA parametrized
by the 70 scalars of the N = 8 theory). In [17], two main classes of embeddings of the STU
model were defined (the embeddings within each class being related by S-, T -dualities):
7 Among the main goals of this analysis is the geometrical characterization, using solvable Lie algebra (SLA)
techniques, of the scalar and vector fields in the N = 8 theory in terms of type IIA and IIB fields.
399
one in which the vector fields derive from NSNS ten-dimensional forms and the other
in which the vector fields have a RR origin (and the scalar fields are NSNS, see in
particular Section 2 of that paper). In the latter class two representative embeddings were
considered: if we denote by x 4 , x 5 , . . . , x 9 the coordinates of T 6 and by x 0 , x 1 , x 2 , x 3
the non-compact spacetime coordinates, in one embedding, which was characterized from
the type IIB point of view, all the three axions of the model come from metric tensor
components (G45 , G67 , G89 ), as opposite to the other, characterized from a type IIA point
of view, in which all the axions come from B field components (B45 , B67 , B89 ). In both
cases the three dilatons are related to three combinations of the radii of the torus. The two
embeddings are related by an operation of T -duality on the compact directions x 5 , x 7 , x 9 .
As we shall show in the sequel these two embeddings provide an interpretation of the
generating solution (2.7) in terms of two T -dual microscopic configurations: a system of
D3-branes at angles (type IIB embedding) and a system of D0- and D4-branes (type IIA
embedding) with a magnetic flux in the worldvolume of the latter (giving therefore extra
D2 and D0 charge, [33]). The magnetic flux (or, equivalently, the non-trivial angle in the
dual type IIB configuration) will be the microscopic extra degree of freedom related to the
fifth parameter q1 characterizing the supergravity solution.
Analyzing the two embeddings from a SLA point of view, one can deduce, in the same
way as it was done in [17] for the type IIA case (see in particular Eqs. (3.14) and (3.15)
of that paper), the subset of the weight basis of the 56 of E7(7) in terms of which the
magnetic y n () (n = 0, . . . , 3) and electric xn () dressed charges 8 of the two STU model
truncations are expressed. According to [17], the symplectic vector of dressed charges
(y n , xn ) is defined in the following way:

n
n
p
y ()
= L1 ()
,
(3.1)
qn
xn ()
where denotes a point in the scalar manifold and L() is the coset representative of the
scalar manifold computed in the same point. As far as the type IIA embedding is concerned,
a basis of weights for (y n , xn ) was found in [17], and from Table 3 of the same work the
correspondent RR vectors may be read off:
type IIA:
(y n ) (A456789, A6789 , A4589, A4567),
(xn ) (A , A45 , A67 , A89 ).
(3.2)
By performing a T -duality along x 5 , x 7 , x 9 according to the geometric recipe given in [17],

we may find the corresponding weights for the type IIB embedding and read from Table 3
of the same work their RR interpretation:
type IIB:
(y n ) (A468 , A568 , A478, A469 ),
(xn ) (A579 , A479 , A569, A578 ).
(3.3)
8 The dressed charges are the physical charges of the interacting theory, which take into account the dressing
of the D-brane naked charges provided by the moduli.
400
Now let us consider our generating solution and compute the dressed charges on the point
of the moduli space defined in Eq. (2.8). Implementing Eq. (3.1), one finds:

y 0 , y 1 , y 2 , y 3 = (0, p1 , p2 , p3 ),

p1 q1
p1 q1
,
,0 .
(3.4)
(x0 , x1 , x2 , x3 ) = q0 ,
p1 + p2 p1 + p2
From the above expressions we may deduce consistent microscopic configurations
corresponding to the generating solution with the chosen boundary condition on the scalar
fields at infinity . From the type IIB viewpoint we may think of a system of D3-branes
intersecting at non-trivial angles, but in such a way to preserve 1/8 supersymmetry; this
can be achieved if the relative rotation between each couple is a SU(3) rotation, [35]. The
configuration is depicted in Table 1.
Using Eqs. (3.3) the first three sets of branes (N0 , N1 , N2 ) may be associated with the
charges x0 , y 1 , y 2 , respectively, i.e., with charge along the 3-cycles (579), (568), (478),
while the fourth set, N4 , with y 3 , i.e., the charge along (469). In fact, due to the nontrivial angle , the fourth set of N3 branes induces D3-brane charge on the cycles (579)
(contributing to x0 ), (479) (represented by x1 ) and (569) (represented by x2 ).
As far as the type IIA microscopic interpretation is concerned, we may consider the
configuration of D0- and D4-branes obtained by T -dualizing the type IIB one described
above along the directions x 5 , x 7 , x 9 . The corresponding system may be deduced from
Eqs. (3.2) and (3.4) and consists of a set of coinciding D0-branes with electric charge
x0 (the minus sign is required by consistency with the construction in [17] and will
be discussed in the sequel) and three sets of coinciding D4-branes along the four-cycles
(6789), (4589) and (4567) with magnetic charges y 1 , y 2 , y 3 . In addition there is a magnetic
flux (related to the angle in the T -dual type IIB configuration, [37,38]) switched on the
worldvolume of the latter brane (i.e., along (4567)). This flux induces an effective D0
charge (contributing to x0 ) and effective D2 charges along the two-cycles (45) and (67)
(represented by x1 and x2 , respectively). The presence of this flux is also consistent with
the fact that the axions in the type IIA embedding are interpreted as coming from the
Table 1
The position of the D3-branes on the compactifying torus; i is the angle on the (x 2i+2 , x 2i+3 )
torus and is a generic non-trivial angle. For each couple of constituent D3-branes, it follows
P
()
()
that 3i=1 (i i ) = 0 mod 2 , = 1, . . . , 4, this ensuring that it is a configuration of 4
(bunches of) D3-branes at SU(3) angle, [35,36]. Notice that the above configuration has been chosen
in such a way that setting = 0 one recovers a four parameters solution, namely 4 bunches of D3branes orthogonally intersecting
N0
N1
N2
N3
/2
/2
/2
0
/2

/2
0
/2
401
BMN tensor in ten dimensions. Indeed, let us briefly recall the general argument relating
the presence of a flux on one D4-brane with an effective D2-brane charge (electric in our
framework) and a non-trivial Bij background field. As well known, B field components
enter non-trivially in the Dp-brane action via the WZ term:
Z

(3.5)
C eF p+1 ,
p
Wp+1
b (B
b being the pullback of
where F is the gauge invariant combination F = 2 0 F + B
the B field). Hence, from the supergravity point of view, one would indeed expect new
charges representing extra D(p 2) effective charges at the microscopic level as well
as non-trivial bulk B field components in the solution (see for instance [39]). In fact,
this is precisely what we get. As shown for instance in [36,41,42], for a suitable choice
of the flux (which, albeit giving smaller brane charges via worldvolume ChernSimons
coupling, modifies the supersymmetry projections imposed by the D-brane background)
the above configuration can preserve 1/8 of the original supersymmetry. Consider the D4brane configuration described above in the general situation in which the magnetic fluxes
are non vanishing on all the three planes (45), (67), (89) (i.e., F45 , F67 , F89 6= 0). From
Eq. (3.5) we may deduce the effective (electric) D2- and D0-brane charges. For instance,
the effective D2-brane charge along (45) (which is the electric dual object of the D4-brane
wrapped on (6789)) is:
!
Z
Z
1
Tr F67 + Tr F89 ,
(3.6)
# of D2-brane (along cycle 45) =
2
T67
T89
which in our conventions is represented by the electric dressed charge x1 . Similarly, we

may compute the effective D2 charges along the other two-cycles. Notice that the only
non-vanishing components of F can only be F45 , F67 , F89 (a F56 6= 0 component would
imply, for instance, a new 4-dimensional magnetic effective charge out of the four at
disposal in the central charge normal gauge), and the three axions come precisely from
those components of the B field.
On our particular solution the D2-brane charge along (89), i.e., x3 , is zero, while (45)
and (67) charges are opposite one to each other (x1 = x2 ), see (3.4). In fact, considering
Eq. (3.6) written also for the other 2-cycles, one can easily see that switching on a magnetic
flux, as we do, only on the D4-branes lying along (4567) and not on the other two bunches
of D4-branes, is consistent with having no D2-brane charge along (89), i.e., x3 = 0, and
opposite D2-brane charge along (45) and (67), i.e., x1 = x2 . These charges turn out to
be proportional to the fifth parameter q1 : sending q1 to zero the fluxes vanish together with
the B fields (axions) and we recover the four parameter solution discussed in [17].
Let us now come to our final goal that is to make the macroscopic/microscopic
correspondence precise, namely to give the precise matching between the parameters
characterizing the microscopic and the macroscopic configurations, respectively:
N0 , N1 , N2 , N3 , q0 , q1 , p1 , p2 , p3 .
402
The type IIA and IIB D-brane configurations discussed above as the microscopic
counterparts of our generating solution, were suggested in [16] as candidates for the
microscopic representation of the 5-parameter solution, whose macroscopic description
was then missing. Let us focus for the moment on the type IIA embedding. In order to
make contact with this literature, let us use an equivalent representation of the dressed
charges, related to the central charges by an SO(8) 9 transformation:

1
zij = xij + iy ij = AB ij ZAB ,
2
(3.7)
where the couple (ij ) indicizes the two times antisymmetric representation of SO(8). The
real and imaginary parts of zij are the N = 8 electric and magnetic dressed charges in the
basis of weights of the 56 of E7(7) defined in [17] and listed in Table 3 of the same paper.
When ZAB is skew-diagonal the matrix ( AB )ij has the form:
1 1 1 1
1 1 1 1
(3.8)
=
1 1 1
1
1 1 1 1
the electric (x) and magnetic (y) charges are non vanishing only for (ij ) equal to (12),
(34), (56) and (78). From Eqs. (2.11) and (3.7) we may read off the values of the charges
xij and y ij :
1
x78 + iy 78 (Z1 + Z2 + Z3 + Z4 ) = 0 + iq0 ,
2
1
q1 p1
ip1 ,
x12 + iy 12 (Z1 Z2 Z3 + Z4 ) =
p1 + p2
2
1
q1 p1
ip2 ,
x34 + iy 34 (Z1 + Z2 Z3 + Z4 ) =
p1 + p2
2
1
x56 + iy 56 (Z1 Z2 + Z3 + Z4 ) = 0 ip3 .
2
(3.9)
The relation between this representation of the dressed charges and the one in Eq. (3.4),
which is related to the choice of the STU model in which the generating solution has been
worked out, is the following:

12 34 56 78 1 2 3
= y , y , y , x0 ,
y ,y ,y ,y

(3.10)
{x12 , x34 , x56, x78 } = x1 , x2 , x3 , y 0 ,
where symplectic transformation x0 = y 78 , y 0 = x78 , as discussed in [17], is related to
the feature of our STU model (both in type IIA and type IIB cases) of being embedded
non-perturbatively in the larger N = 8 theory and therefore is required in order for the
truncation to be consistent. 10 The actual D0-brane effective charge is thus y 78 = q0 .
9 The subgroup of SU(8) which does not mix electric and magnetic charges.
10 Indeed the dimensionally reduced RR vector A in the STU model is an electric potential, while it is
magnetic in the N = 8 from the type IIA viewpoint, see Table 3 of [17].
403
Table 2
The correspondence between type IIB and type IIA charges on the different cycles of the
compactifying torus. Notice how a 6= 0 contribution induces D2-brane and D0-brane effective
charges while for = 0 one gets a four parameter configuration
Type IIB D3-branes
Charge
Type IIA D-branes
3-brane(468)
3-brane(568)
3-brane(478)
3-brane(469)
3-brane(579)
3-brane(479)
3-brane(569)
3-brane(578)
0
N1
N2
N3 cos2
N0 + sin2 N3
sin cos N3
sin cos N3
0
6-brane
4-brane(6789)
4-brane(4589)
4-brane(4567)
0-brane
2-brane(45)
2-brane(67)
2-brane(89)
y0
y1
y2
y3
x0
x1
x2
x3
x78
y 12
y 34
y 56
y78
x12
x34
x56
The precise correspondence between the dressed charges (in the two representations)
and the parameters associated with the microscopic configurations previously discussed
(that is those characterizing the type IIB configuration of Table 1: N0 , N1 , N2 , N3 , ) is
represented in Table 2.
Finally, according to relations (3.4) or (3.9) and Table 2 we finally get the precise
macroscopic/microscopic correspondence: 11
N0 = q0
(q1 p1 )2
,
p3 (p1 + p2 )2
N3 cos2 = p3 ,
tan =
N1 = p1 ,
N2 = p2 ,
q1 p1
.
p3 (p1 + p2 )
(3.11)
Through Eqs. (3.11) all the microscopic parameters, namely N0 , N1 , N2 , N3 and the
angle , are expressed in terms of the quantized charges (p , q ) characterizing the
macroscopic solution and this finally allows us to characterize quantitatively its
microscopic structure. An alternative microscopic system which could reproduce our
generating solution can be given from the M-theory point of view and consists of three
M5-branes with magnetic flux on their worldvolumes and intersecting on a (compact) line
along which N0 units of momentum has been put.
With the above definitions, the expression of the E7(7) quartic invariant J4 in the
(y ij , xij ) basis [43,44], is:

2
J4 = 4 x78x12 x34 x56 + y 78y 12 y 34 y 56 x78y 78 + x12 y 12 + x34y 34 + x56 y 56
+ 4 x78 y 78 x12y 12 + x78 y 78 x34y 34 + x78 y 78 x56y 56 + x12 y 12 x34y 34
11 Our normalizations are the following. In general the 4-dimensional charge of a wrapped Dp-brane is Q =
p
p Vp / V6 where p = 2 (2 0 )3p is the normalized Dp-brane charge density in ten dimensions.
Provided the asymptotic values of the dilatons bi (r), which parameterize the radii of the compactifying torus
and which has been taken to be unitary (see previous section), it turns out that, in units where
0 = 1, the four
dimensional fundamental quanta of charge for any kind of (wrapped) Dp-brane is equal to 2 and our quantized
charges (p , q ) have been taken in units of that quanta, i.e., they are integer valued.
404
+ x12y 12x56 y 56 + x34 y 34x56 y 56
(3.12)
and consequently, upon use of Table 2, one can easily work out the expression of the
entropy S = J4 written in terms of the microscopic parameters:

q

(3.13)
Smicro = 2 cos2 N0 N1 N2 N3 14 sin2 N32 (N1 N2 )2 .
A derivation of the above formula via microscopic counting techniques should be
performed extending the analysis of [5,11] to tori (also the results of [45] could possibly
shed some light in this direction). However, we do not try to perform it here. Let us just
notice that for = 0 one recovers the usual entropy of the four parameters solution whose
derivation via microscopic counting has been carried out, for instance, in [41,42,46,47].
4. Discussion
In the present paper we have worked out the generating solution of four-dimensional
N = 8 BPS black holes in a form which could be easily described, applying the results
of [17], in terms of pure D-brane configurations upon toroidal compactification of string
(or M) theory. As a result we were able to pinpoint the precise correspondence between
the microscopic parameters characterizing one of these configurations and the supergravity
parameters entering the macroscopic description of the solution.
The relevance of this achievement relies on the possibility on one hand to reconstruct
the whole 56-parameter U -duality orbit of N = 8 BPS black holes, by acting on our
solution by means of E7(7) transformations, and on the other hand to study in a precise
fashion the action of dualities on their corresponding microscopic realizations. Starting
from the type IIA configuration described in the previous section and performing a T duality transformation on the whole T 6 , one ends up, for instance, with a configuration
made of N0 D6-branes, 3 bunches of (N1 , N2 , N3 ) D2-branes along the planes (45),
(67), (89) plus effective D4-brane charge. But, more generally, we may also unravel the
microscopic properties of pure NSNS black hole solutions in the same orbit, starting
from the corresponding embedding of the ST U model defined in [17], or even of mixed
NSNS/RR solutions. The important point is that having now both a macroscopic and
a microscopic description of the generating solution one can follow its trasformation
throughout the full U -duality orbit.
In this respect a challenging problem is to recover the expression of Eq. (3.13) from
a microscopic entropy counting point of view, performed on the corresponding D-brane
configuration. Knowing how to act on it by means of U -duality can help to shed some
light on the actual microscopic degrees of freedom of general BPS black holes, since the
generating solution encodes, by definition, all of them. This project is left for future work.
Acknowledgements
We would like to thank M. Bill, P. di Vecchia, P. Fr, T. Harmark and N. Obers for
discussions and R. Russo and C. Scrucca for useful email correspondence. We are also
405
grateful to the organizers of the TMR Torino school on String Theory and Branes physics
during which this work has been brought to an end. We acknowledge partial support by
ECC under contracts ERBFMRX-CT96-0045 and ERBFMRX-CT96-0012.
Appendix A
The general expression of the N = 8 central charge eigenvalues Z in terms of the scalar
fields and the charges characterizing the STU model can be worked out making explicit the
first order equations (2.3) taking into account relations (2.4). Following [19] we have the
following:
1
Re Z1 =
2 2b1b2 b3
b1 q1 b2 q2 b3 q3 + (a2 a3 b1 a1 a3 b2 a1 a2 b3 b1 b2 b3 )p0

+ (a3 b2 + a2 b3 )p1 + (a1 b3 a3 b1 )p2 + (a1 b2 a2 b1 )p3 ,
1
Im Z1 =
2 2b1b2 b3
a1 q1 + a2 q2 + a3 q3 + (a1 a2 a3 + a3 b1 b2 + a2 b1 b3 a1 b2 b3 )p0

(a2 a3 b2 b3 )p1 (a1 a3 + b1 b3 )p2 (a1 a2 + b1 b2 )p3 + q0 ,
Re Z2 = (1, 2, 3) (2, 1, 3),
Im Z2 = (1, 2, 3) (2, 1, 3),
Re Z3 = (1, 2, 3) (3, 2, 1),
Im Z3 = (1, 2, 3) (3, 2, 1),
1
Re Z4 =
2 2b1b2 b3
b1 q1 + b2 q2 + b3 q3 + (a2 a3 b1 + a1 a3 b2 + a1 a2 b3 b1 b2 b3 )p0

(a3 b2 + a2 b3 )p1 (a3 b1 + a1 b3 )p2 (a2 b1 + a1 b2 )p3 ,
1
Im Z4 =
2 2b1b2 b3
a1 q1 + a2 q2 + a3 q3 + (a1 a2 a3 a3 b1 b2 a2 b1 b3 a1 b2 b3 )p0

(a2 a3 b2 b3 )p1 (a1 a3 b1 b3 )p2 (a1 a2 b1 b2 )p3 + q0 ,
(A.1)
where it is meant that all axions and dilatons are r-dependent, i.e., ai = ai (r) and bi =
bi (r).
References
[1]
[2]
[3]
[4]
A. Strominger, C. Vafa, Phys. Lett. B 379 (1996) 99.

J. Maldacena, A. Strominger, Phys. Rev. Lett. 77 (1996) 428.
C.V. Johnson, R.R. Khuri, R.C. Myers, Phys. Lett. B 378 (1996) 78.
D. Kaplan, D. Lowe, J. Maldacena, A. Strominger, Phys. Rev. D 55 (1997) 4898.
406
[5] J. Maldacena, A. Strominger, E. Witten, JHEP 12 (1997) 2.

[6] A. Strominger, JHEP 02 (1998) 009.
[7] P. Claus, M. Derix, R. Kallosh, J. Kumar, P. K. Townsend, A. Van Proeyen, Phys. Rev. Lett. 81
(1998) 4553.
[8] J.R. David, G. Mandal, S.R. Wadia, Nucl. Phys. B 544 (1999) 590.
[9] S. Carlip, Phys. Rev. Lett. 82 (1999) 2828.
[10] G.W. Gibbons, P.K. Townsend, Phys. Lett. B 454 (1999) 187.
[11] G. Lopes Cardoso, B. de Wit, T. Mohaupt, Macroscopic entropy formulae and non-holomorphic
corrections for supersymmetric black holes, hep-th/9906094, Area law corrections from state
counting and supergravity, hep-th/9910179.
[12] F. Larsen, E. Martinec, Currents and moduli in the (4, 0) theory, hep-th/9909088.
[13] M. Cvetic, A. Tseytlin, Phys. Rev. D 53 (1996) 5619; Phys. Rev. D 55 (1997) 3907, Erratum.
[14] M. Cvetic, D. Youm, Nucl. Phys. B 472 (1996) 249.
[15] M. Bertolini, P. Fr, M. Trigiante, Class. Quant. Grav. 16 (1999) 2987.
[16] V. Balasubramanian, How to count the states of extremal black holes in N = 8 supergravity,
hep-th/9712215.
[17] M. Bertolini, M. Trigiante, Regular RR and NSNS BPS black holes, hep-th/9910237.
[18] L. Andrianopoli, R. DAuria, S. Ferrara, P. Fr, M. Trigiante, Nucl. Phys. B 509 (1998) 463.
[19] M. Bertolini, P. Fr, M. Trigiante, Class. Quant. Grav. 16 (1999) 1519.
[20] M. Cvetic, C.M. Hull, Nucl. Phys. B 480 (1996) 296.
[21] M. Shmakova, Phys. Rev. D 56 (1997) 540.
[22] K. Behrndt, D. Lust, W.A. Sabra, Phys. Lett. B 418 (1998) 303.
[23] K. Behrndt, M. Cvetic, W.A. Sabra, Phys. Rev. D 58 (1998) 084018.
[24] S. Ferrara, R. Kallosh, Phys. Rev. D 54 (1996) 1514.
[25] M.J. Duff, J.T. Liu, J. Rahmfeld, Nucl. Phys. B 459 (1996) 125.
[26] K. Behrndt, R. Kallosh, J. Rahmfeld, M. Shmakova, W.K. Wong, Phys. Rev. D 54 (1996) 6293.
[27] K. Behrndt, G.L. Cardoso, B. de Wit, D. Lust, T. Mohaupt, W.A. Sabra, Phys. Lett. B 429 (1998)
289.
[28] S. Ferrara, J. Maldacena, Class. Quantum Grav. 15 (1998) 749.
[29] S. Ferrara, R. Kallosh, A. Strominger, Phys. Rev. D 52 (1995) 5412.
[30] A. Strominger, Phys. Lett. B 383 (1996) 39.
[31] S. Ferrara, R. Kallosh, Phys. Rev. D 54 (1996) 1525.
[32] L. Andrianopoli, R. DAuria, S. Ferrara, P. Fr, R. Minasian, M. Trigiante, Nucl. Phys. B 493
(1997) 249.
[33] M. Douglas, Branes within branes, hep-th/9512077.
[34] M. Green, J.A. Harvey, G. Moore, Class. Quant. Grav. 14 (1997) 47.
[35] N. Ohta, P.K. Townsend, Nucl. Phys. Lett. B 418 (1998) 77.
[36] V. Balasubramanian, R. Leigh, Phys. Rev. D 55 (1997) 6415.
[37] M. Li, Nucl. Phys. B 460 (1996) 351.
[38] C.G. Callan, I.R. Klebanov, Nucl. Phys. B 465 (1996) 473.
[39] J.C. Breckenridge, G. Michaud, R.C. Myers, Phys. Rev. D 55 (1997) 6438.
[40] M.S. Costa, G. Papadopoulos, Nucl. Phys. B 510 (1998) 217.
[41] V. Balasubramanian, F. Larsen, Nucl. Phys. B 478 (1996) 199.
[42] V. Balasubramanian, F. Larsen, R. Leigh, Phys. Rev. D 57 (1998) 3509.
[43] E. Cartan, O Euvres completes, Paris, Editions du Centre national de la Recherche Scientifique,
1984.
[44] R. Kallosh, B. Kol, Phys. Rev. D 53 (1996) 5344.
[45] M.S. Costa, M.J. Perry, Nucl. Phys. B 524 (1998) 333; Nucl. Phys. B 520 (1998) 205.
[46] K. Behrndt, T. Mohaupt, Phys. Rev. D 56 (1997) 2206.
[47] J. Maldacena, Phys. Lett. B 403 (1997) 20.

Transverse dynamics of hard partons in nuclear

media and the QCD dipole
Urs Achim Wiedemann
Theory Division, CERN, CH-1211 Geneva 23, Switzerland
Abstract
We derive the non-abelian generalization of the Furry approximation which describes the
transverse dynamical evolution of a hard projectile parton inside a spatially extended colour target
field. This provides a unified starting point for the target rest frame description of the nuclear
dependence of a large class of observables. For the case of the virtual q q photoabsorption cross
section, we investigate then in detail under which conditions the nuclear dependence encoded in the
Furry wavefunctions can be parametrized by a q q QCD dipole cross section. The important condition
is colour triviality, i.e., the property that for arbitrary N-fold rescattering contributions the only nonvanishing colour trace is Nc CFN . We give proofs for the colour triviality of the inelastic, diffractive
and total photoabsorption cross section measured inclusively or with one jet resolved in the final
state. Also, we list examples for which colour interference effects remain. Colour triviality allows us
to write the q q contribution to the DIS nuclear structure function F2 for small Bjorken xBj in
terms of a path integral which describes the transverse size evolution of the q q pair in the nuclear
colour field. This expression reduces in an opacity expansion to the N = 1 result of Nikolaev and
Zakharov, and in the eikonal approximation to the Glauber-type rescattering formulas first derived
by Mueller. In the harmonic oscillator approximation of the path integral, we quantify deviations
tot L which relate the
from the eikonal limit. Their onset is characterized by the scales L/ lf and E
longitudinal extension L of the nuclear target to the coherence length lf and the total transverse
tot accumulated by the q-q-pair.
energy E

PACS: 12.38.Mh; 24.85.+p; 25.75.-q
Keywords: Colour dipole; Furry approximation; Colour triviality; Opacity expansion
1. Introduction
The partonic interpretation of physical processes and their nuclear dependence is
Lorentz frame dependent. DrellYan, e.g., is a q-q -fusion in the infinite momentum
frame [1], but it becomes a -bremsstrahlung radiation off the projectile quark if viewed
in the target rest frame [2,3]. Similarly, nuclear shadowing is due to the recombination
of partons from different nucleons, if viewed in the infinite momentum frame [4]. In the
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 8 6 - 8
410
U.A. Wiedemann / Nuclear Physics B 582 (2000) 409450
target frame, it arises from non-additive contributions of the rescattering of the hadronic
-Fock states in a spatially extended medium [59]. The descriptions in different Lorentz
frames are equivalent, of course. But depending on the physical problem at hand, a wellchosen Lorentz frame may provide a particularly simple partonic interpretation. To study
the nuclear dependence of physical processes, the target rest frame provides arguably the
most intuitive picture. QCD factorization theorems can be expected to hold for the nuclear
dependence of high-pt processes only [1012], and thus a large variety of other approaches
to the nuclear dependence of hard processes exists in the literature [1,6,1325]. These
exploit that in the target rest frame, the nuclear dependence of a physical process P with
transition amplitude hi |P |f i can be attributed often to the multiple rescattering of the
hard in- and outgoing partons inside the soft spatially extended target colour field. In the
simplest cases, this rescattering effect can be described by multiplying the free in- and
outgoing wavefunctions with straight eikonal Wilson lines which account for the leading
medium-induced colour rotation of projectile quarks [1,26].
In the present work, we derive and study a more refined approximation scheme for inand outgoing wavefunctions i and f which includes the leading transverse dynamical
evolution of the projectile partons inside the target. To this end, we derive in Section 2
explicit expressions for i and f which approximate to leading order 1/E in the norm
and next to leading order in the phase the solution of the Dirac equation in the presence of a
spatially extended colour field. The 1/E-corrections included in these solutions are known
to provide the leading contribution for observables which are essentially determined by the
destructive interference between different production amplitudes as, e.g., the non-abelian
LandauPomeranchukMigdal (LPM) effect [1322] or the nuclear dependence of Drell
Yan pair production [3,21]. Indeed, our derivation of i and f draws on exactly the
same approximation schemes which were used in recent studies [13,27] of the nuclear
dependence of these observables. In Section 2, however, we discuss this rescattering
effect without reference to a particular observable. In this way, we obtain a compact
expression which turns out to be the non-abelian generalization of the abelian Furry
approximation [21,22] and which may serve as building block in the calculation of very
different nuclear dependencies.
The main technical complication in working with non-abelian Furry wavefunctions is
that they involve a path-integral over a path-ordered non-abelian Wilson line. In Sections 3
and 4, we show for the example of the q q photodissociation process how explicit and
exact calculations can be done despite this complication. The key step is a diagrammatic
technique first used by Mueller and collaborators [5,1619,28] which allows to establish
the colour triviality of certain cross sections involving N -fold rescattering processes.
Here, colour triviality means that the contribution to a medium-dependent observable to
N th order in the opacity involves only colour traces which reduce to the N th power of
the Casimir. This renders the problem essentially abelian. In Section 4.3, we derive the
corresponding diagrammatic identities for our configuration space formulation of partonic
rescattering. We then use these identities in explicit proofs of the colour triviality of the
inelastic and diffractive part of the inclusive and one-fold differential (i.e., one jet resolved)
photoabsorption cross section. At least for the q q photodissociation cross section,
411
all colour trivial observables turn out to be given in terms of the transverse dynamical
evolution of a QCD dipole which is described by a simple path integral K. We discuss the
eikonal limit in which the transverse dynamical evolution is neglected, and we quantify the
leading corrections to this limiting case.
We choose for all explicit calculations the photodissociation process q q mainly
since it is the simplest example which allows to illustrate the main technical difficulties
associated with the non-abelian Furry approximation and colour interference effects. The
q q process is a quantitatively significant contribution to the nuclear structure
function F2 , but phenomenological applications require the inclusion of processes with
initial gluon radiation (e.g., q qg).
These are known to contribute in the aligned jet

region, where the transverse q q separation is large, to leading order in 1/Q2 . Discussion of
these radiative contributions involves the quarkgluon vertex and complicates the analysis
of colour interference effects significantly. This lies beyond the scope of a first illustration
of the non-abelian Furry approximation, given here. We shall address the corresponding
additional technical problems in a subsequent work where we study colour interference in
the non-abelian LPM-effect. In the present paper, we discuss only two aspects of the very
general question to what extent our approach can be applied to other or more exclusive
observables or to other models of the medium: First, we give in Sections 4.2 and 4.5
examples of (more exclusive) photoabsorption processes which are not colour trivial, thus
indicating the limitations of the approach advocated here. Second, we argue in Section 5
that the specific model ansatz for in-medium rescattering is not instrumental for our results.
In the Conclusions, we summarize our main results and we shortly comment on further
perspectives.
2. Asymptotic wavefunctions for rescattering particles

We want to calculate the nuclear dependence of some physical process P with transition
amplitude hi |P |f i by describing how the medium affects the propagation of the in- and
outgoing wavefunctions i , f . For the case of the photodissociation process q q
depicted in Fig. 1(a) this means that we write the corresponding transition amplitude in the
form
Z
(2.1)
hi |P |f i = ie d4 x u (x, p1 ) 0 ei kx v (x, p2 ).
Here, the physical process P is determined by the photonquark vertex ie eikx . In this
section, we derive explicit expressions for the medium-dependence of these asymptotic
wavefunctions irrespective of a particular process. Section 2.1 explains the solution of
the corresponding abelian problem, Section 2.2 establishes the non-abelian extension.
In the remaining sections of this paper, we shall discuss then for the example of the
photodissociation (2.1) how explicit calculations can be done.
412
Fig. 1. (a) Contribution to the photodissociation process q q with N - and M-fold final state
rescattering for the outgoing quark and antiquark. (b) Sum over rescattering diagrams for an outgoing
antiquark which leaves the production vertex with momentum p1 and colour a1 . To leading order in
opacity, this resums all rescattering effects and defines the non-abelian Furry wavefunction (2.12).
2.1. Abelian Furry approximation

We consider a relativistic positron with momentum (E2 , p2 ) (E2 m). Prior to its
detection, this positron undergoes multiple small-angle scattering in a spatially extended
medium, described, e.g., by a collection U (x) of single scattering potentials i (x) = (x
x i ) localized at spatial positions x i ,
U (x) =
(x x i ).
(2.2)
i=1
The asymptotic electron wavefunction is a solution to the Dirac equation in this spatially
extended field:

(2.3)
i U (x) m 0 + i (x, p2 ) = 0.
t
Which approximation for (x, p2 ) keeps the leading medium-dependence? An expansion
of in powers of the coupling constant does not since it amounts to an expansion in
powers of the single scattering potential rather than containing the leading order effect
of U . In QED, the leading U -dependence is kept in the Furry approximation [21,22] which
is a high energy expansion of the solution of the Dirac equation. For the outgoing positron
wavefunction, it reads
b2 F (x, p2 ) v(p2 ).
F (x, p2 ) = eiE2 t ip2 z D
(2.4)
413
bi denotes a differential operator

which is exact to order O(U/E) and O(1/E 2). Here, D
bi = 1 i (pi npi ) ,
D
2 Ei
2 Ei
= 0 , z = n x, pi = |pi |,
(2.5)
and the unit vector n specifies the longitudinal direction. The differential operator acts
on the transverse wavefunctions F . For very late times, i.e., for far forward longitudinal
distances xL , this wavefunction satisfies plane wave boundary conditions
(
)
)2
(p
2
xL ,
(2.6)
F (x , xL , p2 ) = exp ip
2 x + i
2 p2
Z
(2.7)
F (y , yL , p2 ) = dx G(y , yL ; x , xL |p2 )F (x , xL , p2 ).
Its evolution to finite longitudinal distances is determined by the retarded Greens
function G whose path integral representation reads (for z0 > z)
0
G r, z; r , z | p =
(Zz0
)

ip 2
r ( ) iU r( ),
Dr( ) exp
d
.
2
(2.8)
Here, r = dr/d and G satisfies the boundary conditions r(z) = r, r(z0 ) = r0 , with
G(r, z; r0 , z0 = z|p) = (r0 r). The Greens function (2.8) describes the Brownian motion
of the projectile particle in the plane transverse to the beam. This is the leading medium
effect on the propagation of the projectile. Starting from the abelian Furry wavefunction
(2.4), the KST-formalism [21,22] then allows to describe, e.g., the medium dependence of
the LPM-bremsstrahlung spectrum [21,22]. One of the main motivations for what follows
is the question to what extent the application of the same approach to non-abelian problems
is justified.
2.2. Non-abelian Furry approximation
To find the non-abelian generalization of the Furry approximation, we consider a
spatially extended static colour potential of the form
A (x) = 0
ia (x) T a ,
(2.9)
i=1
ia (x) = (x x i ) a ai ,
(2.10)
where T a (a = 1, . . . , Nc2 1), denote the generators of the SU (Nc ) colour representation
of the projectile parton. The ith scattering center is located at x i and exchanges a specific
colour charge a = ai .
For the spatial support of the potentials ia we take the rapid fall-off of a Yukawa
potential with some Debye screening mass M. We assume that the mean free path of the
projectile parton in the medium is taken to be much larger than 1/M. This ansatz for
(2.10) is known as GyulassyWang model [13,14] and was originally introduced to mimic
414
rescattering effects of hard partons in the colour-deconfined matter created in the early
phase of a relativistic heavy ion collision.
To leading order O(1/E), the Feynman diagram in Fig. 1(b) is the only sN rescattering
term. Rescattering contributions involving 3- or 4-gluon vertices are known to come with
spatial suppression factors in the GyulassyWang model [13,17]. The N -fold rescattering
of a hard parton in the colour field (2.9) is thus determined by
N Z
Y
p + m) 0
d3 pi 3 i(/
(N)
ip1 y
P
d xi 2 i
I (y) = e
3
(2)
pi m2 + i
i=1
!
ix (p p ) (r)

(2.11)
iA0 (xi ) e i i+1 i v (p).
Here, the quark leaves some production vertex with momentum p1 and colour a1 . It
undergoes N gluon exchanges with the spatially extended colour potential A , emerging
in the final state with momentum p and colour a. I (N) (y) is a matrix in the colour
representation of the projectile. The path-ordering P implies that A0 (xi+1 ) stands to
of this matrix.
the right of A0 (xi ). The diagram in Fig. 1 denotes the component Ia(N)
1a
The momentum transfers to the quark line are written as Fourier transforms of the static
scattering potential with respect to the relative momenta pi+1 pi . Finally, instead of
an explicit production vertex, we have introduced in (2.11) the incoming plane wave
exp[ip1 y]. This factor allows to glue the above Feynman diagram via y-integration
onto another subprocesses P without specifying P at the present stage.
In Appendix A, we approximate the norm of I (N) to leading order O(1/E) and the
phase to next to leading order. Summing then over contributions for arbitrarily many N
rescatterings in the medium, we find the non-abelian Furry wavefunction

0 X
0
b (y, p)v (r) (p).
I (N) (y) = eiEy ipyL DF
v y 0 , y, p = eiEy
(2.12)
N=0
b
This expression compares directly to the abelian Furry approximation (2.4). In (2.12), D
is the operator (2.5) with pi = p, and F is the outgoing transverse wavefunction evolved
from its asymptotic plane wave form to y with the Greens function G,
Z
(2.13)
F (y, p) = dx G(y , yL ; x , xL |p)F (x , xL , p).
The Greens function G is the non-abelian generalization of the Greens function (2.8).
It can be defined explicitly by its expansion in powers of the scattering potential A0 :

G r, z; r0 , z0 |p G0 r, z; r0 , z0 p
Zz0
i
Z
d

d G0 (r, z; , |p)A0 (, )G0 , ; r0 , z0 p
ZxL
+P
ZxL
d1
zL
Z
d2
d 1 d 2 G0 (r, z; 1 , 1 |p)
iA0 ( 1 , 1 )G0 ( 1 , 1 ; 2 , 2 |p)

iA0 ( 2 , 2 )G 2 , 2 ; r0 , z0 p .
Here, G0 is the free non-interacting Greens function

ip(r r0 )2
p
exp
,
G0 r, z; r0 , z0 p
2i(z0 z)
2(z0 z)
415
(2.14)
(2.15)
and the path ordering P in (2.14) ensures that the potential A0 ( 2 , 2 ) stands to the right
of the potential A0 ( 1 , 1 ). We shall use the Greens function G in (2.14) for z0 > z and
z > z0 by defining

for z0 > z.
G r0 , z0 ; r, zp G r, z; r0 , z0 p
(2.16)
For a hermitian scattering potential, this definition is compatible with the representation
(2.14). Hermitian conjugation automatically inverts the path-ordering. A very compact
notation for the expansion (2.14) can be given in terms of a path-ordered Wilson line
W ([r]; z, z0) which follows the non-abelian potential A0 from initial position (r(z), z) to
final position (r(z0 ), z0 ) along the path r( )
( Zz0
)
Z

ip
0 0
2
G r, z; r , z p = Dr( ) exp
d r ( ) W [r]; z, z0 ,
(2.17)
2
z
0
Zz0
W [r]; z, z = P exp i
d A0 r( ),
)
.
(2.18)
Again, operators at larger longitudinal distances (z0 > z) stand to the right. Expanding
the Wilson line to fixed order O(An0 ) coincides with the representation (2.14). The nonabelian Furry approximation thus differs from the abelian one essentially by path-ordering
in the Greens function G which describes the dynamical evolution of the transverse
wavefunction.
To obtain the result (2.12), it is crucial that we approximate first the N -fold rescattering
diagram I (N) , keeping the phase factor to order O(1/E) and then resum contributions for
arbitrary N . If one parallels the derivation of the abelian case by directly starting from the
solution of the non-abelian Dirac equation in the presence of a spatially extended colour
field, one arrives at the recursive solution of Buchmller and Hebecker [29]. The starting
point for their recursion formula is a straight eikonal Wilson line, and the recursion includes
corrections to fixed order in energy. An explicit n-fold iterated expression is thus correct
to fixed order O(1/E n ), but it does not contain the O(1/E)-corrections to the phase of
the wavefunction, which is characteristic for the Furry approximation. Our diagrammatic
approach allows to keep this phase. For large but finite incident energy, our solution is
thus characteristically different from that given in Ref. [29]. [We have checked, e.g., that
the abelian version of [29] does not lead to the correct medium dependence of the QED
bremsstrahlung spectrum while the abelian version of (2.12) does. This difference of both
solutions is expected since the bremsstrahlung spectrum is sensitive to interference effects
which stem from the O(1/E)-contributions to the phase of the wavefunctions [22].]
416
In the infinite energy limit, the solution (2.12) reduces to the straight eikonal Wilson line

lim G(r, z; x , xL |) = W [rs ]; z, xL ,
(2.19)
rs ( ) = x
xL
z
+r
,
xL z
xL z
(2.20)
which is the leading order input for the recursion formula of Ref. [29]. This expression
is often used to describe the leading colour rotation of the hard parton in a soft colour
field [1,26].
The Furry wavefunction (2.12) derived here allows for the description of the nuclear
dependence of a large class of physical observables. To discuss its limitations, we note
that QCD factorization theorems for observables including rescattering effects have been
established (and can be expected to hold) only for processes for which kinematical
constraints ensure very high transverse momentum transfers ( QCD ) [1012]. The
description of soft nuclear rescattering of hard partons always depends on additional
assumptions and different approaches may be taken [1,13,16]. In particular, we assume
(as all other treatments do) that the leading nuclear dependence can be obtained using
the idealization of asymptotic parton wavefunctions, i.e., without considering parton
fragmentation. Since parton fragmentation and soft parton rescattering involve transverse
momenta of the same order, this is a requirement on the spatial separation of the two
phenomena and amounts to an energy-dependent limit on the longitudinal extension L
of the nuclear target up to which this description can be expected to apply. Moreover,
the ad hoc separation of the transition amplitude into a production P and a final state
wavefunction including rescattering effects may be oversimplified for some processes. We
think, e.g., of the production and rescattering of a heavy quarkonium state with specific
quantum numbers. Taking P to be the production of the heavy q-q-pair,
it is not clear a
priori to what extent final state rescattering modifies the quantum numbers of this state.
First studies indicate [30] that the discussion of this problem requires a classification of
the hardness of medium-induced momentum transfers which lies outside the scope of
the present calculation. To sum up: the discussion of the nuclear dependence of hard
observables in terms of hi |P |f i seems justified if no other soft scale (introduced, e.g., by
parton fragmentation or by the binding energy of the final state) interferes significantly with
the final state rescattering effect described by the Furry wavefunctions. The non-abelian
LPM-effect [13,14,1619], the nuclear dependence of DrellYan yields [3,21] and nuclear
shadowing [6,7] are prominent examples for which a description in terms of hi |P |f i
seems suitable.
3. Photodissociation
In this section, we give a closed expression for the q q photodissociation cross
section q q in terms of the non-abelian Greens function G and the squared ingoing
wavefunction of a freely evolving q-q-pair.
This photodissociation cross section
417
contributes to the total virtual photoabsorption cross section total which is related to the
deep inelastic structure function F2 [31]

F2 x, Q2 =

Q2
total x, Q2 .
2
4 em
(3.1)
In the target rest frame, one can show that photodissociation as depicted in Fig. 2(b)
dominates for small Bjorken x over the -q fusion process shown in Fig. 2(a) [7]. The
q q
come from initial state gluon radiation [7,33] which we
leading corrections to total
neglect in what follows.
p
We consider a virtual photon of four momentum k = (, 0 , 2 + Q2 ) which
dissociates into a quark-antiquark pair in the time interval T /2 < t < T /2. The
dissociation cross section is
Z
1
q q
=
(3.2)
dX |Sf i |2 ,
total
2 |k L| T
where the phase space element is written in terms of the on-shell momenta of the outgoing
quarks,
dX =
d3 p2
d3 p1
.
(2)3 2 |E1 | (2)3 2 |E2|
(3.3)
The rescattering of these quarks in the spatially extended colour field is described by the
Furry wavefunctions in the dissociation amplitude
Sf i Sf i ,
Z
Sf i = ie d4 x u (x, p1 ) 0 e |z| eikx v (x, p2 ).
(3.4)
(3.5)
Fig. 2. Two possible time orderings for the interaction of a virtual photon with a nuclear target. At
small Bjorken x, the q q photodissociation process (b) dominates over the -q-fusion, see
Ref. [7].
418
To discuss the case of different photon polarizations, we write this dissociation amplitude
as a contraction of Sf i with the polarization vector . To shorten our notation, we do not

keep track of the fractional charge eq of the quark and of the number of flavours. To do this,
all our final results have to be supplemented by a sum over the available flavour channels
weighted by eq2 .
Inserting the Furry wavefunctions for quark and antiquark, see (2.12), we can separate in
(3.5) an energy conserving -function, Sf i = ie (2) (E2 E1 ) Mf i . We introduce the

fraction of the initial energy carried by one quark, E2 = , E1 = (1 ). Here, E1
is negative, since the corresponding 4-momentum p1 flows into the quark-photon vertex,
see Fig. 1(a). After coordinate transformation (p1L , p2L ) (, k L ), the photodissociation
cross section reads
Z
Z

dp
dp
d
q q
1
2
= em
|Mf i |2 .
(3.6)
total
2
2
2
4 (1 )
(2) (2)
Here, we have used Fermis golden rule and the k L -integration to eliminate the two energyconserving -functions. The brackets h. . .i denote an in-medium average over the colour
field A which will be specified below. The probability |Mf i |2 can be written in terms of
the full interacting Greens functions (2.17) as
Z
|Mf i |2 = d3 y d3 y ei q (yL yL ) e(|yL |+|yL |)
Z
0
0
dx dx0 dx dx0 ei p1 (x x ) ei p2 (x x )

G(x; y|p2 ) b G y; x0 p1 G x 0 ; y p1 b G y ; x |p2 .
(3.7)
In general, all spatial coordinates can be different for Mf i and Mf i . We characterize those
for Mf i by a bar. The spinor structure of the Furry wavefunctions (2.12) is contained in
(3.7) in the vertex function
0
b2 v (r) (p2 ).
b1 0 D
b = u(r ) (p1 )D
(3.8)
In Appendix B, we discuss how these vertex functions combine with free Greens functions
G0 to the incoming q-q Fock state. In terms of the square (1z; 1z; ) of this Fock state,
the differential photodissociation cross section (3.6) takes the form
Z
q q

total
em
1 db 2 1b; 1b;

=
db
d
b
db
1
2
(2)4
d dp
1 dp2
Z
0
0
dx dx0 dx dx0 ei p1 (x x ) ei p2 (x x )
D

E
G b 2 ; x p2 G x; b2 p2 G b1 ; x0 p1 G x 0 ; b 1 p1 ,
(3.9)
where 1b = b1 b2 and 1b = b 1 b 2 . The explicit form of the squared incoming
wavefunction is derived in Appendix B.
In Eq. (3.9), we assume that the nuclear target has nonvanishing density only for
some longitudinal positions zL > 0. The integration variables bi , b i denote the transverse
419
Fig. 3. Representation of the q q photodissociation cross section (3.6), (3.7). In the amplitude
Mf i , the photon dissociates at position (y, yL ) into a quark and an antiquark. These propagate with
free non-interacting Greens functions (dotted lines) up to the front end of the target at longitudinal
position zL . From zL onwards, the quark and antiquark are propagated inside the colour field
with interacting Greens functions (full lines) up to their asymptotic on-shell states (cut). For the
complex conjugate amplitude, the analogous evolution of the hadronic q q Fock state is depicted on
the other side of the cut.
boundary values of the Greens functions at the front end zL = 0 of the nuclear target.
Fig. 3 gives a graphical representation for (3.9) which we shall heavily use in what
follows. According to Fig. 3, the virtual photon wavefunction starts interacting with the
nuclear medium at zL = 0 in both the amplitude and complex conjugate amplitude after
dissociating at longitudinal positions yL , yL with yL , yL < 0. We emphasize, however, that
despite appearance, Eq. (3.9) as well as its graphical representation in Fig. 3 also include
the case that the photon vertex dissociates inside the target at yL > 0. Technical details of
how this is handled are given in Appendix B.
4. Opacity expansion and colour triviality

In this section, we analyze the photoabsorption cross section (3.9) by expanding the
interacting Greens functions G in powers of the scattering potential A0 (x). In a first step,
we focus on the total photoabsorption cross section
Z
Z

q q

= em d db1 db2 db 1 db 2 1b; 1b;
total
Z

dx G b 2 ; xp2 G x; b2 p2

Z

0
0
0
(4.1)
dx G b1 ; x p1 G x ; b1 p1 .
To make sense of an expansion in powers of A0 , we have to specify the in medium average
h. . .i in terms of A0 . To this end, we specify for the colour potential (2.9) the contribution
of a single scattering potential, centered at (ri , z i ),
Z
d3
L
a0 ( ) ei(x ri ) ei( zi ) .
(4.2)
ia (x , ) = a ai
3
(2)
420
Here, we have approximated the argument of the single scattering potential a0 ()

a0 ( ). This implies that the momentum transfer occurs at a fixed longitudinal position
= z i . It is the standard approximation in the high energy limit where the dominant
momentum transfer from the medium is transverse, and it motivates the use of time-ordered
perturbation theory to rescattering problems [13,16].
Starting from (4.2), we define the medium average h. . .i as an average over the transverse
and longitudinal positions (ri , z i ) of the scattering potentials and the colour factors ai ,
!
N XZ
Y

1
(4.3)
dri dzi f r1 , . . . , rN ; z 1 , . . . , z N ; a1, . . . , aN .
hf i
A
a
i=1
Here, A is a total transverse area which we divide out to regain the cross section per unit
transverse area. N is the number of different single scattering potentials up to which the
function f is expanded.
To simplify notation, we replace in what follows the discrete sum over z i in (2.9) by an
integral over the density n of scattering centers
Z

(4.4)
A0 (x , ) = dzi n z i ia (x , ) T a .
Since the effective momentum transfer from the single potential ia (x , ) occurs at = z i ,
we shall often work with as integration variable. The expansion of the photoabsorption
cross section to N th order in A0 is an expansion in the N th order of the opacity parameter
R
s2 d n( ). In what follows, we study the perturbative expansion in this parameter.
4.1. N = 1 result of Nikolaev and Zakharov
As a first illustration of the above formalism, we expand the integrand of the total
R
photoabsorption cross section (4.1) to first order in the opacity parameter s2 d n( ),
thereby reproducing the well-known nuclear shadowing result of Nikolaev and Zakharov.
For N = 1, we have to expand all Greens functions in (4.1) and then to collect all
contributions to order O(A20 ).
q q
vanishes due to
The zeroth and first order contributions O(A00 ) and O(A10) to total
energymomentum conservation: without momentum transfer to the medium, the q-q-pair
2
cannot appear on-shell. To second order O(A0 ), there are four different terms which we
depict diagrammatically in Fig. 4(a): for each term, the full lines correspond to the full lines
shown in Fig. 3 and describe the full Greens functions in the total photoabsorption cross
section (4.1). To simplify the representation, we have dropped in comparison to Fig. 3 the
photon lines and incoming q-q wavefunctions. For the total photoabsorption cross section,
the transverse positions at the cut are equal for the amplitude and the complex conjugate
amplitude.
Let us consider the first term on the r.h.s. of Fig. 4. Here, the Greens functions of
argument p2 are free while both Greens functions with argument p1 are expanded to first
order. Averaging over the position (r1 , z 1 ) of the center of the scattering potential, we find
421
Fig. 4. (a) Diagrammatic representation of the N = 1 contribution to the total photoabsorption cross
section (4.1). For each term, the upper (lower) line denotes the Greens function with energy p1 (p2 )
and the cusps denote the position of the cut. (b) Definition of the diagrammatic shorthand used in
(a) and calculated in (4.5). Crosses stand for one power of the scattering potential A0 . The medium
average ensures that the momentum flow k through the cut is conserved and that the potential A0 is
linked to the q-q-system
in Mfi and Mfi at the same longitudinal position.
Z

Z

(1) 0
0
(1)
0

dx G0 b2 ; x p2 G0 (x; b2 |p2 ) dx G b1 ; x p1 G x ; b1 p1
Z
Z
2

d
1
a0 ( ) Ta1 Ta1 (2) b 2 b2 (2) b 1 b1 .
d1 n(1 )
=
2
A
(2)
(4.5)
This term does not depend on the separation between the q-q pair since the potential
touches only one of the quark lines. In contrast, the second term on the r.h.s. of Fig. 4(a)
depends on this separation and takes the explicit form
Z
Z
2
d
a0 ( ) Ta1 Ta1 ei (b1 b2 )
d1 n(1 )
(2)2

2
ei 2(1) (1 zL ) (2) b 2 b2
(1 zL )
p2

(1 zL )
.
(4.6)
(2) b 1 b1
p1
A
This shows explicitly that the transverse Brownian motion induced by rescattering is taken
into account in our formulation: the q-q dipole does not propagate at fixed transverse
separation b 1 b1 but changes its size in response to an interaction at the longitudinal
position 1 by terms of the form p1 (1 zL ). For N = 1, the first and only interaction takes
place by definition at the position zL = 1 and these terms vanish. For N > 1, however, they
do not vanish and lead to a nontrivial dynamical evolution of the transverse dipole size, see
below.
Adding up the four real contributions in Fig. 4(a), and doing the colour trace, we find
Z
Z
Z
q q
(4.7)
total (N = 1) = Nc em d n( ) d db (b; b; ) (b),
422
2

d
a0 ( ) 1 ei b .
(4.8)
2
(2)
This is the result of Nikolaev and Zakharov [6]. The total cross section (4.1) is the product
of the Born probability (b; b; ) for the incoming hadronic Fock state times a dipole cross
section (b), integrated over the transverse size b of the dipole and the energy distribution
between the quark and antiquark.
(b) = 2CF
4.2. N =2 tagged: non-trivial colour interference effects

For more than N = 1 scattering center, there are in general non-vanishing interference
terms between amplitudes containing different powers of A0 , see Section 4.3 below.
Such contributions appear in the generic situation in which one knows the distribution of
scattering centers but no information is available on whether these centers have participated
with finite momentum transfer in the scattering process. At least in a gedanken experiment,
however, this additional information can be obtained from the recoil of each scattering
center, i.e., by tagging the scattering centers. For this very exclusive observable, all
contributions to the cross section contain the same power of A0 in both amplitudes, and all
scattering centers contribute to the amplitude with finite momentum transfers. We say, such
scattering centers contribute with real terms, in contrast to the contact term contributions
which we discuss in Section 4.3. Tagged scenarios were considered recently in the study
of the non-abelian LPM-effect [24], and there is some interest in comparing results of
tagged and untagged scenarios to understand the origin of colour triviality. To this aim,
we consider here as a second application of the opacity expansion the case of two tagged
scattering centers: (i) two scattering centers are involved in the process (4.1) and (ii) both
scattering centers provide real terms, i.e., transfer
finite transverse momentum on the
q q
amplitude level. The 16 contributions to this tagged (N = 2) are listed in Fig. 5. Summing
them up, we find
Z
Z
Z
Z
q q
tagged (N = 2) = em d1 n(1 ) d2 n(2 ) d db
Z
2
2
d 1, d 2,
a0 ( 1, ) a0 ( 2, )
2
2
(2) (2)

Tr [a a b b ] A(1) + Tr [a b a b ] A(2) ,
where we have introduced the notational shorthands

A(1) = 4(b, b; ) 1 ei 1, b ,

2, (2 1 )
2, (2 1 )
(2)
,b +
;
A = 4 b
2 (1 )
2 (1 )

(2 1 )
i 1, 2, (12) (21 )
ei 1, b e 2(1)
ei 1, 2,
ei 2, b .
(4.9)
(4.10)
(4.11)
Here, the term A(2) contains information about the non-trivial dynamical evolution of the
q-q-separation
which is determined by the relative transverse mass
423
Fig. 5. The 16 contributions to the N = 2 photoabsorption cross section (4.9). In this tagged
case, exactly two real momentum transfers occur on the amplitude level. The tagging gives rise
to non-trivial colour interference effects.

=
1
1
p1 p2
1
= (1 ).
via a classical
The momentum transfer 2, determines this change of the q-q-size
In
the
high
energy
limit
,
this transverse
equation of motion: 2,
(
).
2
1
motion can be neglected and the q-q-pair
propagates at fixed separation. We find

q q
(N = 2)
lim
tagged
Z
Z
Z
Z
= em d1 n(1 ) d2 n(2 ) d db
Z
2

d 1, d 2,
a0 ( 2, )2 (b, b; ) 4 1 ei 1, b
a
(
)
0
1,
(2)2 (2)2

Tr [a a b b ] Tr [a b a b ] ei 2, b .
(4.12)
Even in this limiting case, the colour algebra does not factorize from the momentum
dependence. This is a consequence of the non-trivial colour interference between different
photodissociation amplitudes. The photoabsorption cross section cannot be written as a
function of the dipole cross section (b). We note that this complication stems entirely
from the non-abelianess of the problem. If we replace both colour traces in (4.12) by the
same constant, the two i, -integrations combine to the square of the dipole cross section
(4.8). To sum up: in the tagged case, non-trivial colour interference effects prevent us from
representing the photoabsorption cross section as a function of the dipole cross section.
4.3. Contact terms
In the calculation of the N = 1 cross section (4.7), we did not include all terms of the
Taylor expansion of (4.1) to order O(A20 ). A second order contribution which was not
included is, e.g.,

Z
Z

(2) 0
0
0

dx G0 b1 ; x p1 G x ; b1 p1
dx G0 b2 ; x p2 G0 x; b2 p2
424
1
=
2
Z
d1 n(1 )
2

d
a0 ( ) Ta1 Ta1 (2) b 2 b2 (2) b 1 b1 . (4.13)
2
(2)
Here, the potential A0 is linked two times to the complex conjugate amplitude and zero
times to the amplitude. The medium average h. . .i results in two important constraints:
at the same longitudinal position 1 , and

(i) both powers of A0 couple to the q-q-pair
(ii) no net momentum is transferred to the q-q-system.
Scattering contributions with these

properties were called virtual by Mueller and collaborators [5,27]. We refer to them as
contact terms, all other interactions are referred to as real. Diagrammatically, the six
contact terms for the N = 1 case are given in Fig. 6(a), and Eq. (4.13) is represented
by the first diagram on the r.h.s.
For the N = 1 photoabsorption cross section (4.7), contact terms do not contribute since
at least one interaction with the q-q-system
is needed in both the amplitude and complex

conjugate amplitude in order to get the final state on shell. For N > 2, the same argument
about energy-momentum conservation does not imply the absence of contact terms. We
have to distinguish
the following cases:
q q
inelastic: the cross section involves at least one real interaction.
1. inel.
q q
diffractive (or elastic): the medium interacts with the q-q-pair
only via
2. diff.
contact
terms.
q q
total: the cross section involves at least one (real or contact) interaction in
3. total
both Mf i and Mf i .
The notion inelastic is justified since any real interaction changes the colour of the
target and thus affects the hadronic activity between projectile and target rapidity (in
this sense, the target breaks up). The notion diffractive or elastic involves an
additional assumption: by construction, contact terms transfer exactly zero net transverse
momentum to the q-q-pair.
Also, we regard in all calculations and for arbitrary scatterings

the longitudinal momentum transfer as sufficiently small to justify the approximation
a0 () a0 ( ). Our assumption in the above definitions of the diffractive and total
photoabsorption cross sections is that despite neglecting the longitudinal momentum
Fig. 6. (a) The six contact terms for the N = 1 scattering cross section. (b) Definition of the
diagrammatic shorthand used in (a). The medium average ensures that (i) the momentum flow k
is conserved and no net momentum is transferred to the q-q-system,
and that (ii) both powers of the

scattering potential act at the same longitudinal position z .
425
transfer in all explicit calculations, it is sufficient to put the (extremely weakly off-shell)
partonic contributions of on-shell. This is usually assumed in the calculation of the
total [5] and diffractive [9] cross sections.
For N > 1, contact terms obviously play an important role in the calculation of the
inelastic, diffractive and total photoabsorption cross section. In the remainder of this
subsection, we summarize some of their important properties and shorthands, which will
be heavily used in the following subsections:
4.3.1. Identities involving contact terms
First we observe that in (4.13), the Greens functions of momentum p2 result in the
transverse -function (2) (b 2 b2 ). Dropping them on both sides of (4.13), and observing
that (4.13) and (4.5) differ by a factor (1/2), we have checked the identity in Fig. 7(a).
By a similar calculation, one also checks the identity Fig. 7(b).
4.3.2. A path integral from iterating contact terms
We now consider the case in Fig. 7(c), where the q-q-state
is evolved with exactly

one contact term between longitudinal positions z and z0 from a transverse separation
r1 r2 to a transverse separation r01 r02 . The sum over the three possible contact terms at
longitudinal position , represented on the r.h.s. of Fig. 7(c), takes the explicit form
Z
Z

Ta Ta
d n( ) dr1 ( ) dr2 ( ) r1 ( ) r2 ( )
2Cf

G0 r2 , z; r2 ( ), p2 G0 r2 ( ), ; r02 , z0 p2

(4.14)
G0 r0 , z0 ; r1 ( ), p1 G0 r1 ( ), ; r2 , zp1 .
1
Since Fig. 7(c) is only part of a more complicated Feynman diagram, we can say nothing
about the positioning of the colour factors Ta . The corresponding expression Ta Ta in
(4.14) is purely formal. However, for the following, we anticipate an argument explained
Fig. 7. At the cut, there are simple identities between real interaction terms and contact terms (a),
and amongst real interaction terms (b). (c) A combination of the three possible contact terms at the
same longitudinal position leads to a dipole cross section (r1 ( ) r2 ( )) whose transverse size is
evolved further with free Greens functions, see (4.14). (d) Iterating the combination of contact terms
(c) leads to a dynamical evolution of the dipole size between initial transverse size r1 r2 and final
size r01 r02 . This evolution is described by the path-integral (4.15).
426
in the next subsection: for our purposes, the colour factors reduce to a Casimir factor CF ,
a
1. Then, Eq. (4.14) is seen to correspond to the
i.e., we can substitute in (4.14) TaCT
f
first order n( ) density expansion of the double path integral (for z0 > z)
r1 (z0 )=r01
r2 (z0 )=r02

M r1 , r2 , z; r01 , r02 , z0 =
Dr1
r1 (z)=r1
Dr2
r2 (z)=r2
)
( Zz0

1
exp i d r22 + (1 ) r21 + i n( ) r1 ( ) r2 ( ) .
2
2
(4.15)
Moreover, the expansion of (4.15) to mth order in the density n( ) corresponds exactly
to the mth term on the r.h.s. of Fig. 7(d). This shows that the iteration Fig. 7(d) of the
one contact term interaction Fig. 7(d) is described by the path-integral (4.15). Since this
path integral plays an important role in what follows, we simplify it further. Using the
coordinate transformation
ra ( ) = (1 ) r1 ( ) + r2 ( ),
(4.16)
rb ( ) = r1 ( ) r2 ( ),
(4.17)
we can write

M r1 , r2 , z; r01 , r02 , z0

= K0 ra (z0 ), z0 ; ra (z), z K rb (z0 ), z0 ; rb (z), z(1 ) .
(4.18)
The path-integral on the r.h.s. of this expression is given by (for z0 > z)

K r(z0 ), z0 ; r(z), z =
( Zz0
)
2
1
Dr exp i
r + i n( ) (r) .
d
2
2
(4.19)
Its opacity expansion reads

K r0 , z0 ; r, z = K0 r0 , z0 ; r, z
Zz0
Z
d

d K0 r0 , z0 ; r, (, )K0 (, ; r, z)
Zz0
+
Zz0
d1
Z
d2

d 1 d 2 K0 r0 , z0 ; 2 , 2 ( 2 , 2 )
K( 2 , 2 ; 1 , 1 )( 1 , 1 )K0 ( 1 , 1 ; r, z),
(4.20)
where (, ) = n( ) (). Here, we have suppressed the explicit -dependence in K.

The corresponding free Greens function K0 reads

i(r0 r)2
0 0
exp
.
(4.21)
K0 r , z ; r, z =
2i (z0 z)
2 (z0 z)
1
2
427
In analogy to our definition of the Greens function G in (2.16), we define

for z0 > z.
K r(z), z; r(z0 ), z0 K r(z0 ), z0 ; r(z), z
A path integral K of the form derived here was first used by Zakharov in the abelian
problem of calculating the passage of ultrarelativistic positronium through matter [32].
More recently, the same path integral was shown to appear in the final expression of the
medium-dependence of the LPM-bremsstrahlung spectrum [15,22].
4.4. N = arbitrary, untagged: contact terms remove colour interference effects
What happens if contact terms are included in the N = 2 calculation of the elastic,
inelastic and total photoabsorption cross section? In Fig. 8, we have classified into
four subsets all terms of order O(n( )2 ) which have to be considered for the N = 2
photoabsorption cross sections. There are in Fig. 8(a) the 4 4 = 16 terms for which both
interactions are real, in Fig. 8(b) the 4 6 = 24 terms for which only the second interaction
is contact, in Fig. 8(c) the 3 4 2 terms for which only the first interaction is contact,
and in Fig. 8(d) the 2 3 6 = 36 which involve two contact interactions. All together,
these are 100 terms. For the total and the inelastic cross section, the terms in Fig. 8(a)
and Fig. 8(b) both contribute, but they cancel each other exactly. More precisely, the first
diagram in Fig. 8(a) cancels the first diagram in Fig. 8(b), the second diagram in Fig. 8(a)
cancels the second diagram in Fig. 8(b), etc. This is a consequence of the identities Fig. 7(a)
and 7(b) which we apply to the last interaction before the cut.
The same argument cannot be made for the cancellation of the diagrams in Fig. 8(c) and
Fig. 8(d): energy momentum conservation restricts the occurence of double contact terms,
since at least one interaction is needed on the amplitude level to get the final state onshell. Depending on whether we calculate the diffractive, inelastic or total photoabsorption
cross section, different combinations of these two sets of diagrams contribute. The real
interaction Fig. 4(a) at the cut is determined by minus the combination Fig. 6(a) of contact
terms. This allows us to express all three contributions to the photoabsorption cross section
in terms of contact terms, as shown in Fig. 8. Most importantly, all contributions to these
photoabsorption cross sections have the colour trace Tr [a a b b]: for the untagged case
for which contact terms are included, colour interference terms vanish in the diffractive,
inelastic and total N = 2 photoabsorption cross section.
The above argument can be generalized easily to arbitrary N . As shown in Fig. 9(a),
whenever at least one of the (N 1) first interactions is real, the sum of all real and virtual
terms for the N th interaction cancels. Inelastic, diffractive and total photoabsorption cross
section can thus again be described in terms of (N 1) contact terms and a characteristic
N th contribution at the cut. As an immediate consequence, the colour trace of all these
contributions reads

(4.22)
Tr Ta1 Ta2 . . . TaN TaN TaN1 . . . Ta1 = Nc CFN .
In contrast to the tagged photoabsorption cross section, the colour structure factorizes from
the momentum dependence: Inclusion of the contact terms results in colour triviality for
428
Fig. 8. The contributions to the N = 2 photoabsorption cross section: (a) the 16 real contributions
which are present in the tagged case. (b) Contributions for which the first interaction is real and
the second a contact term. (c) Contributions for which the first interaction is a contact term and
the second interaction transfers momentum. (d) Terms which contain two contact interactions. Here,
energymomentum conservation ensures that the two contact terms do not stand on the same side of
the cut. (e) Inelastic, diffractive and total photoabsorption cross section are determined by different
combinations of the contributions (c) and (d). Rewriting the real interaction in (c) as the sum of two
negative contact term contributions, a particularly simple representation is obtained.
the N -fold diffractive, inelastic and total photoabsorption cross section. This justifies the
assumption made in deriving the path integral (4.15) in the last subsection.
According to Fig. 9(b)(d), the inelastic, diffractive and total photoabsorption cross
section can be written in terms of the path-integral K of (4.19) which describes the iteration
of contact terms. For the inelastic photoabsorption cross section, this leads to the expression
q q
inel.
Z
= Nc em
Z
d

db db b; b;
ZL
d n( )
dr( )
nl

0; r( ), (1 ) r( ) K r( ), ; b, 0(1 ) . (4.23)
K b,
Here, the Greens functions K describe the propagation of the q-q-dipole
from the front

end of the target at z = 0 up to the position of the last interaction. The lower boundary nl
of the integration over is defined by the position of the next to last (nl) interaction. This
429
Fig. 9. (a) Cancellation of real and contact terms for any combination of interactions [I] which
contains at least one real or contact interaction. (b)(d) The identity (a) allows for the generalization
to arbitrary N of the N = 2 photoabsorption cross section of Fig. 8.
is an awkward property, since we have to expand the Ks in the number N of interactions

to determine nl order for order in N . This makes it important to find for (4.23) with the
help of the opacity expansion (4.20) the equivalent representation
Z
Z

q q

= Nc em d db db b; b;
inel.

Z

0; re , L K re , L; b, 0 .
(4.24)
(2) b b dre K b,
For the diffractive contribution, one finds
Z
Z
Z

q q
diff.
dre
(2)

(2)

0; re , L (re b) K re , L; b, 0 .
b re K b,
(4.25)
The total photoabsorption cross section is the sum of both

Z
Z

q q

total

0; b, L K b,
L; b, 0 .
2 (2) b b K b,
(4.26)
Before turning in the next subsection to the discussion of these expressions, we remark
shortly on a phase convention implicitly used in the above results:
The first (N 1) interactions in the above photoabsorption cross sections are contact
terms. Contact terms stand either to the right or to the left of the cut. The first interaction
thus occurs for each non-vanishing contribution to (4.24) at different longitudinal positions
za , zb in the amplitude and complex conjugate amplitude. As a consequence, the free
incoming wavefunction has to be evolved to these different positions in Mfi and Mfi . We
explain in Appendix B following (B.8) that the expression for the squared incoming
430
wavefunction is only correct as long as za = zb . If one insists on using for za 6= zb , one

is forced to adopt the phase convention:
Z

za ; r, zb (1 ) = b; r; eiq (zb za ) .
K0 b,
(4.27)
db b; b;
Strictly speaking, using one has done the yL - and yL -integrations in the photodissociation probability (3.7) before specifying the true endpoints of these integrations which are
given by the positions za , zb of the first interactions in Mfi and Mfi . The phase convention
(4.27) corrects for the part of the yL integral which one misses in assuming za = zb . Here,
we shortly illustrate the consequences of (4.27):
With the help of (4.27), one checks immediately that the N = 1 contribution to (4.24)
agrees with the result (4.7) of Nikolaev and Zakharov. There is no additional phase factor
in this case since za = zb . However, there is an additional phase factor for all contributions
N > 1. In the notationally simplest case, N = 2, we find
Z
Z
Z
q q
inel. (N = 2) = Nc em Re d d 0 n( 0 ) d n( )
Z

0
dr dr (r; r; ) eiq( ) (r)K0 r, 0 ; r, (r).
(4.28)
The phase factor exp{iq( 0 )} is sensitive to the difference between the longitudinal
position of the first scattering center in Mfi and Mfi . The scale is set by the inverse
coherence length
q=
m2
1
Q2
= .
2
2(1 ) lf
(4.29)
It can be neglected in the high energy limit where the longitudinal extension of the
target becomes small compared to 1/q.
4.5. Colour triviality of the one-fold differential photoabsorption cross section
Colour triviality of the inelastic, diffractive and total photoabsorption cross section
(4.24)(4.26) is the result of a complete diagrammatic cancellation which arises due to
the identities in Fig. 7(a) and (b). These identities are stronger than implied by the optical
theorem since they are based only on the transverse momentum integration of one of the
two quarks in the final state. This makes it possible to use them in the simplification of
more differential observables. Here, we study the one-fold differential photoabsorption
cross section
Z

em
d q q

=
db1 db2 db 1 db 2 1b; 1b;
2
(2)
d dp1
Z
0
0
0
dx0 ei p1 (x x )
dx dx
D

E

G b 2 ; xp2 G(x; b2 |p2 )G b1 ; x0 |p1 G x 0 ; b 1 p1 .
(4.30)
431
Fig. 10. (a) The non-vanishing contributions to the inelastic part of the one-fold differential
photoabsorption cross section (4.30) as classified in Appendix Appendix C. (b) Definition of
diagrammatic shorthands used in (a). (c) Order N = 1 contribution to (a1).
We shall derive for the inelastic part of (4.30) an expression in terms of

Z
2
d
a0 ( ) ei b ,
(b) = 2CF
2
(2)
which is closely related to the dipole cross section,

r1 r1 = (0) r1 r1 .
(4.31)
(4.32)
The term (0) corresponds to a contact term for which both vertices are linked to the same
quark of the q-q-system.
In intermediate steps of our calculation, is a useful bookkeeping

device to evaluate the diagrams in Fig. 10(a). Our final result, however, will depend only
on the dipole cross section.
4.5.1. Inelastic one-fold differential cross section
The diagrammatic book-keeping of the inelastic contributions to the one-fold differential
cross section (4.30) are involved. In Appendix C, we classify all contributing diagrams
and we use the identities Fig. 7(a) and (b) to show that many of these diagrams cancel
each other. As a result of this analysis, the remaining non-vanishing contributions can be
represented by the diagrams in Fig. 10(a). The colour trace for the N th order terms of all
these diagrams is Nc CFN : the inelastic part of the cross section (4.30) is thus free of colour
interference terms, it is colour trivial.
432
We start with the N = 1 contribution to the diagram Fig. 10(a1).

e
0 x0 )
ip
1 (x
ZL
d

n( )
r1 r1 (0) r r =x0 x0 .
1
1
2
(4.33)
This contribution is shown explicitly in Fig. 10(c). The relative transverse separation r1 r1
between the quark in the amplitude and complex conjugated amplitude does not change
with longitudinal position . This is a consequence of evolving on both sides of the cut
with Greens functions of same energy p1 . It makes the iteration of the N = 1 contribution
to arbitrary high orders particularly easy. The total contribution to Fig. 10(a1) reads
R L n() 0 0

RL
n()
0
0
(4.34)
eip1 (x x ) e 0 d 2 (x x ) e 0 d 2 (0) .
The first term is just the exponentiation of the (N = 1)-contribution. The second term
subtracts those N th order contributions which involve only contact terms and thus do not
add to the inelastic contribution.
To evaluate the three contributions in Fig. 10(a2) and (a3), we assume that the real
momentum transfer occurs at an arbitrary but fixed longitudinal position . The evolution
of the free incoming Fock state up to is then described by the path-integral (4.18). The
part of the expression for > is given by
RL
n()
(0)
2
eip1 (r1 ()r1 ()) e

r2 ( ) r1 ( ) + r2 ( ) r1 ( ) (0) .
(4.35)
Here, the real interactions at combine to the second line of (4.35) and the arbitrary
number of contact terms at > leads to the exponential of (0). Adding up all
contributions of Fig. 10(a), we find for the differential photoabsorption cross section (4.30):
Z

em Nc

=
db db b; b;
2
(2)
d dp
"1
R L n()
RL
ei p1 (bb) e 0 d 2 (bb) e 0 d
q q
dinel.
ZL
+
n( )
2
RL
(0)

0; r, (r) + (r) (0) K r, ; b, 0
dr dr K b,
nl
n()
2
#
d
n()
2
(0) i p (rr)
1
(4.36)
The first term in the wide brackets stems from the contribution Fig. 10(a1) given in (4.34),
the second term denotes the diagrams Fig. 10(a2) and (a3). Here, the Greens functions
K describe the dynamical evolution of the q-q dipoles up to the real interaction at ,
from which point onwards the dynamics is given by (4.35). In analogy to (4.23), the lower
boundary nl of the integration over is determined by the last interaction point before the
433
real interaction. To remove this indirectly defined variable from our solution, we use again
the opacity expansion (4.20) to show that
q q
dinel.
d dp
1

Z
R
i p (bb)
em Nc
L d n() (bb)
2
0
1
=
e
db
d
b
b;
b;
e
(2)2

Z

0; r, L K r, L; b, 0 .
dr dr ei p1 (rr) K b,
(4.37)
It is easy to check that the p

1 -integration of this expression coincides with the inclusive
inelastic photoabsorption cross section (4.24).
4.5.2. Diffractive and total photoabsorption cross section with one jet resolved
The diffractive one-fold differential photoabsorption cross section involves by definition
only contact terms. Due to energy momentum conservation, at least one contact term is
required on the amplitude level. An arbitrary non-vanishing number of contact terms in
the amplitude translates into a factor K, see the discussion of Fig. 7(d). Hence, the
diffractive contribution to the cross section reads
Z
Z
q q

ddiff.
em Nc
=
db db b; b;
dr dr ei p1 (rr)
2
(2)
d dp1

0; r, L (2) (b r) K(r, L; b, 0|) . (4.38)
(2) b r K b,
The one-fold differential total cross section is given by the sum of the contributions (4.37)
and (4.38),
q q
dtotal
d dp
1
Z

em Nc

=
db db b; b;
2
(2)
Z

R L n()
ei p1 (bb) e 0 d 2 (bb) + 1 dr ei p1 (rb) K(r, L; b, 0|)

Z

0; r, L .
dr ei p1 (br) K b,
(4.39)
4.5.3. Non-trivial colour interference in one-fold and two-fold differential

photoabsorption cross sections
Colour triviality of differential cross sections cannot be taken for granted. It has to be
established by the diagrammatic techniques used in the above subsections. To illustrate
this point, we give in the following simple examples of untagged differential cross sections
which show colour interference effects. We calculate the inelastic part of the N = 2
photoabsorption cross section (3.9) in the limit in which the q-q state propagates
along straight lines through the nuclear target. For a target of thickness L and homogeneous
density n( ) = n0 ,
434

q q
dinel.
(N = 2)
d dp
1 dp2
em
(2)4
db1 db2 db 1 db 2 b1 b2 ; b 1 b 2 ;
n20 L2 1
2CF2 4

b1 b 1 + b2 b 2 b1 b 2 b2 b 1

(4.40)
Tr [a a b b]B (1) + Tr [a b a b]B (2) ,
ei p1 (b1 b1 ) ei p2 (b2 b2 )
where we have introduced the notational shorthands

B (1) = b1 b 1 + b2 b 2 + b 1 b 2 + (b2 b1 ) 4 (0),

B (2) = b 1 b 2 + (b1 b2 ) b1 b 2 (b2 b1 ).
(4.41)
(4.42)
For the cross section (4.40), colour triviality is the condition that B (2) vanishes. However,
B (2) 6= 0, and colour interference terms remain in (4.40). In contrast, the elastic part of the
N = 2 photoabsorption cross section is colour trivial since it involves by construction only
contact terms. Colour triviality (non-triviality) of the inelastic part of the photoabsorption
cross section thus implies automatically the colour triviality (non-triviality) of the
corresponding total cross section.
In general, the more inclusive the cross section, the more likely it is colour trivial. In this
sense, tagged cross sections are very exclusive since they require detailed knowledge of the
final state of the medium. On the other hand, calculating more inclusive observables from
(4.40), one recovers colour trivial examples. Especially, the p2 -integral (or p1 -integral)
of (4.40) turns out to be colour trivial. This is expected, of course, since these integrated
versions of (4.40) determine the N = 2 term of (4.37) in the -limit.
Based on these results, one may ask whether all one-fold differential photoabsorption
cross sections are colour trivial. This is not the case as can be seen, e.g., from the inelastic
parts of the one-fold differential cross sections
q q
dinel.
d d p
1 p2
,
(4.43)
.
(4.44)
q q
dinel.
d d p
1 + p2
For these cross sections, the arguments of in the expression (4.42) for B (2) are constraint
by b1 b 1 = b2 b 2 . It is easy to check that B (2) 6= 0, i.e., both results are not colour
trivial for N = 2 and thus they cannot be colour trivial for N = arbitrary. For a colour
trivial expression, it is crucial that (3.9) is not integrated over some average or relative
momentum but over the transverse momentum of a final state particle. This is necessary
for using the identities of Fig. 7(a) and (b).
4.6. Muellers dipole formulas as eikonal approximation
Explicit calculations of the p -integrated and one-fold differential photoabsorption
cross sections (4.24)(4.26) and (4.37)(4.40) respectively require an explicit represen-
435
tation of the path integral K. This necessarily involves an approximation. In this section,
we study the eikonal approximation which substitutes the path integrals K by their ultrarelativistic limit

(4.45)
lim K L, r; 0, b = (2) (b r)S(b).
We consider a homogeneous density distribution n( ) = n0 for a target of longitudinal

extension L and we write for the Glauber-type suppression factor
1
S(b) = e 2 nL (b) .
(4.46)
Also, we neglect the phase factors exp{iq( 0 )} in this -limit, see our discussion
of (4.28), (4.29) above. In the ultrarelativistic limit, all photoabsorption cross sections
reduce to standard Glauber-type expressions:
Z
Z

q q
= Nc em d db (b; b; ) 1 S 2 (b) ,
(4.47)
lim inel.
Z
Z

2
q q
= Nc em d db (b; b; ) 1 S(b) ,
(4.48)
lim
diff.
Z
Z

q q
= 2Nc em d db (b; b; ) 1 S(b) .
(4.49)
lim total
This total photoabsorption cross section is Muellers dipole formula first derived in Ref. [5].
Also, the elastic cross section is known to describe the diffractive contribution to the deep
inelastic scattering structure function F2 in the ultrarelativistic limit [9].
The one-fold differential cross sections read in the eikonal limit
Z
q q

dinel.
em Nc
ei p1 (bb)
=
db db b; b;
lim
2
d dp
(2)
1

S(b)
S(b) ,
S(b b)
(4.50)
Z
q q

ddiff.
em Nc
ei p1 (bb)
=
lim
db db b; b;
2
d dp
(2)
1

1 S(b) ,
1 S(b)
(4.51)
Z
q q

dtotal
em Nc
ei p1 (bb)
=
db db b; b;
lim
2
d dp
(2)
1

S(b)
S(b) .
1 + S(b b)
(4.52)
The elastic contribution (4.51) has been obtained previously in calculations of the
diffractive component of DIS electron nucleon scattering with one resolved jet in the final
state [9]. Furthermore, the total cross section (4.52) was previously obtained by Mueller
(see Eq. (27) of Ref. [8]) from a one-loop calculation arguing then for its general validity.
In passing, we recall Muellers interpretation of the four terms in (4.52): for the case
2
p1 Q2 , the first two terms turn out to be of equal size while the last two terms are
negligible [8]. Looking at the diagrammatic contributions one concludes that in this limit
gives the probability that a quark which gets many random
the term proportional S(b b)
436
kicks is found with relatively small transverse momentum. The term proportional to 1, on
the other hand, corresponds to the case of no scattering and can be viewed as the quantum
[8].
mechanical shadow of the term proportional to S(b b)
4.7. Corrections to the eikonal approximation
To go beyond the eikonal limit, we discuss now the Gaussian dipole approximation for
the path-integrals K. It is based on the observation that the main support in the path integral
(4.19) comes from small transverse distances r = |r|, where the cross section (r) in (4.8)
has a leading quadratic dependence:
(r) C(r) r 2 .
(4.53)
In explicit model calculations, one finds [15,22] that the r-dependence of C(r) is a slow
logarithmic one, and can be neglected. For the dipole cross section in (4.8), an expansion
to order r 2 confirms this feature. One finds
Zc 2
2
d 2
CF
a0 ( ) ,
(4.54)
C=
2
2
(2)
where the -integral depends logarithmically on the ultraviolet cut-off c which one has
to introduce in this approximation. We note that C provides a measure of the average
transverse momentum transfer h 2 i in a single scattering. For sufficiently small r, when the
r-dependence of C can be neglected, the path integral (4.19) reduces to that of a harmonic
oscillator [15]
A

exp iAB r21 + r22 2 iA r1 r2 ,
(4.55)
Kosz r2 , L; r1 , 0 =
i
,
(4.56)
A=
2 sin( L)
B = cos( L),
(4.57)
with the oscillator frequency
s
1 i n0 C
.
=
(4.58)
For what follows, it is important that the -limit of Kosz coincides with the eikonal
limit (4.45) of the unapproximated path-integral. To see this, we expand the norm of Kosz
to leading order in and the phase to next to leading order,
+ O(0 )
Kosz r2 , L; r1 , 0 =
2iL

Ln0 C 2
i
(re r)2
re + r2 + re r + O 1
exp
2L
6
(2)(re r) e 2 n0 LCr ,
1
(4.59)
which coincides with (4.45) for the quadratic ansatz (r) = Cr2 . The harmonic oscillator
approximation thus provides an explicit representation for the path-integral K which preserves the correct high energy limit. At least numerically, this allows for an explicit study
437
of the 1/-corrections to the eikonal limit of the cross sections (4.24)(4.26) and (4.37)
(4.39). To explore some qualitative features of these 1/-corrections analytically, we take
here recourse to a Gaussian approximation G of the incoming Born wavefunction:
G (r, r; ) = (r) (r),

1
exp r2 /R 2 ,
(r) =
R 2
1
= (1 )Q2 + m2 ,
R2
(4.60)
(4.61)
(4.62)
where the radius R is chosen to reproduce the characteristic width of the fall-off of the
exact Born probability given in (B.8) and (B.14).
The time-evolved final state wavefunction reads
Z

1 + O(1 )
f (r) = dr Kosz r, L; r, 0 (r) =
R 2
2

r
1
r2
r2
exp 2 n0 LC r2 + iqL + i c1 2 c2 + O 3 .
(4.63)
R
2
Here, we have used the phase convention (4.27) to obtain the term iqL in the exponent.
The shorthands c1 and c2 denote the leading real and imaginary 1/-corrections to the
final state wavefunction:

L 2 2 2
n0 LC
12
+
n0 L C + 6
,
(4.64)
c1 =
6
R2
R4

n20 L2 C 2
L2 3 3 3
n0 LC
60
+ 40
+ 6 .
n L C + 10
(4.65)
c2 =
15 0
R2
R4
R
Deviations from the eikonal limit can be calculated with the help of (4.63). For example,
we find for the total photoabsorption cross section (4.26):
Z
Z
q q
q q
lim total
= Nc em d dr G (r, r; )S(r)
total
2

r
2
2
c1 + q L .
(4.66)
2 1 er c2 / cos
In the -limit, this difference vanishes by construction. To understand which scales

determine the deviations from the eikonal limit, we recall that 1/(R 2 ) = q has an
interpretation as inverse coherence length 1/ lf and n0 LC is the total transverse momentum
squared accumulated during the rescattering over a distance L. The eikonal approximation
is thus justified if the phases determined by the coherence length and the total transverse
tot are negligible:
energy E
qL 1,
(4.67)
n0 LC
tot
L 1.
L=
E
2
(4.68)
We conclude this section with two technical remarks:
438
1. The explicit form of the incoming q-q wavefunctions (B.6), (B.13) is given in terms
of the Bessel function K0 which has an integral representation in terms of a Gaussian in r,
1
K0 |r| =
2
Z0
dyL i 4y2 r2 +iyL

L
e
.
yL
(4.69)
The evolution of the Gaussian wavefunction (r) in (4.63) can thus be used for the exact
calculation if 1/R 2 is taken to be the width 2 /(4yL ) of the integrand of (4.69) and the yL integration is done afterwards. However, to order 1/, terms proportional to 1/R 4 appear
in (4.63) and thus the yL -integration cannot be done analytically if 1/-corrections are
included. Only the -limit can be accessed analytically in this way.
2. A surprising difficulty occurs in the calculation of the photoabsorption cross section
(4.24) from the harmonic oscillator approximation (4.55) if one tries to do the re integration first:
Z

(4.70)
Fosz (r, r) = dre Kosz r, 0; re , L Kosz re , L; r, 0 .
After a lengthy but straightforward calculation, we find for the -limit
lim Fosz (r, r) = (2) (r r) e 4 n0 LCr ,
1
(4.71)
whose exponent differs from that of the eikonal dipole formula (4.45) by a factor 1/4. This
mismatch is an artefact stemming from a calculation which performs simplifications on
the cross section level before completing the dynamical evolution on the amplitude level.
Starting from (4.63) for f (re ) and calculating f (re ) f (re ), one obviously reproduces
for (4.24) the eikonal expression (4.47) in the -limit.
5. Photodissociation via non-abelian Stokess theorem

In this section, we argue that the dipole cross section (4.8) parametrizes the transverse
components of the chromoelectric field strength correlations hF F i in the nuclear medium.
This suggests an interpretation of (r) which holds model-independent, i.e., irrespective
of whether we have build up (4.8) from the model ansatz (2.10), or from some other model
parametrization of the colour target field.
We first recall the non-abelian Stokess theorem. This relates the integral of the nonabelian vector potential A along a closed contour C to the area integral of the field
strength tensor F [3436],
!
Z

I
dz A (z) = Tr Parea exp

d (y) UAy F (y) UyA .
(5.1)
Tr P exp
C
Here, the UAy are parallel transporters. They ensure that the r.h.s. is gauge-invariant. Eq.
(5.1) does not depend on the choice of the position A, since cyclicity of the trace allows
to change the reference point by substituting UAy UBA UAy . The area ordering Parea
439
is defined by disecting the area enclosed by C in many (ultimately: infinitesimally small)

areas in such an order that the parallel transporters Ci along the infinitesimal areas combine
to C. As a consequence of the non-abelian Gauss theorem, (5.1) turns out to be independent
of the orientation of the surface d (y) [36].
The total photoabsorption cross section (4.1) is closely related to the Stokess theorem,
as can be seen from the representation
q q
total
= em

db1 db2 db 1 db 2 1b; 1b;
Z
Dr1 Dr2 e
i
2
( r22 +(1)r12 )

i
Dr1 Dr2 e 2
2
2
( r 2 (1)r1 )
!+
Tr P exp
dz A (z)
(5.2)
C(r1 ,r2 ,r1 ,r2 )
Here, C is defined by the transverse paths r1 ( ), r1 ( ), r2 ( ), r2 ( ) which denote the

positions of the rescattering quark and anti-quark in the path integral representation (2.17)
of the corresponding Greens functions. We can consider C as a closed path since we
work for a static potential A = 0 A0 for which the Wilson line along purely transverse
directions dx A (x) vanishes. The photoabsorption cross section (5.2) is thus given by
a closed Wilson loop which sums up the gluon field strength of the medium in the area
determined by the paths of the rescattering q- and q-quarks.
The connection between total hadronic cross sections as (5.2) and closed Wilson loops
has been studied extensively, see, e.g., [26,38]. Phenomenological applications of the nonabelian Stokess theorem (5.1) were pioneered by Dosch and coauthors [37,38]. In Doschs
approach, one takes recourse to an axial gauge in which the parallel transporters on the r.h.s.
of (5.1) reduce to unit operators. The remaining expression is then expanded in powers
of F . Our discussion of (5.2) borrows from this strategy. Clearly, we cannot choose an
axial gauge to remove the parallel transporters in (5.1), since our gauge freedom is already
exhausted by the definition (2.10). However, we can invoke the proof of colour triviality
of (5.2) to discuss the equivalent abelian problem with a rescaled vector potential A
CF A . We find
*
exp ig CF
*
= exp ig
*
s
ds A (s )
s
!+
CF
s s
F (s )
ds dx
s x
Zz2
= exp ig CF
rZ2 ( )
i
dx
Ei (x , )
d
z1
r1 ( )
!+
!+
440
Zz2
=1
g 2 CF
d n( )
2
z1
*Z
rZ2 ( )
0j
i
dx
dx Ei (x , )Ej
x0 ,
r1 ( )
+O n ( ) .
2
(5.3)
Here, s denotes the path around the contour C, and the notation is specified in Fig. 11.
We have used the particular form of a static colour potential A = 0 A0 to rewrite the
r.h.s. of (5.1) in terms of transverse electric fields. The detailed arguments for the shift of
integration variables needed to arrive at the last equation of (5.3) are given in Appendix B
of Ref. [22]. The same appendix contains the derivation of the dipole cross section (4.8)
from a closed abelian Wilson loop:
!+
*
p I
s
ds A (s )
exp ig CF
s
C
Zz2
d A0 r1 ( ), A0 r2 ( ),
= exp ig CF
z1
Zz2
d n( ) r1 ( ) r2 ( )
= exp
!+
!
.
(5.4)
z1
Comparing (5.4) to (5.3), we can relate the dipole cross section to the two-point correlation
function of the transverse electric field strength,
r1 ( ) r2 ( ) =
*Z
rZ2 ( )
0j
i
dx
dx Ei (x , ) Ej
0
, +
x
+
.
(5.5)
r1 ( )
As long as we parametrize this dipole cross section by the Gaussian approximation (r) =
Cr 2 , only the average of the colour field strength enters our final results in form of one
single parameter C. The final result is model-independent in the sense that it does not
depend on the model-specific way in which the colour field strength giving rise to C was
modelled.
Fig. 11. The closed Wilson loop between longitudinal positions z1 and z2 , as considered in (5.3).
441
6. Conclusion
Colour triviality provides a crucial simplification of an otherwise untractable problem. In
general, the N -fold rescattering of a parton in a colour target field leads in the cross section
to colour traces over 2N generators. For sufficiently exclusive processes, the huge number
of colour interference terms thus limits explicit calculations of the soft nuclear dependence
of hard partonic processes to the case of very few rescatterings [24]. Colour triviality is
the consequence of a complete diagrammatic cancellation between these different colour
interference terms. For colour trivial observables, all non-abelian complications reduce
to the rescaling of coupling constants by appropriate powers of Casimirs. The multiple
scattering problem thus becomes an abelian one. The soft nuclear dependence of hard
colour-trivial observables can be described in terms of a QCD dipole cross section which
absorbs the leading medium-dependence in a one-parameter estimate C of the average
target colour field strength. This makes the identification of colour-trivial observables of
particular interest for relativistic heavy ion collisions at RHIC and LHC where very little is
known about the medium a priori but many hard probes are expected to receive sizeable
nuclear modifications due to rescattering effects.
In the present work, we have derived an explicit general expression for the rescattering
effects on a hard coloured parton inside a spatially extended nuclear medium. We have
then studied in detail for the simple example of virtual photodissociation under which
conditions this general rescattering formula results in colour-trivial observables which can
be parametrized by a QCD dipole cross section. In detail:
The non-abelian Furry wavefunction (2.12) derived in Section 2 is a high-energy
approximation to the solution of the non-abelian Dirac equation in a spatially extended
colour field. In contrast to other approximate solutions [29], it is accurate up to order
O(1/E 2 ) in the phase. This is known to be indispensable for calculating the nuclear
dependence of observables which are determined by the destructive interference between
different production amplitudes as, e.g., the LPM-effect or the nuclear dependence of
DrellYan yields. The Furry wavefunction (2.12) thus provides a unified starting point
for the description of the nuclear dependence of a large class of observables.
Based on the Greens function G which describes the dynamical evolution of the
non-abelian Furry wavefunction, we have derived in Section 4 a set of diagrammatic
identities which play the key role in proofs of colour triviality. These identities exploit
the integration over the transverse momentum of a single final state particle only. They are
thus much stronger than diagrammatic cancellations implied by the optical theorem. As a
consequence, they ensure for the example of the q q process a colour trivial result not
only for the total, elastic and inelastic inclusive photoabsorption cross section but also for
cases in which one jet is resolved in the final state. All these observables can be described
in terms of the same one-parameter QCD dipole cross section.
Earlier applications of these diagrammatic techniques as well as statements about the
colour triviality of the photoabsorption cross section exist [5,8,9,1619,28]. Here, we have
given relatively short and complete proofs of these statements by exploiting the advantages
of a new and compact configuration-space notation implied by the Furry wavefunction.
442
In contrast to previous discussions, our formulation includes the transverse dynamical

evolution of the partons. For the example of the photoabsorption cross section studied
here, this allowed us to quantify the leading deviations from the eikonal limit which grow
at fixed energy proportional to L2 . Moreover, the compactness of our notation made it
possible to discuss in detail technically rather involved processes as, e.g., the inelastic and
total photoabsorption cross section with one jet resolved in the final state. This points to the
strength of the present approach which we believe to be suited for the explicit discussion
of more complicated processes as, e.g., further studies of the transverse momentum
dependence of the non-abelian LPM-effect [18,22] or photodissociation including initial
state gluon radiation [33].
Acknowledgements
I am indebted to Miklos Gyulassy for many helpful and inspiring discussions, a
critical reading of this manuscript, and the hospitality extended to me at Columbia
University. I thank Al Mueller for several discussions about contact terms which have
strongly influenced Sections 4.44.6. Thanks go to Yuri Kovchegov for his patience in
explaining to me many technical details of Refs. [9,28]. Helpful comments by E. Iancu,
P. Levai, E. Levin, A. Leonidov, L. McLerran, R. Venugopalan and I. Vitev are gratefully
acknowledged. This work was supported by a CERN Fellowship and by the Director, Office
of Energy Research, Division of Nuclear Physics of the Office of High Energy and Nuclear
Physics of the U.S. Department of Energy under Contract No. De-FG-02-92ER-40764.
Appendix A. Non-abelian Furry approximation

In this appendix, we derive the non-abelian Furry wavefunction (2.12) from the set of
N -scattering Feynman diagrams (2.11). For an N -fold rescattering diagram, we use the
notation pN+1 p for the final state momentum. To simplify I (N) (y) in (2.11), we do the
longitudinal momentum integrals by contour integration
Z
dpiL i (/
pi + m) 0 i pL (x L x L )
e i i i1
2
(2) pi m2 + i
i pL (x L x L )
(/
p + m) 0
L
= i L
xiL xi1
(A.1)
e i i i1 .
2pi
We note that, strictly speaking, the position variables xi in this expression are integration
variables and do not coincide with the center of the ith scattering potential. A more detailed
analysis of (2.11) involves for a particular model also the study of the piL -poles of the
Fourier-transformed single scattering potentials. For a Yukawa-type potential 1/[(pi
pi1 )2 + M 2 ], e.g., these poles give additional contributions to the piL -integrals, which
are however exponentially suppressed due to the Debye-screening mass M. These details
are discussed explicitly in Refs. [13,16] and for contact terms in Ref. [39]. In the end, the
443
only O(1/E)-contribution to the piL -integration turns out to be given by (A.1). On the
r.h.s. of this equation, piL is determined by the pole value to order O(1/E), piL = p
where p2 = E 2 m2 . Using x0 y, we can rewrite (2.11) as
N Z
Y

d2 p
L
i
d3 xi xiL xi1
I (N) (y) = e ipyL P
2
(2)
i=1
!
2
i p (x x ) i pi (x L x L )
(/
pi + m) 0
iA0 (xi ) e i i i1 e 2 p i i1
2 piL
eipxN +ip xN v (r) (p).
L
pi
2p
(A.2)
We now consider the spinor structure of this expression. To order

propagator introduces a numerator
O(1/E 2 ),
each quark
(/
pi + m)0 E(0 3 )0 p
i 0 + 3 0
pi
.
2p
(A.3)
The normalization of I (N) (y) is needed to leading order in energy only. This allows us to
neglect the mass term in (A.3). For the same reason, it is sufficient to keep only the leading
term E(0 3 )0 for all numerators (A.3) with i > 2. Further simplification of (A.2) is
then possible, using
n
(A.4)
E(0 3 )0 = 2n1 E n (0 3 )0 .
The case i = 1 is different: depending on the explicit form of the production vertex, the
leading order contribution of this numerator can cancel. We have to keep the numerator
(/
p 1 + m)0 to order O(1/E). To this aim, we substitute in (A.2)
(A.5)
(/
p1 + m)0 E(0 3 )0 i 0 ,
y
where the differential operator acts on the big bracket in (A.2). Equation (A.5) is equivalent
to (A.3) with the transverse components written in configuration space. In a coordinate
system with pkn, the above equation takes the form

2

(p )2
L (p1 )
p
/ 1 0 = p(0 3 )0 + p
1
2p
2p
(p pn).
(A.6)
p (0 3 )0 + i
y
In the same coordinate system, 0.5 (1 3 0 ) v(p) = v(p). Acting with (A.6) on v(p), we
b v(p). With these steps, the
thus find to leading order in E an expression proportional to D
amplitude (A.2) takes the form
!
Z Y
N

L
bP
d3 xi xiL xi1
I (N) (y) = eipyL D
i=1
N
Y

i=1
iA0 (xi ) G0 (xi1 , xi |p)
444

dx G0 (xN , x|p)F x , xL , p v (r) (p).
(A.7)
Here, we have introduced two elements of the abelian Furry approximation: the outgoing
transverse plane wave F and the free Greens function G0 of (2.15) which we have
identified here with the Gaussian p
i -integrals,

L
L
G0 x
i1 , xi1 ; xi , xi p
Z
=
p
d2 p
i 2i p
i p
i
i (xi xi1 ) e
e
(2)2
L )
(xiL xi1
(A.8)
For the path-ordered product in (A.7) which involves these Greens functions, we introduce
the shorthand
G(N) (y , yL ; x , xL |p)
!
N+1
YZ

3
L
L
L
d xi xi xi1 G0 y , yL ; x
=P
1 , x1 p
i=1
!
N
Y

L
L
iA0 (xi ) G0 xi , xi ; xi+1 , xi+1 p ,
(A.9)
i=1
where xN+1 = x . Allowing for arbitrary many gluon exchanges, we have to sum over N .
G( y , yL ; x , xL |p) =
G(N) (y , yL ; x , xL |p).
(A.10)
N=0
From this one sees easily that G(N) corresponds exactly to the N th order O(AN
0 ) term in
(2.14). The sum (2.12) over N -fold scattering diagrams takes the form of a non-abelian
extension of the Furry approximation.
We emphasize that the approximations used in our derivation and especially the different
treatment of the quark propagators at the production vertex and in the final state rescattering
part of the amplitude I (N) leads to a consistent high energy expansion with a norm accurate
to leading order in O(1/E) and a phase factor accurate up to order O(1/E 2 ). The same
approximations were employed in recent calculations of rescattering amplitudes for which
the production vertex P is a photon or gluon emission vertex [13,27].
Appendix B. qq Fock states

In this appendix, we give details of the derivation of transverse and longitudinal
components of the squared incoming Fock state

+ L 1b; 1b;
.
= 1b; 1b;
(B.1)
1b; 1b;
2
We start with the longitudinal polarization L , satisfying L k = 0 and L = 1,

L =

1 p 2
+ Q2 , 0 , .
Q
(B.2)
The corresponding vertex function reads to leading order in energy

p
L = L b = Q (1 ) rr,r 0 ,
445
(B.3)
r0
where r, are the helicities of the quark and antiquark. The square of this longitudinal
vertex function, summed over the spin of the final state particles, reads
X
L L = 2Q2 (1 ) .
(B.4)
r,r 0
This is a kinematical prefactor which factorizes in the integrand of (3.7). For the first
interaction in the photodissociation amplitude in (3.7), we take some longitudinal position
za . The y-integration in (3.7) can then be done analytically. We find
IL (b2 b1 |yL , za )
Z
= d2 y G0 (b2 , za ; y|p2 )G0 (y; b1 , za |p1 )

i (b2 b1 )2
exp
,
=
2i (za yL )
2 (za yL )
(B.5)
where = (1 ). The virtual photon has to dissociate at longitudinal position yL < za

in order to make an interaction at za possible. This limits the range of the yL -integral:
Zza
IL (1b|za ) =
dyL IL (b2 b1 |yL , za )
=
=

(1 ) i q za
e
2K0 |1b| ,
2i
(1 ) Q2 + m2 .
(B.6)
(B.7)
If the first interaction occurs at za in both the amplitude and complex conjugate amplitude,
then the square of the incoming longitudinal q q Fock wavefunction is defined as the
combination of the squared emission vertex (B.4) and the free time evolution (B.5),
P
L L

r,r 0
a
IL (1b|za )IL 1b|z
L 1b; 1b; =
2
4 (1 )

Q2 2 (1 )2
.
4K0 |1b| K0 |1b|
(B.8)
=
2
2 (2)
This expression appears for longitudinal polarization in the photodissociation cross section
(3.9).
For our calculations in Section 4, we have to consider the more general case that the first
interaction vertex occurs at za in the complex conjugated amplitude Mfi , but at zb > za in
the amplitude Mfi . This implies the further free evolution of (B.5) from za to zb ,
Z

IL (b2 b1 |yL , zb ) = d2 b01 d2 b02 G0 b2 , zb ; b02 , za p2

IL b02 b01 yL , za G0 b01 , za ; b1 , zb p1
446

drb (za )K0 rb (zb ), zb ; rb (za ), za (1 )

IL rb (za )yL , za .
(B.9)
Here, we have used the notation introduced in equations (4.16)(4.19) with rb (za ) =
b02 b01 and rb (zb ) = b2 b1 . Equation (B.9) allows us to circumvent a notational problem
which stems from the identity:
Zzb
dyL I (1b|yL , zb ) = e
iq(zb za )
Zza
dyL I (1b|yL , za ).
(B.10)
Equation (B.10) shows that if the points of first interaction differ in Mfi and Mfi , then
phase factors exp(iq(zb za )) arise in the squared free incoming wavefunction (B.8).
Hence, strictly speaking, we cannot do the yL -integral before specifying the first points
of interaction. Previous discussions of the q-q-dipole
do not have this difficulty, since they

neglect these phase factors a praxis which is justified for nuclear targets of size L
1/q. If we want to keep the phases exp(iq(zb za )) and yet write the photoabsorption
cross section (3.9) in terms of L , we can use (B.9) for a simple convention: whenever the
free Greens function K0 (rb (zb ), zb ; rb (za ), za |(1 )) acts on the first argument 1b
), the result is a phase exp(iq(zb za )), whenever it acts on the second
of L (1b; 1b;
argument, the result is the complex conjugated phase. This allows us to expand the cross
section (4.23) in powers of the opacity without neglecting phase factors and without giving
up the simple representation of q q in terms of .

We now turn to the transverse polarizations
1
() = (0, 1, i, 0),
2
T
(y) = ()

1
=
r,r 0 iy () r 0 (1 2) +
2 (1 )

m
+ r,r 0 1 + r 0 .
2
(B.11)
(B.12)
In contrast to the longitudinal case, this vertex does not factorize in the integrand of (3.7).
However, the y-integration can be done in analogy to (B.6),
I (1b|za ) =
Zza
Z
dyL
d2 y eiqyL |yL | G0 (b2 ; y|p2 )T (y)G0 (y; b1 |p1 )

(1 ) iqza
e
2T (1b)K0 |1b| .
2i
The squared transverse wavefunction reads then
1P

,r,r 0
I 1bza I 1b za
1b; 1b; = 22
4 (1 )
=
(B.13)

2
1b 1b
K1 1b
K1 1
b 2 + (1 )2
2
2 (2) |1b| |1b|

2

m
K0 ( 1b)K0 1b .
+
2
2(2)
447
(B.14)
For the case that the points of first interaction za and zb are different in the amplitude and
complex conjugated amplitude, the convention explained below (B.10) applies.
The arguments of the modified Bessel functions K0 and K1 in (B.8) and (B.14) specify
the transverse
p separation of the q-q dipole pair. This separation is given approximately
by 1/ = 1/ (1 )Q2 + m2 . The higher the virtuality of the photon, the smaller the
transverse size of this dipole. In general, the approximation of the virtual photon by the
q-q Fock state is reasonable at sufficiently high virtuality, Q2 1 GeV2 , above the vector
meson dominance region [7]. There, the q-q separation is small enough to render the
effects from string tension negligible and the quarks undergo independent scatterings. For
a recent attempt to model interactions between the quarks, see Ref. [25].
Appendix C. Classification of diagrams for (4.30)

In this appendix, we use the identities Fig. 7(a) and (b) to simplify the diagrammatic
contributions to the inelastic part of the differential photoabsorption cross section (4.30).
For each N th order term in the opacity expansion of (4.30), we use the notational shorthand
N
Y
Aini ,mi ,o .
(C.1)
i=1
Here, the index i labels the scattering potentials linked to the N th order term, i = 1 being
the first scattering potential. The superscripts ni , mi , o denote how the ith scattering center
is connected to the q-q-system.
Each scattering center is linked twice to the q-q-system,
once a momentum flows into the system, once it flows out. ni [1, 2] specifies whether
the inflowing momentum is transfered to the upper quark line of momentum p1 (ni = 1) or
the lower quark line of momentum p2 (ni = 2). Analogously, mi [1, 2] specifies whether
the outflowing momentum comes from the upper (mi = 1) or lower (mi = 2) quark line.
The remaining superscript o [r, v, w] specifies the position of the two vertices from the
ith interaction w.r.t. the cut: either both vertices stand to the right of the cut (o = v) or to
the left of the cut (o = w), or they are a real contribution (o = r) where one vertex stands
to the right and the other to the left of the cut. We illustrate this notation with the examples
in Fig. 12.
All N th order contributions to the differential photoabsorption cross section (4.30) fall
into exactly one of the following three classes:
in the diagram (C.1).
1. There is at least one factor A11,r
j
or A21,r
in the diagram (C.1), but no factor A11,r
2. There is at least one factor A12,r
j
j
k
(arbritrary k) and no factor A22,r
with k < j .
k
448
Fig. 12. Example of our notation (C.1) for diagrammatic contributions to the photoabsorption cross
section. For more details, see the text following (C.1).
3. (a) There is no factor A11,r

(arbritrary k) but at least one factor A12,r
or A21,r
in the
k
j
j
with k < j ,
diagram (C.1), and there is at least one factor A22,r
k
or
12,r
or A21,r
in the diagram (C.1).
(b) There are no factors A11,r
j , Aj
j
We now use the identities of Fig. 7(a), (b) to show that many of the diagrams in these three
classes cancel each other. In this way we determine the only remaining contribution for
each of the three classes:
Class 1
, A12,o
or
The only non-vanishing contributions are those which contain no terms A22,o
l
l
21,o
Al with o [r, v, w]. They are shown in Fig. 10(a1).
, A12,o
or A21,o
.
Argument. Consider the subclass of diagrams containing factors A22,o
l
l
l
In case that there is more than one such factor in the diagram, consider the term with
the largest index l, l = k say. Leaving all terms j 6= k unchanged, we find in this class
22,v
and A22,w
on the kth position. These
of diagrams exactly one diagram with A22,r
k , Ak
k
three diagrams cancel each other due to the identity Fig. 7(a). Also, for all terms j 6= k
and A21,v
unchanged, we find exactly one diagram with A21,r
k
k , which cancel due to the
and A12,w
cancel
identity Fig. 7(b). For the same reason, the two diagrams with A12,r
k
k
12,w
21,w
= Ak
specifies the same kth term in the same diagram). As
each other (note that Ak
, A12,o
or A21,o
gives a nona consequence, no contribution which contains terms A22,o
l
l
l
vanishing contribution.
Class 2
or A21,r
with
The only non-vanishing contributions contain exactly one real term A12,r
j
j
12,v
12,w
or A22,r
, A22,v
no real term A11,r
k
k , k arbitrary, and with no contact terms Ak , Ak
k , or
22,w
for k > j . They are shown in Fig. 10(a2).
Ak
or A21,r
with the lowest index j .
Argument. Choose in each diagram the term A12,r
j
j
12,o
or A21,o
Consider diagrams which contain terms A22,o
k , Ak
k , with k > j and o = r, v, w.
Taking k maximal and leaving all other terms unchanged, these diagrams cancel due to the
identities Fig. 7(a) and (b).
Class 3
but no real
The only non-vanishing contributions contain exactly one real term A22,r
j
12,r
21,r
22,w
and no contact terms A22,v
, A12,v
or A12,w
with k > j .
terms A11,r
j , Aj , or Aj
k , Ak
k
k
They are shown in Fig. 10(a3).
449
Argument. Consider in each diagram the term A22,r

with lowest index j . Look then at
j
the largest index k > j linking to the p2 -line. Leaving all other terms unchanged, the sum
of the contributions with different configuration at position k ensures cancellation: A22,r
k ,
22,v
22,w
21,r
21,v
12,r
cancel each other due to identity Fig. 7(a). Ak , Ak and Ak , A21,w
Ak and Ak
k
cancel each other due to identity Fig. 7(b). Thus only contributions with exactly one real
survive cancellation.
term A22,r
j
References
[1] G.T. Bodwin, S.J. Brodsky, G.P. Lepage, Phys. Rev. D 39 (1989) 3287.
[2] B.Z. Kopeliovich, Soft component of hard reactions and nuclear shadowing (DIS, Drell
Yan reaction, heavy quark production), in: H. Feldmeier, W. Nrenberg (Eds.), Proc. of the
Workshop Hirschegg95: Dynamical Properties of Hadrons in Nuclear Matter, Hirschegg,
January 1621, 1995, Darmstadt, 1995, p. 102; hep-ph/9609385.
[3] S.J. Brodsky, A. Hebecker, E. Quack, Phys. Rev. D 55 (1997) 2584.
[4] A.H. Mueller, J. Qiu, Nucl. Phys. B 268 (1986) 427.
[5] A.H. Mueller, Nucl. Phys. B 335 (1990) 115.
[6] N.N. Nikolaev, B.G. Zakharov, Z. Phys. C 49 (1991) 607.
[7] G. Piller, W. Weise, hep-ph/9908230.
[8] A.H. Mueller, Nucl. Phys. B 558 (1999) 285303.
[9] Y.V. Kovchegov, L. McLerran, Phys. Rev. D 60 (1999) 054025.
[10] M. Luo, J.W. Qiu, G. Sterman, Phys. Rev. D 49 (1994) 4493; Phys. Rev. D 50 (1994) 1951.
[11] X. Guo, Phys. Rev. D 58 (1998) 036001; Phys. Rev. D 58 (1998) 114033.
[12] R.J. Fries, A. Schfer, E. Stein, B. Mller, Phys. Rev. Lett. 83 (1999) 4261; hep-ph/0002074.
[13] M. Gyulassy, X.-N. Wang, Nucl. Phys. B 420 (1994) 583.
[14] X.-N. Wang, M. Gyulassy, M. Plmer, Phys. Rev. D 51 (1995) 3436.
[15] B.G. Zakharov, JETP Lett. 63 (1996) 952; JETP Lett. 65 (1997) 615.
[16] R. Baier, Y.L. Dokshitzer, A.H. Mueller, S. Peign, D. Schiff, Nucl. Phys. B 483 (1997) 291.
[18] R. Baier, Y.L. Dokshitzer, A.H. Mueller, D. Schiff, Phys. Rev. C 58 (1998) 1706.
[19] R. Baier, Y.L. Dokshitzer, A.H. Mueller, D. Schiff, Nucl. Phys. B 531 (1998) 403.
[20] B.G. Zakharov, Phys. Atom. Nucl. 61 (1998) 838 [Yad. Fiz. 61 (1998) 924]; hep-ph/9807540.
[21] B.Z. Kopeliovich, A. Schfer, A.V. Tarasov, Phys. Rev. C 59 (1999) 1609.
[22] U.A. Wiedemann, M. Gyulassy, Nucl. Phys. B 560 (1999) 345382.
[23] R. Baier, Y.L. Dokshitzer, A.H. Mueller, D. Schiff, Phys. Rev. C 60 (1999) 064902.
[24] M. Gyulassy, P. Levai, I. Vitev, Nucl. Phys. B 571 (2000) 197.
[25] B.Z. Kopeliovich, A. Schfer, A.V. Tarasov, hep-ph/9908245.
[26] O. Nachtmann, Ann. Phys. (N.Y.) 209 (1991) 436.
[28] Y.V. Kovchegov, A.H. Mueller, Nucl. Phys. B 529 (1998) 451.
[29] W. Buchmller, A. Hebecker, Nucl. Phys. B 476 (1996) 203.
[30] P. Hoyer, S. Peign, Phys. Rev. D 57 (1998) 1864; Phys. Rev. D 59 (1999) 034011.
[31] R.D. Field, Applications of Perturbative QCD, Addison-Wesley, 1989.
[32] B.G. Zakharov, Sov. J. Nucl. Phys. 46 (1987) 922.
[33] W. Buchmller, T. Gehrmann, A. Hebecker, Nucl. Phys. B 537 (1999) 477.
[34] N.H. Christ, Phys. Rev. Lett. 34 (1975) 355.
[35] I.Ya. Arefeva, Theor. Math. Phys. 43 (1980) 353.
[36] P.M. Fishbane, S. Gasiorowicz, P. Kaus, Phys. Rev. D 24 (1981) 2324.
[37] H.G. Dosch, Yu.A. Simonov, Phys. Lett. B 205 (1988) 339.
450
[38] H.G. Dosch, E. Ferreira, A. Krmer, Phys. Rev. D 50 (1994) 1992.

[39] M. Gyulassy, P. Levai, I. Vitev, work in preparation.

Aspects of high density effective theory in QCD

Deog Ki Hong a,b,1
a Lyman Laboratory of Physics, Harvard University, Cambridge, MA 02138, USA
b Physics Department, Boston University, Boston, MA 02215, USA
Received 31 May 1999; revised 10 April 2000; accepted 22 May 2000
Abstract
We study an effective theory of QCD at high density in detail, including the finite temperature
effects and the leading order correction in 1/ expansion. We investigate the Cooper pair gap
equation and find that the colorflavor locking phase is energetically preferred at high density. We
also find the color-superconducting phase transition occurs in dense quark matter when the chemical
potential is larger than 250 100 MeV and the temperature is lower than 0.57 times the Cooper
pair gap in the leading order in the hard-dense-loop approximation. The quarkneutrino four-Fermi
coupling and the quarkaxion coupling receive significant corrections in dense quark matter. 2000
PACS: 12.20.Ds; 12.38.Mh; 11.10.Gh
Keywords: Finite density/temperature QCD; Effective theory; Cooper pair
1. Introduction
Effective field theory is indispensable to describe physical processes at energy lower
than the characteristic scale of theory, especially when the microscopic physics is too
complicated to be used as in the case of chiral perturbation theory [1,2] or just unknown as
in the standard model. Since effective field theory has been very useful and successful in
explaining the low energy data, it has now become an everyday language in physics ([35],
for pedagogical lectures on effective field theories see [3]).
Color superconductivity in cold and dense quark matter [6] has been studied quite
intensively in recent years by using effective field theory methods [712]. Most important
ingredient in those studies was the existence of effective four-quark interactions that mimic
the four-fermion interaction generated by phonon exchange in BCS superconductivity,
which are assumed to be generated in the effective Lagrangian either by exchange of
1 On leave from Department of Physics, Pusan National University, Pusan 609-735, South Korea. E-mail
address: dkhong@hyowon.cc.pusan.ac.kr
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 3 0 - 8
452
D.K. Hong / Nuclear Physics B 582 (2000) 451476
massive gluons [11] or by strongly coupled instantons [79]. For low density quark matter,
where strong interaction is no longer weak, such four-Fermi interactions may describe
QCD at low density as model interactions, thus leading to color superconducting gap of
QCD scale ( QCD ). But, it was recently argued quite convincingly by Son [13] that
at least for high density quark matter, where perturbative QCD is applicable, magnetic
gluons are not Debye-screened but only dynamically screened or Landau damped. Such
long-range magnetic gluon interaction
is more important and leads to a bigger Cooper
5
2
pair gap, gs exp(3 / 2gs ), than the usual BCS gap, which is subsequently
confirmed by the SchwingerDyson analysis for the Cooper-pair gap [1417].
Recently, an effective field theory of QCD at high density [14] is derived systematically
in the power expansion of 1/ by integrating out the anti-quarks which are decoupled
at low energy in dense quark matter. In the effective theory, four-quark operators are
generated at the one-loop matching and the electric gluons are screened due to quarks in the
Fermi sea while the magnetic gluons are not at least in perturbation [13,1821]. At scales
below the electric screening mass, the relevant interactions for quarks are the coupling with
magnetic gluons and the four-quark interaction with opposite momenta. Both interactions
for quarks in the color anti-triplet channel with opposite momenta are shown to lead to
color superconductivity [13,14].
According to the quarkhadron phase continuity in QCD, conjectured recently by
Schfer and Wilczek [22], the confining phase of nuclear matter at low density is
complementary to the Higgs phase quark matter at high density. We support this conjecture
by showing that in the effective theory, which is equivalent to QCD in the asymptotic
density, the colorflavor locking diquark condensate is energetically more preferred.
Furthermore, in the effective theory, the higher order corrections are systematically
calculable and one can easily estimate quantities like the critical density, which are
determined by the sub-leading operators in 1/ expansion.
In this article, we study the effective Lagrangian of QCD at high density derived in [14]
in detail, including the higher order corrections in 1/ expansion and the finite temperature
effects. We then analyze the Cooper-pair gap equation more rigorously and calculate the
gap, the critical temperature, and the critical density. As applications, we examine the
effect of marginal four-quark operators on the neutrinoquark four-Fermi coupling and
the quarkaxion coupling in super-dense quark matter.
2. High density effective theory

The idea of effective field theory is quite simple and its construction can be performed in
a systematic way. First, one identifies or guesses the right degrees of freedom to describe
the low energy processes. Then, one integrates out the irrelevant degrees of freedom,
which results in nonlocal interactions. By expanding the nonlocal interactions in powers of
momentum at low energy, one derives a (Wilsonian) effective Lagrangian, which is usually
done in practice by matching all the one light-particle irreducible amplitudes in the full
theory with those amplitudes in the effective theory in loop expansion. Since the higher
453
dimensional operators have smaller effects at low energy, the effective theories provide a
very useful approach to describe low energy physics.
A system of degenerate quarks with a fixed baryon number is described by the QCD
Lagrangian density with a chemical potential ,
1 a a
D
0 ,
/ F
F
+
(2.1)
LQCD = i
4
where the covariant derivative D = + igs Aa T a and we neglect the mass of quarks for
simplicity. The chemical potential is introduced as a Lagrangian multiplier to constrain the
system to have a fixed baryon number, NB . At zero temperature, all the states up to the
Fermi surface are filled,
Z

d3 p
NB
= 2Nf
|
p
E
|
,
(2.2)
F
V
(2)3
where V is the volume, Nf the number of quark flavors, and pEF is the Fermi momentum.
The chemical potential , which defines the energy at the Fermi surface, provides another
scale for QCD. Here we consider a super-dense quark matter where quarks are very dense
so that the average inter-quark distance is much smaller than the characteristic length of
QCD, 1/ 1/QCD . For high chemical potential, the pair creation of a particle and an
anti-particle is suppressed at low energy because of the energy gap , provided by the
Fermi sea. Therefore, the relevant degrees of freedom at low energy will be just particles
and holes, excited near the Fermi surface, together with gluons.
We will derive a low energy effective Lagrangian for cold and dense quark matter by
integrating out the high frequency or irrelevant modes, la Wilson, as we run QCD from
high energy E > to low energy E QCD . Since QCD is asymptotically free, quarks
will interact weakly at energy E > ( QCD ) and their spectrum will be approximately
given by the free Hamiltonian:

(2.3)
E pE = E ,
E
where E = 0 E and denote the energy eigenfunctions with eigenvalues E = |p|,
satisfying E pE = |pE | , respectively. At low energy E < , the states + near the
Fermi surface, |pE | , are easily excited, while , which correspond to the states in the
Dirac sea, are completely decoupled due to the presence of the energy gap provided by
the Fermi sea. Therefore, the right degrees of freedom below consist of gluons and +
only.
If = 0, integrating out the high frequency modes in QCD just renormalizes the
coupling constants without generating any new interactions as it should be for a
renormalizable theory like QCD. However, when 6= 0, the energy spectra of particles
and anti-particles are asymmetric. Anti-particles ( ), which correspond to the absence of
states ( ) in the Dirac sea, have energy larger than and will be decoupled at energy
scale lower than . Therefore, the minimal quarkgluon coupling that connects with
+ (or vice versa), has to be removed at low energy, generating a new interaction among
+ . We will see later that integrating out the modes whose frequency is higher than
results in an effective Lagrangian which is drastically different from QCD.
454
We now integrate out the antiparticles ( ) or the states ( ) in the Dirac sea to derive
the low energy effective Lagrangian. We first decompose the quark momentum into the
Fermi momentum and a residual momentum as
p = v + l ,
(2.4)
= (0, vEF ) with Fermi velocity vEF . The

where the residual momentum |l | < and
q
2
magnitude of Fermi velocity, |E
vF | pF / pF + m2 , will be taken to be unity, for we
assume the quark mass m = 0.
Since we will integrate out the high frequency modes such that we will be left with
excitations only near the Fermi surface at low energy, it is convenient to decompose the
quark field by the Fermi velocity 2
X
(x) =
eiEvF Ex (E
vF , x),
(2.5)
v
vEF
where
(E
vF , x) =
|l |<
d4 l
(E
vF , l)eilx
(2)4
(2.6)
carries the residual momentum l . We introduce projection operators P = (1 E vEF )/2

and define
vF , x) P (E
vF , x).
(E
(2.7)
vF , x) corresponds to the states in the Dirac sea and + (E

vF , x)
In the limit l/ 0, (E
to the states above the Dirac sea.
In terms of the new fields, the QCD Lagrangian for quarks at high density becomes
iD
/ + 0 (x)
Lquark = (x)
X
vF , x)i 0 V D+ (E
vF , x) + (E
vF , x) 0 (2 + iV D) (E
vF , x)
+ (E
=
vEF
vF , x)i D + (E
vF , x) + + (E
vF , x)i D (E
vF , x) ,
+ (E
= (1, vEF ),
where
have used
V
= (1, E
vF ),
with
( 0 , vEF vEF
(2.8)
E ), and we
vF , x) + (E
vF , x) = + P P+ + = V + 0 + ,
+ (E
vF , x) (E
vF , x) = P+ P = V 0 ,
(E
vF , x) (E
vF , x) = + P P = + ,
+ (E
vF , x) + (E
vF , x) = P+ P + = + .
(E
(2.9)
In Eq. (2.8) the incoming quark and the outgoing quark have the same Fermi velocity vEF ,
since, at low energy, the momentum carried away by gluons can be compensated by the
2 For a given quark momentum, the corresponding Fermi velocity is determined up to reparametrization; v
EF
vEF + El / and lE El El, where El is a residual momentum perpendicular to the Fermi velocity. As in
the heavy quark effective theory [23], the renormalization of higher-order operators are restricted due to this
reparametrization invariance.
455
Fig. 1. The tree-level matching condition. Wiggly lines denote gluons; solid lines, states near the
Fermi surface; and double solid line, states in the Dirac sea.
residual momentum of the quarks to conserve the momentum, without changing the Fermi
velocity.
As we can see from Eq. (2.8), the propagation of quarks at high density is (1 + 1)-dimeni / is negligible. This dimensional reduction at high density can be also seen if
sional if l
we decompose the quark propagator as following:
i
pE E + 0

i
i
V
/
V
/
i
=
+
+ O l
/ ,
2 l V + il0
2 2 + l V + il0
iSF (p) =
(1 + i)p0 0
(2.10)
where quark momentum p = v + l and the i prescription is chosen such that not
only the anti-particles but also the holes correspond to the negative energy states moving
i / 0, 1/2V
/ 0 = 1/2(1 + E vEF ) is the projection
backward in time. Note also that, if l
/ 0 = 1/2(1 E vEF )
operator that projects out particles and holes (+ ), while 1/2V
projects out the states in the Dirac sea ( ). Since the excitation of quarks along the
perpendicular direction to the quark velocity does not cost any energy in the leading order,
the perpendicular momentum is nothing but an index that labels the degeneracy in quark
just as the degeneracy in the Landau level under external constant magnetic field is labeled
by the electron momentum perpendicular to the external magnetic field.
vF , x) is tantamount to eliminatAt tree-level, integrating out the irrelevant modes (E
ing (E
vF , x) by the equations of motion, given as
i 0
D
/ + (EvF , x)
2 + iD k

iD k n
i 0 X
D
/ + (EvF , x),
=
2
2
vF , x) =
(E
(2.11)
n=0
/ = D . Plugging Eq. (2.11) into the Lagrangian for quarks

where D k = V D and D
Eq. (2.8), we obtain the tree-level effective Lagrangian of QCD at high density,

1 a a X
F
+
vF , x)iV D+ (E
vF , x)
+ (E
L0eff = F
4
vEF

(D
/ )2 X
iD k n
+ (E
vF , x) .
2
2
n=0
(2.12)
456
Fig. 2. The one-loop matching condition for a four-quark amplitude.
3. One-loop matching and a four-Fermi operator

As we have seen in the previous section, integrating out the irrelevant modes generates a
new coupling between quarks and gluons. This can be also seen diagramatically as shown
in Fig. 1: in the leading order, the propagator of fast modes is replaced by a constant
matrix
i 0
i 0
+ O(l/),
=
2
2 + iD k
(3.1)
and the exchange of generates a new interaction given as, using P P = P ,

L0eff 3
g2 X
/ 0A
/ + + ,
+A
2
(3.2)
vEF
where the ellipsis denotes terms containing more powers of gluons and derivatives. By
writing the interaction in a gauge-invariant fashion, we recover the interaction terms in
Eq. (2.12).
The effect of integrating out fast modes not only generates new interactions at low
energy but also gives rise to quantum corrections to the tree-level couplings which can
be calculated explicitly by matching the loop amplitudes. Usually the new interactions
arise at tree-level upon integrating out the fast modes, but sometimes they may arise at
quantum level, especially when they are marginal, as in the low energy effective theory of
QED under external (strong) magnetic field [24].
As fermions under external fields [2426], in the presence of a Fermi sea [5] the scaling
dimension of fields at low energy changes due to the change in the particle spectrum;
the scaling dimension of quarks becomes 1/2 instead of 3/2, the canonical scaling
dimension of fermions in 3 + 1 dimensions. The scaling dimension of fields is determined
by the kinetic term in the action, which is for quarks in dense quark matter given in the
high density limit as
X Z d4 l
vF , l)(l0 vF lk )+ (E
vF , l)
(E
S0 =
(2)4 +
vEF
(E
vF , l)(2 + l V ) (E
vF , l) ,
(3.3)
+
where lk vEF lE is the residual momentum parallel to the Fermi velocity. Under a
scale transformation l0 sl0 and lk slk with s < 1, the quark fields transform as
vF , l) s 3/2 + (E
vF , l) and (E
vF , l) s 1 (E
vF , l). As we scale the momentum
+ (E
toward the Fermi surface or we decrease s 0, + scales like free massless fermions in
457
1 + 1 dimensions, while scales like heavy fermions in 1 + 1 dimensions and becomes

irrelevant at low energy. Since four-Fermi interactions are marginal for (1 +1)-dimensional
(light) fermions, we expect that a marginal four-quark operator will arise in the high density
effective theory of QCD by quantum effects if it is absent at the tree-level. To see this,
we consider one-loop matching of a four-quark amplitude for scattering of quarks with
opposite Fermi velocities, shown in Fig. 2, because only when the incoming quarks have
opposite Fermi momenta the four-quark operators are marginal as we scale toward the
Fermi surface [5]. Since the effective theory amplitude is ultra-violet (UV) divergent while
the amplitude in full QCD is UV finite by the power counting, we need a UV counter
term for the one-loop matching in the effective theory, which is nothing but a four-quark
operator. The one-loop four-quark amplitude of our interest in the effective theory is,
reverting the notation for + henceforth,

1
igs2 2
=
Aeff
4
2
2
Z
X
/ 0A
/ (EvF , x) A
/ 0A
/ (EvF0 , y)vE1 , l1 ; vE2 , l2
vE3 , l3 ; vE4 , l4
A
vEF ,E
vF0
x,y
g4
= s 2 (2)4 4
4
X
0
4 ) 0 T c T d (p2 )
3 ) 0 T a T b (p1 ) (p
pi (p
ac bd g g + ad bc g g
2

i 1
l i
ln
,
16 2
42
(3.4)
v2 ,
where vEi s are the Fermi velocities of incoming and outgoing particles with vE1 = E
P
vi + li , the Euler number, 0 pi p3 + p4 p1 p2 , l = l3 l1 , and is the
pi = E
renormalization point.
In QCD, the quarkquark scattering amplitude is
4

Z
X

gs4
QCD

p3 , p4
/ (x)(EvF , x) p1 , p2
(E
vF , x)A
A4 =
4!
vEF x
X
0
pi (p3 ) T a T b (p1 ) (p4 )

= gs4 (2)4 4
i
where
I1 =
q
Z
I2 =
q

I1 T a T b + I2 T b T a (p2 ),
(3.5)
1
1
1
,
2 + V (l3 + q) 2 + V (l4 q) (q + l)2 q 2
(3.6)
1
1
1
.
2 + V (l1 q) 2 + V (l4 q) (q + l)2 q 2
(3.7)
458
In the limit that the residual momenta li 0, we find I1 = I2 , if we rotate the q0 axis into
vEF qE axis, and thus the color and Lorentz structures of the amplitudes in both theories
are the same. To perform the integration, we take, for convenience, the external residual
momenta are perpendicular to the Fermi velocity, V li = 0. Then, we get for li / 0
Z1
I1 =
dx
0
1x
Z
Z
dy
q
2
[(1 x y)q 2 + x(q + l)2 + yq02 y(E
vF qE + 2)2 ]3

2
i
l
2
2
=
2 ln
+ O(l / ) .
64 2 2
42
(3.8)
is IR divergent as li 0, which is the same

Note that the scattering amplitude AQCD
4
infrared divergence of the amplitude in the effective theory, Eq. (3.4), as it should be
because the low energy effective theory is equivalent to the microscopic theory in the
infrared limit by construction. Since both amplitudes have to be same at the matching
scale = , we need a four-quark operator in the effective theory, given by the difference
of the amplitudes at the matching scale :

X
0
igs4
1
eff
4 4
4
(2)
A
=
pi (p3 ) T a T b (p1 )
AQCD
4
4
64 2 2
i

(p4 ) T a T b + T b T a (p2 ) + O(li /)
Z

(3.9)
= p3 , p4 i d4 x L14f p1 , p2 + O(li /),
where, at the matching scale , the new effective four-quark operator 3 generated at oneloop is given as
L14f =
g1bare X
vF , x) T a T b (E
vF , x)
(E
22
vEF

vF , x) T a T b + T b T a (E
vF , x) .
(E
(3.10)
In the modified minimal subtraction, the renormalized four-quark coupling is g1ren = 2s2
at the matching scale . It is convenient to rewrite the marginal four-quark operator by
Fierz-transforming the product of gamma matrices and SU(3)C generators in the amplitude
Eq. (3.10). Using (T a )t u (T a )vs = 1/(2t s uv ) 1/(6t uvs ) and the Fierz transformation,
we get

P+ P+ j i T a T b ut P T a T b + T b T a P lm;vs

= P+ P+ j i T a T b ut

1 ab
abc c
abc c
g P + d T + (P P )f T
3
lm,vs
3 If we had matched a quarkquark scattering amplitude with momenta not opposite to each other, we would
get similarly a four-quark operator in the effective Lagrangian. But, since it is irrelevant as we scale toward the
Fermi sea, it will not affect the low energy dynamics significantly.
459

11
5
ut vs + us t v
18
6

3
1
+ (P+ 5 )j i (P 5 )lm ut vs + us t v
2
2

13 S
2 A
uv;t s uv;t
= (P+ )j i (P )lm
s
9
9 2

A
1 S
2 uv;t s uv;t
(3.11)
(P+ 5 )j i (P 5 )lm
s ,
2
where i, j, l, m denote the Dirac indices and t, s, u, v the color indices. In the last line,
the color indices in the operator are further arranged to decompose the amplitude into the
irreducible representations of SU(3)C by introducing invariant tensors in color space,
S
A
and
uv;t
uv;t
s (ut vs us vt )/ 2.
s (ut vs + us vt )/ 2
= (P+ )j i (P )lm
The four-quark operator in the effective theory becomes then

1 X
vF , x)s (E
vF , x)v (E
vF , x)u (E
vF , x)
gus;t v t (E
L14f =
22
vEF
vF , x)5 s (E
vF , x)v (E
vF , x)5u (E
vF , x)
+ hus;t v t (E
A
gus;t v = g3 us;t
v
S
g6 us;t
v,
with
the matching scale are given
(3.12)
S
A
hus;t v = h3 us;t
h6 us.s;t
of couplings
v . The value
2 v
2
as g3 = 4 2s /9 = 2g6 /(13) and h3 = 4 2s = 2h6 .
at
4. Screening mass
As discussed in [1921,27], the quark loop correction to the vacuum polarization tensor
gives rise to color screening in quarkgluon plasma, while the gluon loop renormalizes the
color charge. We first calculate the vacuum polarization tensor in QCD and then compare it
with the one in the effective theory when we perform the one-loop matching for the gluon
two-point amplitude.
Since the Feynman propagator for quarks of mass m in matter at zero temperature is
given by
Z
d4 p ipx
i
,
(4.1)
e
iSF (x) =
(2)4
(1 + i)p0 0 pE E + 0 m
we find, performing the integration over p0 ,
Z
d3 p p
/ +m
(p0 )eipx+ix0 (x0 )
iSF (x) = (x0 )
3
(2) 2p0

Z
p
/ m ipx+ix0
/ +m
d3 p p
ipx+ix0
p
)e
+
e
,
(4.2)
0
(2)3 2p0
2p0
q
E 2 + m2 . We see that this propagator agrees with the one obtained by the
where p0 = |p|
canonical quantization [20] except the overall phase factor due to the shift in the energy by
to set the energy of the Fermi surface to be zero.
460
In terms of this full propagator Eq. (4.1), the quark-loop contribution to the one-loop
vacuum polarization tensor becomes
Z

(4.3)
ab full(p) = gs2 d4 x eipx Ja (x)Jb (0)
Z
iNf 2
d3 q 1 1
gs ab
=
2
(2)3 2q0 2k0

(q0 ) ( k0 )
( q0 ) (k0 )
T (q, k)
T (q, k)
q0 k0 + p0 i
q0 k0 + p0 + i
(q0 )
T (q, k 0 )
+
q0 + k0 + p0 i

(k0 )
00
T (q, k ) ,
(4.4)
q0 k0 + p0 + i
E kE0 = E
q p,
E and kE00 = E
q + p.
E
where T (q, k) = Tr [ q/ k/], kE = qE + p,
If we rewrite the step functions as (q0 ) = 1 ( q0) and (k0 ) = 1 (
k0 ) for the last two terms in Eq. (4.4), one can easily see that the quark-loop contribution
of the vacuum polarization tensor consists of two parts, one due to the matter and the other
due to the vacuum, ab full(p) = ab mat (p) + ab vac (p), where ab vac is the quark-loop
contribution when there is no matter (i.e., = 0),
Z
iNf 2
d3 q 1 1
gs ab
ab vac =
2
(2)3 2q0 2k0

T (q, k 00 )
T (q, k 0 )
,
(4.5)
q0 + k0 + p0 i q0 k0 + p0 + i
where Nf is the number of light quark flavors. For |p | , the matter part of the vacuum
polarization becomes, with M 2 = Nf gs2 2 /(2 2 ),

Z
dvEF 2pE vEF V V
iM 2
V V + V V
ab
, (4.6)
+ g
ab mat =
2
4
p V + i pE vEF
2
which is transversal, p ab mat (p) = 0 [20].

Now, we calculate the quark contribution to the one-loop vacuum polarization tensor
in the effective theory. Quarks in the (high density) effective theory are almost on-shell
at low energy and the quark current, which consists of states near the Fermi surface, is
proportional to its velocity, V = (1, vEF ):
X
+ (E
vF , x) T a + (E
vF , x)
J a (x)
vEF
V + (E
vF , x) 0 T a + (E
vF , x),
(4.7)
vEF
vF , x) = + (E
vF , x) and P P+ = P 0 V . Therefore, the
where we used P+ + (E
gluons transversal to the quark velocity is decoupled at low energy and the gluons
longitudinal to the quark velocity is coupled to the quark color charge density in a
combination given as
Aa (x)J a (x) =
+ (E
vF , x) 0 T a + (E
vF , x) V Aa (x).
461
(4.8)
vEF
Therefore, in the leading order in 1/ expansion, the quarkgluon coupling does not
involve the spin of quarks like a scalar coupling. For a given quark, moving nearly at a
Fermi momentum E
vF , only one component of gluons, namely A0 vEF AE combination,
couples to the quark in the leading order.
With the quark current given in Eq. (4.7), we obtain the quark-loop contribution to the
vacuum polarization in the effective theory as
Z
X

vF , x)Jb (E
vF0 , 0)0
0T Ja (E
ab (p) = gs2 d4 x eipx
= gs2 Nf ab
X
vEF
vEF
V V

1 E vEF
i
i
0
0
2
q/ k + i q/ k + p
/ k + i
(2)

0
2
X
(p + pE vEF )
i
M 2,
V V 1
= ab
2 + i
8
p
k
vE
Z
d4 q

Tr
4
(4.9)
(4.10)
pk
(p0 , vEF vEF
=
pE ). We note that the quark-loop contribution to the vacuum
where
polarization tensor in the effective theory is exactly same as the sum of the first and second
terms in the QCD vacuum polarization, Eq. (4.4). This is what it should be, since the
effective theory does reproduce the contribution by quarks and holes in QCD when the
external gluon momentum p 0. The third and fourth terms in Eq. (4.4), due to the quark
and anti-quark pair creation, is absent in the effective theory because the anti-quarks are
integrated out. The effect of anti-quarks will be taken into account when we match the
gluon two-point amplitudes.
To match the gluon two-point amplitudes in both theories, we therefore need to add a
term in the one-loop effective Lagrangian,
Leff 3
M 2 X a a
A A ,
16
(4.11)
vEF
which also ensures the gauge invariance of the effective Lagrangian at one-loop. The static
screening mass can be read off from the vacuum polarization tensor in the limit p0 0,
which is in the effective theory
ab (p0 0, pE ) ' iM 2 ab 0 0 .
(4.12)
We see that the electric gluons, Aa0 , have a screening mass, M, but the static magnetic
gluons are not screened at one-loop due to the added term Eq. (4.11), which holds at all
orders in perturbation as in the finite temperature [18,27].
Finally, by matching the quark two-point amplitudes, we get a one-loop low-energy
(Wilsonian) effective Lagrangian density;
462

1
M 2 a a
a 2
D
A A + (1 + b1 )i
Leff = (1 + a1 ) F
k
4
16
1
(1 + c1 ) ( D)2
2

1
A
S
vF , x)s (E
vF , x)v (E
vF , x)u (E
vF , x)
+ 2 g3 us;t
v g6 us;t v t (E
2

A
S
vF , x)5 s (E
vF , x)v (E
vF , x)5 u (E
vF , x)
+ h3 us;t
v h6 us;t v t (E
+ ,
(4.13)
where the summation over vEF is suppressed and the coefficients a1 , b1 , c1 are dimensionless and of order s (). The ellipsis denotes the irrelevant four-quark operators and terms
with more external fields and derivatives.
5. RG analysis and Gap equation

As we scale further down, the effective four-quark operators will evolve together with
other operators, which can be seen by further integrating out the high frequency modes,
s < |l | < . The scale dependence of the four-quark operators has three pieces. One is
from the one-loop matching condition for the four-quark amplitudes and the other two are
from the loop corrections to the four-quark operators. Putting all contribution together, we
find the one-loop renormalization group equations for the four-quark operators [14] to be
s
1 2 ln 2
gi = i s2
i gi s ,
g
s
12
4 2 i
(5.1)
g6 = g6 , g = g and i = ( 2/9)(13/4, 1/2) and i = (1, 2). Since

where i = (6, 3),
3
3
in the high density limit the quarkgluon coupling is (1 + 1)-dimensional, the quarks do
not contribute to the running of the strong coupling. The one-loop function for the strong
coupling constant at high density is (s ) = [11/(2)]s2 .
To solve the RG equation, Eq. (5.1), we introduce a new variable

1
ln 2
11
i s ()
s () ln s.
and t =
(5.2)
gi +
yi
22s ()
6
2
Then, the RG equations becomes
ai
dyi
= yi2
,
dt
(1 + t)2
(5.3)
where ai = (132i ln 2 + 144i i2 )/(17424). Further, letting yi = fi (t)/(1 + t), we get

dfi
= fi2 fi + ai .
dt
Integrating Eq. (5.4), we get
(1 + t)
fi (t) =
C2 (fi (0) C1 )(1 + t)(C1 C2 ) C1 (fi (0) C2 )

,
(fi (0) C1 )(1 + t)(C1 C2 ) (fi (0) C2 )
(5.4)
(5.5)
where
C1 =
1+
1 4ai
= 1 C2
2
and fi (0) =
463
gi () + ln 2i s ()/6
.
22s ()
Since ai 1, we take C1 ' 1, C2 1. And also, for QCD , fi (0) ' i ln 2/(132)
1. Therefore, for 1 + t 0+ or s exp[2/(11s ())], we get
gi (t) + ln 2i s (t)
' 22C2 s (t).
6
Namely,

2
(ln 2)2 2
s () i
i .
gi () '
11
144
(5.6)
At a scale much less than the chemicalpotential, , g6 () ' 0.29 s () and g3 () '
0.04 s (). Similarly for hi , i = ( 2/2)(2, 1) and i = (1, 2) and we get h6 () =
0.81 s () and h3 () ' 0.4 s ().
At a scale below the screening mass, we further integrate out the electric gluons, which
will generate four-quark interactions. For quarks moving with opposite Fermi momenta,
the electric-gluon exchange four-quark interaction is given as
L1g 3
gs2 (M) X
0 Ta (E
0 Ta (E
vF , x)
vF , x).
2M 2
(5.7)
vEF
= (1/2)t s uv (1/6)
the four-quark couplings are
Using
t 2uvs , we find that
2 (M) and g (M) g (M) +
(M)
g
(M)
+
2
2
g
(M)/3
'
0.95g
shifted
as
g
6
6
s
s
3
3
2
2 gs (M)/3 ' 0.49gs2 (M), while hi s are unchanged since Eq. (5.7) does not involve 5 .
As we approach further to the Fermi surface, closer than the screening mass M, the
four-quark operators in the color anti-triplet channel become stronger, because the
function for the attractive four-quark operators is negative, (g3 ) = g32 /(4 2 ). If the fourquark interaction is dominant at low energy, it leads to vacuum instability in the infrared
region by forming a color anti-triplet condensate or Cooper pair, but, since the long-range
color-magnetic interactions also become strong at low energy, we need to consider both
interactions to determine the Cooper-pair gap.
Since both relevant interactions are attractive for a pair of quarks in color anti-triplet
channel with opposite Fermi momenta, they may lead to condensates of quark pairs in
color anti-triplet channel. To describe the Cooper-pair gap equation, we introduce a charge
conjugate field,
a
Ttau Tvs
vF , x) = Cij j (E
vF , x),
(c )i (E
(5.8)
where i and j are Dirac indices and the matrix C satisfies C 1 C = T . Then, we can
vF , x), c (E
vF , x))T as
write the inverse propagator for 9(E
vF , x) ((E

(lk )
Z(lk )l V
vF , l) = 0
,
(5.9)
S 1 (E
(lk ) Z(lk )l V
where Z(lk ) is the wave function renormalization constant and (lk ) is the Cooper-pair
gap. The Cooper-pair gap is nothing but a Majorana mass for quarks. When Cooper-pairs
464
form and Bose-condense, the quarks get a Majorana mass dynamically and the Fermi sea
opens up a gap.
SchwingerDyson (SD) equations are infinitely coupled integral equations for Green
functions. In order to solve the SchwingerDyson equations, we need to truncate them
consistently. Since gs (M) is still weak at high density M QCD , one may use the
ladder approximation which is basically the weak coupling expansion of the SD equations,
consistently with the WardTakahashi identity. Another gauge-invariant truncation is socalled hard-dense-loop (HDL) re-summation, which can be derived gauge invariantly from
the transport equation [20] as in the finite temperature [2830]. When the gluon momentum
is of order of the screening mass M gs , the quark one-loop vacuum polarization is also
same order as we can see from Eq. (4.10). Therefore, if the important momentum scale
of the diagrams is of order of M, one needs to re-sum all the bubble diagrams to get
the consistent gluon propagator. Since the HDL re-summed gluon propagator correctly
incorporates the medium effects like screening at high density, we will solve the SD
equations for the quark propagator in the HDL approximation to calculate the Cooper-pair
gap.
In the HDL approximation, the SD equations for the quark propagator are given as,
neglecting small hi four-Fermi couplings,
Z

T a Z(lk )l V T a T
Z(pk ) 1 p V = (igs )2 V D (p l)V
Z 2 lk2 (lk )2
l
Z
g
Z(lk )l V
+ 32
,
(5.10)
2
Z lk2 (lk )2
l
(pk ) = (igs )2
g
+ 32
Z
l
V D (p l)V
T a (lk )T a T
Z 2 lk2 (lk )2
(lk )
,
(lk )2
Z 2 lk2
(5.11)
where the gluon propagator D is given in the HDL approximation as 4 , following the
notations used by Schfer and Wilczek [16],
T
P
L
P
k k
4 ,
k2 G k2 F
k
where is the gauge parameter and the projectors are defined by
iD (k) =
ki kj
T
,
P00
= 0 = P0iT ,
E2
|k|
k k
T
= g + 2 P
.
k
(5.12)
PijT = ij
(5.13)
L
P
(5.14)
4 In the SchwingerDyson equation the loop momentum should take the whole range up to the ultraviolet
cutoff, which is the chemical potential in the case of high density effective theory. Hence the gluon propagator
includes both magnetic and electric gluons.
465
Before solving the SD equations, we first show that the preferred solution of the
SD equations supports the colorflavor locking condensate predicted in [10]. Since the
gluon interaction is vectorial, the gluon exchange interaction in the gap equation does not
distinguish the handedness of quarks and thus it will generate same condensates regardless
of handedness; |hL L i| = |hR R i| = |hL R i|, suppressing other quantum numbers.
But, the four-Fermi interaction in the effective Lagrangian, Eq. (4.13), is chiral, which can
be seen if we rewrite it in the handedness basis as following:
L14f 3

g3 + h3 A
us;t v Lt (E
vF , x)L s (E
vF , x)Lv
(E
vF , x)L u (E
vF , x) + (L R)
2
2
g h A
+ 3 2 3 us;t
v
2

Lt (E
vF , x)L s (E
vF , x)R v (E
vF , x)R u (E
vF , x) + (L R) . (5.15)
We see that the LL (or RR) four-Fermi coupling is bigger than the LR four-Fermi
coupling, since h3 > 0. Therefore, the gap in LL or RR channel will be bigger than the
one in LR channel due to the difference in the four-Fermi couplings. Thus, the LL or RR
condensate is energetically more preferred than the LR condensate. We also note that in
the effective theory the gluons are blind not only to flavors but also to the Dirac indices
of quarks because they couple to quark currents in the combination of V A like a scalar
field, as given in Eq. (4.8). Therefore, in the SD equations, the diquark Cooper-pair can be
decomposed into color anti-triplet and color sextet. But, since quarks of opposite momenta
are attractive in the anti-triplet channel while repulsive in the sextet channel, without loss
of generality, we can write the Cooper-pair gap in color anti-triplet 5 . Since quarks are
anti-commuting, the only possible way to form diquark (S-wave) condensate is either in
spin-singlet or in spin-triplet:

a
vF , x)L bj (E
vF , x) = R ai (E
vF , x)R bj (E
vF , x)
(5.16)
L i (E
= ij abc K[]c (pF ) + ij abc K{}c (pF ),
(5.17)
where a, b, c = 1, 2, 3 are color indices, , , = u, d, s, . . . , Nf flavor indices, and i, j =

1, 2 spinor indices. Indices in the bracket and in the curled bracket are anti-symmetrized
and symmetrized, respectively. But, the spin-one component of the gap, K{}c , vanishes
vF , x) and (1 + E vEF )il (1 E vEF )lj
algebraically, since (E
vF , x) = 1/2(1 + E vEF )(E
= 0.
When Nf = 3, the spin-zero component of the condensate becomes (flavor) anti-triplet,
K[ ]c (pF ) = Kc (pF ).
(5.18)
Using the global color and flavor symmetry, one can always diagonalize the spin-zero
condensate as Kc = c K . To determine the parameters, Ku , Kd , and Ks , we need to

minimize the vacuum energy for the condensate. By the CornwallJackiwTomboulis
formalism [33], the vacuum energy in the HDL approximation is given as
5 At high but finite density, the Cooper-pair gap contains a small component of color-sextet [31,32]. But we
will ignore this for simplicity.
466
Fig. 3. The 2PI diagram in the leading order HDL approximation.

V () = Tr ln S 1 + Tr ln / + Tr S 1 / S + (2PI diagrams)

9 Z
lk2
2i (lk )
1
d2 lk
2 X
+ 2
+ h.o.,
ln 2
=
4
2 lk + 2i (lk )
(2)2
lk + 2i (lk )
i=1
(5.19)
where h.o. are the higher order terms in the HDL approximation, containing more
powers of coupling gs , and i s are the eigenvalues of color anti-symmetric and flavor
anti-symmetric 9 9 gap, ab
. The 2PI diagrams are two-particle-irreducible vacuum
diagrams. There is only one such diagram (see Fig. 3) in the leading order HDL
approximation.
Since the gap depends only on energy in the leading order, one can easily perform the
momentum integration in (5.19) to get 6 ,
!
Z
2i
2i
1
2
q
dl0 q
+ q
V () =
4 2
l 2 + l 2 + 2 4
l 2 + 2
0
2
2 X
i (0) ,
' 0.43 2
4
(5.20)
where in the second line we used an approximation that

i (0) if |l0 | < |i (0)|,
(5.21)
i (l0 ) '
0
otherwise.
Were i independent of each other, the global minimum should occur at i (0) = const
for all i = 1, . . . , 9. But, due to the global color and flavor symmetry, only three of them
are independent. Similarly to the condensate, the gap can be also diagonalized by the color
and flavor symmetry as
ab = abc c .
(5.22)
Without loss of generality, we can take |u | > |d | > |s |. Let d /u = x and

s /u = y. Then, the vacuum energy becomes
V () ' 0.43
2
|u |2 f (x, y),
4 2
(5.23)
6 If the condensate forms, the vacuum energy due to the gluons also depends on the gap due to the Meisner
effect. But, it turns out to be subleading, compared to the quark vacuum energy; Vg () M 2 2 ln(/)
gs 2 2 [36].
467
where f (x, y) is a complicate function of 1 6 x, y 6 1 that has a maximum at x = 1 = y,

f (x, y) 6 13.4. Therefore, the vacuum energy has a global minimum when u = d =
s , or in terms of the eigenvalues of the gap
i = u (1, 1, 1, 1, 1, 1, 1, 1, 2).
(5.24)
Among nine quarks, a , eight have (Majorana) mass u and one has mass 2u .
Since the condensate is related to the off-diagonal component of the quark propagator at
high momentum as, suppressing the color and flavor indices,
Z

d4 l il(xy) (lk )
vF , x) lim
e
(E
vF , x)(E
yx
(2)4
lk2 2 (lk )

(0)
x yE ) 2
+
,
(5.25)
= lim 2 (E
yx
4 |xk yk |m
where m is the anomalous dimension of the condensate and the ellipsis are less singular
terms. Being proportional to the gap, the condensate is diagonalized in the basis where the
gap is diagonalized. Thus, we have shown that in the HDL approximation the true ground
state of QCD with three massless flavors at high density is the colorflavor locking phase,
K = K for all = u, d, s. The condensate takes

a
vF , x)L bj (E
vF , x) = R ai (E
vF , x)R bj (E
vF , x)
L i (E
= ij abI I K(pF ),
(5.26)
breaking the color symmetry, U (1)em , the chiral symmetry, and the baryon number
symmetry. The symmetry breaking pattern of the CFL phase is therefore
SU(3)c SU(3)L SU(3)R U (1)em U (1)B 7 SU(3)V U (1)Q Z2 , (5.27)
where SU(3)V is the diagonal subgroup of three SU(3) groups and the generator of U (1)Q
is a linear combination of the color hypercharge and U (1)em generator,
Q = cos Qem + sin Y8 ,
(5.28)
where tan = e/gs .

Now, we analyze the SD gap equation to see if it admits a nontrivial solution. Since the
colorflavor locking gap is preferred if it exists, we may write the gap as
abI
I .
ab
=
Then the gap equation becomes, neglecting small hi couplings,

Z
a
a T
d4 l
T (lk )(T )
D
(p
l)V
V
(pk ) = (igs )2
(2)4
lk2 2 (lk )
Z
g
(lk )
d4 l
,
+ i 32
4
2
(2) lk 2 (lk )
(5.29)
(5.30)
where we use the bare vertex and take Z(pk ) = 1 in the leading HDL approximation.
468
E and thus
In the weak coupling limit, |k0| |k|

F k0 , kE ' M 2 ,

k0
G k0 , kE ' M 2 .
E
4
|k|
Since the gap has to be fully antisymmetric in color indices, we get

1
1
2
a
a T
t v us t u vs uv = t s .
Tt u uv (T )vs =
2
6
3
After Wick-rotating into Euclidean space, the gap equation becomes

Z
d4 q
2 2
V P T V
g
(pk ) =
s
2 + M 2 |p q |/|p
3
(2)4
(p q)2k + qE
E qE |
0
0
4

(p q)2k
g3
(qk )
1
.
2
2
4
2
2
2
(p q)
qk + 2 (qk )
(p q)k + qE + M
(5.31)
(5.32)
(5.33)
Note that the main contribution to the integration comes from the loop momenta in the
q | M 2/31/3 . Therefore, we find that the leading contribution is
region qk2 2 and |E
by the first term due to the Landau-damped magnetic gluons. For this momentum range,
we can take |pE qE | |E
q | and
4/3
j
.
(5.34)
V P T V = vFi vF ij ki kj = 1 + O
M 4/3
We also note that the term due to the four-Fermi operator is negligible, since g3 gs4 at the
matching scale .
Neglecting (p q)2k in the denominator, the gap equation becomes at the leading order
in the weak coupling expansion and 1/ expansion

Z
(p q)2k
1
2gs2
d4 q
1
+ 2
+
(pk ) =
2 + M 2 |p q |/|E
3
(2)4 qE
|E
q |4
qE + M 2
0
0 q |
4
(qk )
qk2 + 2 (qk )
(5.35)
The qE integration can now be performed easily to get

2

Z 2
d qk (qk )
g2
3
3
3
ln
. (5.36)
+
ln
+
(pk ) = s
2
9
(2)2 qk2 + 2
2
M2
2
4 M |p0 q0 |
We see that in this approximation (pk ) ' (p0 ). Then, we can integrate over vEF qE to
get
g2
(p0 ) = s 2
36

(q0 )
,
dq0 q
ln
|p0 q0 |
q02 + 2
(5.37)
where = (4/) (/M)5 e(3/2) . If we take ' (0) for a rough estimate of the gap,

2

6
gs2
or ' exp
1=
ln
.
(5.38)
36 2
gs
469
As was done by Son [13], one can convert the SchwingerDyson gap equation (5.37) into a
differential equation to take into account the energy dependence of the gap. Approximating
the logarithm in the gap equation as

ln |p0 | if |p0 | > |q0 |,
(5.39)
ln |p0 q0 | '
ln |q0 | otherwise,
we get
p00 (p) + 0 (p) +
2s (p)
p
= 0,
9 p2 + 2
(5.40)
where p p0 . When
with boundary conditions p0 = 0 at p = and = 0 at p = ,
p (p), the equation becomes
p00 + 0 +
r 2 (p)
= 0,
4 ||
(5.41)
r2
= 0,
4 p
(5.42)
where r 2 = 2gs2 /(9 2 ) and || is the gap at p = 0. We find (p) = ||J0 (r p/|| ) for
p k|. When p , the differential equation (5.40) becomes
p00 + 0 +
we get for
which has a power solution, pir/2 . Since the gap vanishes at p = ,
p

r
ln
.
(5.43)
(p) = B sin
2 p
By matching two solutions at the boundary p = || we get
/r .
B ' || and || = e
(5.44)
The gap is therefore given as at the leading order in the weak coupling expansion 7

3 2
|| = c 5 exp
,
(5.45)
gs
2 gs
5/2
where c = 27 4 Nf e3/2+1 . This agrees with the RG analysis done by Son [13] (see
also [37]) and also with the SchwingerDyson approach in full QCD [1517]. The 1/gs
behavior of the exponent of the gap at high density is due to the double logarithmic
divergence in the gap equation (5.33), similarly to the case of chiral symmetry breaking
under external magnetic fields [25,3841]. In addition to the usual logarithmic divergence
in the quark propagator as in the BCS superconductivity, there is another logarithmic
divergence due to the long-range gluon exchange interaction, which occurs when the gluon
loop momentum is colinear to the incoming quark momentum (E
q 0).
7 The gauge-parameter dependent term is subleading in the gap equation (5.33). Since the gap has to be gaugeindependent, the gauge parameter dependence in the prefactor will disappear if one includes the higher order
corrections.
470
6. Temperature effects and higher order corrections

When the quark matter is not very dense and not very cold, the effects of finite
temperature and density become important. In this section we calculate the critical density
and temperature. First, we add the 1/ corrections to the gap equation Eq. (5.30) to see
how the formation of Cooper pair changes when the density decreases. As derived in [14],
the leading 1/ corrections to the quarkgluon interactions are
1 X
(E
vF , x) ( D)2 (E
vF , x)
L1 =
2
vEF

X D2

F
+ gs
.
(6.1)
=
2
4
vEF
In the leading order in the HDL approximation, the loop correction to the vertex is
neglected and the quarkgluon vertex is shifted by the 1/ correction as
igs vFi 7 igs vFi igs
i
l
,
(6.2)
where li is the momentum carried away from quarks by gluons. Then, the gap equation
(5.35) becomes

Z
2 /2 )
2 /2
1 l
|lE |(1 l
2g 2
d4 l
+
(pk ) = s
2 + M2
3
(2)4 |lE |3 + (/4)M 2 |l0 p0 |
l
2 /2 )
lk2 (1 l
(lk )
+
.
(6.3)
l4
lk2 + 2
For a constant gap approximation, (pk ) ' , the gap equation becomes in the leading
order, as p 0,

Z
d2 lk
gs2
3
1
gs2
=
ln
ln
3 .
(6.4)
ln
1=
9
(2)2
|l0 |
2 lk2 + 2 36 2
Therefore, we see that, when < c ' e3 , the gap due to the long-range color magnetic
interaction disappears. Since the phase transition for color superconducting phase is
believed to be of first order [4244], we may assume that the gap has the same dependence
on the chemical potential as the leading order. Then, the critical density for the color
superconducting phase transition is given by

3 2
3
.
(6.5)
c = e c exp
2gs (c )
Therefore, if the strong interaction coupling is too strong at the scale of the chemical
potential, the gap does not form. To form the Cooper pair gap, the strong
coupling at
the scale of the chemical potential has to be smaller than gs (c ) = 2 / 2. By using the
one-loop function for three light flavors, (gs ) = [9/(16 2)]gs3 , and the experimental
value for the strong coupling constant, s (1.73 GeV) = 0.32+0.031
0.053(exp) 0.016(theo)
471
Fig. 4. The contour for integration over the Matsubara frequency. The circles are poles.
[45,46], we get 0.13 GeV . c . 0.31 GeV, which is about the same order as the one
estimated by the instanton induced four-Fermi interaction [44,47] or by general effective
four-Fermi interactions [42,43]. But, this should be taken as an order of magnitude, since
for such a small chemical potential the higher order terms in 1/ expansion, which we
have neglected, are as important as the leading term.
So far we have not included the temperature effect in analyzing the gap. The temperature
effect is quite important to understand the heavy ion collision or the final stage of the
evolution of giant stars because quarks will have to have high energy to bring together to
form a super dense matter. The super dense and hot quark matter will go through a phase
transition as it cools down by emitting weakly interacting particles like neutrinos.
At finite temperature, T , the gap equation (5.35) becomes, following the imaginary-time
formalism developed by Matsubara [48,49],

+ Z
X
(n )
gs2
dq
T
,
(6.6)
ln
(n0 ) =
9 n= 2 n2 + 2 (n ) + q 2
|n0 n |
where n = T (2n + 1) and q vEF qE . We now use the constant (but temperaturedependent) gap approximation, (n ) ' (T ) for all n. Taking n0 = 0 and converting
the logarithm into integration, we get
Z 2
+ Z
X
(T )
1
dq
gs2
T
dx 2
.
(T ) =
18 n= 2
n + 2 (T ) + q 2 x + (n 0 )2
(6.7)
Using the contour integral (see Fig. 4) [50], one can in fact sum up over all n to get
g2
1 = s 2T
36
I
Z 2
1
d
dq dx
2i 1 + e/T
0
1
.
(6.8)
2
( q 2 2 (T ))[( i0 )2 + x]
Since the gap vanishes at the critical temperature, (TC ) = 0, after performing the contour
integration in Eq. (6.8), we get
472
Z 2
dq dx
g2
1= s 2
36
[(TC
(TC )2 + x q 2
tanh[q/(2TC )]
+ x q 2 ]2 + (2TC q)2
2q
)2

coth[ x/(2TC )]
(TC )2 + q 2 x
.
[(TC )2 + q 2 x]2 + (2TC )2 x

x
(6.9)
At high density TC , the second term in the integral in Eq. (6.9) is negligible, compared
to the first term, and integrating over x, we get
Zc
g2
1= s 2
36
dy
0
g2
= s2
36
" Z1

2
2c
tanh y
y
ln
+
O
2
2
y
(/2) + y
2c
tanh y
ln 2c +
dy
y
Zc
tanh y 2c
ln 2 +
dy
y
y
#
(6.10)

g2
= s 2 (ln c )2 + 2A ln c + const ,
36
where we have introduced y q/(2TC ) and c /(2T

C ) and A is given as
Z1
A=
tanh y
+
dy
y

4
tanh y 1
= ln
dy
+ .
y
(6.11)
Therefore, we find, taking the EulerMascheroni constant ' 0.577,

TC =
eA
' 1.13 .
2
(6.12)
As comparison, we note in the BCS case, which has a contact four-Fermi interaction with
strength g,
the critical temperature is given as
Z c
1 = g
dz
tanh z
z
#
Z1
Z
tanh z
1 tanh z
dz
+ dz
dz
' g
z
z
z
1
0
1

= g ln eA c
" Z c
(6.13)
where c ( 1) is determined by the Debye energy, c = D /(2TC ). It shows that the ratio
between the critical temperature and the Cooper-pair gap 8 in color superconductivity is
same as the BCS value, e / ' 0.57 [17,51,52].
8 In the literature, the BCS gap is defined as twice of the dynamical mass, 2 [56].
473
Fig. 5. The corrections to the quarkneutrino four-Fermi coupling.
7. Applications
It is believed that the core of compact stars like neutron stars may be dense enough
to form quark matter and may shed some lights on understanding QCD at high density.
The properties of compact stars can be investigated by studying the emission of weakly
interacting particles like neutrinos or axions, which is the dominant cooling process of the
compact stars [5355]. Since the emission rate depends on the couplings of those particles,
it is important to understand how the interaction and the coupling of neutrinos or axions
change in dense quark matter.
Neutrinos interact with quarks by the exchange of neutral currents, which is described,
at low energy, as four-Fermi interaction,
GF
L 9L L L ,
Lq = 9
2
(7.1)
where GF = 1.166 105 GeV2 is the Fermi constant. Again, by decomposing the quark
fields as in Eq. (2.5) and integrating out modes, the four-Fermi interaction becomes
GF X
+ L (E
vF , x)+ L (E
vF , x)L (x)V
/ L (x) + ,
(7.2)
Lq =
2 vE
F
where the ellipsis denotes the higher order terms in the power expansion of 1/. Since
the four-fermion interaction of quarks with opposite momenta gets enhanced a lot at
low energy, as we have seen in the previous section (Section 5), it may have significant
corrections to the couplings of those weakly interacting particles to quarks. We first
calculate the one-loop correction to the neutrinoquark four-Fermi coupling by the
marginal four-quark interaction:
GF
vF , x)+ L (E
vF , x)L (x)V
/ L (x)
Lq = +
L (E
2
Z

ig
vF0 , y) 0 s (E
vF0 , y) v (E
vF0 , y) 0 u (E
vF0 , y)
t (E
32 tAv;us
2M
y
4 g3 GF
(E
vF , x)+ L (E
vF , x)L (x)V L (x),
=
3 2 2 + L
(7.3)
where vEF and vEF0 are summed over and g3 is the value of the marginal four-quark coupling
at the screening mass scale M. Now, in fact, because of the kinematical constraint due to
474
the presence of the Fermi surface, only the cactus diagrams, shown in Fig. 5, contribute to
the coupling corrections, which can be summed up as

g3
4GF X
vF , x)+ L (E
vF , x)L (x)L (x)
+
Lq =
L (E
2
g
3 2 vE
3
F

g3
(E
v
,
x)
(E
v
,
x)
(x)E
v
(x)
.
(7.4)
+
F
+
F
L
F
L
L
L
2 + g3
Similarly, for axions which couple to quarks as
Laq =
1
5 9,
a 9
2fP Q
(7.5)
where a is the axion field and fP Q is the axion decay constant, we find the correction to
the axionquark coupling to be

g3 0 a
2 X
(E
vF , x)5 + (E
vF , x)
+
Laq =
3fP Q
2 g3
vEF

E
g vEF a
vF , x)5+ (E
vF , x) .
+ (E
3
2 + g3
Therefore, as the marginal four-quark coupling approaches 2 , the quark-neutrino and
quarkaxion couplings become divergent. Since at low energy g3 is quite large in dense
quark matter, we argue that quark matter will produce neutrinos or axions copiously if the
density of quark matter is high enough ( QCD ) such that the marginal four-quark
interaction gets enhanced sufficiently.
8. Conclusion
We have studied in detail an effective field theory of QCD at high density, constructed
by integrating out the anti-quarks to describe the low energy dynamics of dense quark
matter. In the effective theory, the dynamics of quarks is effectively (1 + 1)-dimensional;
the energy of quarks does not depend on the perpendicular momentum to the Fermi
velocity, which just serves to label the degeneracy. At energy lower than the screening
mass of electric gluons, the relevant interactions are four-Fermi interactions with opposite
incoming momenta and the coupling with magnetic gluons.
Because of the dimensional reduction at high density, both four-Fermi interaction and
magnetic gluon exchange interaction lead to Cooper-pair condensate of quarks in color
anti-triplet channel for arbitrarily weak coupling. In the HDL approximation, the gap
formed by the magnetic gluon exchange interaction is much bigger than the one by fourFermi interaction. The colorflavor locking condensate is found to be energetically more
preferred for high density quark matter with three light flavors, because the colorflavor
locking gap is bigger than the spin one gap, if we include the four-Fermi interactions.
Further, including the 1/ corrections and the temperature effects, the critical density and
475
the critical temperature for color superconducting phase are calculated by solving the gap
equation in the HDL approximation.
Finally, we have calculated the corrections to the quark-neutrino four-Fermi interaction
and to the quarkaxion coupling in dense quark matter due to the marginal four-quark
interaction. And we found the correction is quite significant and thus dense quark matter
will copiously produce neutrinos and axions.
Acknowledgements
The author is grateful to Sekhar Chivukula, Andy Cohen, Raman Sundrum for useful
comments, and wishes to thank Roman Jackiw, Steve Hsu, Cristina Manuel, Mannque Rho,
Dirk Rischke, Tom Schfer, and Andrei Smilga for enlightening discussions. The author
wishes to acknowledge the financial support of the Korea Research Foundation made in the
program year of 1998 (1998-15-D00022). This work was also supported by Pusan National
University Research grant, 1998.
References
[1] S. Weinberg, Physica A 96 (1979) 327.
[2] J. Gasser, H. Leutwyler, Ann. Phys. (N.Y.) 158 (1984) 142.
[3] D.B. Kaplan, Effective field theories, nucl-th/9506035, Lectures given at the Seventh Summer
School in Nuclear Physics, Institute for Nuclear Theory, June 1930, 1995.
[4] A. Manohar, Effective field theories, hep-ph/9508245, Lectures presented at the Lake Louise
Winter Institute, February, 1995.
[5] J. Polchinski, in: J. Harvey, J. Polchinski (Eds.), Proceedings of the 1992 TASI, World Scientific,
Singapore, 1993.
[6] D. Bailin, A. Love, Phys. Rep. 107 (1984) 325.
[7] R. Rapp, T. Schafer, E.V. Shuryak, M. Velkovsky, Phys. Rev. Lett. 81 (1998) 53.
[8] M. Alford, K. Rajagopal, F. Wilczek, Phys. Lett. B 422 (1998) 247, hep-ph/9711395.
[9] J. Berges, K. Rajagopal, Nucl. Phys. B 537 (1999) 443, hep-ph/9804403.
[10] M. Alford, K. Rajagopal, F. Wilczek, Nucl. Phys. B 538 (1999) 443, hep-ph/98044233.
[11] N. Evans, S. Hsu, M. Schwetz, Nucl. Phys. B 551 (1999) 275, hep-ph/9808444; Phys. Lett.
B 449 (1999) 281, hep-ph/9810514.
[12] K. Langfeld, M. Rho, Quark condensation, induced symmetry breaking and color superconductivity at high density, hep-ph/9811227.
[13] D.T. Son, Phys. Rev. D 59 (1999) 094019, hep-ph/9812287.
[14] D.K. Hong, Phys. Lett. B 473 (2000) 118, hep-ph/9812510.
[15] D.K. Hong, V.A. Miransky, I.A. Shovkovy, L.C.R. Wijewardhana, Phys. Rev. D 61 (2000)
056001, hep-ph/9906478.
[16] T. Schfer, F. Wilczek, Phys. Rev. D 60 (1999) 114033, hep-ph/9906512.
[17] R. Pisarski, D. Rischke, nucl-th/990741; nucl-th/9910056.
[18] R. Pisarski, D. Rischke, Phys. Rev. Lett. 83 (1999) 37, nucl-th/9811104.
[19] J. Kapusta, T. Toimela, Phys. Rev. D 37 (1988) 3731; Phys. Rev. D 39 (1989) 3197.
[20] C. Manuel, Phys. Rev. D 53 (1996) 5866.
[21] G. Alexanian, V.P. Nair, Phys. Lett. B 390 (1997) 370.
[22] T. Schfer, F. Wilczek, Phys. Rev. Lett. 82 (1999) 3956, hep-ph/9811473; Phys. Rev. D 60
(1999) 074014, hep-ph/9903503.
476
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
M. Luke, A.V. Manohar, Phys. Lett. B 286 (1992) 348.

D.K. Hong, Phys. Rev. D 57 (1998) 3759, hep-ph/9707432.
D.K. Hong, Y. Kim, S.J. Sin, Phys. Rev. D 54 (1996) 7879.
V. Gusynin, D.K. Hong, I. Shovkovy, Phys. Rev. D 57 (1998) 5230.
J. Kapusta, Field Theory at Finite Temperature and Density, Cambridge University Press,
Cambridge, England, 1989, and references therein.
R. Pisarski, Phys. Rev. Lett. 63 (1989) 1129.
E. Braaten, R. Pisarski, Nucl. Phys. B 337 (1990) 569.
J. Frenkel, J.C. Taylor, Nucl. Phys. B 334 (1990) 119.
M. Alford, J. Berges, K. Rajagopal, Nucl. Phys. B 558 (1999) 219, hep-ph/9903502.
R.D. Pisarski, D.H. Rischke, nucl-th/9907094.
J.M. Cornwall, R. Jackiw, E. Tomboulis, Phys. Rev. 2428 (1974).
K.S. Tam, Phys. Lett. B 152 (1985) 238.
M. Peskin, Recent developments in quantum field theory and statistical mechanics, in: Les
House Summer School on Theoretical Physics, 1982.
N. Evans, J. Hormuzdiar, S.D.H. Hsu, M. Schwetz, hep-ph/9910313.
S.D.H. Hsu, M. Schwetz, hep-ph/9908310.
V.P. Gusynin, V.A. Miransky, I.A. Shovkovy, Phys. Rev. D 52 (1995) 4747; Nucl. Phys. B 462
(1995) 249.
C.N. Leung, Y.J. Ng, A.W. Ackley, Phys. Rev. D 54 (1996) 4181.
D.-S. Lee, C.N. Leung, Y.J. Ng, Phys. Rev. D 55 (1997) 6504.
V.P. Gusynin, V.A. Miransky, I.A. Shovkovy, Phys. Rev. Lett. 83 (1999) 1291, hep-ph/9811079.
J. Berges, K. Rajagopal, Nucl. Phys. B 538 (1999) 215, hep-ph 9804233.
M.A. Halasz et al., Phys. Rev. D 58 (1998) 096007, hep-ph 9804290.
M. Stephanov, K. Rajagopal, E. Shuryak, Phys. Rev. Lett. 81 (1998) 4816, hep-ph/9806219.
J. Ellis et al., Phys. Rev. D 54 (1996) 6986.
M. Schmelling, in: ICHEP 96, International Conference on High-Energy Physics, July 2531,
Warsaw, Poland, 1996; hep-ex/9701002.
R. Rapp et al., hep-ph/9904353.
T. Matsubara, Prog. Theor. Phys. 14 (1955) 351.
A.A. Abrikosov, L.P. Gorkov, I.E. Dzyaloshinski, Methods of Quantum Field Theory in
Statistical Physics, Pergamon Press, Oxford, 1965.
V.P. Gusynin, I.A. Shovkovy, Phys. Rev. D 56 (1997) 5251.
W.E. Brown, J.T. Liu, H.-c. Ren, hep-ph/9912409.
D.K. Hong, Talk given at TMU-Yale Symposium on Dynamics of Gauge Fields, December
1315, 1999, Tokyo.
M.S. Turner, Phys. Rep. 197 (1990) 67.
G. Raffelt, Phys. Rep. 198 (1990) 1; Stars as Laboratories for Fundamental Physics, University
of Chicago Press, Chicago, 1996.
D.K. Hong, Phys. Lett. B 445 (1998) 36.
J.R. Schrieffer, Theory of Superconductivity, W.A. Benjamin, New York, 1964.

QCD-like theories at finite baryon density

J.B. Kogut a , M.A. Stephanov b,c , D. Toublan d , J.J.M. Verbaarschot d ,
A. Zhitnitsky e
a Loomis Laboratory of Physics, University of Illinois, Urbana-Champaign, IL 61801, USA
b Department of Physics, University of Illinois, Chicago, IL 60607-7059, USA
c RIKEN-BNL Research Center, Brookhaven National Laboratory, Upton, NY 11973, USA
d Department of Physics and Astronomy, SUNY, Stony Brook, New York 11794, USA
e Department of Physics and Astronomy, University of British Columbia, Vancouver, BC V6T 1Z1, Canada
Received 24 February 2000; accepted 18 April 2000
Abstract
We study QCD-like theories with pseudoreal fermions at finite baryon density. Such theories
include two-color QCD with quarks in the fundamental representation of the color group as well
as any-color QCD with quarks in the adjoint color representation. In all such theories the lightest
baryons are diquarks. At zero chemical potential they are, together with the pseudoscalar mesons,
the Goldstone modes of a spontaneously broken enlarged chiral symmetry group. Using symmetry
principles, we derive the low-energy effective Lagrangian for these particles. We find that a second
order phase transition occurs at a value of equal to half the mass of the Goldstone modes. For values
of beyond this point the scalar diquarks Bose condense and the diquark condensate is nonzero. We
calculate the dependence of the chiral condensate, the diquark condensate, the baryon charge density,
and the masses of the diquark and pseudoscalar excitations on at finite bare quark mass and scalar
diquark source. The relevance of our results to lattice QCD calculations and to real three-color QCD
at finite baryon density is discussed. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.30.Rd; 12.39.Fe; 12.38.Lg; 71.30.+h
Keywords: QCD partition function; Finite baryon density; QCD with two colors; Adjoint QCD; QCD Dirac
operator; Lattice QCD; Low-energy effective theory; Chiral perturbation theory
1. Introduction
The study of strong interactions at finite baryon density has a long history. Phenomenological knowledge of nuclear forces allows one to obtain a good understanding of equilibrium nuclear matter which is relatively dilute. However, these results cannot be simply
extended to higher densities relevant to neutron stars, supernova explosions and relativistic heavy ion collisions, where the microscopic degrees of freedom of QCD, quarks and
gluons, become important.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 4 2 - X
478
J.B. Kogut et al. / Nuclear Physics B 582 (2000) 477513
Understanding QCD at finite density has been a tremendous challenge [13]. Unlike
finite temperature QCD, where significant progress has been achieved using lattice Monte
Carlo simulations, QCD at finite baryon density is not amenable to such a numerical
approach. The primary reason is that the determinant of the Euclidean QCD Dirac operator
is not real at finite baryon chemical potential, .
Recent progress has been achieved analytically by studying QCD at infinite density and
by studying models of the NambuJonaLasinio type as well as the instanton liquid model
[46]. It was found that the QCD vacuum at sufficiently high baryon density could become
a color superconductor [46]. In other words, due to the attraction in the color anti-triplet
isosinglet channel the quark Fermi surface becomes unstable towards the formation of a
condensate of diquark pairs.
No first principle lattice calculation methods exist at this moment to study the
phenomenon of color superconductivity. However, the mechanisms which lead to the
formation of diquark condensates (gluon exchange or instanton induced attraction) can be
investigated for QCD with two colors. This presents a tremendous advantage. QCD with
two colors can be studied numerically on the lattice [710] the determinant of the Dirac
operator is real. Another class of QCD-like theories with diquark condensation that can be
studied on the lattice at nonzero chemical potential is QCD with any number of colors and
of quarks in the adjoint color representation. Clearly, the physics of both types of theories
is different from that of the three-color QCD, but the differences are easy to understand
and classify. We hope that understanding the behavior of such theories at finite baryon
number density will provide us with an additional insight into the phenomenon of diquark
condensation in three-color QCD. In addition, numerical simulations of both QCD with
two colors and QCD with adjoint quarks are now being pursued by several groups [911].
The third class of theories which can be studied in the same way is QCD with the phase
of the quark determinant quenched. In other words, for each quark such theory contains
a conjugate quark, with opposite baryon charge. This leads to the appearance of colorless
diquark states baryonic pions. The zero-flavor limit Nf 0 of such a theory is the
quenched approximation of QCD, as has been demonstrated analytically in [12] using a
random matrix theory at nonzero .
The unifying property of all such theories is the pseudo-reality of the quark representations, which manifests itself in the fact that the determinant of the Dirac operator is real.
Another unifying property (which is ultimately related to this pseudo-reality) is the fact
that diquarks can make color singlets these are the baryons of such theories. The Bose
Einstein condensation of diquarks, with nonzero baryon charge, may be viewed as baryon
charge superconductivity (rather than color superconductivity as in three-color QCD).
In this paper we construct the low-energy chiral Lagrangian describing mesons and
baryons (diquarks) at finite baryon density for two-color QCD and QCD with adjoint
quarks. As was pointed out in [13] the -dependence of this Lagrangian describing mesons
and baryons can be fixed by global and certain local flavor symmetries. Within the domain
of validity of this Lagrangian we study both its vacuum properties and the mass spectrum
of the Goldstone modes. Our goal is twofold: (i) to understand and describe quantitatively
479
the physics associated with diquark condensation; (ii) to provide lattice theorists with the
qualitative and quantitative predictions aiding their data analysis.
This paper is organized as follows. In Section 2 we give an overview of the different
patterns of chiral symmetry breaking at = 0 for three-color QCD, two-color QCD and
for QCD with adjoint quarks. We relate these patterns to the antiunitary symmetries of the
Dirac operator, which provide us with a convenient classification in terms of the Dyson
index . Section 3 reviews the global symmetries of the theories, with the emphasis on
the enlarged SU(2Nf ) symmetry. The spontaneous breaking of this symmetry by the
chiral condensate is the subject of Section 4, where we identify the Goldstone excitations.
The effective Lagrangian at nonzero bare quark mass m is introduced in Section 5. In
Section 6 we review the introduction of the chemical potential into the effective Lagrangian
by means of a local gauge principle [13,14]. The vacuum alignment dictated by the
effective Lagrangian is analyzed in Sections 7 and 8. In Section 9 we expand the static
part of Lagrangian around the minimum and in Section 10 we determine the masses of
the excitations as functions of and m from the pole of their respective propagators. In
Section 11 we introduce the diquark source j and find its effect on the vacuum and the
mass spectrum. In Section 12 we present the dependence of the vacuum condensates and
the baryon number density on , m and j . Finally, in Section 13, we rederive the equation
of state of the dilute Bose gas of diquarks with repulsion and show that it exactly matches
the equation of state obtained using our mean field analysis of the previous sections.
Concluding remarks and discussion are presented in Section 14.
2. Overview and classification

At zero chemical potential the spontaneous breaking of the chiral symmetry is believed
to be an essential low-energy property of three color QCD with fundamental fermions. In
the limit of massless quarks, the QCD Lagrangian is invariant under UL (Nf ) UR (Nf )
transformations, but the ground state is not. The analysis of the hadron spectrum and
numerical simulations on the lattice strongly support this assertion [15,16]. The order
parameter of the spontaneous breaking of chiral symmetry is the chiral quarkantiquark
condensate. In the case of two-color QCD with fundamental fermions and in the case of
any-color QCD with adjoint fermions the symmetry of the Lagrangian is enlarged from
UL (Nf ) UR (Nf ) to U (2Nf ). In the case of Nc = 2, this symmetry is sometimes referred
to as the PauliGrsey symmetry [17,18]. Also in these cases there is strong evidence both
from lattice simulations [19,20] and from arguments based on supersymmetry [21] that
chiral symmetry is broken by a nonzero vacuum expectation value of the chiral condensate.
The pattern of chiral symmetry breaking is determined by two ideas. The VafaWitten
theorem [22] which tells us that vector symmetries cannot be spontaneously broken, and
the idea of maximum breaking of the axial symmetry [2326]. For each of the three classes
of theories chiral symmetry is broken spontaneously according to different patterns [23
27]. The axial U (1)A subgroup of the global symmetry is broken explicitly by the axial
anomaly in all three cases. The vector-like U (1)B symmetry, corresponding to baryon
480
charge conservation, is intact. The remaining symmetries in QCD with three or more colors
with fundamental fermions are broken according to SU R (Nf ) SU L (Nf ) SUV (Nf ).
For two-color QCD with fundamental fermions the symmetry is broken according to
SU(2Nf ) Sp(2Nf ), whereas for any-color QCD with adjoint fermions the pattern of
symmetry breaking is given by SU(2Nf ) O(2Nf ).
We can classify these above three cases by the Dyson index, , of the Dirac operator
with a value of = 2, = 1 and = 4, respectively. The value of is given by the
number of independent degrees of freedom per matrix element and is determined by the
antiunitary symmetries of the Dirac operator. It is a concept that originated in Random
Matrix Theory [2830], and is important for the Cartan classification of symmetric spaces
[31].
In the case of two-color QCD the pseudo-real nature of SU(2)color [2327,32] can be
expressed as the antiunitary symmetry of the Dirac operator D = D + m,
[D, 2 C5 K] = 0
or D2 C5 = 2 C5 D
( = 1),
(1)
where 2 is the color symmetry generator, C is the Dirac charge conjugation matrix and K
is the complex conjugation operator. Since (2 CK)2 = 1 it is always possible to find a
basis in which the Dirac operator becomes real which gives = 1.
The symmetry (1) persists even at 6= 0 [33]. The reality of the Dirac determinant and
the feasibility of lattice Monte Carlo simulations is the consequence of (1). This property
also allows us to use QCD inequalities at finite to show that condensation can only occur
in the scalar diquark channel [13].
In the case of QCD with adjoint quarks the antiunitary symmetry of the Dirac operator
is
[D, C5 K] = 0 or DC5 = C5 D
( = 4).
(2)
Since (CK)2 = 1, it is always possible to find a basis in which the Dirac operator can
be organized into selfdual (pseudo-real) quaternions. The value of the Dyson index is thus
= 4.
In both above cases, = 1 and = 4, the antiunitary symmetry leads to enlargement of
the global symmetry to U (2Nf ) (at = 0).
There is no antiunitary symmetry in the case of QCD with three or more colors with
fundamental quarks. The Dirac operator is a complex matrix, thus = 2.
In this paper we shall study two classes of theories: = 1 and = 4, at finite chemical
potential . The main starting point of our analysis is the observation of the fact that in
these theories the lowest lying baryons belong to the set of Goldstones of the spontaneously
broken extended flavor symmetry SU(2Nf ). As a result the dependence on can be
described in the Chiral Perturbation Theory framework [3438].
We shall construct the effective Lagrangian governing the low-momentum modes in
the theory. These modes are the Goldstone particles of the spontaneously broken global
symmetries. The symmetry of the theory is largest at = 0, m = 0, and so is the number
of true Goldstone modes. A nonzero chemical potential and/or a bare quark mass m
removes part of the symmetry and some of the Goldstone modes acquire masses. Our main
481
goal is to find the functional dependence of the masses of such pseudo-Goldstones on

and m. As we proceed we learn about many other properties of the theory: condensates and
vacuum alignment, phase transitions, etc.
The two cases: two-color fundamental quarks ( = 1) and any-color adjoint quarks ( =
4) can be analyzed in a similar way. We shall perform this analysis as follows. At each step
we shall begin with = 1 case, explaining the concepts and ideas. Then we follow it
immediately with the same analysis for the = 4 case with emphasis on the comparison
between the two cases, which will help understand both cases better. We shall use similar
notations for the objects which are conceptually the same in both cases. In many instances
we need not rewrite the formulas, only changing the meaning of the notations suffices.
All the formulas which use any properties specific to either = 1 or = 4 are explicitly
tagged. The formulas without such tags are general and apply to both cases. As we shall
quickly see the two cases naturally complement each other and are, in a certain sense, dual
to each other.
3. Global symmetries at m = = 0
3.1. = 1
The fermionic part of the QCD Lagrangian with 2 fundamental colors is given by
T

0
D
L
D = i L
.
L =
R
R
0
D
(3)
We are working in Euclidean space with hermitian -matrices and we use spin matrices
= (i, k ). The quark flavor (as well as color and spin) indices are suppressed and
the sum over Nf flavors is implied. The symbol D denotes color covariant derivative,
+ iA , an antihermitian operator, with A being a matrix in color algebra, A = Aa a /2.
As usual = 0 = T 0 and in the Euclidean partition function and are
independent integration variables.
The fact that the Lagrangian (3) has a higher flavor symmetry than the apparent U (Nf )
U (Nf ) is related to the pseudoreality of the two-color Dirac operator. In particular, the
conjugate field R = 2 2 R transforms similarly (in the same color representation)
to L . Rewriting the Lagrangian (3) we find

T
D
L
0
L
(4)
= i D ,
L=i
R
R
0
D
where we have used the well-known properties of the Pauli matrices, 2 2 = T and
2 k 2 = kT , taken into account anticommutativity of Grassman variables, dropped total
derivatives, and introduced the spinor of dimension 2Nf ,

L
L
( = 1).
(5)
2 2 R
R
In this form the U (2Nf ) symmetry becomes manifest. Due to the axial anomaly the symmetry in the corresponding quantum theory is only SU(2Nf ) (up to discrete symmetries).
482
3.2. = 4
QCD with quarks in the adjoint representation of the color group is described by the
Lagrangian as in (3), but with different notations. The fields are now transforming
according to the adjoint representation of the color group. The color covariant derivative is
again given by D = + iA , but now A is given by the antisymmetric matrix (A )bc =
Aa fabc , where fabc = f abc are the generators of the adjoint representation, i.e., the structure
constants.
The antisymmetric property of the structure constants now replaces the property of the
fundamental generators:
2 k 2 = kT fabc = facb .
Using this property we can again recast the Lagrangian using spinors of length 2Nf and
obtain (4), but with the spinors (and R ) defined by

L
L
( = 4).
(6)
2 R
R
The only difference from (5) is the absence of color 2 matrix in the definition of R ,
and, of course, the fact that the spinors carry an adjoint, instead of a fundamental, color
index. Similarly, in terms of the spinors (6) the SU(2Nf ) symmetry of the theory becomes
manifest.
4. Spontaneous symmetry breaking and Goldstones

4.1. = 1
Let us now understand transformation properties of hi,

the order parameter of the
chiral symmetry breaking, with respect to the SU(2Nf ) symmetry of the theory. We can
as
rewrite

T
1 T
0 1
L
0 1
L
= 2 2
+ h.c. ( = 1). (7)
=
R
R
1 0
1 0
2
The Pauli matrices 2 and 2 ensure antisymmetrization in spin and color indices, to
produce a spin and color singlet. We see that the condensate is not invariant under all
SU(2Nf ) rotations. The subgroup which leaves (7) invariant is Sp(2Nf ). The Goldstone
manifold is therefore given by SU(2Nf )/Sp(2Nf ) with Nf (2Nf 1) 1 independent
degrees of freedom.
Another way of counting the total number of Goldstone modes is starting from
the observation that the condensate (7) is a product of two fundamental SU(2Nf )
flavor representations, antisymmetric in flavor indices. Therefore, the condensate belongs
to an antisymmetric tensor representation the dimension of which is Nf (2Nf 1).
Condensation can occur in any of these Nf (2Nf 1) directions; the fluctuations along
the remaining Nf (2Nf 1) 1 directions then become Goldstone modes.
483
The effective theory for the Goldstone modes can be written in terms of the fluctuations
of the orientation of the chiral condensate, , which, according to the previous paragraph,
is an antisymmetric unimodular (det = 1) unitary matrix (exactly Nf (2Nf 1) 1
independent components). If we denote the equilibrium value of the orientation of the
chiral condensate by c , the Goldstone manifold given by SU(2Nf )/Sp(2Nf ) can be
parameterized, according to the transformation of under SU(2Nf ), by
= U c U T ,
(8)
where

Xa
i
and = a p
.
U = exp
2F
2Nf
(9)
The fields a are the Goldstone modes. The implied sum (over a) is over the Nf (2Nf
1) 1 generators of the coset SU(2Nf )/Sp(2Nf ) and in order to simplify the algebra in
later sections we are using the normalization
Tr Xa Xb = 2Nf ab .
(10)
The construction of the Goldstone manifold (9) corresponds to the classification of the
(2Nf )2 1 generators of the SU(2Nf ) with respect to a fixed antisymmetric antiunitary
matrix, in our case given by c , into Tk and Xa [26]. The Tk generators leave c invariant,
exp(ik Tk )c exp(ik Tk )T = c ,
i.e., Tk c = c TkT .
(11)
By definition, they are the generators of the symplectic group Sp(2Nf ). The remaining
generators, Xa , form the coset SU(2Nf )/Sp(2Nf ). They obey the relation
Xa c = c XaT ,
and thus U c U T = U 2 c .
(12)
The partition of generators in generators of the coset, Xa , and generators of the invariant
subgroup Sp(2NF ), Ti , depends on the matrix c . The defining relations (12) are left
unaltered by rotation of c according to
c V c V T ,
(13)
where V is an SU(2Nf ) matrix, and a simultaneous rotation of the generators by

Xa V Xa V .
(14)
This means that the set of broken generators, Xa , changes, if the matrix c is changed.
If we use, as we do below, the following choice for c :

0 1
( = 1),
(15)
c =
1 0
the generators Xa can be written in the block representation as

T
P
Q
,
=
Q P
(16)
with Nf Nf matrices P and Q such that Tr P = 0, P = P , and QT = Q. It is then

easy to see that the number of independent components in P and Q are
NP = Nf2 1
and NQ = Nf (Nf 1)
( = 1).
(17)
484
4.2. = 4
What are the transformation properties of the chiral condensate with respect to the
as
SU(2Nf ) symmetry in the = 4 case? We can rewrite

T
0 1
L
= L
R
R
1 0

1 T
0 1
+ h.c. ( = 4).
(18)
= 2
1 0
2
The Pauli matrix 2 ensures the antisymmetrization with respect to spin indices, to
produce a spin singlet. There is no antisymmetrization in color (unlike in the case of
fundamental colors = 1) the color singlet is a symmetric product of two adjoint
representations. The Pauli principle now demands that the SU(2Nf ) flavor indices must
be symmetric, and they are, as is evident in (18). Therefore the condensate belongs to the
symmetric (as opposed to antisymmetric for = 1) second rank tensor representation of
SU(2Nf ), which has dimension Nf (2Nf + 1). The condensation can occur in any of the
Nf (2Nf + 1) directions. The fluctuations in the remaining Nf (2Nf + 1) 1 directions
become Goldstone bosons. Alternatively, since the chiral condensate (18) is invariant
under SO(2Nf ), the Goldstone manifold is SU(2Nf )/SO(2Nf ), which gives us the same
number of Goldstone modes.
The Goldstone manifold SU(2Nf )/SO(2Nf ) should now be parameterized by symmetric unimodular unitary matrices . The Goldstone fields are introduced in the same
way as in (8) and (9), i.e., = U c U T , where c is now also a symmetric unimodular unitary matrix and the implied sum in (9) is over the generators Xa of the coset
SU(2Nf )/SO(2Nf ).
The classification of the generators is also similar to = 1 case. The generators of
the coset, Xa obey the commutation relation (12) with a given symmetric unitary matrix,
also denoted by c , whereas the Ti generators leave c invariant as in (11). The Ti are
now the generators of an SO(2Nf ) subgroup (as opposed to Sp(2Nf ) in the = 1 case).
The remaining generators Xa form a coset SU(2Nf )/SO(2Nf ), which is the Goldstone
manifold in this case.
In this case the standard choice for the matrix c is

0 1
( = 4).
(19)
c =
1 0
p
With this choice, the matrix = a Xa / 2Nf in the coset can be split into 4 blocks of
size Nf Nf as in (16). The matrix P is again hermitian and traceless, but the matrix Q
is now symmetric. Therefore, the counting of the independent degrees of freedom in the
= 4 case is
NP = Nf2 1
and NQ = Nf (Nf + 1)
( = 4).
(20)
In both cases, = 1 and = 4, the kinetic term of the effective Lagrangian describing
the Goldstone modes should be invariant under the global SU(2Nf ) group and under
Lorentz transformation. The corresponding SU(2Nf ) nonlinear sigma-model is given by
F2
Tr ,
2
where F is the pion decay constant.
Leff =
485
(21)
5. Bare quark mass m

5.1. = 1
If the bare quark mass m (in this paper, the same for all quarks) is not zero, an explicit
SU(2Nf ) breaking term in the effective Lagrangian (21) appears. To determine its form
we first rewrite the bare mass term using SU(2Nf ) notations,

1
1
0 1
T
m = m 2 2
+ h.c. = T 2 2 M + h.c. ( = 1), (22)
1 0
2
2
where the mass matrix M is given by

b and M
b= 0 1
M = mM
1 0
( = 1).
(23)
We see explicitly that the bare mass term is only invariant under an Sp(2Nf ) subgroup of
SU(2Nf ). The full SU(2Nf ) invariance can be restored if M is also transformed together
with according to
V
and M V MV .
(24)
This extended symmetry must be also manifest in the effective theory,

V V T
and M V MV .
(25)
The lowest order term induced by the quark mass must therefore have the form
b
Lq.mass = G Re Tr(M) = mG Re Tr(M).
(26)
b
One can view Re Tr(M)
as a generalized cosine of the angle between unitary matrices
b . Therefore the direction of

b
and M . It is maximal when is aligned with M
minimizing (26) is given by
b ,
c = M
(27)
which leads us to our choice of c (15). The mass term comes with a phenomenological
coefficient, which we denote by G. It is given by the derivative of the vacuum energy
with respect to m and is, therefore, proportional to the chiral condensate in the chiral limit
m 0 at = 0 (see Section 12),
G=
h i0
.
2Nf
(28)
The resulting Lagrangian with the mass term,

Leff =
F2
b
Tr mG Re Tr(M),
2
(29)
486
is the familiar Chiral Perturbation Theory Lagrangian at lowest order in the momentum
expansion [3437]. Expanded to second order in the pion fields according to (8), (9), it
yields a spectrum with Nf (2Nf 1) 1 degenerate (pseudo-)Goldstones with masses
given by the usual Gell-MannOaksRennen relation
m2 =
mG
.
F2
(30)
We can use this relation to trade G for another parameter, m , and write
Leff =

F2
b
Tr 2m2 Re Tr(M)
.
2
(31)
The symmetry of the theory is reduced from SU(2Nf ) to Sp(2Nf ) by the mass term.
Since the chiral condensate does not break any more symmetries in this case we do not
have any true Goldstones when m 6= 0.
5.2. = 4
To determine the form of the term in the effective Lagrangian induced by a small bare
quark mass m we rewrite the quark mass term as

= 1 m T 2 0 1 + h.c. = 1 T 2 M + h.c. ( = 4), (32)
m
1 0
2
2
where we have used the spinors of length Nf introduced in Section 3. In this case the mass
matrix is given by

0 1
b
b
M = mM with M =
( = 4).
(33)
1 0
Using the same arguments as in Section 5.1 we find that the mass term in the effective
Lagrangian as dictated by the extended flavor symmetry is given again by (26), with
now being a symmetric matrix and M given by (33). Similarly, this term is minimized
b now taken from (33) we arrive at our choice of c (19).
b . With M
when = M
The symmetry of the theory is reduced from SU(2Nf ) to SO(2Nf ) by the mass
term. Since chiral the condensate does not break any more symmetries in this case
we do not have any truly massless Goldstones when m 6= 0. The masses of all
Nf (2Nf + 1) (pseudo-)Goldstones are equal and are given by the Gell-MannOaks
Rennen relation (30).
6. Chemical potential
In this section we review the introduction of the chemical potential in the effective
partition function following the approach of [13]. This approach relies only on the
SU(2Nf ) symmetry of the theory and the resulting -dependent terms are the same for
both cases, = 1 and = 4.
487
6.1. Global symmetries and

At nonzero chemical potential the microscopic Lagrangian is given by
0 + m.
D
L =
(34)
As was the case for the mass term, we can also rewrite the baryon charge density in terms
of the SU(2Nf ) spinors.
T

1 0
L
1 0
0 = L
=
B ;
R
R
0 1
0 1

+1 0
B
.
0 1
(35)
The physical meaning of +1 and 1 in the baryon charge matrix B is simple: they are the
baryon charges of the quarks L and conjugate quarks R . We see that this term is not a
singlet under SU(2Nf ). It transforms in the adjoint representation of this group, in other
words, the baryon charge is one of the (2Nf )2 1 generators of this group. In terms of the
spinors of length 2Nf , the microscopic Lagrangian is thus given by

1
2 2
L = i (D B ) T
M + h.c. ,
(36)
2
2
where the upper branch corresponds to = 1 and the lower branch to = 4. For reasons
that will become clear in the next subsection, we have introduced the four-vector B =
(B, 0).
As in the case of the quark mass term, the chemical potential term, B violates
the SU(2Nf ) symmetry. Similarly, we can maintain this symmetry by accompanying the
rotation of by a corresponding rotation of B,
V
and B V BV .
(37)
Such an extended symmetry must be manifest in the effective Lagrangian,

V V T
and B V BV .
(38)
This restricts the lowest order nonderivative term in to a linear combination of

2 Tr(B T B)
and 2 Tr(BB)
(39)
with arbitrary coefficients. Only the first of these terms contains a dependence on the
Goldstone fields. 1
At m = 0 the chemical potential breaks the global symmetry of the theory from
SU(2Nf ) down to the usual (as in three-color QCD) SU(Nf )L SU(Nf )R U (1)B .
If m 6= 0 also, the symmetry of the theory is only SU(Nf )V U (1)B .
1 Such type of symmetry breaking terms also occur in the context of the non-hermitian Random Matrix Theory
[12,3942].
488
6.2. Local symmetry and the coefficient of the 2 term

The coefficients of the terms (39) in the effective Lagrangian can be related to baryon
number susceptibility, i.e., the second derivative of the vacuum energy with respect
0 , these parameters are not
to . However, unlike G, i.e., the chiral condensate hi
independent. They are related to the pion decay constant F by virtue of a local symmetry
[13].
As was observed in [13], one can further extend the symmetry (37) of the microscopic
Lagrangian (36) to include local SU(2Nf ) flavor transformations
V
and B V B V
1
V V .
(40)
We can also recover the Lorentz (Euclidean) symmetry by transforming B as a four-vector.

To make such an extended local symmetry (and also Lorentz symmetry) manifest in
the effective Lagrangian (31) we must replace the normal derivative by a flavor covariant
derivative,

= B + BT ,

(41)
= + B + BT .
Thus we arrive at the effective Lagrangian of lowest order in the momentum expansion

F2
b
Tr 2m2 Re Tr(M)
2
F2
Tr + 2F 2 Tr B 0
=
2

b
F 2 2 Tr B T B + BB F 2 m2 Re Tr(M).
Leff =
(42)
It is important to note that the dependence on comes with no additional parameters. It is

completely fixed, by the local symmetry, in terms of an already existing parameter F .
7. Vacuum alignment
The static part of the Lagrangian (42) determines the vacuum alignment of the field
as well as the masses of the excitations. This part of the Lagrangian has the form

b
Lst () = F 2 2 Tr B T B + BB F 2 m2 Re Tr(M)

x2
F 2 m2
T
b
(43)
Tr B B + BB 2 Re Tr(M) ,
=
2
2
where we introduced x = 2/m . The 2 and m2 terms in (43) compete for the direction
of the condensation which we denote by . For x = 0 the orientation of is determined by
b . This value we denoted by c . This is the orientation of the usual
the mass matrix = M
chiral condensate which carries no baryon charge. When x = [13] the static Lagrangian
489
is minimized on a manifold in the space of , from which we choose the following value
and denote it by d ,

iI 0
( = 1),
(44)
d =
0 iI
where I is an antisymmetric Nf Nf matrix, which, written in (Nf /2) (Nf /2) blocks,
has the form

0 1
I=
.
(45)
1 0
This minimum (even for m 6= 0, unlike c ) has a degeneracy, which for Nf = 2, is simply a
rotation by the generator B. The condensate d breaks spontaneously the baryon number
symmetry. The degeneracy leads to the appearance of a massless Goldstone boson. For
Nf > 2 this condensate breaks more than just the U (1)B symmetry. For m = 0 it also
breaks SU(Nf )L SU (Nf )R down to Sp(Nf )L Sp(Nf )R given by the rotations leaving
the block matrix I in (44) invariant [13]. When m 6= 0 it breaks the SU(Nf )V symmetry
down to Sp(Nf )V . 2
The situation is similar in the = 4 case. The only difference is that now the matrix
is a symmetric SU(2Nf ) matrix. It rotates between the value of c given by (19) and d ,
which we chose as

i 0
( = 4).
(46)
d =
0 i
The discussion of the preceding paragraph carries over to the = 4 case with the already
familiar substitution of Sp(Nf ) by SO(Nf ). The condensate d breaks SU(Nf )L,R flavor
symmetries down to SO(Nf )L,R , in addition to breaking U (1)B .
At intermediate values of x the orientation of the condensate rotates, as a function of
x, from c to d . We shall prove that it can always be written as the linear combination
= c cos + d sin ,
(47)
where = 0 for x = 0 and /2 for x = .

The angle is a function of x. It is determined by substituting the value of given by
(47) into (43). This results in the static Lagrangian

2
x
(cos 2 1) 2 cos .
(48)
Lst ( ) = F 2 m2 Nf
2
Minimizing with respect to we find
= 0,
when x < 1,
1
(49)
cos = 2 , when x > 1.
x
Note that = 0 is always an extremum of (48) but it becomes a maximum for x > 1.
2 Similarly to the alignment of to the bare quark mass matrix M , the direction of is determined by an
c
d
external diquark source, J , which, as we shall see in section 11, breaks the degeneracy. When J is zero, we can
choose any orientation of d within the manifold of minima, and the results will not change due to the symmetry
relating all such minima.
490
We find that the condensate is a non-analytic function of x = 2/m . There is a second

order phase transition as a function of at x = 1. For small < m /2 the vacuum does not
change. At = m /2 a transition occurs and the direction of the condensate starts rotating
from that of c to that of d . A nonzero value of the projection on d , proportional to
sin , means that the vacuum breaks the baryon number symmetry spontaneously. This
happens at the value of equal to 1/2 of the mass of the lightest baryon in the system
the diquark, or, the baryonic pion.
8. Global minimum
In this section we show that the minimum of the static potential is given by defined
in (47). The argument in this section shows that it is an absolute, or global, minimum
of Lst (). In the next section we expand the Goldstone fields to second order about this
minimum.
8.1. = 1
We decompose the antisymmetric unitary matrix into 4 blocks of size Nf Nf ,

A C
( = 1).
(50)
=
B
CT
The antisymmetric matrices A and B satisfy the following unitarity constraints:
AA + CC = 1,
BB + (C C)T = 1,
AC = CB
( = 1).
Taking this into account we can express Lst entirely in terms of the matrix C,

1
1
1
C 2 Nf x 2 + 2 .
Lst () = F 2 m2 x 2 Tr C 2
x
x
x
(51)
(52)
Ignore, for the moment, the constraints on the matrix elements of C. The trace in (52) can
be viewed as the distance in the 2Nf2 dimensional space of real and imaginary parts of the
matrix elements of C from the point given by diagonal matrix 1/x 2 . For x > 1 the absolute
minimum is achieved
when C = 1/x 2 . The matrices A and B can be chosen, for example,
p
as A = B = iI 1 1/x 4 , to satisfy all the constraints (51). If we define cos = 1/x 2 , we
observe that the resulting matrix in (50) is given by (47).
When x < 1, we notice that unitarity constraints (51) demand that Tr CC 6 1. This
means that we have to consider the points in the space of C only within the unit hypersphere
around C = 0. It is easy to see that the minimum distance within this sphere is at the surface
point closest to 1/x 2 , i.e., at C = 1. The constraints (51) can be satisfied only by A = B = 0
and the resulting matrix in (50) is c , the minimum of the static potential at = 0.
8.2. = 4
The same analysis applies in the = 4 case. Since is now a symmetric matrix its
block decomposition is given by

=
A
CT
+C
B
491

( = 4),
(53)
with symmetric matrices A and B and with unitarity constraints

AA + CC = 1,
BB + (C C)T = 1,
AC = CB
( = 4).
(54)
The static part of the Lagrangian is the same as in (52). The minimum is given by C = 1
for x < 1 and C = 1/x 2 for x > 1. The matrices A andpB are not determined uniquely by
the unitarity constraints. One can check that A = B = i 1 1/x 4 (as in (47)) satisfy these
constraints. 3
9. Curvatures at the minimum

We have established that the global minimum of the effective Lagrangian is achieved at
the value of the field given by (47) and (49). The orientation of the condensate rotates
as a function of x = 2/m (nonanalytic at x = 1). In this section we shall expand the
effective Lagrangian up to the second order in the fluctuations of which will help us to
determine the (pseudo-)Goldstone masses and their dependence on x.
9.1. Normal phase: < m /2
When x < 1 the vacuum orientation of does not depend on x and is given by = c .
Expanding around c using the Goldstone fields defined in (9) according to

2
i
c ,
(55)
= U c U T = U 2 c = 1 +
F
2F 2
we find
Lst () = Lst (c ) +

m2 x 2
Tr[B, ]2 + Tr 2 + ,
2 4
(56)
where ellipsis denotes terms of higher powers of . The commutator in this Lagrangian
can be written as
3 As in the case of = 1, the case of odd number of flavors has to be investigated separately. Let us consider
Nf = 1. Then a symmetric unitary matrix has two degrees of freedom and can be parameterized as

i sin ei
cos
=
.
cos
i sin ei
b (33). We thus have that

Notice that the determinant of has to be equal to the determinant of the mass matrix M
the matrix C in (53) is just a number parameterized according to C = cos , leading to the effective potential (up
to constants)

1
1
1
x 2 Tr C 2
C 2 x 2 2 = x 2 cos2 2 cos x 2 .
x
x
x
The minimum is at = 0 for x < 1 and at cos = 1/x 2 for x > 1. In the diquark condensation phase (x > 1), we
find one massless excitation (the -mode), a nonzero baryon density and nonzero chiral and diquark condensates.
492
[B, ] = b,
(57)
with b the baryon charge of the pseudo-Goldstone modes. Since all our pseudo-Goldstone
modes are quarkantiquark or (anti-)diquark states the values of b are b = 0, 2. The
curvature of the P , Q and Q (16) modes, thus depends on through the baryon charge
and is given by m2 (b)2 . For example, in the = 1 case of Nf = 2, there are 3
pseudoscalar mesons, 1 diquark and 1 antidiquark. Using the block decomposition of the
generators (16) we find

(58)
Lst () = Lst (c ) + m2 Tr P 2 + (1 x 2 )QQ + .
Two comments are in order here. First, note that, the curvature (and the mass) of the
meson modes P does not depend on in the normal phase. Second, at x = 1 the curvature
of the diquark modes Q vanish, which signals a phase transition and the onset of diquark
condensation.
9.2. Diquark condensation phase: > m /2.
In this phase the condensate begins to rotate according to (47), (49). This rotation can
be also written as
= V c VT = V2 c ,
where V2 = eiX2 ,
(59)
and X2 is the generator that rotates c into d . Comparing (59), (47) we find
d = iX2 c .
(60)
This generator belongs to the set of broken generators with respect to c (as well as with
respect to d ) since it satisfies (12).
We could parameterize fluctuations of around the vacuum value given by as
= U UT ,
(61)
where U are unitary matrices generated by rotated (in the sense of (14)) generators,
V Xa V , instead of Xa (8), (9). However, as we shall find, the meson mass matrix is
diagonal in the basis given by the parametrization
= V U c U T VT ,
(62)
where we have rotated the coset generators Xa back to their = 0 values using (13), (14)
and (59).
Before expanding the static Lagrangian, we substitute (62) into (43) to obtain
2
T
2

x
F 2 m2
Tr U 2 c V BV c U V BV + BB
Lst () =
2
2

2
T b
(63)
2 Re Tr V MV U c .
The rotated values of B and c (by angle in the sense of (59)) can be expressed as
493
V BV = V BV
= B cos + X1 sin ,
T
b = (V M
b V
) =
= c cos d sin ,
VT MV
(64)
where we have introduced X1 by

X1 = iBX2 .
(65)
This generator, similarly to X2 , belongs to the coset of broken generators with respect to
c (12). It is the generator into which B is rotated while c is rotated into d .
Substituting (64) into (63) and expanding to second order in (55) we find

2
x
Lst () = Lst ( ) + F m2 sin 2 + sin Tr(X2 )
2
2
2

m x
Tr[B, ]2 cos2 Tr[X1 , ]2 sin2
+
2 4

2
(66)
+ Tr cos + .
As should be expected, the linear term vanishes due to (49) and we shall concentrate now
on the quadratic term.
9.2.1. = 1
At this point we need the explicit form of X1 for = 1. According to (60) we have

0 I
( = 1).
(67)
X2 = id c =
I 0
For X1 we therefore find

0 iI
X1 = iBX2 =
iI 0
( = 1).
(68)
We then use the block decomposition of (16) and find

Lst () = Lst ( ) + m2 Tr x 2 QQ cos2 PS2 + QR QR sin2

+ P 2 + QQ cos + ( = 1).
(69)
We have introduced the following projections of P and Q:

1
1
and PA = P I P T I ,
P + IPT I
2
2

1
1
and QI = Q I Q I
( = 1).
QR = Q + I Q I
2
2
PS =
(70)
These projections are orthogonal, Tr PS PA = Re Tr QR QI = 0. Using this fact and the

relation cos = 1/x 2 we obtain
Lst () = Lst ( )

+ m2 x 2 Tr QR QR sin2 + PA2 cos2 + PS2 +
( = 1).
(71)
From (71) we can now read off the curvatures for the different multiplets of the
(pseudo-)Goldstone modes. We see that there is a true flat direction, QI , which describes
494
true massless Goldstones of the diquark condensation phase. These fields are the phases
(a single U (1)B phase for Nf = 2 and a set of U (1)B SU (Nf )V /Sp(Nf )V phases
for other Nf ) of the diquark condensate. As we shall see in the next section, the linear
derivative terms in the effective Lagrangian mix QI and QR , thus the actual true Goldstone
excitations are certain linear combinations of QI and QR .
The degeneracies of the multiplets follow from the definitions (70) and are given by
Nf (Nf + 1)
Nf (Nf 1)
,
NPA =
1,
2
2
Nf (Nf 1)
( = 1).
(72)
NQR = NQI =
2
They correspond to representations of the group Sp(Nf ), which is the residual symmetry
remaining intact after spontaneous breaking. The number of the flat directions QI exactly
matches the number of the broken generators in SU(Nf )V U (1)B Sp(Nf )V .
NPS =
9.2.2. = 4
The derivation of the curvatures of the effective potential Lst in the = 4 case follows
the same lines as in the = 1 case. All the differences stem from the fact that is a
symmetric unitary matrix in this case. Consequently, c and d are given by (19) and (46)
instead of (15) and (44). In particular, the generator which rotates c into d is now given
by

0 1
( = 4),
(73)
X2 = id c =
1 0
and the generator X1 , into which B rotates and which is responsible for the splitting of
both P and Q branches (in the block decomposition (16)), is given by

0 i
( = 4).
(74)
X1 = iBX2 =
i 0
It is now straightforward to rewrite the quadratic part of the Lagrangian (66) in terms of
the P Q-block decomposition. The mass matrix becomes diagonal in terms of the following
projections of P and Q:

1
1
P + PT
and PA = P P T ,
2
2

1
1
QR = Q + Q and QI = Q Q ( = 4).
2
2
The static part of the effective Lagrangian in terms of these fields is given by
PS =
(75)
Lst () = Lst ( )

+ m2 x 2 Tr QR QR sin2 + PS2 cos2 + PA2 +
( = 4),
(76)
which is similar to (71), but S and A labels exchange places.

The main difference is in the dimension of the degenerate multiplets which matches the
representations of the corresponding residual symmetry group. By inspection one finds
Nf (Nf 1)
Nf (Nf + 1)
,
NPS =
1,
2
2
Nf (Nf + 1)
( = 4).
= NQI =
2
495
NPA =
NQR
(77)
10. Spectrum
In order to complete our program and determine the spectrum of low-lying excitations
we must take into account the derivative terms in the Lagrangian (42). They contribute in
a non-trivial way to the quadratic form in the Goldstone fields. Due to the non-Lorentz
invariant nature of the system we study at finite , the dispersion laws 4 do not have
a simple form, E 2 = p2 + m2 . The mass, as measured on the lattice and given by the
exponential fall-off of the propagator at large Euclidean time, is the value of E, i.e., ip0 ,
at the pole of the propagator at p = 0. These pole masses, or rest energies, we shall now
evaluate.
10.1. Normal phase: < m /2
Expanding the derivative terms in the Lagrangian (42) in the same way as we expanded
the static part in the previous section we obtain
2

1
Tr [B , ] + m2 2 + ,
(78)
2
where B = B0 , as defined earlier in Section 6.2. Using the Fourier decomposition of ,
X
X
p exp(ipx) =
p exp(Ex0 ipx),
(79)
(x) =
L() = L(c ) +
we find that the dispersion law has the generic form

q
(E + b)2 = p 2 + m2 or E = b + p2 + m2 ,
(80)
where b is the baryon charge of the given excitation. This is in agreement with the fact that
the effect of on each state is simply an energy shift by b. In particular, the dispersion
law of the P -type Goldstone modes, which carry no baryon charge, are not affected by
at all. The rest energies of the diquarks Q and antidiquarks Q are shifted according to
their charges. To summarize, the dispersion laws in the normal phase are given by 5
1/2
P : E = m2 + p 2
,
2
2 1/2
+ 2,
Q : E = m + p

1/2
2.
(81)
Q: E = m2 + p 2
4 In Euclidean field theory the dispersion relation is given by the poles of the propagator in the E ip -plane.
0
5 Note that though the mass of the diquarks Q vanishes at the transition point, = m /2, the dispersion law is
not linear, but quadratic, E p 2 . This is related to the well-known critical slowing down at a second order phase
transition.
496
Table 1
Spectrum (masses and degeneracies) of two-color QCD with fundamental quarks ( = 1)
< m /2: SU(Nf ) U (1)
mass
degen.
rep.
PS
Nf2 1
PA
Nf2 1
m + 2
1 N (N 1)
f
2 f
m 2
1 N (N 1)
f
2 f
> m /2: Sp(Nf )

mass
degen.
1 N (N + 1)
f
2 f
m2 /2
1 N (N 1) 1
f
2 f
p
2 1 + 3(m /2)4
1
2 Nf (Nf 1)
1+
1 N (N 1)
f
2 f
1+
rep.
Table 2
Spectrum (masses and degeneracies) of any-color QCD with adjoint quarks ( = 4)
< m /2: SU(Nf ) U (1)
mass
degen.
rep.
PA
Nf2 1
PS
Nf2 1
m + 2
1 N (N + 1)
f
2 f
m 2
1 N (N + 1)
f
2 f
> m /2: SO(Nf )

mass
degen.
1 N (N 1)
f
2 f
m2 /2
1 N (N + 1) 1
f
2 f
p
2 1 + 3(m /2)4
1
2 Nf (Nf + 1)
1+
1 N (N + 1)
f
2 f
1+
rep.
The masses of these excitations as well as their degeneracies are given in Tables 1 and 2.
A schematic picture of the mass dependence is shown in Figs. 1 and 2.
10.2. Diquark condensation phase: > m /2.
In this phase the ground state changes and we need to expand around the rotated value
of the condensate = . We find 6

1
L() = L( ) + Tr V B V , V B V ,
2

(82)
+ m2 2 cos + .
6 A useful observation which helps to write this in such a compact form is that Tr[X , ] = 0 for any .
1
0
This can be checked explicitly using the block decomposition of , but it is easier to see that once you realize
that the commutator of two X-like generators is a T -like generator, and its trace with another X-like generator,
such as 0 is zero.
497
Fig. 1. Spectrum of two-color QCD ( = 1) at finite and m (schematic). The branches are labeled
according to (81), (70) and 75) and by Youngs diagrams showing the multiplet structure under the
corresponding residual symmetry. The star denotes the conjugate fundamental representation. The
residual symmetry groups are marked above the plot. The degeneracies are given by (17) and (72).
For example, for the case of Nf = 2 we find NP = 3, NQ = NQ = 1, and NPS = 3. The branch
PA does not exist for Nf = 2. On the left/right of the plot the residual symmetries and multiplet
structure of the corresponding limiting cases = 0 and m = 0 are shown.
Expanding the product we find

L() =

1
Tr ( )2 2 cos [B, ]0 + Lstat () +
2
(83)
with the static part given by (66). We observe, first of all, that the linear derivative term
contains only the charged fields, i.e., Q and Q . The dispersion laws of the P fields
remain unaffected by the linear term. It remains Lorentz invariant, with mass given by the
curvature of the static part of the Lagrangian. In order to determine the dispersion laws for
the Q and Q fields we need to solve a secular equation obtained by substituting Fourier
decomposition of Qs into

L() = L( ) + Tr QR QR + QI QI
498
Fig. 2. Spectrum of any-color QCD with adjoint quarks ( = 4) at finite and m (schematic). The
branches are labeled according to (81), (70) and (75) and by Youngs diagrams showing the multiplet
structure under the corresponding residual symmetry. The residual symmetry groups are marked
above the plot. The degeneracies are given in Table 2. For example, for the case of Nf = 2: NP = 3,
NQ = 6, NQ = NQ = 3, NPS = 2 and NPA = 1. On the left/right of the plot the residual symmetries
and multiplet structure of the corresponding limiting cases = 0 and m = 0 are shown.

4 cos QI 0 QR + QR 0 QI + 42 QR QR sin2 + ,
(84)
where we only wrote the Q-dependent terms. This expression is the same for either = 1
or = 4; only the definition of Qs is different: Eqs. (70) or Eqs. (75). The secular equation
has the form

4E cos
E 2 p2
= 0.
(85)
det
4E cos E 2 p2 42 sin2
We see that QR and QI are mixed by the linear derivative term. The mixing is maximal
when = 0, and vanishes when = /2, i.e., when m . For any there is a solution
e
for which the rest energy E(0) = 0 the Goldstone boson branch, which we denote Q.
e = Q = QR + QI (at rest, p = 0). For = /2 it is entirely QI .
For = 0 it is given by Q
The other solution of the secular equation, the linear combination of QR and QI orthogonal
is massive. We denote it by Q
e .
to Q,
499
The dispersion laws for the meson multiplets PS , PA and the mixed diquarkantidiquark
e and Q
e are given by (in the = 1 case) the following relations:
multiplets Q
PS :
E 2 = p2 + m2 x 2 ,
PA :
E 2 = p2 + m2 x 2 cos2 ,
e :
Q
e
Q:
q
E 2 = p2 + 22 (1 + 3 cos2 ) + 2 2 (1 + 3 cos2 )2 + 4p2 cos2 ,
q
2
2
2
2
E = p + 2 (1 + 3 cos ) 2 2 (1 + 3 cos2 )2 + 4p2 cos2 ( = 1).
(86)
The only difference in the dispersion laws for = 4 is the interchange of PS and PA , and
also that the degeneracy of the multiplets as given by (72) and (77) is different. Note that
e is linear: E p. The slope is a
the dispersion relation for the lowest lying mode, i.e., Q,
function of : it vanishes at the transition point, x = 1, and approaches 1 for large . The
linear slope of low energy excitations is characteristic of superfluidity.
The pole masses of the excitations, i.e., the position of the pole at zero momentum,E(0),
are given in Table 1 for = 1 and Table 2 for = 4. The representations of the residual
symmetry groups are denoted by their Young diagrams. Because mesons are quarkquark
or quarkantiquark pairs they transform as rank two tensors (unless they are singlets). The
representations can be uniquely identified by their dimensions found in (17), (72) and (20),
(77). An explicit form of the representations is given in Appendix A. Figs. 1 and 2 show
schematic pictures these results, together with the residual symmetry groups in the different
phases. In particular, one observes that the spectrum is continuous at the transition point
x = 1.
11. Diquark source
In three-color QCD the diquark condensate is not a color-singlet, therefore, one cannot
study the phenomenon of diquark condensation by applying an external diquark source:
such a source term will not be gauge invariant. In the theories which we study in this
paper, e.g., in two-color QCD, the diquark condensate is colorless. Therefore, one can add
a gauge-invariant source j to the theory. This source plays a role similar to the role the
quark mass m plays with respect to the chiral condensate. Such a nonzero source term
j is in fact used in lattice simulations [9] in order to measure the value of the diquark
condensate in the limit j 0. In this section we show how the results of the previous
sections are modified in the presence of a nonzero j .
11.1. = 1
Let us first rewrite the scalar diquark source term in the macroscopic theory using our
SU(2Nf ) spinor notations. In the notations of Section 3 we find

j T
1
j T
iI 0
+ h.c. T 2 2 J + h.c.,
i C5 2 I + h.c. = 2 2
0
iI
2
2
2
500
J = j J
and J =
iI
0
0
iI

( = 1),
(87)
the antisymmetric matrix I is defined in (45), and the summation over the Nf flavor indices
has been suppressed. Comparing with the quark mass term (22) we find that the two belong
to the same multiplet (i.e., they transform into one another) under the SU(2Nf ) rotations. 7
We can write the sum of the diquark source and the bare mass term as

i j T C5 2 I + h.c. = 1 T 2 2 M ( = 1),
(88)
m
2
2
where
q

b + j J = m2 + j 2 M
b cos + J sin
M = mM
q
b .
(89)
= m2 + j 2 M
We have used a notation similar to (47) with
j
.
(90)
m
It is easy to see how the diquark source term modifies
the effective Lagrangian (42): we
p
b . The Gell-MannOaks
b in (29) by m2 + j 2 GM
need to replace the mass matrix mGM
Rennen relation at = 0 becomes
p
m2 + j 2 G
2
.
(91)
m =
F2
Using this fact we can write the effective Lagrangian for the theory with both mass term
and diquark source (compare to (42)) as
tan =

F2
b ) .
Tr 2m2 Re Tr(M
(92)
2
Now we can repeat the steps we performed in Section 9, but for the Lagrangian (92).
Since the source term we introduced favors the direction of d , we expect that the
minimum of the Lagrangian (92), , is again given by a linear combination of c and d
as in (47),
Leff () =
= = c cos + d sin = V c VT ,
where V = eiX2 ,
(93)
and X2 is the generator that rotates c into d as before. The value of , as determined
by the saddle-point equations, now depends on the value of the diquark source. The proof
of Section 8 is not easily extendable to this case. We shall continue on the assumption that
the minimum is global. We shall prove, however, that it is a minimum when we expand to
second order in fluctuations of and find no linear terms and a positive quadratic form. If
we substitute this ansatz into the effective Lagrangian (92) we find
7 There is a freedom in the choice of the orientation of the diquark source J. It is precisely the same freedom
as the one corresponding to the SU(Nf ) U (1)/Sp(Nf ) degeneracy of the diquark condensation vacuum d .
An infinitesimal diquark source j determines the vacuum orientation of d . Our choices reflect this fact. Indeed
d = J .

Lst ( ) = F 2 m2 Nf

x2
(cos 2 1) 2 cos( ) .
2
501
(94)
The dependence of the angle on x follows from minimizing this Lagrangian and is given
by
x 2 cos sin = sin( ),
(95)
which is different from (49), but coincides with it in the limit 0, of course. For finite
the angle is already nonzero ( = ) at x = 0 as it should be, since a diquark source
drives a nonzero diquark condensate. At x = , the value of is /2 independently
of . The major difference from (49) is that the dependence of on x is analytic. There
is no phase transition as a function of when the diquark source j is nonzero. This is
to be expected: the diquark source plays the role an external magnetic field plays in a
ferromagnet. The diquark source explicitly breaks SU(Nf )V U (1)B symmetry down to
Sp(Nf )V and there is only one phase (the one with residual Sp(Nf )V symmetry) for all
values of x.
Our next step is to expand in powers of fluctuations of around . This can be done
following the algebra in Section 9 with minimal modifications. The Goldstone manifold is
again parameterized by
= V U c U T VT ,
but the second identity in (64) is now given by
T
b V = V M
b V
=
= c cos( ) d sin( ),
VT M
(96)
(97)
whereas the first identity is the same as before. Using this algebra, the effective Lagrangian
to second order in the pion fields can be expressed as

L() = Lst ( ) + F m2 x 2 cos sin sin( ) Tr(X2 )

1
+ Tr ( )2 2 cos [B, ]0
2

m2 x 2
Tr[B, ]2 cos2 Tr[X1 , ]2 sin2
+
2 4

(98)
+ Tr 2 cos( ) + .
Inspecting the linear terms we find that they vanish if is chosen according to (95). This
means that our ansatz (93) is indeed an extremum.
In order to find the mass spectrum we have to study the term of second order
in the -fields in the Lagrangian (98), and determine the dispersion laws of the
different (pseudo-)Goldstone fields. Using block decomposition of the generators (16),
projections (70) of P and Q and the relationship (95) between x and we find

L() = Lst ( ) + Tr QR QR + QI QI

4 cos QI 0 QR + QR 0 QI
502

sin
sin
+ QR QR x 2 sin2 +
+ m2 Tr QI QI
sin
sin

sin
+ Tr PA PA + PA2 m2 x 2 cos2 +
sin

sin
+ ( = 1).
+ Tr PS PS + PS2 m2 x 2 +
sin
(99)
As in the case of a vanishing diquark source studied in the previous section, the linear
derivative term contains only the charged fields Q and Q . Therefore the dispersion
relations for the neutral fields P are not affected by this term: the masses of the P fields
are given by the curvature of the static part of the Lagrangian at the minimum. However
the linear derivative term mixes the charged Goldstones. In order to obtain the dispersion
laws for the Q fields, as in Section 10, we have to solve a secular equation obtained by
substituting the Fourier decomposition of these fields into the effective Lagrangian (99).
The secular equation reads

det
E 2 p 2 m2 sin
sin
4E cos
4E cos
2
2
E p 42 sin2 m2 sin
sin

= 0.
(100)
The dispersion laws for the different Goldstone modes are therefore found to be given
by
PS :
PA :
e :
Q
e
Q:

sin
2
E =p
,
x +
sin

sin
2
2
2
2
2
E = p + m x cos +
,
sin
sin
+ 22 (1 + 3 cos2 )
E 2 = p2 + m2
sin
s

sin
2
2
2
2
2
2
,
+ 2 (1 + 3 cos ) + 4 cos p + m
sin
2
+ m2
sin
+ 22 (1 + 3 cos2 )
E 2 = p2 + m2
sin
s

sin
2
2
2
2
2
2
( = 1), (101)
2 (1 + 3 cos ) + 4 cos p + m
sin
e are no longer
where is an implicit function of x and given by (95). Note that the Q
true Goldstone modes, since the symmetry, which is broken spontaneously in the diquark
condensation phase x > 1 at j = 0, is now broken explicitly by the diquark source term.
The masses of the different multiplets are given by the value of E at p = 0. The positivity
of all masses shows that the minimum given by the saddle-point equation (95) is at least
a local minimum. This mass spectrum at a small nonzero diquark source is depicted in
Fig. 3.
503
Fig. 3. Mass spectrum for a small non-vanishing source j = 0.1m.
11.2. = 4
In the = 4 case the diquark source term has the form 8

j T
1 T
j
i 0
i C5 + h.c. = 2
+ h.c. T 2 J + h.c.,
0
i
2
2
2

i
0
J = j J and J =
( = 4).
0 i
(102)
The analysis follows the same lines as in the = 1 case. We again introduce the combined
mass matrix
q
b ,
b + j J = m2 + j 2 M
(103)
M = mM
b and J defined for = 4 in (33) and (102). In the expansion of the Lagrangian
with M
to second order in the -fields for = 1 we have only used the general commutation
b which, with a proper redefinition of the generators X1 and X2 , are the
properties of M
8 As before, our previous choice of corresponds to our choice of J , i.e., = J .
d
d
504
same in the case = 4. We thus obtain the same second order Lagrangian (98) as for = 1
resulting in the same saddle point equation (95) and the same mean field value of the free
energy. However, the mass spectrum is slightly different. We express the quadratic part of
the Lagrangian in terms of the block-decomposition (16) and the projections (75). The final
result is

L() = Lst ( ) + Tr QR QR + QI QI

4 cos QI 0 QR + QR 0 QI

sin
sin
+ QR QR x 2 sin2 +
+ m2 Tr QI QI
sin
sin

sin
+ Tr PS PS + PS2 m2 x 2 cos2 +
sin

sin
+ ( = 4).
(104)
+ Tr PA PA + PA2 m2 x 2 +
sin
As in the = 1 case, the linear derivative term mixes the Q and Q fields. In order
to get the dispersion relations of these fields, the same secular equation as before (100)
has to be solved. The dispersion laws are the same as in (101) with the familiar exchange
of labels S A, and the degeneracy of the multiplets given by (77) instead of (72). The
corresponding mass spectrum at a small nonzero j is shown in Fig. 3.
12. Condensate and baryon charge densities

In this section we derive the dependence of the chiral condensate, the diquark condensate
and the baryon charge densities. These densities are obtained by differentiating the vacuum
energy density with respect to m, j and ,
Evac
Evac
Evac
,
hi =
,
nB =
.
(105)
m
j
The shorthand symbol denotes the l.h.s. of Eq. (87) in the = 1 case and of Eq. (102)
in the = 4 case without factor j .
In the effective theory the vacuum energy density is given by the value of the effective
Lagrangian (42) at the minimum,
h i =

F2
22 Tr B T B + BB 2m2 Re Tr
2
(106)
= 4Nf F 2 2 sin2 2Nf G(m cos + j sin ),
Evac = Leff ( ) =
where we used (90) and (91). By differentiating 9 we find

h i = 2Nf G cos ,
hi = 2Nf G sin ,
nB = 8Nf F 2 sin2 ,
(107)
where the angle of the rotation of the condensate, , depends implicitly on , m and j
through the solution of (95). Note, in particular, that the sum h i2 + hi2 does not
9 The implicit dependence of on m, j and is not contributing because E / = 0.
vac
505
Table 3
The values of the chiral condensate, hi,

the diquark condensate, hi, and the baryon density nB
in the two phases of the theory
phase
hi
hi
nB
< m /2
0
hi
2
0 m
hi
2
r0
4
0 1 m
hi
2
0

4
8Nf F 2 1 m
2
> m /2
and the diquark hi condensates in units of

Fig. 4. The magnitudes of the chiral hi
0 = 2Nf G as a function of /m for zero diquark source. Also the density of the baryon
hi
charge in units of 32Nf F 2 m is shown.
Fig. 5. The magnitudes of the chiral hi

and the diquark hi condensates in units of
0 = 2Nf G as a function of /m for zero diquark source. Also the density of the baryon
hi
charge in units of 32Nf F 2 m is shown.
506
depend on , m or j (to the order in Chiral Perturbation Theory we are working), which
reflects the fact that the condensate rotates.
For j = 0 the dependence of on is simple: = 0, for < m /2, and =
arccos[m2 /(42 )] otherwise (49). That means the densities (107) are constant for <
m /2, as they should since the vacuum state does not change until reaches m /2. The
results for the condensate and baryon charge densities are summarized in Table 3.
For j 6= 0 the angle varies between the values and /2 as a function of and m,
determined by the saddle point equation (95) x 2 sin cos = sin( ). The saddle point
is a quartic equation which, in principle, can be solved analytically for arbitrary j . The
dependence of the densities (107) on is shown in Fig. 4 for j = 0 and in Fig. 5 for j 6= 0.
13. Equation of state of the non-ideal diquark gas

We now take a different look at the system we study (quarks with two colors or any-color
adjoint quarks) near the threshold, m /2. First of all, we observe, that for very close
to m /2 the density of baryons is very (arbitrarily) small. This means that we can describe
the system by an almost ideal gas. It should be a Bose gas, since the baryons, i.e., diquarks,
are bosons. The equation of state of a dilute non-ideal Bose gas is a textbook problem
[43]. The interaction between the bosons is crucial here (the dependence on its strength
is non-analytic, e.g., the attractive Bose gas is unstable). The baryons in our system repel
each other, as we shall see. Moreover, we shall also see that the strength of this repulsion
is exactly the one that gives the correct equation of state (107), (49) at small densities:

m
+ .
(108)
nB = 32Nf F 2
2
This is a very nontrivial check of the consistency of our effective Lagrangian!
Since we are working at very small densities, we can expand the Lagrangian (29) up to
next (quartic order) in the Goldstone fields. The quartic terms will describe the interaction
energy of the Goldstones due to two-body scattering,
Eint =
1
m2
Tr[,
][,
Tr 4 + .
24F 2
24F 2
(109)
We are now working in Minkowski space and the extra minus sign in front of the first term,
compared to (29), is a consequence of this. The presence of the time derivatives should not
confuse the reader. Since we are dealing with a non-relativistic system, m /2 m ,
we only need the leading term in the time dependence of the relativistic fields, i.e., 0
im . The spatial derivatives give only a subleading contribution. The dominant effect
is that of the s-wave amplitude.
We now consider the ground state of such a system with the baryon density fixed (unlike
before, when we fixed ). Most of the particles will occupy the lowest energy, i.e., zeromomentum, level (the population of the excited levels is the next order effect). These
particles form a condensate, whose amplitude we denote by . The energy of the system
507
is different from zero only because of the interaction, and, as a function of , it is given
to us by (109). Looking at the block decomposition of (16) we note that only Q and Q
(diquark and antidiquark) fields carry nonzero baryon number. Therefore, we set P = 0 in
our case. Thus, from (109), we find for the dependence of the energy on the magnitude of
the Q condensate
2
m2
Tr QQ + ,
(110)
2
4F
where we neglected subleading non-relativistic corrections.
Now we need to calculate nB as a function of Q. Since Q carries baryon charge 2, we
can write 10

nB = 2 Tr Q i0 Q Q i0 Q = 4m Tr QQ + ,
(111)
Eint =
where we neglected subleading corrections. Since all flavors of diquarks are equivalent 11
the matrix Q is proportional to I for = 1, or to 1 for = 4. Thus we can relate traces
in (110) and (111) by Tr(QQ )2 = (Tr QQ )2 /Nf . Therefore, we find for the dependence
on nB of the vacuum energy at fixed nB ,
Evac =
n2B
nB
nB
m + Eint =
m +
+ .
2
2
64F 2 Nf
(112)
The first term is the rest energy of the diquarks. This determines the equation of state
nB
Evac m
+
=
+ .
nB
2
32F 2 Nf
(113)
The fact that the equations of state (108) and (113) are identical is a very nontrivial property
of the effective Lagrangian (42). It is intimately related to the symmetries (global and local)
which determine the form of this Lagrangian.
This calculation also shows that the diquarks Bose-condense [45], and thus form a
superfluid. Since the diquarks are charged we can call this phase a superconducting phase.
14. Conclusions and discussion

In this paper we have derived the low-energy effective Lagrangian for certain QCD-like
theories at finite chemical potential . It contains the same number of phenomenological
parameters as at = 0: the pion decay constant F and the vacuum value of the chiral
condensate h i0 . Using this effective Lagrangian we have completely determined its
low-energy properties. In particular, we have established the dependence of the ground
state, the condensates, and the masses of the excitations on the chemical potential , the
bare quark mass m and the diquark source j . The theories to which our analysis applies
10 Another way to see (111) is to use the fact that n = L/. Thus, from (42) we find that n =
B
B
2F 2 Tr B 0 = 2i Tr B0 , which coincides with (111).
11 There is only one flavor of diquarks for = 1, N = 2. For larger N the flavor SU(N ) symmetry ensures
f
f
f
the equivalence of all diquarks.
508
include: (i) two-color QCD with quarks in fundamental color representation, and (ii) anycolor QCD with quarks in adjoint color representation. The unifying feature of these
theories is that the low-energy excitations, i.e., the Goldstone bosons of spontaneously
broken symmetries, include baryons in the form of diquark states. Since the chiral
Lagrangian describes the dynamics of such low-energy excitations, we can use the effective
theory to describe phenomena associated with the condensation of diquarks.
Since we are working only to the lowest order in Chiral Perturbation Theory, we must
remember that our approach has a limited region of applicability. First of all, the momenta
of the particles, described by the Lagrangian (42) must be smaller than the scale M of
the masses of the non-Goldstone excitations, such as vector mesons, for example. In other
words, the expansion parameter of the chiral perturbation theory is p/M. The bare quark
mass is also included to leading order in m. It is obvious from (42) that m m2 p2 in
the momentum power counting. The chemical potential is an external parameter with
the dimension of mass. The interesting behavior of our theory is at values of m p,
and is counted as order p in the momentum power counting. Our effective theory thus
applies as long as m and are much smaller than M.
Our effective theory predicts a phase transition at a value of = m /2. It is important
to note that this transition occurs within the domain of applicability of the effective theory,
as long as m is small, as discussed above, unlike, for example, the transition in QCD with
three colors. The transition at = m /2 is a robust prediction of our effective theory. The
existence of such a transition is confirmed by lattice Monte Carlo simulations [7,9].
The fact that our effective Lagrangian (42) predicts the transition at a value of equal
to m /2 is a direct consequence of the symmetries of the theory. The coefficient of
the -dependent term in the effective Lagrangian, responsible for the transition, is not
an independent phenomenological parameter, but is fixed rigidly by the local symmetry
[13]. This symmetry corresponds to the conservation of the baryon charge (among other
generators of SU(2Nf )). This ensures that the charge of a composite state in the effective
theory is an additive sum of the charges of the constituents (non-renormalization). Indeed,
the factor two in the relation 0 = m /2 is the baryon charge of the diquark. This is in
agreement with the value of , below which no transition can occur at zero temperature,
given by the minimum value of the mass per baryon number among all baryons in the
theory (see, e.g., [44]). In three-color QCD this would correspond to a value of 0
approximately equal to 1/3 of the nucleon mass (offset by the binding energy of nuclear
matter). 12
Another way of looking at the -dependence in the theory is the following. In the
microscopic theory, given by the Lagrangian (36), enters as (a timelike component of)
an Abelian gauge potential. As such, it can be completely removed from the Lagrangian
by the time-dependent gauge transformation
e
and e ,
(114)
12 We remind the reader that we measure the baryon charge in units of the U (1) charge of a single quark.
B
509
where is the Euclidean time. 13 However, this does not mean that the partition function
does not depend on . Let us consider first the case of finite temperature. In this case, the
boundary conditions in the Euclidean time direction for the quarks change after (114): they
are no longer antiperiodic. The dependence on in the partition function comes entirely
from the boundary conditions on the fermion fields

(115)
=1/T = e/T =0 and =1/T = e/T =0 .
The -independent Lagrangian with such boundary conditions is completely equivalent
to the Lagrangian (36) and usual antiperiodic boundary conditions. This fact can be
conveniently used in lattice simulations. The corresponding effective theory can also be
defined with all -dependence in the boundary conditions. The boundary conditions for a
given effective (composite) bosonic field should then read (1/T ) = exp(b/T )(0),
where b is the baryon charge of the field [46]. In the case of the matrix valued field
we have

T
B =0 exp
B ,
(116)
=1/T = exp
T
T
where B is the baryon charge matrix (35).
In the limit of zero temperature, our intuition suggests that the dependence of the
partition function on boundary conditions should weaken and disappear. This is, however,
not completely true for the boundary conditions such as (115) due to their singular nature in
the limit 1/T . There is indeed an interval of : 0 < < 0 , where the partition
function does not depend on . However, outside this interval, the singular nature of the
boundary conditions starts playing a role: a phase transition occurs and a -dependence
appears.
Why is the approach of this paper not directly applicable to real three-color QCD? The
main reason is that at = 0 the effective theory described by the chiral Lagrangian does
not contain excitations with nonzero baryon number (pions do not carry baryon charge).
It is easy to see that, as a consequence, applying either the method of local symmetry of
Section 6.2 or the gauge transformation described in this section, one finds no dependence
on in the effective chiral Lagrangian of QCD. Related to that is the fact that the value of
0 mnucleon/3 is large, and the approach of this paper, based on an expansion in small
, will not reach it.
As we emphasized, the theories we considered are related by the fact that the fermion
representations are pseudo-real. This is a consequence of the antiunitary symmetries of
the Dirac operator (1), (2). We identified two such symmetry classes, distinguished by the
Dyson index = 1 and = 4. The effective theories for these two classes of theories
are very similar and dual, or complementary, to each other, from the point of view of the
residual global symmetries. In the diquark (finite density) phase, for the = 1 theories
the residual flavor symmetry is given by Sp(Nf ), while for the = 4 case it is SO(Nf ).
The excitations form multiplets which correspond to symmetric or antisymmetric second
13 Note that, in Euclidean formulation, the and are independent variables and the global flavor symmetry
group is in fact Gl(2Nf ) [38].
510
rank tensor representations of these groups (see Figs. 1, 2 and Tables 1, 2). One can see
that the = 1 and = 4 cases mirror each other with respect to Sp(Nf ) SO(Nf )
and symmetric antisymmetric. On the other hand, the Dirac operator for three-color
QCD with fundamental quarks does not have any antiunitary symmetries, and the fermion
representations are complex. This case falls into the third remaining Dyson class with the
index = 2.
It would be interesting to apply the approach of this paper to lattice theories with
pseudoreal fermions, for example to two-color QCD with fundamental quarks. The
symmetries of such theories are different from their continuum counterparts and were
analyzed in [9]. In particular, the transition to continuum limit may turn our to be nontrivial.
This can be related to the antiunitary symmetry of the lattice Dirac operator for staggered
fermions in fundamental representation

1X
x, Ux, x+,y
,
(117)
Dxy =
Ux, x,y
2
where x, = (1)x1 ++x1 , and U are SU(2) color matrices. The antiunitary symmetry
of this lattice Dirac operator is given by 2 D = D 2 , or [D, 2 K] = 0. Since (2 K)2 =
1, we conclude that such Dirac operator belongs to the class = 4. However, in the
continuum limit, the Dirac operator must be in the class = 1. Such an observation
was also made in [9,10,30,47,48] from the point of view of global symmetries and their
breaking, where it was pointed out that it is not yet known how the apparent pattern of
SU(2Nf ) O(2Nf ) breaking becomes SU(2Nf ) Sp(2Nf ) in the continuum. Even
though the symmetry may modify many of the details of our analysis, when it is applied
to lattice theories, some features should be robust. Such would include the phase transition
at = m /2, the relation similar to h i2 + hi2 = const between the chiral and the
diquark condensates, the linear dependence of nB on near the transition, the existence
of several branches in the spectrum, similar to Figs. 1, 2, 3, and many other qualitative
features.
Acknowledgements
S. Hands, H. Leutwyler, M.-P. Lombardo, S. Morrison, E. Shuryak and D.K. Sinclair are
acknowledged for useful discussions. J.B.K. is supported in part by the National Science
Foundation, NSF-PHY96-05199. D.T. and J.J.M.V. are partially supported by the US DOE
grant DE-FG-88ER40388. D.T. is supported in part by Holderbank-Stiftung and by
Janggen-Phn-Stiftung. A.Z. is supported, in part, by the National Science and Engineering
Research Council of Canada (NSERC).
Appendix A
In this appendix we analyze the representations of the remaining symmetry groups
shown in Table 1 for = 1. They are all different subgroups of Sp(2Nf ).
511
Under the symmetry group Sp(2Nf ) the Goldstone fields transform as

V V T .
(118)
With parameterized as U c U T (with U SU(2Nf )) and V c V T = c (the 2Nf

2Nf antisymmetric unit matrix is denoted by c ) it follows that the generators transform
according to

(119)
Xc V XV 1 c = V Xc V T or Xc ij Vik Vj l Xc kl .
From the transposition relation (12) it follows that (Xc )T = Xc . If the symmetry
group is Sp(2Nf ) the generators transform according to an antisymmetric rank two
representation of Sp(2Nf ). The degeneracy is thus given by 2Nf2 Nf 1. If the symmetry
group is SU(Nf ) U (1) the symmetry transformation is given by

U1 0
,
(120)
V=
0 U1
the P -type generators transform according to P T U1 P T U11 and the Q-type generators
as Q U1 QU1T . Since the Q are antisymmetric they transform according to an
antisymmetric rank two representation of SU(Nf ). The fields Q transform according to
the conjugate representation. The dimension of both representations is equal to Nf (Nf
1)/2. Since the P -type generators are traceless, the degeneracy is given by Nf2 1.
In the diquark condensation phase the symmetry group Sp(2Nf ) is with respect to the
rotated antisymmetric unit matrix, , with symplectic transformations defined by
V V T = .
(121)
According to the argument at the beginning of this appendix, the transformation properties
of the generators are given by

(122)
X ij Vik Vj l X kl ,
where the X are the rotated generators defined by
X = V XV1 ,
(123)
and V = exp(iX2 /2) defined in (59).

With Sp(Nf ) as symmetry group the symmetry transformations are given by (120)
with U1 = I U1 I (the Nf Nf antisymmetric unit matrix is denoted by I ). Since in
this case {V , X2 } = 0 and using that = V U c U T VT (see Eq. (62)) we find that the
representations can be discussed in terms of the fields U c U T with the familiar P Q-block
structure of the generators. The P -type generators thus transform as P T I U1 P T I U1T .
The symmetric and and antisymmetric components of P I corresponding to PS I and
PA I , respectively, transform independently. The dimensions of the representations are
given by Nf (Nf + 1)/2 and Nf (Nf 1) 1, respectively. The generators Q and I Q I
e and Q
e transform in the
transform in the same way, and thus the linear combinations Q
same way. Since Q is antisymmetric they transform according to an irreducible rank two
representation with dimension equal to Nf (Nf 1)/2.
512
Finally, for symmetry group Sp(Nf ) Sp(Nf ) the symmetry transformation is given
by

U1 0
,
(124)
V=
0 U2
with both U1 and U2 symplectic transformations. In this case it is imperative to consider the
rotated generators. However, we only need the rotated generators for . In terms of
the block structure of the unrotated generators, we obtain after a straightforward calculation
for = /2,

I PA I + iQI I iI PS + QR
.
(125)
X =
iPS I + I QR I
PA iI QI
Therefore the combination iI PS + QR transforms as
iI PS + QR U1 (iI PS + QR )U2T .
(126)
Indeed, this transformation corresponds to the Young tableaux given in upper right corner
e QR . The 11- and 22-blocks of
of Fig. 1. Notice that for we have that Q
the matrix (125) correspond to two different linear combinations of PA and QI . One
combination transforms as a rank two tensor with respect to U1 , and the other combination
as a rank two tensor with respect to U2 . Since both combinations are antisymmetric with
respect to transposition, they transform according to the Young tableaux given in the
lower right corner of Fig. 1. The additional singlet terms arise because the irreducible
representations are traceless.
In case of Sp(Nf ) symmetry, U2 = U1 = I U1 I , the transformation properties of PS ,
PA , QR and QI can be obtained from (125) by combining the transformation properties of
the diagonal blocks and of the off-diagonal blocks. The results are in agreement with the
discussion in the paragraph following Eq. (123).
The same analysis can be performed for = 4. In fact, because the matrix I is absent,
this case is somewhat simpler, and we leave it as an exercise to the reader.
References
[1] F. Karsch, M.-P. Lombardo (Eds.), Proceedings of QCD at Finite Baryon Density, Bielefeld,
April 1998, Nucl. Phys. A 642 (1998).
[2] J.B. Kogut, M.P. Lombardo, D.K. Sinclair, Phys. Rev. D 51 (1995) 1282; Nucl. Phys. B, Proc.
Suppl. 42 (1995) 514.
[3] I.M. Barbour, S.E. Morrison, E.G. Klepfish, J.B. Kogut, M.P. Lombardo, Nucl. Phys. Proc.
Suppl. A 60 (1998) 220.
[4] D. Bailin, A. Love, Phys. Rept. 107 (1984) 325.
[5] M. Alford, K. Rajagopal, F. Wilczek, Phys. Lett. B 422 (1998) 247; Nucl. Phys. B 537 (1999)
443.
[6] R. Rapp, T. Schfer, E.V. Shuryak, M. Velkovsky, Phys. Rev. Lett. 81 (1998) 53; Ann. Phys. 280
(2000) 35.
[7] E. Dagotto, F. Karsch, A. Moreo, Phys. Lett. B 169 (1986) 421.
[8] E. Dagotto, A. Moreo, U. Wolff, Phys. Rev. Lett. 57 (1986) 1292; Phys. Lett. B 186 (1987) 395.
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
513
S. Hands, J.B. Kogut, M.-P. Lombardo, S.E. Morrison, Nucl. Phys. B 558 (1999) 327.
S. Hands, S.E. Morrison, hep-lat/9902012; hep-lat/9905021.
T. Schfer, Phys. Rev. D 57 (1998) 3950.
M.A. Stephanov, Phys. Rev. Lett. 76 (1996) 4472; Nucl. Phys. Proc. Suppl. 53 (1997) 469.
J. Kogut, M.A. Stephanov, D. Toublan, Phys. Lett. B 464 (1999) 183.
R.F. Alvarez-Estrada, A. Gomez Nicola, Phys. Lett. B 355 (1995) 288; Phys. Lett. B 380 (1996)
491 (Erratum).
C. DeTar, Quarkgluon plasma in numerical simulations of QCD, in: R. Hwa (Ed.), Quark
Gluon Plasma 2, World Scientific, 1995.
A.V. Smilga, Phys. Rep. 291 (1997) 1.
W. Pauli, Nuovo Cimento 6 (1957) 205.
F. Grsey, Nuovo Cimento 7 (1958) 411.
S. Duane, J.B. Kogut, Nucl. Phys. B 275 (1986) 398.
S. Duane, A.D. Kennedy, B.J. Pendleton, D. Roweth, Phys. Lett. B 195 (1987) 216.
V.A. Novikov, M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Nucl. Phys. B 260 (1985) 157.
C. Vafa, E. Witten, Nucl. Phys. B 234 (1984) 173.
S. Dimopoulos, Nucl. Phys. B 168 (1980) 69.
M. Vysotskii, Y. Kogan, M. Shifman, Sov. J. Nucl. Phys. 42 (1985) 318.
D.I. Diakonov, V.Yu. Petrov, Lecture Notes in Physics, Vol. 417, Springer, 1993.
M.E. Peskin, Nucl. Phys. B 175 (1980) 197.
A. Smilga, J.J.M. Verbaarschot, Phys. Rev. D 51 (1995) 829.
F.J. Dyson, J. Math. Phys. 3 (1962c) 1199.
E.V. Shuryak, J.J.M. Verbaarschot, Nucl. Phys. A 560 (1993) 306.
J.J.M. Verbaarschot, Phys. Rev. Lett. 72 (1994) 2531; Phys. Lett. B 329 (1994) 351.
M. Zirnbauer, J. Math. Phys. 37 (1996) 4986.
J.J.M. Verbaarschot, hep-ph/9902394, Lectures given at APCTPRCNP Joint International
School on Physics of Hadrons and QCD, Osaka, Japan, 1998 and the 1998 YITP Workshop
on QCD and Hadron Physics, Kyoto, Japan, 1998.
M.A. Halasz, J.C. Osborn, J.J.M. Verbaarschot, Phys. Rev. D 56 (1997) 7059.
S. Weinberg, Physica A 96 (1979) 327.
J. Gasser, H. Leutwyler, Ann. Phys. 158 (1984) 142.
J. Gasser, H. Leutwyler, Nucl. Phys. B 250 (1985) 465.
H. Leutwyler, Ann. Phys. 235 (1994) 165.
D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 560 (1999) 259.
Y.V. Fyodorov, B.A. Khoruzenko, H.-J. Sommers, Phys. Lett. A 226 (1997) 46.
Y.V. Fyodorov, M. Titov, H.-J. Sommers, Phys. Rev. E 58 (1998) 1195.
K.B. Efetov, Phys. Rev. Lett. 79 (1997) 491; Phys. Rev. B 56 (1996) 9630.
A.V. Kolesnikov, K.B. Efetov, Waves Random Media 9 (1999) 71.
L.D. Landau, E.M. Lifshitz, L.P. Pitaevskii, Statistical Physics, Pergamon Press, Oxford, New
York, 1980, part 2, ch. 25.
M.A. Halasz, A.D. Jackson, R.E. Shrock, M.A. Stephanov, J.J.M. Verbaarschot, Phys. Rev. D 58
(1998) 096007.
E. Shuryak, Lectures given at NATO Advanced Study Institute, in: Confinment, Duality and
Nonperturbative Aspects of QCD, Cambridge, 1997, p. 307.
T.S. Evans, hep-ph/9510298.
J.B. Kogut, H. Matsuoka, M. Stone, H.W. Wyld, S. Shenker, J. Shigemitsu, D.K. Sinclair, Nucl.
Phys. B 225 (1983) 93.
S. Hands, M. Teper, Nucl. Phys. B 347 (1990) 819.

Fragmentation functions for pions, kaons, and

protons at next-to-leading order
B.A. Kniehl a , G. Kramer a , B. Ptter b
a II. Institut fr Theoretische Physik, Universitt Hamburg, Luruper Chaussee 149, 22761 Hamburg, Germany
b Max-Planck-Institut fr Physik (Werner-Heisenberg-Institut), Fhringer Ring 6, 80805 Munich, Germany
Abstract
We present new sets of fragmentation functions for charged pions, charged kaons, and protons,
both at the leading and next-to-leading orders. They are fitted to the scaled-momentum distributions
of these hadrons measured in e+ e annihilation on the Z-boson resonance at CERN LEP1 and SLAC
SLC. These data partly come as light-, charm-, bottom-quark-enriched and gluon-jet samples, which
allows us to treat all partons independently, after imposing the SU(2) flavour symmetry relations. In
order to gain sensitivity to the scaling
violation in fragmentation, we also include data from SLAC
PEP, with center-of-mass energy s = 29 GeV, in our fits. This allows us to also determine the
strong-coupling constant, with a competitive error. LEP1 data on the longitudinal cross section as
well as DESY DORIS and PETRA data at lower energies nicely agree with theoretical predictions
based on our fragmentation functions. 2000 Elsevier Science B.V. All rights reserved.
PACS: 13.65.+i; 13.85.Ni; 13.87.Fh
1. Introduction
The inclusive production of a single charged hadron, h, in the annihilation process
e+ e ( , Z) h + X,
(1)
has been measured at many different e+ e colliders over a wide range of center-of-mass
(CM) energies, s, between 3 and 183 GeV [1,2]. Here, h may either refer to a specific
hadron species, e.g., pion, kaon, or proton, or to the sum over all charged-hadron species.
A large amount of precise data has now become available from various experimental
collaborations at CERN LEP1 and SLAC SLC, who started taking data several years ago.
The process (1) is particularly suitable in order to study the fragmentation of quarks and
gluons into hadrons. The information contained in fixed-target, hadron-collider, and epscattering data is less useful, since it is obscured by theoretical uncertainties from the
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 0 3 - 5
B.A. Kniehl et al. / Nuclear Physics B 582 (2000) 514536
515
parton distribution functions and the choice of factorization scales connected with the
initial states.
The partonic cross sections pertinent to process (1) can be entirely calculated in
perturbative QCD with no additional input, except for the strong-coupling constant s .
They are known at next-to-leading order (NLO) [3,4] and even at next-to-next-to-leading
order (NNLO) [5,6]. The subsequent transition of the partons into hadrons takes place at
an energy scale of the order of 1 GeV and can, therefore, not be treated in perturbation
theory. Instead, the hadronization of the partons is described by fragmentation functions
(FFs), Dah (x, Q2 ). Their values correspond to the probability that the parton a, which
is produced at short distance, of order 1/Q, fragments into the hadron h carrying
the fraction x of the momentum of a. In the case of process (1), Q2 is typically of
order s. Given their x dependence at some scale Q20 , the evolution of the FFs with Q2
may be computed perturbatively. At present, the evolution equations are only known
through NLO. Consistency requires that we do not include the NNLO corrections to the
partonic cross sections [5,6] in our analysis of process (1) via unpolarized photons and Z
bosons.
The theoretical analysis of the fragmentation process in e+ e annihilation is of interest
for several reasons. First, it allows us to test QCD quantitatively within one experiment
observing hadrons at different values of s and to determine the strong-coupling constant,

s (Q), to be compared with determinations from other observables and/or processes.
Second, since according to the factorization theorem the FFs are independent of the
process in which they have been determined, they can be used for quantitative predictions
of inclusive single hadron cross sections in other processes, like pp,
ep, p, and
scattering. Third, the exact knowledge of the FFs, in particular for identified hadrons, may
help to elucidate the fundamental features of hadronization and to constrain models used
for the calculation of complete final states in various high-energy reactions.
After the pioneering leading-order (LO) analyses of pion, kaon [711], and charmedmeson [12] FFs in the late 70s, there had long been no progress on the theoretical side of
this field. In the mid 90s, NLO FF sets for 0 [14], , K , and mesons [15,16] have
been constructed through fits to data of e+ e annihilation, mostly generated with Monte
Carlo event generators.
In 1994/95, two of us, together with Binnewies, determined the FFs of quarks and gluons
into and K mesons at LO and NLO from genuine experimental data, adopting two
different strategies [17,18]. In the first analysis [17], we included in the fit only the data on
and K production taken by the TPC/Two-Gamma Collaboration [19] at SLAC PEP,
with energy s = 29 GeV. These data combine small statistical errors with fine binning
in x and are thus more constraining than data collected by other experiments in the energy
range from 5.2 GeV to 44 GeV. Charged-pion and charged-kaon data at other energies
were only used for cross checks. Furthermore, theoretical predictions obtained using these
FFs were compared with data on the inclusive production of unidentified charged hadrons
taken at the PEP, DESY PETRA, and KEK TRISTAN colliders making an assumption
on the p/p contribution based on information extracted from the TPC data [19]. In 1994,
when the first analysis [17] was performed, data with identified quark flavours did not exist.
516
Therefore, we had to make the assumption that the s, c, and b (d, c, and b) quarks fragment
into (K ) mesons in the same way.
In our second analysis [18], we could remove this assumption because new data on
charged-hadron production with partial flavour separation had been released by the ALEPH
Collaboration [20] at LEP1 (see also Refs. [2,21,22]). Apart from the usual data sample
without flavour separation, they also presented light- and b-quark-enriched samples. Here
and in the following, we consider u, d, and s quarks as light. In addition, we included in
our fits accurate and K data from ALEPH [23] and information on tagged three-jet
events from OPAL [24] at LEP1, which constrained the gluon FFs. In order to describe
the production of charged hadrons in terms of the and K FFs, we needed some
information on the fragmentation of quarks and gluons into protons and antiprotons. To
that end, we introduced a function f (x) which parameterizes the ratio of the p/p and
production cross sections as measured by ALEPH [23].
It is clear that this analysis [18] suffered from the lack of specific data on the
fragmentation of tagged quarks and gluons to , K , and p/p hadrons, rather than
just to charged hadrons of unidentified species. This lack has been remedied in 1998 by
the advent of new data from DELPHI [25] at LEP1 and SLD [26] at SLC. Besides other
things, SLD [26] measured the differential production cross section, as a function of the
scaled momentum x = 2p/ s, for the identified hadron species , K , and p/p in
Z-boson decays to light, c, and b quarks and in such decays without flavour separation.
Similarly, DELPHI [25] presented , K , and p/p production data, covering the full
s s , Z b b,
and
momentum range up to 45.6 GeV, discriminating between Z uu,
d d,
Z q q events. There is also new specific information that allows us to better control the
gluon FF. In fact, new gluon-tagged three-jet data of inclusive charged-hadron production
was released by ALEPH [27] and OPAL [28]. In the OPAL [28] analysis, the gluon jets
were identified using the method proposed by Gary [29]. This method is based on rare
incl in which the q- and q-quark
jets appear in the same

events of the type e+ e q qg
hemisphere of a multihadronic e+ e annihilation event. The object gincl , which is taken
to be the gluon jet, is defined by all hadrons observed in the hemisphere opposite to that
containing the q- and q-quark
jets. This method allows for the extraction of the gluon FF

with as little theoretical bias as possible. It is supposed to be superior to the earlier OPAL
[24] measurement of the gluon FF based on a jet-finding algorithm to define the gluon jets,
which entered the BKK analysis [18]. The new OPAL [28] data set comprises more points
than the previous one [24], and the energy of the gluon jets is now well defined, with a
mean value of Ejet = 40.1 GeV.
It is the purpose of this work to make full use of all the recent data from ALEPH [27],
DELPHI [25], OPAL [28], and SLD [26] in order to construct new sets of LO and NLO FFs
without imposing relations between the FFs of different flavours, except that we identify
the FFs of u and d quarks into mesons and those of u and s quarks into K mesons,
as in Refs. [17,18]. In the case of protons and antiprotons, we fix the d-quark FFs to be
half of the u-quark FFs. For completeness, we also include in our fits the older , K ,
and p/p production data without flavour separation from ALEPH [23], which also entered
the BKK analysis [18]. In order to be sensitive to the running of s , i.e., to be able to
517
fit the asymptotic scale parameter, (5) , we also include the TPC data [19], as in Refs.
MS
[17,18]. There is also precise data on charged-hadron production with flavour separation
from ALEPH [21,22], who provided light-, c-, and b-quark-enriched samples, and from
OPAL [30], who provided light- and b-quark-enriched samples. To be on the conservative
side, we exclude this data from our fits because the contribution due to charged hadrons
other than , K , and p/p is unknown and may be comparable or larger than the
experimental errors. In fact, it turns out that this data yields rather sizeable 2 values
relative to the theoretical predictions obtained from our FFs. We also test our FFs against
data from e+ e colliders with lower CM energies, namely, from DESY DORIS [31,32] and
PETRA [33,34]. Furthermore, we study the longitudinal cross section for the production of
charged hadrons on the Z-boson resonance and compare the results with data from ALEPH
[21], DELPHI [35], and OPAL [36].
This work is organized as follows. In Section 2, we summarize the data included in
our fits, outline the theoretical framework, describe our fitting procedure, present our FFs,
and discuss their goodness. Furthermore, we check our FFs against e+ e data which
are not included in our fits. In Section 3, we investigate for charged-hadron production
how well our FFs satisfy the momentum sum rule, which we did not impose on our fits.
Furthermore, we examine to what extent our FFs also account for the longitudinal cross
section of charged-hadron production on the Z-boson resonance, and how our NLO gluon
FF compares with the one of the BKK set [18] and with the one recently measured by
DELPHI [37]. Our conclusions are summarized in Section 4.
2. Analysis and results

We start by summarizing the experimental data which are used for our fits or for
comparisons. We include in our fits the , K , and p/p data with flavour separation
from DELPHI [25] and SLD [26] and those without flavour separation from ALEPH
[23], DELPHI [25], SLD [26], and TPC [19]. Furthermore, we use the charged-hadron
data with flavour separation from DELPHI [25] and those without flavour separation from
SLD [26]. In order to constrain the gluon FFs, we employ the gluon-tagged three-jet data
without hadron identification from ALEPH [27] and OPAL [28]. The charged-hadron data
without and with flavour separation from ALEPH [21,22] and OPAL [30] are only used for
comparison. All these data sets are summarized in Table 1.
The key observable of the experimental analyses and our study is the scaled-momentum
distribution normalized to the total hadronic cross section, (1/tot )d h /dx. In general,
the scaled momentum is defined as x = 2p/ s, where p is the three-momentum of the

observed hadron in the CM frame. In three-jet events, one often defines x = p/Ejet , where
Ejet is the gluon-jet energy in the CM frame. By charge-conjugation invariance, the e+ e
cross sections for + , K + , and p production should be the same as those for , K , and
p production, respectively. Therefore, one usually sums over both charges, which is also
true for the production of unidentified charged hadrons. In turn, also our FFs refer to the
sums of particles and antiparticles.
518
Table 1
2 values obtained at LO and NLO for various data samples.
CM energies, types of data, and DF
Samples not used in the fits are marked by asterisks
s [GeV]
2 in NLO
DF
Data type
29.0
(all)
K (all)
p (all)
0.64 [19]
1.86 [19]
0.79 [19]
91.2
h (all)
1.28 [25]
1.32 [26]
0.20 [25]
h (uds)
h (c)
h (b)
0.43 [25]
(all)
(uds)
(c)
(b)
1.28 [23]
K (all)
K (uds)
0.30 [23]
p
p
p
p
(all)
(uds)
(c)
(b)
0.71 [19]
1.40 [19]
0.70 [19]
48.1 [22]
14.4 [30]
89.4 [21]
80.1 [21]
221 [21]
5.10 [30]
0.51 [30]
6.63 [30]
0.58 [25]
0.72 [25]
3.09 [26]
1.87 [26]
1.36 [26]
1.00 [26]
1.65 [23]
0.44 [26]
0.65 [26]
2.11 [26]
1.21 [26]
0.32 [23]
0.79 [26]
1.29 [26]
0.92 [26]
0.97 [26]
0.80 [23]
0.57 [25]
K (c)
K (b)
0.86 [25]
0.53 [25]
0.14 [25]
0.93 [23]
2 in LO
DF
0.09 [25]
0.11 [25]
0.56 [25]
1.40 [25]
1.44 [26]
0.20 [25]
0.41 [25]
53.1 [22]
15.8 [30]
92.1 [21]
58.9 [21]
220 [21]
4.60 [30]
0.45 [30]
6.33 [30]
0.60 [25]
0.73 [25]
3.13 [26]
2.17 [26]
1.16 [26]
0.99 [26]
0.58 [25]
0.79 [25]
0.60 [25]
0.14 [25]
0.06 [25]
0.14 [25]
0.62 [25]
0.45 [26]
0.64 [26]
1.90 [26]
1.23 [26]
0.70 [26]
1.28 [26]
0.89 [26]
0.89 [26]
Ejet [GeV]
26.2
40.1
Dgh
Dgh
1.19 [27]
1.03 [28]
1.18 [27]
0.90 [28]
The LO and NLO formalisms for extracting FFs from e+ e data were comprehensively
described in Refs. [17,18] and will, therefore, not be reviewed here. The NLO formula for
d h /dx was originally obtained in Refs. [3,4] and is given in Eq. (3) of Ref. [18]. Deviating
from Refs. [17,18], we calculate tot including QCD corrections up to O(s2 ). We work in
the MS renormalization and factorization scheme and choose the renormalization scale
and the factorization scale Mf to be = Mf = s, except for gluon-tagged three-jet

events, where we put = Mf = 2Ejet . For the actual fitting procedure, we use x bins in the
interval 0.1 6 x 6 1 and integrate the theoretical functions over the bins width as is done
in the experimental analyses. The restriction at small x is introduced to exclude events in
the nonperturbative region, where mass effects and nonperturbative intrinsic-transversemomentum (kT ) effects are important and the underlying formalism is insufficient. We
parameterize the x dependence of the FFs at the starting scale Q0 as

(2)
Dah x, Q20 = Nx (1 x) .
519
As in Ref. [18], we use Q0 = 2 GeV for a = u, d, s, g, Q0 = m(c ) = 2.9788 GeV for

a = c, and Q0 = m() = 9.46037 GeV for a = b. We impose the conditions

Du x, Q20 = Dd x, Q20 ,

DuK x, Q20 = DsK x, Q20 ,

p/p
p/p
(3)
Du x, Q20 = 2Dd x, Q20 ,
which are suggested by the valence quark composition of the respective hadrons.
Equations (3) are preserved by the Q2 evolution. The constraint on the K FFs should
be slightly violated by the mass difference between the u and s quarks [7]. In fact, the
s K transition should happen more frequently than the u K one because less
energy is needed for the creation of a uu pair from the vacuum than for a s s pair. However,
we shall neglect this subtlety in our analysis because the presently available experimental
information is not rich enough to resolve such minor differences. For all the other FFs, we
treat N , , and as independent fit parameters. In addition, we take the asymptotic scale
parameter (5) , appropriate for five quark flavours, as a free parameter. Thus, we have a
MS
total of 46 independent fit parameters. The quality of the fit is measured in terms of the 2
2 , for all selected data points. Using a multidimensional
value per degree of freedom, DF
minimization algorithm [38], we search this 46-dimensional parameter space for the point
at which the deviation of the theoretical prediction from the data becomes minimal.
2 values achieved for the different data sets used in our LO and NLO fits may
The DF
2 values lie around 1 or below, indicating that the
be seen from Table 1. Most of the DF
2 values
fitted FFs describe all data sets within their respective errors. In general, the DF
come out in favour of the DELPHI [25] data. The overall goodness of the NLO (LO) fit is
2 =0.98 (0.97). In Table 1, we have also specified the 2 values for those data
given by DF
DF
sets which we have discussed above, but not included in the fits [21,22,30]. Some of these
2 values than those included in the fits, which will be
sets come out with much higher DF
discussed in detail later. The values of the parameters N , , and in Eq. (2) resulting from
the LO and NLO fits are collected in Table 2. They refer to the fragmentation of the partons
where the sum over particles and
a = u, d, s, c, b, g into the hadrons h = , K , p/p,
antiparticles is implied.
Since we included in our fits high-quality data from two very different energies, namely
29 and 91.2 GeV, we are sensitive to the running of s () and are, therefore, able to extract
(5)
= 88 +34
values of (5) . We obtain (5) = 213 +75
73 MeV at NLO and
31 MeV at LO.
MS
MS
MS
+0.006
This corresponds to s (MZ ) = 0.1170 +0.005
0.007 at NLO and s (MZ ) = 0.1181 0.007 at LO.
(5)
2 is increased
Here, the errors are determined by varying in such a way that the total DF
MS
by one unit if all the other fit parameters are kept fixed. They do not include theoretical
errors, which may be estimated through variations of the overall scale = Mf . Our NLO
values for (5) and s (MZ ) perfectly agree with those presently quoted by the Particle
MS
Data Group as world averages, (5) = 212 +25

23 MeV and s (MZ ) = 0.1185 0.002,
MS
respectively [39]. We note, however, that in Ref. [39] the value of (5) is obtained from
MS
s (MZ ) through the next-to-next-to-leading-order formula for s (), while we use the LO
520
Table 2
Values of N , , and in Eq. (2) resulting from the NLO fit. The numbers given in parentheses refer
to the LO fit
Hadron
Flavour
u=d
s
c
b
g
u=s
d
c
b
g
p/p
u = 2d
s
c
b
g
N
0.448 (0.546)
16.6 (22.3)
6.17 (8.76)
0.259 (0.311)
3.73 (6.05)
0.178 (0.259)
4.96 (5.38)
4.26 (5.18)
1.32 (1.57)
0.231 (0.0286)
1.26 (0.402)
4.01 (4.08)
0.0825 (0.111)
24.3 (40.1)
1.56 (0.740)
1.48 (1.47)
0.133 (0.127)
0.536 (0.386)
1.99 (1.93)
0.742 (0.714)
0.913 (1.02)
5.90 (6.14)
5.60 (5.62)
3.53 (3.47)
2.33 (2.92)
0.537 (0.619)
0.0556 (0.00321)
0.241 (0.178)
0.884 (0.841)
1.36 (2.94)
0.759 (0.859)
2.80 (3.08)
4.21 (4.30)
6.15 (6.01)
1.80 (2.73)
0.0712 (0.860)
0.173 (0.0974)
1.61 (1.54)
0.579 (0.742)
0.0157 (0.770)
4.13 (2.80)
5.21 (4.99)
2.01 (2.21)
12.1 (12.4)
3.58 (7.69)
and NLO formulas, depending on the order of our analysis, as is required by consistency.
(5)
Since the errors on s () and quoted in Ref. [39] result from an average of different
MS
kinds of measurements, they are considerably smaller than those obtained here from a
single type of experiment. In this context, we remark that, throughout our analysis, we
assume nonperturbative power corrections proportional to 1/Q to be negligible in the
energy range relevant for our fits.
The goodness of our LO and NLO fits to the ALEPH [23,27], DELPHI [25], OPAL
[28], and SLD [26] data may be judged from Figs. 15. In Figs. 14, we study the cross
section (1/tot)d h /dx of charged-hadron, , K , and p/p production, respectively, at
s = 91.2 GeV as a function of x. In Fig. 1, the contributions from all quark flavours
and the gluon are included, while Figs. 24 refer to the fragmentation of light, c, and b
quarks, respectively. As in the experimental analyses [21,22,25,26,30], we have to arrange
for the latter three contributions to add up to the first one. Therefore, we distribute the
gluon contribution among the various quark contributions according to their electroweak
coupling strengths to the neutral current. In this way, the gluon radiation off a quark line of
a given flavour is accounted towards the contribution of this quark flavour. The theoretical
results are compared with the respective data from ALEPH [23] in Fig. 1, with that from
DELPHI [25] in Figs. 1, 2, and 4, and with that from SLD [26] in Figs. 14. We observe
that, in all cases, the various data are mutually consistent with each other and are nicely
2 values
described by the LO and NLO fits, which is also reflected in the relatively small DF
given in Table 1. The LO and NLO fits are almost indistinguishable in those regions of x,
where the data have small errors. For large x, where the statistical errors are large, the LO
and NLO results sometimes moderately deviate from each other.
521
Fig. 1. Normalized differential cross section of inclusive hadron production at s = 91.2 GeV as a
function of x. The LO (dashed lines) and NLO (solid lines) fit results are compared with data from
ALEPH [23] (triangles), DELPHI [25] (circles), and SLD [26] (squares). The upmost, second, third,
and lowest curves refer to charged hadrons, , K , and p/p,
respectively. Each pair of curves is
rescaled relative to the nearest upper one by a factor of 1/5.
In Fig. 5, we compare the ALEPH [27] and OPAL [28] measurements of the gluon FF
in charged-hadron production, which comprises five and twelve data points, respectively,
with our LO and NLO fit results. The scale choice is = Mf = 2Ejet , with Ejet = 26.2
2 values of order
[27] and 40.1 GeV [28], respectively. The data are nicely fitted, with DF
unity, as may bee seen from Table 1. By the same token, this implies that the data are
mutually consistent.
As may be seen from Table 1, the ALEPH [21,22] and OPAL [30] data on chargedhadron production with and without flavour separation, which are excluded from our fits,
2 values relative to the theoretical predictions based on our FFs.
lead to rather sizeable DF
They are obviously inconsistent with the DELPHI [25] and SLD [25] data included in
our fits. In particular, the flavour-enriched charged-hadron samples from ALEPH [21] give
522
Fig. 2. Same as in Fig. 1, but for the light-quark-enriched samples from DELPHI [25] (circles) and
SLD [26] (squares).
2 values of order 100. The comparison of these ALEPH [21,22] and OPAL [30] data
DF
with our LO and NLO predictions is also illustrated in Fig. 6. We should point out that the
ALEPH [2123] data possess the following property. The sum of the three flavour-tagged
cross sections of charged-hadron production [21], which agrees with the charged-hadron
cross section without flavour separation quoted in Ref. [22], appreciably overshoots the
sum of the , K , and p/p cross sections without flavour separation, which stem from
an independent measurement [23]. The latter three data sets [23] are included in our fits
2 values, as may be seen from Table 1. We remark that
and lead to quite acceptable DF
the charged-hadron data with flavour separation from ALEPH [21] served as one of the
main inputs for the BKK analysis [18]. We attribute this difference to the contribution from
charged hadrons other than , K , and p/p hadrons, which may be comparable or larger
than the relatively small errors on the charged-hadron data [21,22]. We assume that this
interpretation is also valid for the OPAL data [30]. On the contrary, the DELPHI [25] and
SLD [26] analyses are based on the assumption that the sum of the , K , and p/p data
523
Fig. 3. Same as in Fig. 1, but for the c-quark-enriched samples from SLD [26] (squares). The last
two points on the right belong to the charged-hadron sample.
exhaust the full charged-hadron data. In want of a separate measurement of the excess in
cross section due to high-mass charged hadrons, we decided to proceed as in Refs. [25,26].
We now report on comparisons with pre-LEP/SLC data on , K , and pp production
which we did not include in our fits. Specifically, we consider the and K data from
DASP [31] ( s = 5.2 GeV) and ARGUS [32] ( s = 9.98 GeV) at DESY DORIS and
from TASSO [34] ( s = 34 GeV) at PETRA, and the p/p data taken by TASSO at s =
22 [33] and 34 GeV [34]. There is no separation into flavour-enriched or three-jet samples
2 values are summarized in Table 3. As in the fits, they
for this data. The resulting DF
are evaluated only taking into account the data points with x > 0.1. The comparison of
the , K , and p/p data with our LO and NLO predictions is visualized in Figs. 7
9, respectively. For reference, also the corresponding TPC [19] and SLC [26] data are
included in these comparisons. From Table 3 and Figs. 79, we learn that the DASP [31],
ARGUS [32], and TASSO [33,34] data agree quite well with our LO and NLO predictions.
This concludes the description of our fitting procedure and of the quality of the fits.
524
Fig. 4. Same as in Fig. 1, but for the b-quark-enriched samples from DELPHI [25] (circles) and SLD
[26] (squares).
Table 3
2 values obtained at LO and NLO for various pre-LEP/SLC data
samples not included in our fits
s [GeV]
Data type
2 in NLO
DF
2 in LO
DF
5.4
(all)
K (all)
3.10 [31]
1.80 [31]
3.02 [31]
2.33 [31]
9.98
(all)
K (all)
3.27 [32]
3.21 [32]
2.78 [32]
2.81 [32]
22.0
p (all)
1.29 [33]
1.50 [33]
34.0
(all)
K (all)
p (all)
0.80 [34]
0.31 [34]
0.58 [34]
0.88 [34]
0.37 [34]
0.46 [34]
525
Fig. 5. Gluon FF for charged-hadron production as a function of x at Mf = 52.4 and 80.2 GeV.
The LO (dashed lines) and NLO (solid lines) predictions are compared with three-jet data from
ALEPH [27], with Ejet = 26.2 GeV (upper curves) and from OPAL [28], with Ejet = 40.1 GeV
(lower curves). The OPAL data and the pertinent predictions are rescaled by a factor of 1/100.
3. Applications
We now investigate if our FFs satisfy the momentum sum rules. Since a given parton a
fragments with 100% likelihood into some hadron h and momentum is conserved during
the fragmentation process, we have
Z1
dx xDah (x, Q2 ) = 1,
(4)
for any value of Q2 . As is well established experimentally, even for the flavour-enriched
data samples [25,26], the sum of the , K , and p/p production cross sections agrees,
to good approximation, with the one of charged-hadron production. Thus, insertion of our
, K , and p/p FFs in Eq. (4) exhausts the charged-hadron contribution. As for the
526
Fig.
6. Normalized differential cross section of inclusive charged-hadron production at
s = 91.2 GeV as a function of x. The LO (dashed lines) and NLO (solid lines) fit results are compared with data from ALEPH [21,22] (triangles) and OPAL [30] (circles). The upmost, second, third,
and lowest curves refer to the full, light-quark-enriched, c-quark-enriched, and b-quark-enriched
samples, respectively. Each pair of curves is rescaled relative to the nearest upper one by a factor of
1/5.
neutral-hadron FFs, which also enter Eq. (4), we make the following assumptions which
are suggested by SU(2) flavour symmetry. As in Ref. [18], we approximate
1
0
Da (x, Q2 ) = Da (x, Q2 ),
2
K 0 /K 0
K
(x, Q2 ).
Du,d,s,c,b,g (x, Q2 ) = Dd,u,s,c,b,g
(5)
Furthermore, we assume that

1 p/p
n/n
Du (x, Q2 ) = Du (x, Q2 ),
2
n/n
p/p
Ds,c,b,g (x, Q2 ) = Ds,c,b,g (x, Q2 ).
n/n
p/p
Dd (x, Q2 ) = 2Dd
(x, Q2 ),
(6)
527
Fig.
7. Normalized differential cross section of inclusive production as a function of x at
s = 5.2, 9.98, 29, 34, and 91.2 GeV. The LO (dashed lines) and NLO (solid lines) predictions
are compared with data from DASP [31], ARGUS [32], TPC [19], TASSO [34], and SLD [26].
Upper curves correspond to lower energies. Each pair of curves is rescaled relative to the nearest
upper one by a factor of 1/10.
K 0 /K 0
Here, Da (x, Q2 ), Da
(x, Q2 ), etc., refer to the sums of the FFs for the individual
hadron species. Equations (5) are supported by e+ e data of pion [2,40] and kaon [41
44] production, except for very small x. To be on the conservative side, we take the lower
limit of integration in Eq. (4) to be 0.05 rather than 0, so that the FF parameterizations
are not used too far outside the x range which was selected for the fits. This particular
choice may be motivated by observing that the , K , and p/p data sets in Figs. 79
are well described by the theoretical predictions down to x values of order 0.05. Since we
leave out the x range below 0.05, our results for the left-hand side of Eq. (4) are expected
to be somewhat below unity. A more sophisticated approach [45] would be to improve
the description of the low-x region by modifying our ansatz for the FFs according to the
so-called modified leading logarithmic approximation (MLLA) [46].
528
Fig. 8. Same as in Fig. 7, but for K mesons.
In Table 4, we list the results obtained with our LO and NLO FFs for Q = 2, 4,
10, 91, and 200 GeV. The accuracy of our results is limited by the uncertainties in our
assumptions (5) and (6), in particular for the lower values of Q. In the case of the d quark,
we find values slightly larger than unity at small values of Q, both in LO and NLO. On the
other hand, the s-quark values are considerably smaller than unity, in particular in NLO.
The gluon FF gives values appreciably larger than unity at small values of Q, both in LO
and NLO. We believe that the imbalance between the d- and s-quark results is due to our
limited knowledge of the relative magnitude of the valence- and light-sea-quark contents
of the , K , and p/p hadrons. Since the u, d, and s quarks are always combined in the
available data, there are no individual constraints on their FFs. In order to trace the source
to the left-hand side of

of this imbalance,
let us consider the and K contributions
Eq. (4) at Q0 = 2 GeV for a = u, d, s. In the case of , we have
Z1
0.05
dx xDu,d
x, Q20
Z1
= 0.41,
dx xDs
0.05

x, Q20 = 0.23,
(7)
529
Fig.
9. Normalized differential cross section of inclusive p/p production as a function of x at
s = 22, 29, 34, and 91.2 GeV. The LO
(dashed lines) and NLO (solid lines) predictions are
compared with data from TASSO (with s = 22 [33] and 34 GeV [34]), TPC [19], and SLD [26].
Upper curves correspond to lower energies. Each pair of curves is rescaled relative to the nearest
upper one by a factor of 1/10.
which nicely exhibits the valence-like character of the u and d quarks in contrast to the
sea-like character of the s quark. In the case of K , however, we find
Z1

K
dx xDu,s
x, Q20 = 0.19,
0.05
Z1
dx xDdK

x, Q20 = 0.25,
(8)
0.05
which shows that the d quark does not behave sea-like, contrary to expectations. We have
thus identified the non-sea-like behaviour of the d quark in the K mesons as the source
of the violation of the sum rule in Table 4.
530
Table 4
Left-hand side of Eq. (4) at NLO for a = u, d, s, c, b, g and Q = 2, 4, 10, 91, 200 GeV. We sum
over h = , 0 , K , K 0 , K 0 , p, p,
n, n and integrate over 0.05 6 x 6 1. The numbers given in
parentheses are evaluated with our LO set
a
u
d
s
c
b
g
Q [GeV]
10
91
200
0.96 (1.13)
1.05 (1.14)
0.60 (0.82)
1.33 (1.85)
0.96 (1.06)
1.05 (1.07)
0.68 (0.82)
0.91 (0.97)
1.15 (1.32)
0.92 (1.00)
1.00 (1.01)
0.67 (0.78)
0.88 (0.93)
0.73 (0.80)
1.00 (1.10)
0.82 (0.87)
0.89 (0.88)
0.62 (0.69)
0.78 (0.81)
0.65 (0.69)
0.73 (0.74)
0.78 (0.83)
0.85 (0.84)
0.60 (0.67)
0.75 (0.78)
0.62 (0.66)
0.67 (0.67)
The analysis of the the longitudinal cross section offers a unique opportunity to test
the gluon FF. The total cross section of inclusive hadron (h) production, which we have
considered so far, may be decomposed into a transverse (T ) and a longitudinal (L)
component, d h /dx = dTh /dx + dLh /dx, as indicated in Eq. (3) of Ref. [18]. As is well
known, at O(s0 ), only transversely polarized photons and Z bosons contribute to the cross
section. The longitudinal cross section dLh /dx first appears at O(s ). Thus, one needs to
know it through O(s2 ) in order to test the gluon FF in NLO. The O(s2 ) correction to
dLh /dx has been calculated in Refs. [5,6,47]. In the case of inclusive hadron production,
dLh /dx is given by a convolution of the longitudinal partonic cross sections with the
corresponding FFs and has the following form [47]:
!

Z1 ( X
NF
6
1 h x
dz
1 dL,q
1 dLh
=
, Mf2
Qqi (s)
D6
tot dx
z
NF
z
tot dz
i=1

x
x
1 dL,g
, Mf2
+ Dgh
z
tot dz

N
NS
F
X
x
1 dL,q
h
, Mf2
Qqi (s)D(+),i
+
z
tot dz
i=1
)

NF
F
X
d
x
1
x
1
L,q
h
, Mf2 +
, Mf2
. (9)
QFqi (s) D(+),i
Dh
+
z
NF 6 z
tot dz
i=1
Here, NC = 3 and NF = 5 denote the numbers of colours and active quark flavours,
respectively, and Qqi (s) and QFqi (s) represent the effective electroweak couplings of the
quarks to the photon and Z boson. The latter are defined as
Qqi (s) = ee2 eq2i + 2ee ve eqi vqi
s(s MZ2 )
(s MZ2 )2 + MZ2 Z2

+ ve2 + ae2 vq2i + aq2i
s2

2 2
s MZ
+ MZ2 Z2
QFqi (s) =
vq2i
+ aq2i
NF
X

aqj
aqi
j =1
s(s MZ2 )
(s MZ2 )2 + MZ2 Z2
531
(10)
where MZ and Z are the mass and width of the Z boson, respectively, ef is the electric
charge of the fermion f in units of the positron charge,
T3,f 2ef sin2 w
T3,f
and af =
2 sin w cos w
2 sin w cos w
are its vector and axial-vector couplings to the Z boson, respectively, T3,f is the third
component of isospin of its left-handed component, and w is the electroweak mixing
angle. In the limit s MZ2 , only the QED contribution survives, and we have Qqi ee2 eq2i
and QFqi 0. The labels 6 and NS in Eq. (9) indicate the flavour singlet and flavour
non-singlet contributions, which are related to the FF combinations
vf =
h
(x, Mf2 ) =
D6
NF
X

h
Dqi (x, Mf2 ) + Dqhi (x, Mf2 ) ,
i=1
1 h
D (x, Mf2 ).
(11)
NF 6
The contributions proportional to the colour factor CF TR = 4/8 are labeled F , which
is to suggest that they vanish in QED due to Furrys theorem. They only arise from Zboson exchange, provided one does not sum over all quark flavours belonging to one
fermion generation. Furthermore, they are devoid of collinear singularities, so that mass
factorization is not needed. We neglect these contributions, since they amount to less than
6 /dz, d 6 /dz, and
1% over the whole x range [47]. The partonic cross sections dL,q
L,g
NS
2
dL,q /dz through O(s ) may be found in Refs. [5,6,47].
ALEPH [21], DELPHI [35], and OPAL [36] have measured dLh /dx for charged-hadron
production as a function of x without quark flavour separation. In addition, DELPHI [35]
has also presented light- and b-quark-enriched samples. In Fig. 10, these data are compared
with the LO and NLO evaluations of Eq. (9) with our FFs. As in our fits to data of d h /dx,
we choose the scales to be equal to the CM energy s = 91.2 GeV, where the data have
been taken. Although only data with x > 0.1 has been included in our fits, dLh /dx is also
2 values, evaluated from the data points with
well described down to x = 0.01. The DF
2 values for the LO predictions are in average
x > 0.1, are collected in Table 5. The DF
more than a factor of two larger than those for the NLO predictions. This is not surprising,
since it was already found in Ref. [18] that the LO result falls short of the data by a factor
of two. At that time, the common scale had to be reduced to 20 GeV in order to obtain
agreement. We observe from Fig. 10 and Table 5 that this is no longer necessary for the
NLO results. Obviously, the O(s2 ) corrections [5,6,47] are sufficient to produce a K factor
that brings our predictions into good agreement with the data. Good agreement between the
NLO prediction of dLh /dx and the ALEPH [21] and OPAL [36] data was already found
by Rijken and van Neerven [5,6], who employed the BKK gluon FF [17].
The comparisons presented in Fig. 10 provide us with a very useful check of the gluon
FF in the low-x range, where the data have small errors. In particular, the good agreement
h
(x, Mf2 ) = Dqhi (x, Mf2 ) + Dqhi (x, Mf2 )
D(+),i
532
Fig. 10. Normalized

differential longitudinal cross section of inclusive charged-hadron production
as a function of x at s = 91.2 GeV. The LO (dashed lines) and NLO (solid lines) predictions are
compared with data from ALEPH [21], OPAL [36], and DELPHI [35] without flavour separation
and with light- and b-quark-enriched samples from DELPHI [35] (in this order from top to bottom).
Each pair of curves is rescaled relative to the nearest upper one by a factor of 1/30.
Table 5
2 values obtained at LO and NLO for the longitudinal cross
section
s [GeV]
91.2
Data type
2 in NLO
DF
2 in LO
DF
Lh (all)
Lh (uds)
Lh (b)
1.36 [21] 1.74 [35] 0.49 [36]

7.98 [35]
0.51 [35]
11.0 [21] 1.27 [35] 7.94 [36]

1.05 [35]
0.80 [35]
533
Fig. 11. Gluon FFs for charged-hadron production as functions of x at Mf = 10, 52.4, 80.2, and
200 GeV from the DELPHI [28] (dotted lines), BKK NLO [18] (dashed lines), and new NLO
(solid lines) sets. The predictions for Mf = 52.4 and 80.2 GeV are compared with the gluon-tagged
three-jet data from ALEPH [27] and OPAL [28], respectively. Upper curves correspond to lower
energies. Each set of curves is rescaled relative to the nearest upper one by a factor of 1/50.
in the case of flavour separation nicely confirms the relative importance of the light- and
2 values obtained at NLO for the
b-quark FFs relative to the gluon FF. Since the DF
longitudinal cross section are comparable to those achieved in our NLO fit, we do not
expect that the inclusion of the information on the longitudinal cross section in our NLO
fit would significantly change the outcome.
As is well known, the gluon FF only enters the prediction for d h /dx at O(s ), while
at O(s0 ) it only contributes indirectly via the Q2 evolution. Therefore, we have used the
ALEPH [27] and OPAL [28] data on gluon-tagged three-jet events in order to constrain
it. These data were already compared with the fitted gluon FF in Fig. 5. It is of interest to
compare this gluon FF with the BKK [18] one and the one recently extracted by DELPHI
[37] from their three-jet events for scales in the range 6.5 6 Mf 6 28 GeV. In Ref. [37], a
parameterization, valid for this Mf range and for 0.15 6 x 6 0.75, was generated adopting
534
an ansatz similar to the one introduced in Ref. [18]. In Fig. 11, the x dependences of these
three gluon FFs are studied for Mf = 10, 52.4, 80.2, and 200 GeV. The results for Mf =
52.4 and 80.2 GeV are compared with the ALEPH [27] and OPAL [28] gluon-tagged
three-jet data, which are included in our fits. From Fig. 11 the following observations
can be made. Our new gluon FF is rather similar to the BKK one, especially for larger
values of Mf . At Mf = 80.2 GeV, the DELPHI gluon FF is somewhat steeper than the
new and BKK ones, except for large x values, and it agrees slightly better with the OPAL
[28] data. On the other hand, our new gluon FF agrees best with the ALEPH [27] data at
Mf = 52.4 GeV. At Mf = 10 GeV, the DELPHI gluon FF has a flatter x dependence than
the other ones and overshoots them, while at Mf = 200 GeV the situation is similar to the
case of Mf = 80.2 GeV.
4. Conclusions
The LEP1 and SLC experiments have provided us with a wealth of high-precision
experimental information on how partons fragment into low-mass charged hadrons [2,
2028,30,3537,40]. The data partly come as light-, c-, and b-quark-enriched samples
[25,26] or as
without [21,22,30] or with identified final-state hadrons ( , K , and p/p)
gluon-tagged three-jet samples without hadron identification [27,28,37]. This new situation
motivated us to update, refine, and extend the BKK analysis [18] by generating new LO and
NLO sets of , K , and p/p FFs. The x distributions of the resulting FFs at their starting
scales Q0 are given by Eq. (2) with the parameters listed in Table 2. The Q2 evolution is
determined by the timelike AltarelliParisi equations in the respective orders, LO or NLO,
which are summarized in the Appendix of Ref. [48]. 1 The evolution procedure in Mellin
space is described in Ref. [18]. A FORTRAN subroutine which returns the values of the
Dah (x, Q2 ) functions for given values of x and Q2 may be downloaded from the URL
http://www.desy.de/poetter/kkp.html or obtained upon request from the authors.
By also including in our fits , K , and p/p data (without flavour separation) from
PEP [19], with CM energy s = 29 GeV, we obtained a handle on the scaling violation
in the fragmentation process, which allowed us to extract LO and NLO values of s (MZ ).
+0.005
We found s (MZ ) = 0.1181 +0.006
0.007 at LO and s (MZ ) = 0.1170 0.007 at NLO, where the
errors are experimental. These results are in perfect agreement with what the Particle Data
Group currently quotes as the world average, s (MZ ) = 0.1185 0.002 [39].
Our strategy was to only include in our fits LEP1 and SLC data with both flavour
separation and hadron identification [25,26], gluon-tagged three-jet samples with a fixed
gluon-jet energy [27,28], and the , K , and p/p data sets from the pre-LEP1/SLC
era with the highest statistics and the finest binning in x [19]. Other data served us for
cross checks. In particular, we probed the scaling violation through comparisons with ,
K , and p/p data from DORIS and PETRA, with CM energies between 5.4 and 34 GeV
[3134]. Furthermore, we tested the gluon FF, which enters the unpolarized cross section
1 There is an obvious typographical error in the published version of Ref. [48], which was absent in the preprint
version thereof. In the line before the last of Eq. (17), ln ln(1 x) should be replaced by ln(1 x).
535
only at NLO, by comparing our predictions for the longitudinal cross section, where it
already enters at LO, with available data [21,35,36]. Finally, we directly compared our
gluon FF with the one recently measured by DELPHI in three-jet production with gluon
identification as a function of x at various scales Q2 [37]. All these comparisons led
to rather encouraging results. We also verified that our FFs satisfy reasonably well the
momentum sum rules, which we did not impose as constraints on our fits. All these cross
checks make us believe that our FFs should allow for a reliable description of inclusive
charged-hadron production in all kinds of high-energy-collision experiments over wide
ranges of x and Q2 . The very-high-Q2 regime will be accessible with the CERN LHC and
a future e+ e linear collider. Our FFs are bound to break down at very low values of x,
where it becomes necessary to modify the ansatz (2) according to the MLLA [46]. This is
beyond the scope of the present analysis, but should eventually be investigated [45].
Note added in proof

After the submission of this paper, a preprint [49] appeared which also presents NLO
FFs for , K , and charged hadrons. These FFs are fitted to the ALEPH [21,22], SLC
[26], and TPC [19] data assuming certain power laws in the light-quark sector.
Acknowledgements
The II. Institut fr Theoretische Physik is supported in part by the Deutsche Forschungsgemeinschaft through Grant No. KN 365/1-1, by the Bundesministerium fr Bildung und
Forschung through Grant No. 05 HT9GUA 3, and by the European Commission through
the Research Training Network Quantum Chromodynamics and the Deep Structure of Elementary Particles under Contract No. ERBFMRX-CT98-0194.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
P. Mttig, Phys. Rep. 177 (1989) 141.

ALEPH Collaboration, R. Barate et al., Phys. Rep. 294 (1998) 1.
G. Altarelli, R.K. Ellis, G. Martinelli, S.-Y. Pi, Nucl. Phys. B 160 (1979) 301.
R. Baier, K. Fey, Z. Phys. C 2 (1979) 339.
P.J. Rijken, W.L. van Neerven, Phys. Lett. B 386 (1996) 422; Phys. Lett. B 392 (1997) 207;
Nucl. Phys. B 487 (1997) 233.
P.J. Rijken, PhD Thesis, Rijksuniversiteit te Leiden, March 1997.
R.D. Field, R.P. Feynman, Phys. Rev. D 15 (1977) 2590; Nucl. Phys. B 136 (1998) 1.
V. Barger, T. Gottschalk, R.J.N. Phillips, Phys. Lett. B 70 (1977) 51.
J.F. Owens, E. Reya, M. Glck, Phys. Rev. D 18 (1978) 1501.
R. Baier, J. Engels, B. Petersson, Z. Phys. C 2 (1979) 265.
M. Anselmino, P. Kroll, E. Leader, Z. Phys. C 18 (1983) 307.
L.M. Sehgal, P.M. Zerwas, Nucl. Phys. B 108 (1976) 483.
C. Peterson, D. Schlatter, I. Schmitt, P.M. Zerwas, Phys. Rev. D 27 (1983) 105.
P. Chiappetta, M. Greco, J.-Ph. Guillet, S. Rolli, M. Werlen, Nucl. Phys. B 412 (1994) 3.
536
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
M. Greco, S. Rolli, Z. Phys. C 60 (1993) 169; Phys. Rev. D 52 (1995) 3853.

M. Greco, S. Rolli, A. Vicini, Z. Phys. C 65 (1995) 277.
J. Binnewies, B.A. Kniehl, G. Kramer, Z. Phys. C 65 (1995) 471.
J. Binnewies, B.A. Kniehl, G. Kramer, Phys. Rev. D 52 (1995) 4947.
TPC/Two-Gamma Collaboration, H. Aihara et al., LBL-23737 and UC-34D, March 1988; Phys.
Rev. Lett. 61 (1988) 1263.
G.D. Cowan, in: P.J. Bussey, I.G. Knowles (Eds.), Proceedings of the XXVII International
Conference on High Energy Physics, Glasgow, Scotland, UK, 20-27 July 1994, Institute of
Physics Publishing, Bristol, 1995, p. 883.
ALEPH Collaboration, D. Buskulic et al., Phys. Lett. B 357 (1995) 487; Phys. Lett. B 364
(1995) 247 (Erratum).
C. Padilla Aranda, PhD Thesis, Universitat Autnoma de Barcelona, September 1995.
ALEPH Collaboration, D. Buskulic et al., Z. Phys. C 66 (1995) 355.
OPAL Collaboration, P.D. Acton et al., Z. Phys. C 58 (1993) 387.
DELPHI Collaboration, P. Abreu et al., Eur. Phys. J. C 5 (1998) 585.
SLD Collaboration, K. Abe et al., Phys. Rev. D 59 (1999) 052001.
ALEPH Collaboration, R. Barate et al., CERN-EP/98-16, February 1998, submitted to Eur.
Phys. J.C.
OPAL Collaboration, G. Abbiendi et al., Eur. Phys. J. C 11 (1999) 217.
J.W. Gary, Phys. Rev. D 49 (1994) 4503.
OPAL Collaboration, K. Ackerstaff et al., Eur. Phys. J. C 7 (1999) 369.
DASP Collaboration, R. Brandelik et al., Nucl. Phys. B 148 (1979) 189.
ARGUS Collaboration, H. Albrecht et al., Z. Phys. C 44 (1989) 547.
TASSO Collaboration, M. Althoff et al., Z. Phys. C 17 (1983) 5.
TASSO Collaboration, W. Braunschweig et al., Z. Phys. C 42 (1989) 189.
DELPHI Collaboration, P. Abreu et al., Eur. Phys. J. C 6 (1999) 19.
OPAL Collaboration, R. Akers et al., Z. Phys. C 68 (1995) 203.
DELPHI Collaboration, Eur. Phys. J. C 13 (2000) 573.
F. James, M. Roos, Comput. Phys. Commun. 10 (1975) 343.
Particle Data Group, C. Caso et al., Eur. Phys. J. C 3 (1998) 1, and 1999 off-year partial update
for the 2000 edition available on the PDG WWW pages (http://pdg.lbl.gov/).
DELPHI Collaboration, W. Adam et al., Z. Phys. C 69 (1996) 561.
ARGUS Collaboration, H. Albrecht et al., Z. Phys. C 44 (1989) 547.
TASSO Collaboration, W. Braunschweig et al., Z. Phys. C 47 (1990) 167.
ALEPH Collaboration, D. Buskulic et al., Z. Phys. C 64 (1994) 361.
OPAL Collaboration, R. Akers et al., Z. Phys. C 67 (1995) 389.
B.A. Kniehl, G. Kramer, W. Ochs, B. Ptter, in preparation.
V.A. Khoze, W. Ochs, Int. J. Mod. Phys. A 12 (1997) 2949.
J. Binnewies, PhD Thesis, Universitt Hamburg, July 1997, DESY 97-128; hep-ph/9707269.
J. Binnewies, B.A. Kniehl, G. Kramer, Z. Phys. C 76 (1997) 677.
S. Kretzer, DO-TH 00/04; hep-ph/0003177.

Nuclear enhanced higher-twist effects

in the DrellYan process
R.J. Fries a , A. Schfer a , E. Stein a , B. Mller b
a Institut fr Theoretische Physik, Universitt Regensburg, D-93040 Regensburg, Germany
b Department of Physics, Duke University, Box 90305, Durham, NC 27708-0305, USA
Abstract
We calculate the DrellYan cross section, resolving the full kinematics of the lepton pair, at high
transverse momentum for hadron nucleus collisions. We use the general framework of Luo, Qiu
and Sterman to calculate double scattering contributions that are of twist-4 and demonstrate their
nuclear enhancement. By comparing single and double scattering at RHIC energies we find that
double scattering gives contributions of comparable size. We also show that the angular dependence
of the DrellYan pair discriminates between the various double scattering contributions. 2000
PACS: 25.75.-q; 12.38.Bx; 13.85.Qk
Keywords: QCD; Structure functions; Power corrections; Nuclear effects
1. Introduction
In recent times the interest in understanding hard processes involving nuclei, like
electron nucleus collisions (e + A) and hadron nucleus collisions (h + A), has rapidly
increased. The relativistic heavy ion collider (RHIC) will provide a new tool for
experimental studies of h + A and A + A physics. This experimental advance provides
a challenge to theorists for developing a description of such reactions strictly in terms of
the underlying fundamental field theory, i.e., quantum chromodynamics (QCD).
The usual way to proceed in QCD is to use factorization theorems [1] that permit
a rigorous treatment of hard processes. In this approach, incalculable soft contributions
are factorized into universal, process independent distribution functions. Measurement of
the distribution functions in one process then allows the prediction of cross sections for
other processes. An example is the DrellYan (DY) process where two hadrons scatter to
produce a lepton pair of large invariant mass Q2 . If Q2 is large enough the factorization
theorems state, that the inclusive DrellYan cross section can be expressed as the product
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 9 1 - 1
538
R.J. Fries et al. / Nuclear Physics B 582 (2000) 537570
of a calculable hard part and universal twist-2 parton distribution functions that can
be measured, for instance in deeply inelastic lepton nucleon scattering. Such universal
distribution functions have been successfully extracted from existing data. The natural
question arises how this picture is changed for hadronnucleus collisions. Differences
between the twist-2 distribution of a single nucleon and those of a nucleus of mass number
A > 1 were first discovered by the European Muon Collaboration. This phenomenon is
known as the EMC effect at large values of the Bjorken scaling variable x and as shadowing
at lower values of x [2].
However, strong interaction dynamics cannot solely be described in terms of twist-2
structure functions. These functions are simple one-particle correlators and have the interpretation of a probability to find a parton in the hadron with certain momentum fraction x.
These distribution functions are measured when the hard probe scatters off a single parton
in the nucleon. Of course, the probe may scatter off more than one parton in the target. Such
processes are related to multi-parton correlators and describe corrections to the singleparton scattering process. In general they are suppressed by powers of the large scale, and
are called contributions of higher twist.
If we consider processes involving large nuclei it is quite obvious that multiple scattering
will be enhanced and even dominate in the limit A . The classical example for
enhancement of multiple scattering in h + A processes is the Cronin effect, i.e., the
anomalous A-dependence of the DrellYan cross section at large values of the transverse
momentum of the lepton pair [3,4]. Instead of a scaling of the cross section with the volume
A as expected in the twist-2 case, one observes an A-dependence parametrized by
h+A = A h+p with 4/3 [5]. This is a clear signal for multiple scattering and shows
that higher-twist multiple scattering is grossly enhanced at large momentum transfer.
The literature contains a wide variety of approaches dealing with multiple scattering, see
e.g., [6,7], but a rigorous formulation in the language of perturbative QCD was attempted
only recently. In a series of papers, Luo, Qiu and Sterman (LQS) and recently also Guo,
showed how to extend the factorization theorem beyond leading twist to get a handle on
double scattering in terms of perturbative QCD [8,9]. Their twist-4 matrix elements are
correlators of two partons in the nucleus and can describe scattering off partons from
different nucleons. This approach predicts a stronger dependence on the size of the nucleus
than the linear scaling with A of single scattering. LQS argue that due to the geometrical
structure of the matrix elements some of these twist-4 nuclear matrix elements pick up
an additional factor A1/3 , compensating for some of the inherent suppression of higher
twist corrections. This nuclear enhancement is able to describe the observed scaling with
A4/3 in certain kinematical regions.
In this paper we discuss nuclear enhanced double scattering contributions to DY pair
production off nuclear targets
h(P2 ) + A(P1 A) l + l + X
(1)
in the framework of the LQS approach. It was already shown some time ago by Guo [10,
11] that higher-twist effects in the process (1) are enhanced at large transverse momentum.
Extending the calculations of Guo [10], we have calculated the full kinematical dependence
(a)
539
(b)
Fig. 1. Single- (a) and double-scattering (b) schematically for a hadron h colliding with a nucleus A
producing a + pair. One additional unobserved parton is radiated in the leading contribution to
high transverse momentum q .
2 dy d including the angular distribution of the lepton
of the cross section d/dQ2 dq
pair and evaluated it, in particular, for p + A scattering at RHIC. Our most important
results were already published as a letter [12]. In this contribution we present the complete
analytical results together with a brief review of the LQS approach and some technical
details. We also give some numerical results not contained in [12] and an extended
discussion. We will argue that nuclear enhanced double scattering corrections cannot be
neglected at RHIC.
In Fig. 1 we show typical Feynman diagrams contributing to the DY process. Single
scattering means that, e.g., a quark from the single hadron and an antiquark from the
nucleus annihilate into a virtual photon in a hard process. To obtain large transverse
momentum an additional unobserved parton, here a gluon, has always to be radiated.
Double scattering in our example implies that the quark can pick up an additional gluon
from the nucleus. The generalized twist-4 factorization theorem of LQS states that two
distinct scattering reactions occur for double scattering. One is referred to as doublehard (DH) scattering. Here the parton from the single hadron undergoes two subsequent
independent hard scattering reactions in the target. The other one is called softhard (SH)
scattering. Here the parton from the single hadron interacts with the collective soft gluon
field in the nucleus (it can, e.g., be thought of as feeling the color-magnetic Lorentz
force) before producing the virtual photon via the final hard scattering. At large transverse
momentum interference of these two mechanisms can be neglected.
The double-hard reaction resembles the classical double scattering picture, while soft
hard scattering may have an analogy in plasma physics, where a hard scattering event can
be accompanied by initial- or final-state interaction with the mean field. The first process
might be accounted for by event generators in simulations of heavy-ion reactions whereas
the implementation of softhard scattering has yet to be formulated in a fully consistent
manner (see [13] for a conceptual discussion).
It would be helpful to find observables with vanishing or at least very small
contributions from single scattering. The measurement of such observables would permit
the identification of twist-4 effects at larger values of Q2 where perturbation theory is more
reliable and thus provide a clearer evidence for the occurrence of double scattering. We find
these observables in the angular distribution of DY pairs, namely the helicity amplitude W
and the LamTung relation W = WL /2 [14], both introduced in Section 2. The Lam
540
Tung relation is analogous to the CallanGross relation in deep-inelastic electronnucleon

scattering (DIS). Furthermore it would be helpful to find observables where double-hard
and softhard contributions can be disentangled. These can also be found in the angular
distribution of DY pairs. While SH scattering leads to non-trivial angular patterns, DH
scattering does not. Any anomalous A-dependent angular distribution therefore would be
a clear sign of non trivial twist-4 correlations.
Although we will focus on h + A collisions, our work has been primarily motivated
by the fact that higher-twist effects or multiple scattering should be most prominent in
nucleusnucleus (A + A) collisions at high energy, as they will be routinely studied
at RHIC. Two microscopic, QCD-inspired descriptions of nucleusnucleus collisions at
RHIC energies and beyond have been proposed in recent years. In one, the parton cascade
picture [1517], the collision is modeled as a sequence of incoherent binary scattering
events among partons (minijets), which can be cast in a probabilistic formulation as
a multiple scattering process. In the other one, the random mean-field model [18] the
collision is described as a nonlinear interaction among coherent color fields carried by the
colliding nuclei. We hope that the study of the interplay between softhard and multiplehard processes in a simpler framework, such as h + A interactions, will help to elucidate
the competition between these processes in A + A collisions.
The determination of multiparton correlation functions in nuclei is also an important step
towards the application of QCD to nuclear physics in general. The generic enhancement
of higher-twist effects in reactions involving nuclei facilitates their extraction and give
the chance for a deeper theoretical understanding. They encode the essential aspects of
the difference in the quarkgluon structure between isolated nucleons and nuclei. From
the practical point of view a better understanding of such medium effects is necessary to
be able to interpret the so called hard probes of high-energy heavy-ion collisions in a
convincing manner.
The unintegrated DrellYan cross section is actually the classic example for a typical
two-scale process. Both Q2 and q , the mass and transverse momentum of the photon,
2)
are detected. The perturbative treatment then gives rise to logs of the type log2 (Q2 /q
which can become large if the scales differ substantially. In such a case they have to be
resummed. Resumming large logs in the presence of additional soft gluons contained in
the twist-4 correlators is a delicate task that could not be solved yet. We therefore follow a
less ambitious procedure and require that both scales are of the same order, thus reducing
the original two-scale problem to an effective one-scale problem. In addition the LQS
2 to be not much smaller than Q2 in order to
formalism in its present form requires q
neglect interference terms that are not calculated up to now and which would introduce
additional twist-4 matrix elements. Recently Guo, Qiu and Zhang proposed a method to
describe the region of low transverse momentum and how to match it to the perturbative
calculation at high q [19].
Finally we would like to comment on the experimental situation. There exist data
for angular coefficients on + A scattering, e.g., from the NA10 [20] and E615 [21]
experiments which are not yet completely understood in terms of conventional QCDtreatment. These experiments measure pair masses Q 4 . . . 8.5 GeV and transverse
541
momenta q < 3 GeV which is certainly at the limit of where we could still trust our
calculation. We will focus on future RHIC perspectives where we could hopefully reach
higher transverse momenta.
The paper is organized as follows: In Section 2 we give the general definition of the
DrellYan kinematics, we introduce our observables helicity amplitudes and angular
coefficients and explain possible choices of the photon rest frame. In Section 3 we
remind the reader of the known twist-2 results for the helicity amplitudes. Section 4 is
devoted to the calculation of the double-scattering contribution. We explain some essential
features of the factorization procedure of LQS. In particular we discuss the distinction
between double-hard and softhard scattering and give an introduction of how to calculate
them practically. Finally, Section 5 contains our numerical results along with a discussion
and we close with a summary of our final conclusions. In the appendices we have collected
some technical details referred to in the text.
2. General definitions
2.1. Kinematics
Let a nucleus with momentum P1 per nucleon collide with a hadron with momentum
We study lepton pair decay of virtual photons with momentum q and mass Q2 = q 2 .
In the hadron center-of-mass (c.m.) frame we take

1
1
S, 0, 0, S ,
P2 =
S, 0, 0, S ,
(2)
P1 =
2
2
where S is the total center-of-mass energy. The differential cross section for DrellYan
pair production is then given by
P2 .
2
d4 q
em
L W
d.
(3)
4
2SQ
(2)4
The solid angle is parametrized by the polar and azimuthal angles and of one decay
lepton in the rest frame of the virtual photon. We have the freedom to choose the axes in
the photon rest frame conveniently. We will discuss this point after Eq. (16).
The hadronic tensor follows the standard definition and is given by
Z
(4)
W = d4 x eiqx hP1 P2 | j (x)j (0) |P1 P2 i .
d =
The leptonic tensor is

L = 2 l1 l2 + l2 l1 g Q2 ,
(5)
where l1 and l2 are the momenta of the leptons.

We rewrite the q-dependence of the cross section in terms of rapidity y, transverse
momentum q and mass Q of the virtual photon in the hadron c.m. frame. Introducing
the hadronic Mandelstam invariants
S = (P1 + P2 )2 ,
T = (P1 q)2 ,
U = (P2 q)2,
(6)
542
the rapidity y of the photon is

0
2

q + q3
Q U
1
1
.
(7)
= ln
y = ln 0
2
2
q q3
Q2 T
Likewise we can express the transverse momentum of the lepton pair in the hadron c.m.
system by
(Q2 U )(Q2 T )
Q2
S
and we can rewrite the differential cross section as:
2
em
1
d
=
L W
.
2 dy d
4SQ4
(2)4
dQ2 dq
2
=
q
(8)
(9)
2.2. Helicity amplitudes and their frame dependence

To discuss the angular distribution of the lepton pair it is convenient to introduce helicity
amplitudes W, 0 defined as
W, 0 = ( ) (
0 )
W ,
(10)
where ( ) , {1, 0} is a set of polarization vectors of the virtual photon with respect
to the axes defined by the spatial unit vectors X, Y , Z in its rest frame:
1
(0) = Z .
(11)
() = (X iY ) ,
2
Following [22] it is advantageous to introduce a complete set of structure functions
WL = W0,0 ,
WT = W1,1 ,
1
W = W1,1 ,
(12)
W = (W1,0 + W0,1 ),
2
which are referred to as the transverse, longitudinal, spin flip, and double spin flip helicity
amplitudes. Due to parity conservation other combinations are related to these by W, 0 =
0
() + W, 0 . In terms of these helicity structure functions the cross section can be
written as
2
em
d
=
2 dy d
64 3 SQ2
dQ2 dq

1 3
cos2
WTL 1 + cos2 + WL
2 2

2
+ W (sin 2 cos ) + W sin cos 2 .
The angular dependence can be integrated out which reduces the cross section to

2
em
16
d
WTL
=
2 dy
64 3 SQ2 3
dQ2 dq
(13)
(14)
with WTL = WT + 12 WL = 12 W . The ratio of the angular dependent and angular

integrated cross sections is usually parametrized in two standard forms [20,23]:
543
Fig. 2. Definition of the angles and in the rest frame of the lepton-pair. The frame shown here is
the CollinsSoper frame.
16
3
d
2 dy d
dQ2 dq
d
2 dy
dQ2 dq

A2 2
1 3
cos2 + A1 sin 2 cos +
sin cos 2
= 1 + cos2 + A0
2 2
2

4
1 + cos2 + sin 2 cos + sin2 cos 2
=
+3
2
in terms of two sets of angular coefficients
(15)
WL
W
2 W
,
A1 =
,
A2 =
,
(16)
WTL
WTL
WTL
2A1
2A2
2 3A0
,
=
,
=
.
(17)
=
2 + A0
2 + A0
2 + A0
At this point it is necessary to comment on the proper choice of coordinate axes in the
photon rest system with respect to which the polar and azimuthal angles and are
defined. Clearly the helicity amplitudes and angular coefficients depend on the chosen
frame. Two prominent examples are the CollinsSoper frame (CS) and the Gottfried
Jackson frame (GJ) which are discussed in detail in [22]. When we boost from the hadron
c.m. frame to a lepton c.m. frame, the collinearity of the hadron momenta P 1 and P 2 is
lost and they span a plane which we identify with the XZ-plane. We are still free to fix
Z within this plane. In the CollinsSoper frame we choose Z to bisect the angle between
P 1 and P 2 , see Fig. 2. The angle between Z and P 1 and between Z and P 2 is called
CS . In the GottfriedJackson frame Z lies in the direction of P 1 . We have introduced
the angle GJ between P 1 and P 2 for short notations in that case. Obviously we have
2CS + GJ = . See Appendix A.1 for more details. Also note that for q = 0 both frames
are equal and is no longer defined.
In order to obtain the results in both frames in a straightforward way we use the
projectors given by Mirkes [24]
A0 =
PTL = g ,
PL1 =
P1 P1
E12
544
PL12 =
P1 P2 + P2 P1
,
E1 E2
PL2 =
P2 P2
E22
(18)
where E1 and E2 are the energies of the hadrons in the lepton c.m. frame. In terms of
hadronic Mandelstam variables and the photon momentum these are
q
ey
Q2 T
2 ),
=
S(Q2 + q
E1 =
2Q
2Q
q
ey
Q2 U
2 ).
=
S(Q2 + q
(19)
E2 =
2Q
2Q
Contracting the hadronic tensor with these projectors gives amplitudes T = P W ,

{TL, L1, L2, L12} which are just linear combinations of the helicity amplitudes in the
different frames. These can be found by linear transformations
W ()
TTL
TL
()
WL
() = M() TL1 ,
(20)
TL2
W
()
TL12
W
with {CS, GJ}. The transformation matrices M() follow from simple geometry and
are given in Appendix A.1 for the CS and GJ frames.
3. Leading twist results

To make this paper self contained we give the leading order results for the angular
coefficients. Due to twist-2 factorization the hadronic tensor is given by a convolution
of a perturbatively calculable partonic tensor and two twist-2 distribution functions:
X Z d1 Z d2
H a+b (1 , 2 )fa/A (1 )fb/ h (2 )
W = g 2 eq2
1
2
a,b

(21)
2 s + t + u Q2 .
The sum runs over partons a and b in the nucleus A and the single hadron h, respectively,
and 1 , 2 are their momentum fractions. We denote the strong coupling by g and eq is
the fractional electric charge of the quark flavour coupling to the virtual photon. The f ( )
correspond to twist-2 parton distribution functions defined in the usual way as
Z
1
dy + i P + y 1
+
hP | q(0)
P e
Pq(y ) |P i ,
fq/ h ( ) = +
P
2
2
Z
1
dy i P + y
hP | Fa+ (0)PFa + (y ) |P i ,
e
(22)
fg/ h ( ) =
+
P
2
for quarks and gluons, respectively, in a hadron moving with momentum P + along the
light cone [1]. q and F denote quark fields and gluon field strength tensors and P is the
path-ordered exponential
( Z0
)
+
P = P exp ig dx A (x ) ,
545
(23)
which makes the definition gauge invariant. In the following we will not write this pathordered exponential explicitly, but it is assumed to be inserted between all factors in a
product of operators referring to different spacetime points. In the A+ = 0 gauge this
term will disappear but we will always use covariant gauge. Note that we also suppressed
the dependence of the distribution function on the factorization scale 2 .
a+b represents the hard part of the process. We have introduced
The partonic tensor H
partonic Mandelstam variables by
s = (1 P1 + 2 P2 )2 = 1 2 S,
t = (1 P1 q)2 = 1 (T Q2 ) + Q2 ,
u = (2 P2 q)2 = 2 (U Q2 ) + Q2 .
(24)
The -function in Eq. (21) stems from the on-shell condition for the unobserved emitted
parton with momentum l. It can be used for trivial integration of 2 via

Q2 + 1 (T Q2 )
1
2
2
2 +
(l ) = s + t + u Q =
.
(25)
1 S + U Q2
1 S + U Q2
We can decompose the partonic tensor in a way completely analogous to the
decomposition of the hadronic tensor W in terms of helicity amplitudes. Therefore we
a+b
a+b
, HLa+b , Ha+b , H
which are obtained in the same way
define the four projections HTL
as in Eqs. (10, 12). This allows us to write hadronic tensor and helicity amplitudes as
X Z d1
1
fa/A (1 )fb/ h (2 )Ha+b (1 , 2 ),
1 1 (Q2 T ) Q2
1
W = 8
s eq2
(26)
a,b B
2
1 (T Q )
, = Q +
with B = S+TU
2 , s = g /4 and {, TL, L, , }. There
Q2 2
1 S+U Q
are three kinds of partonic processes contributing to the leading-twist process at large
transverse momentum: quark-antiquark annihilation (q + q) and Compton processes with
a gluon from the single hadron (q + g) or the nucleus (g + q). They are shown in Fig. 3.
The cross section for the leading twist DY pair production at large transverse momentum
is well-known [24,25]. For completeness we give the results in the CollinsSoper frame in
Appendix A.2. The relation H = 12 HL is precisely the so-called LamTung relation
[14]. It can be expressed in terms of angular coefficients as A0 = A2 or 2 (1 ) =
0. This relation holds for all partonic subprocesses and is not specific for the Collins
Soper frame but holds in any photon rest frame as long as the Z-axis lies in the reaction
plane. Corrections arise at next-to-leading order (NLO) [24]. Note that Ha+b picks up an
additional sign compared with Hb+a in contrast to the other helicity amplitudes. This is
due to the inversion of the orientation of the angle CS if we interchange the roles of P1
and P2 . For symmetric collisions, e.g., p + p, this leads to the leading-twist result W = 0.
2
546
(a)
(b)
(c)
Fig. 3. Twist-2 contributions to order s : annihilation (a), and Compton processes (b) and (c). The
blobs L and R stand for all possible tree-level diagrams.
4. Twist-4 factorisation
4.1. Softhard and double-hard scattering
For the calculation of the double scattering contributions we use the approach of Luo,
Qiu and Sterman which is presented, e.g., in [8,10]. We consider it worthwile to repeat
some aspects here. In Fig. 4 we illustrate the forward diagrams that need to be calculated
for double scattering. The box denotes all possible partonic subprocesses. We have to
introduce a cut to obtain the corresponding hadronic tensors. This is not only possible
in the middle but we also have to include interference graphs which appear if the forward
diagram is cut in an asymmetrical way. In Fig. 5 we give concrete examples for graphs
which involve a quarkgluon matrix element in the nucleus and an antiquark coming from
the single hadron. The first diagram is cut in the middle (M), the others are cut left (L) and
right (R).
The starting point for the calculation of the hadronic tensor is the forward scattering
amplitude in fourth order of the strong coupling constant g:
T =
g 4 eq2
4!
Z
d4 x d4 x1 d4 x2 d4 x3 d4 x4 eiqx hP1 P2 | T [J (x3 )J (x1 )j (0)j (x)
J (x2 )J (x4 )A (x3 )A (x1 )A (x2 )A (x4 ) + ] |P1 P2 i ,
(27)
(a)
(b)
(c)
(d)
547
(e)
Fig. 4. Collection of the schematic forward processes we want to consider for double scattering.
Factorization allows separate treatment of twist-4 and twist-2 matrix elements (T and f ) and hard
part (H ).
where j (J ) are electromagnetic (color) quark currents and A denotes gluon fields. T is
the time-ordering operator. Dots indicate further contributions involving more gluon fields
with gluon self interaction and less quark currents such that the power of g remains four.
By application of Wicks theorem we obtain all diagrams which potentially contribute to
the forward process.
We classify diagrams by their partonic subprocess ab + c, where a and b are partons
from the nucleus and c comes from the single hadron. Our notation for the parton momenta
is shown in Fig. 4. We concentrate first on qg + q which can be realized, e.g., by the
diagrams of Fig. 5. As a first step towards factorization we perform Fierz transformations
in color and Dirac space for the fermion lines entering and leaving the blobs:

1
1
j
i
j0
i0
A
A
j 0 i 0 ij + 2tj 0 i 0 tij + .
(28)
q (0)q (z) = ( ) q (z) q (0)
4
Nc
Here the t A are the SU(Nc ) Gell-Mann matrices, and Nc is the number of colors. Latin
548
(a)
(b)
(c)
Fig. 5. Examples for double-scattering diagrams with the partonic process qg + q cut in the middle
(a), right (b) and left (c). Circles in Fig. (a) indicate poles from propagators which lead to finite gluon
momentum (open) or soft gluon momentum (filled).
characters are color indices, Greek one are Dirac indices. We keep only Dirac structures
which give leading contributions to the unpolarized DrellYan process. After Fourier
transformations of the fields we end up with
qg+q
T
Z
=
d4 l d4 r1 d4 r2 d4 K d4 K 0
(2)4 (2)4 (2)4 (2)4 (2)4
Z
d4 y d4 z1 d4 z3 d4 z4
0
(2)4 (4) (r1 + r2 + K l q)eir1 z1 eir2 y eiKz3 eiK z4

1
hP1 | q(0)
q(z1 )A (z3 )A (z4 ) |P1 i
2

1 RS X
1
hP2 | q(0)
q(y) |P2 i eq2 g 4 2 2

S(d)RS

.
2
Nc Nc 1
(29)
We used the fact that forward matrix elements only depend on the relative coordinates. So,
after shifting positions in each matrix element separately, we introduced a new set of space
time coordinates (y,z1 ,z3 ,z4 ). l is the momentum of the unobserved radiated parton. The
index d runs over all diagrams of the class qg + q and S(d)RS
contains all perturbative
propagators and vertices of the hard part of diagram d, with ( ) inserted for the upper
(lower) blob, and traces over color and Dirac indices. For the diagram of Fig. 5(a) we have,
e.g.,

RS
S
=

i(/r2 K
1
i(/r1 +K
/ K
/ 0 q/ )
/ 0)
B
Tr (i )
i
t
4
(r1 + K K 0 q)2 + i
(r2 + K 0 )2 + i

i(/r2 K)
/
i t S i t R
2
(r2 + K) + i

q
/
)
i(g ) AB
i(/
r
1
(i
.
)
i t A
(r1 q)2 + i
l 2 + i
549
(30)
R and S are color indices of gluon fields. Note that we have coupled quark and gluon field
operators in the nuclear matrix elements separately to color singlets. We have omitted a
t E qAR AS |P1 i which would induce color
term proportional to (d RES + if RES ) hP1 | q
forces between quark and gluon in the nucleus. This would not allow to factorize quarks
and gluons in color space to test the size of the nucleus. Such contributions are not expected
to show nuclear enhancement, i.e., scaling by A4/3 .
Next we perform the light cone expansion, i.e. we use that the partons move almost
collinear to the hadron in the infinite momentum frame. We introduce the parton
momentum fractions by setting r1 = 1 P1 , r2 = 2 P2 , K = xP1 + K and K 0 = x 0 P1 +
0 . r 0 = r + K K 0 is fixed by momentum conservation in the upper forward matrix
K
1
1
element. Note that we have to allow one parton (called b here) from the nucleus to be
soft with essentially zero longitudinal momentum. Therefore, we must take into account
0 to get a contribution in higher orders of the collinear
transverse momenta K and K
expansion as we will see later. Now one can integrate the trivial degrees of freedom in
Eq. (29). To complete the factorization we perform a Sudakov decomposition of the matrices in the forward matrix elements on the light cone and retain only the leading terms
=
P1 +
,
P1+
P2
.
P2
Then we can introduce twist-2 distribution functions fc/ h (2 ),
given in Eq. (22), for the lower blob. Finally we cut the forward diagram. This sets the
radiated unobserved parton with momentum l on the mass shell and in the hard part all
expressions right to the cut receive a complex conjugation.
Hence after these first steps we have factorized the hadronic tensor in a straightforward
way. We get
Z
d1 d2
ab+c
0
= (4s )2 eq2
dxP1+ d2 K dx 0 P1+ d2 K
fc/ h (2 )(2)(l 2 )
W
1 2
ab+c

0
0
H 1 , 2 , x, x 0 , K , K
. (31)
T ab 1 , x, x 0 , K , K
We have derived Eq. (31) for our examples of Fig. 5, but this equation holds for all
processes shown in Fig. 4. We have separated the soft part involving the twist-2 distribution
function fc/ h and the nuclear matrix element T ab from the hard part H ab+c
.
For our example qg + q we are left with a quarkgluon matrix element in the nucleus
0
T qg (1 , x, x 0 , K , K
)
Z
2
dz1 dz3 d z3 dz4 d2 z4 ir + z iK + z iK z iK 0 + z iK 0 z

3 e
3 e
4 e 4
e 1 1e
=
2 (2)3
(2)3
1
+
hP1 | q(0)
q(z1 )A (z3 , z3 )A (z4 , z4 ) |P1 i .
2
(32)
550
For other classes of processes we also need the quarkquark and gluongluon correlators
T gg and T qq which are given by Eq. (32) with the last line replaced by

1
+

(0)F +
for T gg ,
(z1 )A z1 , z1 A z4 , z4 |P1
+ P1 |F
1 P1

1
+
hP1 | q(0)
q(z1 )q z3 , z3 q z4 , z4 |P1 for T qq .
(33)
2
Again here the field operators of each single parton form color singlets separately. We have
arranged the first two field operators referring to the nuclear parton a in the way they appear
in the corresponding twist-2 structure function while the remaining two field operators of
parton b are still in their original form.
The hard part for the qg + q-process
is

1X
1
1
qg+q
0

g r1 r2
H 1 , 2 , x, x 0 , K , K
S(d)RR
= 2 2
. (34)
4
N N 1
c
cut
Hard parts for other processes are obtained analogously. The delta function in Eq. (31) can
be rewritten for the cases in which the diagrams are cut in the middle (M), right (R) or left
(L) as
1
(M) :
(l 2 ) = 2 S + T Q2 (1 + x b ),
1
(35)
(R) :
(l 2 ) = 2 S + T Q2 (1 + x x 0 c ),

1
(L) :
(l 2 ) = 2 S + T Q2 (1 c ),
where
b =
2
2 (U Q2 ) + Q2 2qK + K
2 S + T Q2
and

c = b K
=0
(36)
As pointed out in various papers [9,10] there are in principle two contributions to double
scattering: there can be two hard scattering reactions of the parton c from the single hadron
or the outgoing unobserved parton on one side and partons a and b from the nucleus on
the other side. This is called double-hard scattering. Furthermore there can occur processes
where the parton c from the single hadron or the outgoing parton picks up one soft nuclear
parton b in addition to one hard scattering off the nuclear parton a. This is called softhard
scattering. These are not only qualitative notions but they can be rigorously derived from
the equations given above. One can understand them in terms of poles in the hard part of
the scattering tensor. In Eq. (31) we have four momentum integrations d1 d2 dx dx 0 and
four propagators. We will perform the 1 -integration by making use of the delta function
(l 2 ). The propagators provide poles with respect to x and x 0 . In general we expect two
poles in x and two poles in x 0 . Application of the residue theorem will fix the momenta and
it is easy to check that two poles give large momenta x = xhard = x 0 , respectively and the
other ones give small momenta x = xsoft = x 0 with xhard xsoft 1 as long as q Q.
To be more precise the pole structure we expect for our example in Fig. 5(a) is
Z
L(x)
W dx
(x xsoft + i)(x xhard + i)
dx 0
R(x 0 )
,
(x 0 xsoft i)(x 0 xhard i)
551
(37)
where L and R stand for the numerators of the amplitudes left and right of the cut
depending on x and x 0 . The residue theorem implies

L(x = xsoft ) L(x = xhard) R(x 0 = xsoft ) R(x 0 = xhard )
.
(38)
W
xsoft xhard
xsoft xhard
xsoft xhard
xsoft xhard
Soft-hard and double-hard scattering correspond to the quadratic terms proportional to
L(x = xsoft)R(x 0 = xsoft ) and L(x = xhard )R(x 0 = xhard ), respectively. We follow the
arguments of reference [9,10] and neglect the remaining mixed terms of Eq. (38) which
are interference terms between soft and hard rescattering. This is permitted if q is not too
2 Q2 however we obtain x
small. In the case q
soft xhard . The interference will then be
important and eventually spoil nuclear enhancement due to the negative signs in Eq. (38)
[10]. Since the integrated DrellYan cross section does not exhibit nuclear enhancement,
we even expect a suppression effect at low values of transverse momenta [11].
Most diagrams we naively get from Wicks theorem for double scattering show a
different behaviour. For them the propagators do not provide four but only three or two
poles but at least one in each variable x and x 0 . This implies that only softhard,
only double-hard or neither of these contributions can occur. For example it is easy to
understand from this analysis that only symmetrically (M) cut diagrams can contribute to
double-hard scattering. Consider the diagram of Fig. 5(b). The propagator on the right-hand
side has denominator [(r10 q)2 i], from which x and x 0 dependences cancel out. On
the left-hand side, the propagator at the bottom of the diagram contributes a denominator
K2
K2
(r2 + K)2 + i = 2 S(x + 2S + i). This is a soft pole which fixes x to xsoft = 2S
1. The remaining two propagators on the left-hand side fix x 0 to either a soft or a hard
value. Therefore we have two contributions: (i) x = x 0 is small. This means that parton b is
soft in Fig. 5(b) this is the gluon from the nucleus. Together with the production of the
virtual photon which is always a hard process, this constitutes a softhard double scattering
process. (ii) x is soft and x 0 is hard. This is what we called a softhard interference because
parton b has different momenta on the left and the right side of the diagram, i.e., momentum
is transferred between partons a and b. We omit this contribution as explained above. We
conclude that there is no double-hard scattering contribution from the diagram in Fig. 5(b)
since x cannot be hard. On the other hand the diagram in Fig. 5(a) is an example in which
all four poles occur, indicated by small circles.
We now discuss another technical point which has some relevance for nuclear
enhancement. Consider again Eq. (37) and the example of Fig. 5(a) and note that the
+
+
0 +
numerators L and R contain exponential functions eixP1 z3 ei(b x)P1 z1 and eix P1 z4 .
To apply the residue theorem we have to close the integration contours. For z3 z1 > 0
we have to close the x-integration in the upper complex x-plane with no poles encircled
and therefore vanishing integral. On the other hand for z3 z1 < 0 we must close in the
lower complex x-plane with soft and hard pole encircled. Therefore these poles always
introduce additional theta functions in coordinate space, in our example (z1 z3 ) from
the x integration and (z4 ) from the x 0 integral. The presence of these functions in
552
(a)
(b)
(c)
(d)
Fig. 6. Graphs contributing to double-hard scattering: qg + q,

qg + g, q q + g and qq + q.
There are
additional processes with structures similar to the last one, e.g., q q + q. This is discussed in the next
section. The blobs (L) and (R) stand for all possible tree-level diagrams. Propagators providing poles
which are used to fix hard momenta are indicated by small open circles.
the matrix elements is an essential ingredient of the proof of the following rule about the
cancellation of final state interactions for softhard scattering: If at least one soft parton
from the twist-4 matrix element is directly coupled to the outgoing unobserved parton line
these graphs do not contribute to softhard scattering. These graphs exhibit more than one
possibility to set the cut and the nuclear enhancement cancels between the contributions
with different cuts due to the particular arrangement of -functions in that case [9].
Hence, after a careful discussion of the pole structure of each diagram the number of
diagrams we are finally forced to evaluate can be reduced. Figs. 6 and 7 show the diagrams
that contribute to double-hard and softhard scattering, respectively. We marked the poles
used to fix momenta. Also we omitted softhard contributions with soft quarks since they
should be suppressed compared to their counterparts with soft gluons.
4.2. Calculation of the double-hard processes
First we investigate the case of double-hard scattering. Since all parton momenta are
0 immediately. We perform
hard we can work in the collinear limit and set K = 0 = K
the integrations over x and x 0 only taking into account contributions from hard poles. These
poles are marked with circles in Fig. 6. They fix momenta to the values x = xh = x 0 and
1 = xa with
553
(a)
(b)
(c)
Fig. 7. Graphs contributing to softhard scattering: qg + q,

qg + g and gg + q. The blobs (L)
and (R) stand for all possible tree-level diagrams. Only symmetrically cut diagrams (M) are shown.
Corresponding right (R) and left (L) cut diagrams have to be added. Propagators providing poles
which are used to fix soft momenta are indicated by small filled circles.
xa =
Q2
,
Q2 T
2 (U Q2 ) + Q2
Q2
.
xh = a xa =
2 S + T Q2
Q2 T
(39)
(40)
Note that xa is determined by kinematical variables only whereas xh depends on

the integration variable 2 . Two additional -functions in the light cone spacetime
coordinates show up as explained above.
The field operators of the second parton in the nuclear matrix element are selected in
such a way that we get the usual twist-2 light cone operator q
+ q for quarks and field
+
+
strength tensors F F for gluons. In the latter case we have to turn two factors of
P1+ into light cone derivatives (ixh )1 z , (ixh )1 z and perform partial integrations
3
4
to let them act on the gluon fields in the matrix element. Hard gluons carry only physical
polarizations, this allows us to set + A (z3 ) + A (z4 ) = F + (z3 )F + (z4 ) 12 (g +
n n + n n ) with the usual light cone basis vectors n = P1 /P1+ , n = P2 /P2 .

We are now able to give the final result for double-hard scattering. For a partonic
subprocess of the type ab + c we have
554
Wab+c
(2)2 (4s )2 eq2

Q2
Z1
B
d2
1
DH
fc/H (2 )Tab
(xa , xh )
2 2 S + T Q2
1
HDH/ab+c (xa , xh , 2 ),
xh
(41)
with {, TL, L, , }. Note that parton a is always a quark or antiquark for doublehard scattering and eq denotes its electric charge. As in the helicity amplitudes which arise
from contractions of the hadronic tensor W defined in Eq. (10), here we use the same
contractions for the hard partonic part H , defining amplitudes HTL , HL , H and H
depending on the frame. The hard parts for the relevant diagrams can be written as

1 1
1
Tr (/r1 q/ )R /r2 L (/r1 q/ ) /r1 1 =xa
g g ,
2
Q 4
x=x 0 =xh 2

1 1
1
DH,qg+g
g g , (42)
= C 2 Tr (/r1 q/ )R /lL (/r1 q/ ) /r1 1 =xa
H
Q 4
x=x 0 =xh 2

1 1
1
DH,q q+g
g g .
= C 2 Tr (/r1 q/ )RKL
/ (/r1 q/ ) /r1 1 =xa
H
Q 4
x=x 0 =xh 2
DH,qg+q
=C
H
L and R indicate the sum over all possible tree level diagrams for the blobs at the
left and the right side of the corresponding diagram in Fig. 6. Each diagram provides
an individual color factor which is indicated symbolically by the factor C. For partonic
processes with three participating fermions we have to distinguish between different
possible combinations of quark and antiquarks and different flavours. In principle there
are contributions written in our short notation as q q + q, q q + q and qq + q.
Each of
these contributes four diagrams contained in Fig. 6(d). In addition, there are processes of
the type q q + q 0 , qq 0 + q and qq 0 + q 0 , which are each represented by a single diagram.
In our notation q is a quark or antiquark, q the corresponding antiparticle and q 0 denotes
a quark or antiquark of a different flavour than q. As an example we give the hard part for
the q q + q subprocess:

1 1n
DH,q q+q
= C 2 Tr (/r1 q/ )K
/ (/r1 q/ ) /r1 Tr[/r2 /l ](r1 + K q)4
H
Q 8

Tr (/r1 q/ ) /l /r2 K
/ (/r1 q/ ) /r1 (r1 + K q)2 (r2 + K)2
o
+ (K l) 1 =xa g g .
(43)
x=x 0 =xh
Note the relative sign that is due to the different number of fermion loops in the
corresponding diagrams.
DH (x , x ) are the universal nuclear matrix elements for double-hard scattering
The Tab
a h
introduced in [8,26]. They depend on the momentum fractions xa , xh of both hard partons
from the nucleus and are given by
Z
dz dz1
+
+
1 1
DH
(z1 z3 )(z4 )eixa P1 z1 eixh P1 (z3 z4 )
dz4 3
Tqg (xa , xh ) =
xh 2
2 2

+
P1 |F + (z4 )F + (z3 )q(0)
q(z1 )|P1 ,
TqDH
q (xa , xh ) =
1
4
DH
(xa , xh ) =
Tqq
1
4
555
dz dz1
+
+
dz4 P1+ 3
2 2

+
P1 |q(0)
q(z1 )q(z
3 ) + q(z4 )|P1 ,
dz dz1
+
+

dz4 P1+ 3
2 2

+
P1 |q(0)
q(z1 )q(z
4 ) + q(z3 )|P1 .
(44)
We are now ready to explain the origin of nuclear enhancement in these matrix elements.
As mentioned above the fields of both partons are coupled separately to colour singlets.
This allows to probe distances up to the diameter of a large nucleus. Each spatial
integration of the matrix element arises from a Fourier transformation and we expect it
to be accompanied by a rapidly oscillating exponential function (since P1+ ). At
first glance, this seems to rule out coherence over the whole nucleus. However only z1
and z3 R z4 in Eq. (44) are restricted by phase factors while z3 + z4 is not. Hence, we
can set d(z3 + z4 ) to be proportional to the nuclear radius RA r0 A1/3 and obtain an
additional factor of A1/3 . Here r0 1.1 fm is the radius of a nucleon.
To find a reasonable model for the double-hard matrix elements imagine the insertion of
P
a complete set of states P |P i hP | /(2P + V ) = 1 where V is the volume of the nucleus.
Neglecting for a moment the -function and the integration over z3 + z4 we can compare
these definitions with the common definitions of twist-2 structure functions in Eq. (22).
Assuming that the sum over states is saturated by the lowest lying nucleon state double hard
matrix elements reduce to a product of two twist-2 structure functions times a dimensional
factor which scales like A4/3
DH
Tab
(xa , xh ) = CA4/3 fa/A (xa )fb/A (xh ).
(45)
fa/A is the nuclear parton distribution per nucleon in a nucleus with mass number A and
proton number Z, N = A Z, given by
fa/A (x) =
Z
N
fa/p (x) + fa/n (x).
A
A
(46)
In a primitive model one can simply set the remaining free integration with respect to z4 +
z3 equal to RA and obtains C = 3/(8r02) 0.005 GeV2 . Although the factorization of
the double-hard matrix element into a product of two twist-2 distribution functions and the
scaling by A4/3 seems to provide a reasonable model, the choice of the normalization factor
C is a hot matter of discussion, see Section 5. The -functions provide an ordering of both
single scattering events on the light cone. They simply express the fact that scattering off
parton b takes place before the scattering off parton a. This corresponds to z4 < 0 on the
left side and to z3 < z1 on the right side of each diagram.
Now we give the final results for the hard parts. For the evaluation it is important also
to include ghosts, whenever one can close gluon lines through the non-perturbative blobs.
DH = 1 H DH , which enters the angular integrated cross section, we reproduce
For HTL
2
the results of [10]. We have listed them in Appendix A.3 for all subprocesses. The helicity
amplitude WTL is independent of the frame. Due to the special form of all diagrams
556
DH and
contributing to double-hard scattering we easily see that the contractions PL1 H
DH
PL12 H in Eq. (42) always vanish, and for the last contraction we get
DH
=
PL2 H
Q2 S[(Q2 T )(Q2 U ) Q2 S]
2
(Q2 T ) (Q2 U )
DH
HTL
=
2
Q2 q
(Q2
2 )2
+ q
DH
HTL
(47)
for all processes. For the CollinsSoper frame this leads to the simple relations
DH
=
WLDH = sin2 CS WTL
2
q
2
Q2 + q
DH
=
WDH = sin CS cos CS WTL
DH
WTL
,
Qq
DH
WTL
,
2
Q2 + q
(48)
(49)
1
DH
= WLDH .
(50)
W
2
Immediately one observes that the LamTung relation holds in the CS frame and all other
frames considered here. For the GottfriedJackson frame the results get particularly simple.
The double-hard contributions for deviations from the simple 1 + cos2 behaviour vanish
completely:
WLDH = 0,
WDH = 0,
DH
W
= 0.
(51)
At first glance this seems to be rather surprising but it is a direct consequence of the special
form of all double-hard contributions and therefore has a simple explanation.
The double hard process resembles the classical double scattering picture. To understand
this we consider again the process depicted in Fig. 1. The antiquark from the hadron
undergoes a first scattering with a hard gluon from the nucleus. By subsequently radiating
a gluon with large transverse momentum the antiquark returns to the mass shell, but now
with transverse momentum q . The second scattering then corresponds to the lowest-order
annihilation process of two massless quarks. Therefore, the double-hard process can be
factored as the product of two single scattering cross sections, supported by our model that
the double-hard matrix element can in good approximation be written as a product of two
twist 2 distribution functions. This process naturally fulfills the LamTung relation and
depends only on the structure function WTL W . Other helicity amplitudes differ only
by kinematical factors, i.e., can be obtained from WTL in the GottfriedJackson frame by
simple rotations. In this probabilistic picture scaling by A4/3 appears to be very natural.
The cross section of the first scattering obviously scales by the volume A of the nucleus.
Afterwards the projectile parton travels further through the nucleus and has to penetrate
the nuclear matter. The probability whether a second interaction takes place on the way to
the boundary of the nucleus scales then by the nuclear radius A1/3 .
4.3. Calculation of the softhard processes
For softhard scattering we have to keep K in the expansion of the gluon momenta.
The limit K = 0 would lead to K = 0 which gives no contribution to physical double
scattering but just a contribution of the eikonal phase. Note that we are always using
557
covariant gauge where these factors are present. Starting from equation (31) we do a
collinear expansion

1
d2
d e
e
e=H
e
H
H
+
K
+
K
+
(52)
K
H

K =0
K =0
2!
dK
dK dK K =0
0
K =0
of the terms depending on intrinsic transverse momentum

Z

dx dx 0 ir + z iK + z iK 0 + z
e
3 e
4 H 1 , 2 , x, x 0 , K , K 0 .
e 1 1e
H=
(53)
0 which allows us to write

We have used K = K
d

0
= K
+ K
.
0
dK
K
K
(54)
K 0 give the leading contribution

The mixed terms of second order proportional to K
K 0 can be converted into partial derivatives with

to soft hard scattering. The factor K
respect to z3 , z4 . A partial integration and an expansion A = A+ P1 /P1+ + for soft

gluon fields in the nuclear matrix elements turns them into field strength tensors. The K
0 integrals can then be carried out. Note that having soft quarks with transverse
and K
momentum and converting the corresponding K into covariant derivatives acting on the
quark fields would lead to operators contributing only at the level of twist-6 [8].
We perform the pole integrations for the symmetrically cut graphs (M) in Fig. 7 and
the corresponding asymmetrically cut ones (R) and (L). The residues fix the momentum
fractions of the soft gluons to x = xs = x 0 while the hard partons carry 1 = xb for the
symmetric diagrams and 1 = xc for the asymmetric diagrams and
xs =
2
k
,
2 S
(55)
2
2 (U Q2 ) + Q2 2qK k
k2
,
2
2 S
2 S + T Q
2
2
2 (U Q ) + Q
.
xc = c =
2 S + T Q2
xb = b xs =
(56)
(57)
2 = K 2 > 0. When we take a forward diagram

We have introduced the notation k
for softhard scattering then the sum of differently cut diagrams (M), (R) and (L) can
be divided into two terms. This procedure is described in detail in [9,10]. In the first
term, phase factors and -functions arrange in such a way that this term is bounded in
spacetime by geometrical arguments and cannot pick up an additional factor A1/3 . This
contribution shows no nuclear enhancement. The second term is not bounded and only
depends on the hard part of the symmetric diagram (M). We therefore neglect the first term
and arrive at
WSH/ab+c = (2)2 (4s )2 eq2

Z1

d2
g
d2
1

fc/H (2 )
2
2 2 S + T Q
4
dK dK K =0
558
1 SH,ab+c
H
(xb , xs , 2 ),
(58)
xb
where {, TL, L, , }. The hard part for the qg + q process shown in Fig. 7(a) is
SH
Tab
(xb )
1
(2 S)2

1
Tr R (/r2 K)/
/ P 1 /r2P/ 1 (/r2 K)L
/ /r1 1 =xb
4
x=x 0 =xs
SH/qg+q
=C
H
(59)
L and R stand for the tree-level diagrams on the left- and the right-hand side of the
diagram in Fig. 7(a) indicated by the blobs. C symbolizes the color factor of each single
diagram. Hard parts for qg + g and gg + q subprocesses are obtained in an analogous
fashion.
The universal matrix elements appearing for softhard scattering depend only on the
momentum fraction xb of the hard parton and are given by
Z
dz dz3
+
1
SH
(z1 z3 )(z4 )eixb P1 z1
(xb ) =
dz4 1
Tqg
2
2 2

+
P1 |F + (z4 )q(0)
q(z1 )F + (z3 )|P1 ,

Z
+
1
dz1 dz3
SH
(z1 z3 )(z4 )eixb P1 z1
(xb ) =
dz
Tgg
4
2 2
xb P1+

P1 |F + (z4 )F + (z3 )F + (0)F + (z1 )|P1
(60)
for scattering off a quarkgluon and a gluongluon pair in the nucleus, respectively. We
can repeat the arguments given below Eq. (44) for the nuclear enhancement. Since the
longitudinal momentum fraction of the gluons is essentially zero, we even encounter only
+
one phase factor eixb P1 z1 contrary to the case of double hard scattering discussed before.
One therefore might expect that softhard matrix elements lead to an enhancement factor
proportional to the square of the nuclear radius A2/3 since only one out of three integrations
is limited by phase factors. However, the fields at positions z3 and z4 are correlated by
color forces. The anticipated absence of long-range color correlations in a nucleus implies
that integration over z4 z3 is still restricted to a region of the size of a nucleon.
To build a model for the soft hard matrix elements one again can compare Eq. (60) with
the definition of the usual twist-2 matrix elements Eq. (22). Neglecting for a moment the
-functions we see that the twist-4 pieces differ only by an additional xb -independent
renormalization. Assuming that this can be factored into a term proportional to the nuclear
radius and a (possibly scale dependent) constant 2
Z
dz
1/3 2
A dz4 3 F + (z4 )F + (z3 )
(61)
2
the xb dependence of the twist-4 matrix element can be modeled by the usual twist-2
structure function
SH
(x) = 2 A4/3 fa/A (x).
Tab
(62)
559
Eq. (61) can be interpreted as the collective color-magnetic Lorentz force experienced by
the projectile parton while traveling through nuclear matter [8].
The only K -dependence of the hard part will be implicit in xb and xs . Therefore we
can rewrite the derivatives in the form
Z1
WSH,ab+c
= (2)
(4s )2 eq2
B
d2
1
fc/H (2 )
2 2 S + T Q2

2
2q
1 SH,ab+c
1
SH
(x
)
H
(x
,
)
T
c
c
2
ab
2 xc2
xc
(2 S + T Q2 )2

1
2(Q2 T )
SH
+
(xc ) HSH,ab+c (xc , 2 )
Tab
xc
xc
2 S(2 S + T Q2 )

0
1
2
SH
,
(xc )
HSH,ab+c (xc , 2 )
+ Tab
xc
2 S

(63)
where H (xc , 2 ) = H (xc , xs = 0, 2 ) and H 0 (xc , 2 ) = x s H (xc , xs = 0, 2 ). Note that

in [10] the last term containing a derivative with respect to xs is absent since it turns
SH does not depend explicitly on x . We give the results for the hard parts
out that HTL
s
in the CollinsSoper frame in Appendix A.4. There are some subtleties in removing the
unphysical degrees of freedom for processes of the class qg + g. There we have soft quarks
from the nucleus with dominating component A+ and for the hard gluons from the single
SH we reproduce the results
hadron we have to take only the transverse polarizations. For WTL
SH
cannot be
of [10]. Unlike for double-hard scattering the amplitudes WLSH , WSH and W
obtained in a trivial way from WTL . Related to that is an obvious violation of the LamTung
relation.
5. Numerical results
In order to cover different energies S we calculated helicity amplitudes and angular
coefficients for two different settings: for 252 GeV on tungsten as in the Fermilab
E615 experiment [21] and for future p + A collider experiments at RHIC energies with
250 GeV protons and A 100 GeV large nuclei. The main task before doing the numerics
is to insert a suitable model for the twist-4 matrix elements in the nucleus. The models
introduced in the last sections for the double-hard matrix elements
DH
(xa , xh ) = CA4/3 fa/A (xa )fb/A (xh ),
Tab
(64)
and for the softhard matrix elements

SH
(x) = 2 A4/3 fa/A (x)
Tab
(65)
are already given by LQS [9,10]. We tried to argue above that their dependence on the
parton momentum fractions and their scaling behaviour with respect to A seem to be quite
natural.
560
Nevertheless the correct normalization of the models is not yet clear. The hope is that C
and are universal constants. In the early days of the model a value of C = 0.072 GeV2
was taken from theory [10,26] while 2 was taken from comparison with experiment.
This was done by analyzing dijet photoproduction on nuclei [8,27] leading to a value
2 = 0.05 0.1 GeV2 . Later an analysis of q -broadening in hadronnucleon DrellYan
processes [11] implied a value of 2 = 0.01 GeV2 . Obviously there is a discrepancy, which
may either point to a strong scale dependence or questions the universality of the softhard
nuclear matrix elements. Note that the soft interaction besides the hard process in soft
hard double scattering is a final state interaction for dijet photoproduction but an initial
state interaction for the DrellYan process.
In their latest work Guo, Qiu and Zhang [19] used the fact that there is a formal
connection between softhard and double-hard matrix elements via the limit
DH
SH
(xq , xg ) = Tqg
(xq ),
lim xg Tqg
xg 0
(66)
which leads with the models of Eqs. (64), (65) to

C
2
.
xg g(xg )|xg 0
(67)
They take xg g(xg )|xg 0 3 as the gluon momentum density at xg 0. Taking into
account the later value of 2 this leads to a normalization factor C which is an order of
magnitude below the former theoretical value. But it is more consistent with the value of
C 0.005 GeV2 which we obtained in Section 4 from a naive normalization argument.
In another, more geometrical, model one can estimate the two-parton area density that
can be seen by the projectile parton in the nucleus by the product of two single-parton area
densities. Integration of the two-parton area density over the effective transverse area of
the nucleus yields exactly the right-hand side of Eq. (64) with a constant C = 9/(8r02 )
0.01 GeV2 . Obviously C is presently basically unknown and a true improvement can
only be expected if comparisons of a larger number of observables with more precise
experimental data are available.
For softhard scattering we use the smaller value of 2 = 0.01 GeV2 obtained from
DrellYan data to maintain consistency. For double-hard scattering we have chosen the
maximal value C = 0.072 GeV2 to give a feeling how big the effects of double scattering
might be. We take the twist-2 distribution functions from the CTEQ3M parameterization
for the nucleons [29] and from the recent GRS fits for the pion [30]. Where nuclear effects
on twist-2 distribution functions are taken into account the EKS98 parametrization [31] is
used.
Let us first address the question how this approach works with the pion data from E615.
In Fig. 8 we show our results for the angular coefficients , and in the Gottfried
Jackson frame for the process + W at E615 energy. We also give experimental data
from the E615 collaboration. We can fairly well describe the data for at low transverse
momentum but fail for and . It seems to be an essential feature that corrections due to
nuclear enhanced twist-4 matrix elements can be quite large for the helicity amplitudes but
the effects largely cancel for the angular coefficients which are ratios thereof. Therefore
561
(a)
(b)
(c)
Fig. 8. The angular coefficients , and (from (a) to (c)) as functions of q in the
GottfriedJackson frame for 252 GeV on tungsten at Q = 5 GeV, y = 0. The dashed line is
the result for single-scattering and the thick line is given by the full calculation with single- and
double-scattering. We also show results of the E615 experiment at low transverse momentum.
changes of the angular coefficients in the range q Q are small compared with the pure
twist-2 calculation. Experimental values have large error bars above q = 2 GeV. We
already pointed out that data at q /Q 1 would provide a much better test for the twist4 calculation because of the absence of softhard interference terms and the absence of
2 ) which have to be resummed. We
double logarithmic corrections of the type log2 (Q2 /q
have to conclude that nuclear enhanced double scattering contributions are not able to yield
substantial improvement for the old pion data. Deviations of the W data from the naive
perturbative QCD-predictions might be explained by other approaches like the Berger
Brodsky mechanism which takes into account higher twist contributions of the pion wave
function [32,33], in terms of non-trivial spin correlations in the QCD-vacuum [34] or via
contributions of the chiral-odd T -odd distribution function h
1 [35].
We turn to p + Au collisions at RHIC with S = 105 GeV2 per nucleon pair. We give
helicity amplitudes instead of angular coefficients since these provide a clearer measure
of nuclear enhanced higher twist effects as explained above. To make numerical results of
helicity amplitudes W , {TL, L, , } more meaningful, we introduce the prefactor
given in Eq. (13),
w =
2
W ,
64 3 SQ2
(68)
562
(a)
(b)
(c)
Fig. 9. The helicity amplitudes wTL , wL and w for p + Au collisions at S = 105 GeV, Q = 5 GeV
and q = 4 GeV. We show the twist-2 contribution (solid lines), double-hard (dashed lines), and
softhard (dashed-dotted lines). Soft-hard is scaled up by a factor of 10. We give results with (thin
lines) and without (thick lines) EKS98 modifications for nuclear parton distributions.
to convert the quantities W into differential cross sections. In Fig. 9 we give single
scattering, double-hard and softhard results for the helicity amplitudes wTL , wTL and
w as functions of rapidity y in the CollinsSoper frame. One can read off the main
results already given in [12]. First the nuclear enhanced twist-4 contribution is absolutely
important at RHIC energies. Even at quite large values of Q = 5 GeV and q = 4 GeV
used for the calculation its effect can exceed the leading-twist result for the parameters
used. Second the double-hard contribution is by far larger than the softhard part in the
kinematical range under consideration. This can be seen for all helicity amplitudes in
Fig. 9. Even if we use smaller values of C proposed above double-hard scattering would
reach the same order of magnitude as single scattering.
We also give results with EKS98 modifications to nuclear parton distributions. These
parametrizations take into account shadowing, antishadowing and other nuclear effects on
parton distributions, first described by the European Muon Collaboration. The numerical
effect of these modifications on the single scattering results are small, but they turn out
to be quite large for double-hard scattering. Note that it is not clear whether these nuclear
effects for twist-2 parton distributions should be carried over to the models for twist-4
matrix elements. We give results for modified and unmodified distributions to indicate the
further uncertainty of the twist-4 matrix elements. The sensitivity of double-hard scattering
563
Fig. 10. Violation of the LamTung relation. Difference of angular coefficients A0 and A2 is shown.
Calculation are with (solid line) and without (dashed line) EKS98 corrections to nuclear parton
distributions.
(a)
(b)
Fig. 11. Contributions of different subprocesses to wTL for double-hard (a) and softhard (b)
scattering. Double-hard subprocesses are qg + q (solid line), qg + g (dashed line), q q + g (dotted
line), and q q + q plus all pure fermionic processes (dash-dotted line). Softhard processes are qg + q
(solid line), qg + g (dashed line), and gg + q (dotted line).
to these EKS98 modifications is simply related to the fact that they involve a product of
two nuclear parton distributions. The general question about the correct small-x behaviour
of the twist-4 nuclear matrix elements is very interesting since it has a big influence at
RHIC energies.
We return to the question of the LamTung relation. We already saw analytically that
double-hard scattering respects this relation while softhard does not. In Fig. 10 we show
the difference A0 A2 . It only picks up contributions from softhard scattering and
therefore this quantity is very small. That means that violation of the LamTung sum rule
is a tiny effect for the process under consideration.
The characteristic feature of the rapidity distribution is the peak of the double-hard
distribution at negative rapidities (i.e., for the photon emitted in direction of the initial
single hadron). The reason for this is that negative rapidity prefers small Bjorken-x for
partons from the nucleus. Parton distributions at small parton momentum fractions are
large particularly for gluons and they enter quadratically for double-hard scattering. In
Fig. 11 we separated the different partonic subprocesses contributing to double-hard and
564
(a)
(b)
Fig. 12. wTL as a function of transverse momentum q at fixed Q = 5 GeV, y = 0 (left) and as a
function of photon mass Q at q = 4 GeV and y = 0 (right). Curves are shown for single scattering
(solid line), double-hard (dashed line), and softhard scattering (dotted line).
softhard scattering to enable a deeper analysis. One can see that double-hard processes
with a quark and a gluon from the nucleus dominate over those with two quarks at negative
rapidities (i.e., small xh ). On the other hand qg + q is more important than qg + g in
this region since momentum fraction 2 of the parton from the single hadron is large and
quarks dominate over gluons in that case. All the rapidity spectra can be understood in this
way.
Fig. 12 shows dependence of wTL on photon mass Q and transverse momentum q .
It can be seen that softhard and double-hard contributions as expected for higher-twist
processes fall off more rapidly with Q and q than (twist-2) single-scattering.
6. Summary
We have calculated nuclear enhanced double-scattering contributions to DrellYan pair
production in hadron-nucleus collisions using the framework of Luo, Qiu and Sterman.
We found that the nuclear enhancement at RHIC energies and large transverse momentum
makes double scattering a contribution that is equally important as single scattering.
Measurements of angular coefficients and helicity amplitudes can distinguish double-hard
and softhard scattering. Double-hard scattering has the interpretation of two sequential
hard scatterings of onshell partons and is therefore related to probabilistic pictures of
multiple scattering that are widely used in the phenomenology of heavy-ion reactions. In
this spirit further comparison with alternative approaches to multiple scattering [6,7,3739]
would be very interesting.
Double-hard scattering respects the LamTung relation and, in contrast to softhard
scattering, does not show non-trivial deviations from the 1 + cos2 angular distribution.
Numerically double-hard scattering dominates over the softhard contribution. Softhard
scattering violates the LamTung relation but this violation is an effect of order 103 for
the model used.
565
Double-hard scattering can be modeled as the product of two twist-2 structure functions
which give the probability for finding two partons in the nucleus. The softhard matrix
elements on the other hand encodes essential non-perturbative properties of the nucleus,
describing the influence of the collective color field on the transverse momentum of the
produced lepton pair. Inserting reasonable models for the corresponding matrix elements
we predict, that collective color effects are negligible compared to the effects of subsequent
independent hard scatterings in the kinematical range considered here. However, the
absolute normalization of the twist-4 matrix elements in these models is not yet clear.
The double-hard scattering shows an interesting asymmetry in the rapidity distribution of
the DrellYan pair. The pair is predominantly produced in the projectile, i.e., single hadron
rapidity region. Although the absolute counting rates for the DY pairs at large transverse
momentum are not high [12] the predicted effect fits perfectly well in the PHENIX detector
acceptance [36]. This asymmetry therefore should be observable.
An analysis of WL and W , particularly in the GottfriedJackson frame, as well as the
measurement of WTL (i.e., the angular integrated cross section) as a function of rapidity can
check whether double-hard scattering is the dominant contribution. A possible violation of
the LamTung sum rule provides a further possibility to check the picture of softhard
and double-hard scattering in general. If our predictions with respect to the ratio of soft
hard and double-hard scattering are confirmed by experiment it would provide support for
the description of heavy ion reactions in terms of incoherent multiple scattering of free
partons. Note that the prediction of a trivial frame dependence of double-hard scattering
and the conservation of the LamTung relation are independent of the models for the twist4 distribution functions and provide good tests for the entire LQS approach.
Acknowledgements
We are grateful to X. Guo for helpful correspondence. In course of the work we benefited
from discussions with V. Braun, L. Syzmanowsky, and O. Teryaev. E. S. thanks G. Sterman
and X.N. Wang for useful discussions. This work was supported by BMBF and GSI. B.M.
thanks the Alexander-von-Humboldt Stiftung for support.
Appendix A
A.1. Transformation of amplitudes to different frames
Here we list the matrices M() defined in Eq. (20) that map the invariant projections of
the hadronic tensor to the set of helicity amplitudes (WTL , WL , W , W ). They are
1
0
0
0
2
MCS =
0
1
2
1
4 cos2 CS
4 sin CS1 cos CS
2
CS
1+cos
2
8 cos CS sin2 CS
1
4 cos2 CS
1
4 sin CS cos CS
2
CS
1+cos
2
8 cos CS sin2 CS
4 cos12
CS
0
13 cos2 CS
8 cos2 CS sin2 CS
566
0
1
cot GJ
MGJ = 0
1
2
1+cos2
GJ
2 sin2 GJ
0
0
0
1
sin2 GJ
0
0
2 sin1GJ
cos GJ
sin2 GJ
(A.1)
for the CollinsSoper and the GottfriedJackson frame, respectively. CS denotes the angle
between P 1 and the Z axis in the photon rest frame and GJ is the angle between P 1 and
P 2 . Obviously 2CS + GJ = and in terms of hadronic Mandelstam variables we have
s
s
Q2 S
Q2 S
,
sin CS = 1
,
cos CS =
2
2
2
(Q T )(Q U )
(Q T )(Q2 U )
2Q2 S
,
(Q2 T )(Q2 U )
s
2Q2 S
(Q2 T )(Q2 U )
1.
sin GJ =
(Q2 T )(Q2 U )
Q2 S
cos GJ = 1
(A.2)
A.2. Single-scattering results

Here we list the results for the hard parts H of single-scattering helicity amplitudes
in the CollinsSoper frame. They are defined in Eq. (26). We distinguish annihilation and
Compton processes.

2

Q2 u
1
Q t
q+q
q+q
+
,
(Q2 t)2 + (Q2 u)2 ,
H L = C1
HTL = C1
tu
Q2 u
Q2 t
s

Q2 t
1 q+q
Q2 s Q2 u
q+q
q+q
2
,
H = HL ,
H = C1
tu Q2 t
Q u
2
(Q2 s)2 + (Q2 t)2
u (Q2 + s)2 + (Q2 t)2
q+g
,
HL = C2
,
st
s (Q2 t)(Q2 u)
s
1 q+g
Q2 u 2(Q2 t)2 (Q2 u)2
q+g
q+g
,
H = HL ,
H = C2
st
(Q2 t)(Q2 u)
2
q+g
HTL = C2
g+q
HTL = C2
s
g+q
= C2
(Q2 s)2 + (Q2 u)2

,
su
g+q
HL
Q2 t 2(Q2 u)2 (Q2 t)2

,
su
(Q2 t)(Q2 u)
= C2
t (Q2 + s)2 + (Q2 u)2

,
s (Q2 t)(Q2 u)
1 g+q
g+q
H = HL .
2
(A.3)
C1 = CF /Nc = 4/9 and C2 = 1/(2Nc ) = 1/6 are essentially the color factors.
A.3. Double-hard results
DH for double-hard
In this section we give the hard parts HTL of the projections WTL
processes. They do not depend on the frame.
567

2
2 S
2 (Q2 T )
1 (Q2 T ) + (2 S)2
+ 2
=
,
+
27
2 S
Q T
6 2 S + T Q2

Q2 T
1 2 S + T Q2
DH,qg+g
+
=
HTL
36
Q2 T
2 S + T Q2
DH,qg+q
HTL
1 (Q2 T ) + (2 S + T Q2 )
,
16
(2 S)2

2 S
2 2 S + T Q2
DH,q q+g
+
=
HTL
27
2 S
2 S + T Q2
1 (2 S)2 + (2 S + T Q2 )
,
2
6
(Q2 T )

2
2
2
2 (2 S)2 + (2 S + T Q2 )
(Q2 T ) + (2 S + T Q2 )
=
+
2
27
(2 S)2
(Q2 T )
+
DH,q q+q
HTL
DH,q q+
q
HTL
4 (2 S + T Q2 )
,
+
81 2 S(Q2 T )

2
2
2 (2 S)2 + (2 S + T Q2 )
(Q2 T ) + (2 S)2
=
+
2
27
(2 S + T Q2 )2
(Q2 T )
(2 S)2
4
,
81 (Q2 T )(2 S + T Q2 )

2
2
2
2 (Q2 T ) + (2 S + T Q2 )
(Q2 T ) + (2 S)2
=
+
2
27
(2 S + T Q2 )2
(Q2 T )
DH,qq+q
HTL
+
DH,q q+q
0
HTL
DH,qq 0 +q 0
HTL
DH,qq 0 +q
HTL
(Q2 T )
4
,
81 2 S(2 S + T Q2 )
2
2 (2 S)2 + (2 S + T Q2 )
,
2
27
(Q2 T )
2 (Q2 T ) + (2 S + T Q2 )
,
27
(2 S)2
2 (Q2 T ) + (2 S)2
.
27 (2 S + T Q2 )2
(A.4)
A.4. Soft-hard results

SH
SH
SH
SH
, TL1
, TL2
and TL12
Here we list the hard parts H (T ) of the invariant projections TTL
of the hadronic tensor, defined in Section 2.2, for softhard processes. The results can be
projected into a specific frame via Eq. (20) and the matrices given in Appendix A.1. This is
mathematically trivial but it complicates the expressions substantially. For this reason we
give only the invariant projections. Note that especially HTL = H (TTL )/2 for every frame
described in Section 2.2.
568

4 (Q2 T )(xc xa )
2xa 2 S
xc (2 S + T Q2 )
+
+
,
27 xc (2 S + T Q2 )
(xc xa )(2 S + T Q2 ) (Q2 T )(xc xa )
8
xa 2 S
qg+q
=
,
H TL1
27 xc (Q2 T )

8
2 S + T Q2
xa xc S((xc + xs )S + U Q2 )
xs Q2 S
qg+q
=
H TL2
2
2
2
27 (xc + xs )S + U Q (Q U )
xa xc
(Q2 U )2
qg+q
H TTL
xa
S2
xa xc (Q2 U )2

(xc + xs )((xc xs )(Q2 T ) + 2 (Q2 U ) xc 2 S) (xc xs )Q2
,
2 S + T Q2

8
xa
2 S 2
qg+q
=
H TL12
2
27 xa xc (Q T )(Q2 T )
(xc xs )(Q2 T ) + 2 (Q2 U ) Q2 xc 2 S

2 S + T Q2

(xc + xs )(2 S + T Q2 ) + 2 ((xc + xs )S + U Q2 ) Q2 + xc 2 S
xa s
+
,
xc (Q2 U )
2 S + T Q2

1
xc 2 S
2xa (2 S + T Q2 ) (xc xa )(Q2 T )
qg+g
=
+
,
H TTL
8 (xc xa )(Q2 T )
2 S(xc xa )
xc 2 S
qg+g
1 xa (2 S + T Q2 )
,
4
xc (Q2 T )

1
xs Q2
(xc + xs )S + U Q2
qg+g
=
H TL2
4 xc (Q2 U )
2 (Q2 U )
H TL1
xa (xc + xs )S + U Q2 (xc xs )(Q2 T ) + 2 (Q2 U ) Q2 xc 2 S

xa xc
2 (Q2 U )
Q2 U

xa xc ((xc + xs )S + U Q2 )(2 S + T Q2 ) + (Q2 T )(Q2 U ) Q2 S
,
xa xc
2 (Q2 U )2

1
xa 2 S + T Q2 (xc + xs )(Q2 T ) 2 (Q2 U ) Q2 + xc 2 S
qg+g
H TL12 =
4 xa xc (Q2 T )
2 (Q2 U )

xa (xc + xs )(2 S + T Q2 ) + 2 ((xc + xs )S + U Q2 ) + Q2 xc 2 S
,
+
xc
2 (Q2 U )

1
2 S
2Q2 (Q2 T )(xc xa ) 2 S + T Q2
qg+g
=
,
H TTL
18 2 S + T Q2
2 S
xc2 2 S(2 S + T Q2 )
2 xa (xa xc )
qg+g
=
,
H TL
9
xc2

1 Q2
(xc + xs )S + U Q2
xs S
S
qg+g
=
+
H T
2
2
2
9Q U
2 (Q U )
(Q U ) 2 S + T Q2
(xc S + U Q2 )((xc + xs )(2 S + T Q2 ) + 2 ((xc + xs )S + U Q2 ) + Q2 xc 2 S)

xc 2 (Q2 U )(2 S + T Q2 )
569

2xs Q2 ((xc + xs )S + U Q2 )
,
xc2 2 (Q2 U )(2 S + T Q2 )

1 xa 2xc 2 (Q2 U ) + xc (xc + xs )(Q2 T ) + xc2 2 S + 2x2 Q2
qg+g
=
H T
9 2 xc
Q2 U

2 ((xc + xs )S + U Q2 )(2Q2 + xc (Q2 T )) + 3xc Q2 (Q2 T ) + 2Q4 + xc2 22 S 2
.
(Q2 U )(2 S + T Q2 )
References
[1] J.C. Collins, D.E. Soper, G. Sterman, in: A.H. Mueller (Ed.), Perturbative Quantum Chromodynamics, World Scientific, Singapore, 1989, and references therein.
[2] M. Arneodo, Phys. Rept. 240 (1994) 301.
[3] J.P. Boymond et al., Phys. Rev. Lett. 33 (1974) 112.
[4] J.W. Cronin et al., Phys. Rev. D 11 (1975) 3105.
[5] D.M. Alde et al., Phys. Rev. Lett. 64 (1990) 2479.
[6] G.T. Bodwin, S.J. Brodsky, G.P. Lepage, Phys. Rev. D 39 (1989) 3287.
[7] R. Baier, Y.L. Dokshitzer, A.H. Mueller, S. Peigne, D. Schiff, Nucl. Phys. B 484 (1997) 265.
[8] M. Luo, J. Qiu, G. Sterman, Phys. Rev. D 49 (1994) 4493.
[9] M. Luo, J.W. Qiu, G. Sterman, Phys. Rev. D 50 (1994) 1951.
[10] X. Guo, Phys. Rev. D 58 (1998) 036001.
[11] X. Guo, Phys. Rev. D 58 (1998) 114033.
[12] R.J. Fries, B. Mller, A. Schfer, E. Stein, Phys. Rev. Lett. 83 (1999) 4261.
[13] S.A. Bass, B. Mller, W. Pschl, J. Phys. G 25 (1999) 109.
[14] C.S. Lam, W. Tung, Phys. Rev. D 21 (1980) 2712.
[15] K.J. Eskola, K. Kajantie, J. Lindfors, Phys. Lett. B 214 (1988) 613.
[16] K. Geiger, B. Mller, Nucl. Phys. B 369 (1992) 600.
[17] K. Geiger, Phys. Rept. 258 (1995) 237.
[18] L. McLerran, R. Venugopalan, Phys. Rev. D 49 (1994) 2233; Phys. Rev. D 49 (1994) 3352.
[19] X. Guo, J. Qiu, X. Zhang, hep-ph/9911476; hep-ph/9912361.
[20] NA10 Collaboration, S. Falciano et al., Z. Phys. C 31 (1986) 513.
[21] J.S. Conway et al., Phys. Rev. D 39 (1989) 92.
[22] C.S. Lam, W. Tung, Phys. Rev. D 18 (1978) 2447.
[23] J.C. Collins, D.E. Soper, Phys. Rev. D 16 (1977) 2219.
[24] E. Mirkes, Nucl. Phys. B 387 (1992) 3.
[25] J. Cleymans, M. Kuroda, Nucl. Phys. B 155 (1979) 480.
[26] A.H. Mueller, J. Qiu, Nucl. Phys. B 268 (1986) 427.
[27] E683 Collaboration, D. Naples et al., Phys. Rev. Lett. 72 (1994) 2341.
[28] FNAL E866 Collaboration, M.A. Vasilev et al., Phys. Rev. Lett. 83 (1999) 2304.
[29] H.L. Lai et al., Phys. Rev. D 51 (1995) 4763.
[30] M. Glck, E. Reya, I. Schienbein, Eur. Phys. J. C 10 (1999) 313.
[31] K.J. Eskola, V.J. Kolhinen, C.A. Salgado, Eur. Phys. J. C 9 (1999) 61.
[32] A. Brandenburg, S.J. Brodsky, V.V. Khoze, D. Mller, Phys. Rev. Lett. 73 (1994) 939.
[33] K.J. Eskola, P. Hoyer, M. Vanttinen, R. Vogt, Phys. Lett. B 333 (1994) 526.
[34] A. Brandenburg, O. Nachtmann, E. Mirkes, Z. Phys. C 60 (1993) 697.
[35] D. Boer, Phys. Rev. D 60 (1999) 014012.
[36] S.R. Tonse, J.H. Thomas, in: X.-N. Wang (Ed.), Pre-Equilibrium Parton Dynamics, Lawrence
Berkeley Laboratory Report No. LBL-34831, 1993, p. 341.
570
[37] B.Z. Kopeliovich, A.V. Tarasov, A. Schfer, Phys. Rev. C 59 (1999) 1609.
[38] B.Z. Kopeliovich, A.V. Tarasov, A. Schfer, hep-ph/9908245.
[39] U.A. Wiedemann, M. Gyulassy, Nucl. Phys. B 560 (1999) 345.

The nonperturbative color Meissner effect

in a two-flavor color superconductor
G.W. Carter a , D. Diakonov b
a Department of Physics and Astronomy, SUNY Stony Brook, NY, 11794-3800 USA
b NORDITA, Blegdamsvej 17, DK2100 Copenhagen , Denmark
Received 31 January 2000; accepted 17 May 2000
Abstract
Color superconductivity in QCD breaks the SU(3) color gauge group down to SU(2), inducing
masses in five of the eight gluons. This is a dynamical Higgs effect, in which the diquark condensate
acts as the vacuum expectation value of a composite scalar field. In order to analyze this effect at
low quark density, when gaps are large and generated nonperturbatively, we use instanton-induced
quark interactions augmented with gauge-invariant interactions between quarks and perturbative
gluons. The five gluon masses are found from the static limit of the relevant polarization operators,
in which transversality is maintained via the NambuGoldstone modes of broken color symmetry.
Working in the microscopic theory we calculate these masses to one-loop order and estimate their
density dependence. Finally, we speculate that the Meissner effect may postpone the onset of color
superconductivity to higher matter density than estimated previously. 2000 Elsevier Science B.V.
All rights reserved.
PACS: 11.15.Ex; 11.15.Kc; 12.38.Lg
It has recently been demonstrated that the color symmetry of Quantum Chromodynamics
(QCD) might be spontaneously broken through the formation of a diquark condensate. This
phase is metastable at zero and small matter density [1] but becomes stable and replaces
the usual chiral-broken phase at some critical density [2,3].
Condensation of diquarks is analogous to Cooper pairing of electrons in BCS theory, in
which the massive photon is the microscopic embodiment of the Meissner effect. In QCD
the gauge group is SU(3), but the consequences are similar: gauge bosons (gluons) become
massive due to their interaction with the correlated fermions (quarks). Not every gluon
species becomes massive, since the symmetry breaking pattern is SU(3) SU(2) and the
three gluons of the residual SU(2) remain massless. Among the five massive modes, four
are degenerate and the mass of the fifth is 4/3 times that of the others. This is comparable
E-mail addresses: carter@tonic.physics.sunysb.edu (G.W. Carter), diakonov@nordita.dk (D. Diakonov).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 0 9 - 6
572
G.W. Carter, D. Diakonov / Nuclear Physics B 582 (2000) 571590
to the symmetry-broken phase of the electroweak theory. Thus the direct analogy is to the
Higgs mechanism, with the diquark field forming a composite scalar and its nonzero VEV
generating a dynamical Higgs effect.
In this paper this mechanism is analyzed microscopically. As the source of diquark
condensation we will use the effective quark action obtained by averaging over instanton
configurations [4,5]. In this approach to the QCD vacuum, all information from the large
background gauge fields (instantons) is encoded in the non-local form of the t Hooft
interaction and the two instanton parameters, average size () and number density (N/V ).
This procedure replaces dynamical gluons with classical, nonperturbative field solutions,
ignoring quantum gluonic fluctuations. The reintroduction of gluonic fluctuations will
require a gauge-invariant perturbative modification of the t Hooft vertex. A direct
consequence of gauge invariance is the transverse polarization operator, which we will
compute explicitly. This is the main result of the paper as from it we obtain the gluon
masses.
Since the dynamical Higgs effect is interesting regardless of density, we will first
consider the vacuum alone (chemical potential = 0), although in this case the
superconductor is only metastable [1]. The gluon masses are found to be proportional to g
where is the superconducting gap, the scale of which is a nonperturbative one set by the
strength of the instanton background. Extending the formalism to finite chemical potential
is straightforward, given that the density dependence of the propagators and effective
action are known [6]. However, in this paper we avoid exact calculations at 6= 0. Instead,
for simplicity we assume that the in-medium gluon masses are determined by the behavior
of g() up to and somewhat above the critical value of 300 MeV. At this density, the
point at which the color superconductor becomes the stable phase, we find that the gluon
masses are about 120 MeV.
For gluons the color-breaking Meissner masses are a primary ramification of the
superconducting state, however this is not the only effect of a quark medium. There will be
additional contributions, among them a Debye screening mass of the order of g, which do
not break color symmetry. When the density becomes large, the Debye mass increases and
instantons are screened out of the picture. Yet diquark condensation persists, now due to
perturbative gluon exchange [7], and at asymptotically large density the Meissner masses
are also proportional to g [8,28]. So while one still has quark pairing, any matching
between low and high density mechanisms remains unclear.
In this paper we consider the case of two massless flavors, which is expected to be
relevant at finite-density chiral restoration given a relatively large strange quark mass [9,
10]. Instantons are Euclidean pseudoparticles, and thus all calculations will be in Euclidean
space.
In Section 2 we recall those features of the ordinary Higgs mechanism (based on
elementary scalar fields) which will also be relevant for composite scalars. In Section 3
the instanton-induced action including perturbative gluons will be formulated. In Section
4 the diquark gap equation is reviewed, and in Section 5 the color current is determined.
In Section 6 we sum a set of quarkquark correlation functions to recover the Nambu
Goldstone modes of the theory, a necessary exercise as these will mix with the longitudinal
573
gluons. These ingredients are assembled into the gluon polarization operator in Section 7,
which taken in the static limit corresponds to a mass. Section 8 compares the result to chiral
symmetry breaking, and in Section 9 this result is discussed and conclusions are drawn.
1. Higgs mechanism
The dynamical Higgs mechanism, though technically more involved, does not differ
much from the ordinary one. Let us denote the quark bilinear combination

f i gj
(1)
= fg ij L L + (L R) ,
where Greek letters denote color, f, g = 1, 2 flavor, and i, j = 1, 2 are spinor indices of the
left (L) and right (R) components of the quark field . The resulting complex field is
a Lorentz scalar isoscalar field belonging to the fundamental representation of the SU(3)
group. To support gauge invariance it must couple to the gauge potential via the covariant
derivative, ( ) = igAa (a /2) , in which the a denote the eight Gell-Mann
matrices. The kinetic energy term for the composite diquark field is the usual
L = Z1 | |2

g
1
= Z + i Aa a a
2

g 2 a b a b
A A .
+
4
(2)
The only difference with the standard case of the elementary field is that there is, in
principle, a common wave function renormalization factor, Z . For elementary fields
Z = 1 at the tree level, however, it deviates from unity even for the elementary field when
one takes into account the perturbative virtual emission of particles. For the fields that are
composite from the start there is no reason for Z to be unity. This quantity is a priori
unknown and should be determined from a dynamical calculation, as will follow below.
We stress, however, that the relative weights of the three terms in Eq. (2) are fixed by
gauge invariance.
If the scalar field develops a nonzero VEV signaling the diquark condensation,

(3)
= 20 3 ,
(it can be always arranged along the third color axis and made real), then gluons obtain a
mass matrix
a = b = 1, 2, 3,
0, 1

2 Z 2 , a = b = 4, 5, 6, 7,
2
2 1 2 a b 3
2g
(4)
Mab = 2g Z 0 3 =
0
8 2 1 2
g
Z
,
a
=
b
=
8.
0
3
The symmetry breaking pattern is, thus, SU(3) SU(2); three gluons corresponding to
the unbroken SU(2) subgroup remain massless, four gluons obtain masses proportional the
Higgs VEV, and the fifth gluon is 4/3 times as heavy. This relation will be, of course,
574
Fig. 1. Diagrams contributing to the gluon polarization operator in the Higgs effective theory.
reproduced for composite Higgs fields as well, as it follows from symmetry considerations
alone.
This simple elementary-Higgs model also provides an alternative way to find the gluon
masses, which we will generalize to the case of the composite Higgs field. One can
compute the gluon polarization operator 5ab
(q), as seen in Fig. 1. In Fig. 1a we take
the linear coupling of the gauge potential Aa , as given by the second term in Eq. (2), and
iterate it twice. The intermediate state in this diagram is the would-be NambuGoldstone
boson, which contributes to the polarization operator

1
a Z

(NG) ab
q Z1 g b
(q) = 2 Z g q
5
q2
3
q q
(5)
= 2g 2 Z1 20 2 a b 3 .
q
This contribution is purely longitudinal.
Fig. 1b is the contact term; it gives a Kronecker delta contribution:
3
(q) = 2g 2 Z1 20 a b 3 .
5(contact)ab
Combining the two we get a transverse polarization operator,

3
q q
2 1 2
(q)
=
2g
Z
a b 3 .
5(full)ab
2
q
(6)
(7)
The fact that it is transverse reflects the color current conservation, even in case of a broken
symmetry.
In the case of a composite Higgs field the tree diagrams of Fig. 1 will be replaced by
loop diagrams, formed by quarks, however, the general setting will be rather similar to this
simple case: there will be a pole contribution from the would-be Goldstone bosons in the
intermediate state, and there will be a contact term. The resulting polarization operator
will be transverse, provided one takes a conserved color current.
2. Gauge invariant effective action

Following a long tradition of instanton phenomenology, we assume the following:
Instantons are the dominant nonperturbative contribution to low-energy QCD, and
Low and high momentum scales are safely separable for quark and gluon fields.
These assumptions are supported a posteriori by instanton-based phenomenology of the
vacuum, as reviewed in Ref. [11]. Instantons break chiral symmetry spontaneously and
axial U (1) anomalously, both at a satisfactory magnitude. Instanton-induced interactions
also provide the necessary qq attraction for a diquark condensate, nonperturbatively.
575
Here we use a formulation which not only relies on the separation of quark zero and
free modes, but also explicitly includes both instantons and perturbative gluons. Gluons
are separated into classical instantons and quantum corrections, which we write as
X
X
)
A(I
A(I ) (x) + a (x),
(8)
A (x) =
(x) +
where the sums are over all instantons and anti-instantons, each given by the t Hooft
solution in the singular gauge [12]. Although there is a certain distribution of instanton
sizes (see, for example, [13]) we simply use the average value, ' 1/3 fm, in all
calculations.
The main idea of this paper is to examine the effects of color superconductivity on the
gluonic excitations above the instanton vacuum, which means we must retain and analyze
the aa . Hereafter, the term gluon will refer to the perturbative gauge fluctuation above
the instanton background, aa .
The origin the t Hooft interaction is explained in the literature [12] and will not
be repeated here. To summarize, the low momentum quarks are approximated by zero
mode solutions in the presence of one instanton. Averaging over the instanton ensemble
generates a vertex for dynamical quarks, mediated by the zero modes, which can be treated
perturbatively when the instanton liquid is reasonably dilute. The relevant small parameter
is the ratio of average instanton size () to the average inter-instanton spacing (R). From
phenomenological [14], variational [15], and lattice calculations [13] one finds
1
' .
(9)
R 3
More details follow in Section 4.
We now consider the effective action itself. In the quark sector it is a non-local one of
four-fermion operators, since quarks are connected to instantons via the quark zero modes
whose spatial extent is of the order of the instanton size, . Specifically, one has [4]:
Z
SINT = dU d4 z
Nf
Y

e f z, U )/
(xf z, U )(y
f (yf ) ,
d4 xf d4 yf f (xf )/
(10)
where U is the instantons 2 Nc color/spin orientation matrix and z is its position. The
(x) is the zero mode solution for fermions in the field of one instanton; in general it
depends on the chemical potential. At = 0 its Fourier transform is the form factor,

1
,
(11)
f (p) = 2x I1 (x)K0 (x) I0 (x)K1 (x) + I1 (x)K1 (x)
x
x=p/2
such that f (0) = 1. At 6= 0 the form factor is also known explicitly [6].
Because of non-locality Eq. (10) is not gauge invariant. When calculating bulk vacuum
properties one neednt worry about quantum corrections as observables are seldom
sensitive to them, however in this case the dependence is crucial. A limited literature does
exist in which non-local interactions are modified to be gauge invariant [1619] and we
576
follow the same procedure here. In particular, we will minimally modify the effective action
(10) to suit the present needs.
We will take the non-local four-fermion interaction as the starting point. It then
becomes a matter of multiplying each quark operator by a path-ordered exponential in
the background of the perturbative gluon field a , replacing as:
(x) (x)W (x, z),
!
Zx
g
a a
ds a ,
W (x, y) = P exp i
2
(12)
aa
is the perturbative field. As has been pointed out in the cited works, the choice
where
of path integrated over is not unique. Yet as long as these factors transform as
W (x, y) U (x)W (x, y)U (y)
(13)
by virtue of path ordering, the action will be gauge invariant. This remains true when, as in
this case, the interaction involves an explicit color average since overall color is conserved.
We are also concerned only with the static limit, q 2 0, in which results are independent
of any particular choice of path.
Thus we write the modified interaction as a product over flavors,
Z
SINT =
dU d4 z
Nf
Y
4
(xf z)
d xf d4 yf f (xf )W (xf , z)/
f

W (yf , z) f (yf ) ,
(yf z)/
(14)
noting that this includes terms of all orders in ga . Calculating the gluon polarization
operator will require the linear and quadratic contributions.
To this interaction term one must add the usual quark kinetic term, minimally modified
to preserve gauge invariance:

Z
g a a
4
(15)
SKIN = d x i + a .
2
Consequently, the color current obtained from the variation of the action in respect to aa
will have two contributions: one is the ordinary one arising from the minimal coupling (15)
and the other arising from the non-local interaction term (14).
3. Diquark condensation
In this section some details of the color superconductor phase are reviewed. All
expressions are Euclidean and the notation generally follows that of Ref. [6]; the reader
is referred to this reference for details specific to this particular approach. The spontaneous
breaking of chiral symmetry was also a central concern in that work, whereas here we are
considering restored chiral symmetry. The main result of Refs. [6,2023] is a competition
between chiral and diquark condensates at zero temperature and nonzero quark chemical
577
Fig. 2. SchwingerDysonGorkov diagrams for normal and anomalous quark propagators. The labels
refer to the functions in the text.
potential. In all cases, there is a phase transition from a low-density phase of spontaneously
broken chiral symmetry to a high-density one of color superconductivity.
While technically involved, our previous results arise from an effective four-quark
interaction which allows for pairing of quarks as in the BCS theory. The instanton model
retains more of QCDs features than ad hoc models, notably the anomalous breaking of
axial U (1) and transmutation of dimensions, and is constructed in a more systematic way
from that underlying theory.
A prominent feature of the instanton approach used here is the determination of the
effective four-fermion coupling constant . This constant is determined by a saddle-point
evaluation and proves to be nonlinearly dependent on the background instanton density,
N/V . A gap equation obtained from a set of SchwingerDysonGorkov equations are
solved to first order in , done self-consistently to determine the quark pairing gap, 0 .
Standard quark propagators are necessarily split by a nonzero 0 , since this introduces a
color bias and quarks are no longer color degenerate. This detail will be crucial when quark
loops are computed in Section 7.
Since this paper deals only with a phase of chiral-symmetric, diquark condensation the
SchwingerDysonGorkov equations and corresponding diagrams are the simple ones of
Fig. 2. We define the quark propagators as
( f

f i
g S1 (p)ij , , = 1, 2,
(16)
(p)gj (p) =
f
g S2 (p)ij , , = 3,
and the anomalous Gorkov propagator as

f i

f i
gj
gj
L (p)L (p) = R (p)R (p) = fg 3 ij F (p).
(17)
In these expressions indices f and g refer to flavor, i and j to spin, the Greek letters to
color, and L,R are chiral spinors. Written in the chiral L, R basis, the 4 4 propagator
S1 (p) is of the form:

0
Z(p)S0 (p)+
,
(18)
S1 (p) =
Z(p)S0 (p)
0
while S2 (p) is the free propagator for color 3 and hence identical to the above with the
, 1)
function Z(p) absent. The notation is x = x , where the 2 2 matrices = (iE
decompose the Dirac matrices into chiral components. The off-diagonal, bare propagator
is therefore written S0 (p) = [p ]1 .
With these definitions, we have the scalar SchwingerDysonGorkov equations
578
Z(p) = 1 (p)F (p),

(p)Z(p)
.
F (p) =
p2
(19)
The momentum dependence of the gap,

(p) = 0 f (p)2 ,
(20)
is given by the instanton-induced form factor, Eq. (11), which suppresses the interaction
beyond p > 1/ ' 600 MeV. This pair of equations leads to a gap equation for 0 ,
Z
2
d4 p
2
(p)F (p)
0 =
Nc (Nc 1) (2)4
Z
(p)2
d4 p
2
,
(21)
=
Nc (Nc 1) (2)4 p2 + (p)2
which must be self-consistently solved with . This coupling constant is in turn determined
by a saddle-point integration (exactly in the thermodynamic limit N, V ),
=
4Nc (Nc 1) 2
0 .
N/V
Combining these two equations may be eliminated and we obtain

1 Z
(p)2
d4 p
N
.
1=8
V
(2)4 p2 + (p)2
(22)
(23)
The scale of the gap, therefore, is set by the instanton density N/V . To this order N/V
remains at its vacuum value, and although it will be affected by the finite density of quarks
this is an O() correction to the instanton weight and not considered here.
Solved numerically, the gap is 0 ' 400 MeV in vacuum and drops to 0 ' 200 MeV
at the phase transition from chiral broken to color superconducting matter, as detailed in
Ref. [6]. 1
4. Color current
A color-conserving Noether current naturally follows from the modified interaction,
including contributions from both SKIN (15) and SINT (14). Along with the standard quark
gluon coupling piece,

Z
d4 p
a
a +
a
(24)
(p) L (p + q) + L (p) R (p + q) ,
j (q) =
(2)4 R
2
2
the current now includes a four-quark coupling to the gluon as shown in Fig. 3 and written
in terms of four-momentum:
1 In Ref. [6] was defined as half this papers (and the more standard) definition.
0
579
Fig. 3. Quarkgluon vertices of order g (left) and g (right) interactions.

SINT
aa (q) aa =0
Z
f1 f2 g1 g2
ix (p p 0 )+iyf (kf +kf0 )iz(pf0 +kf0 )
Die f f f
=
4
" Zz
g
g
ds eiqs Lf
(p1 )a M p10 , k10 L1 (k1 )Lf
(p2 )M p20 , k20 L2 (k2 )
1
2
a (q) =
x1
Zy1
+
ds eiqs Lf
(p1 )M
1

g
g
p10 , k10 a L1 (k1 )Lf

(p2 )M p20 , k20 L2 (k2 )
2
+ (L R).
(25)
Indices on quarks denote chirality and flavor. The measure is

Nf =2
D = dU d z
4
d4 xf d4 yf
d4 pf d4 pf0 d4 kf d4 kf0
(26)
(2)16
and the form factors lie in the color/spin matrices

l
ki
M(p, k)i
j = Uk j l U f (p)f (k).
(27)
The U are 2 Nc color orientation matrices, averaged in each vertex. Through use of the
Dirac equation for , one can explicitly verify that q J (q) = 0 for J = j + . While
this condition would remain satisfied with a transverse addition to the color current, no
such addition is motivated here.
In practice we are interested in pairing off two of the four quark legs of the second
vertex. Since we are considering a phase where chiral symmetry is unbroken, a chiralityviolating h i loop cannot contribute. Thus we pair either hi or h i and obtain
the effective current,
Z
d4 pf d4 pf0 d4 p0
0
d4 z d4 x
a (q) =
2
(2)12
"
Zz
ix(p1 p 0 )iz(p 0 +p2 )
e
ds eiqs L (p1 )a N (p2 , p0 )L (p2 )
x
+e
ix(p1+p 0 )iz(p 0p2 )
Zx
ds e
iqs
0 a
L (p1 )N (p2 , p ) L (p2 ) ,
+ (L R),
(28)
580
where here the flavor/color/spin structure is

N (p, p0 )ff
0 , 0 ,ii 0
= i ff 3 ii f (p)f (p0 ).
(29)
The two terms in Eq. (28) correspond to pairing either two incoming or two outgoing quark
legs of Fig. 3.
Gluon mass terms are found in the q 2 0 limit. When the gluon couples to nonsingular
composite quark modes one can set q 2 strictly to zero and the path integrals simplify to
e
ipi0 (zxi )
Zz

ds eiqs
xi
= eipi (zxi ) (z xi )
q=0
0
= i eipi (zxi )
0 .
pi
(30)
This substitution, the result of which clearly depends only on the endpoints of the path
s rather than any particular choice of path, leads to differentiation of the form factor. On
the other hand, the vertex can also couple gluons to NambuGoldstone modes which are
singular as q 2 0. Their 1/q 2 behavior must be countered by expanding a (q) to order
q , but before this complication arises the NambuGoldstone modes must be specified.
5. NambuGoldstone modes
A symmetry has been spontaneously broken and NambuGoldstone modes are certain
to follow. With SU(3) being broken to SU(2) there are five massless modes and, directly
analogous to the Higgs mechanism, they do not become additional degrees of freedom.
Instead they mix with and are incorporated into the five massive gluons, thereby relevant
in the color Meissner effect.
Since the associated condensate is hqqi, these massless modes will be quark bilinears.
To determine their quantum numbers one need only perform a gauge rotation on the
diquark condensate and catalog the five orthogonal correlators which appear. Recalling
the condensate direction as chosen in Eq. (17),
fiT
gj
(31)
iij fg L 2 L 0 ,
we can list the five diquark operators which couple to the NambuGoldstone excitations:
fkT 7
fg kl L
fkT 5
ifg kl L
gl
L ,
gl
L ,
fkT 7
gl
fkT 2
gl
ifg kl L
fg kl L
L ,
L .
fkT 5
fg kl L
gl
L ,
(32)
Propagators for these quarkquark modes will exhibit a simple pole at q 2 = 0, behavior
recovered by computing the corresponding correlation functions. In order to obtain
this pole we must sum s and u channel contributions to all orders in the fourquark
coupling . 2 Since connecting these modes to the gluon propagator will be itself of order
2 A related summation could be done for t channel diquark exchange as well, however these do not contribute
the dominant 1/q 2 behavior in the limit of zero transfer momentum.
581
Fig. 4. BetheSalpeterGorkov diagrams for summation of the quarkquark correlation functions.
g 2 , we do not include any gluonic corrections to the instanton vertex. With both standard
and anomalous quark propagators at our disposal we obtain the set of coupled Bethe
SalpeterGorkov equations diagramed in Fig. 4. The diagram shows not only standard
two-body propagators, denoted , but also its anomalous analog, . It is easy to see that
will vanish when any of the external lines are quarks of color 3, since the vertices
conserve color and the internal (Gorkov) propagators involve only colors 1 and 2.
Knowing the quantum numbers of the NambuGoldstone modes, we can immediately
write down anstze for the four-point functions which will be required when we later
compute the gluon polarization operator:

gj
g0 0 j 0 0
0
(k )
f i (p)f
0 0 i 0 (p ) 0 (k) 0
0
ff 0 gg ii 0 jj
[3 a ] 0 [ 3 b ]
Nf2 (Nc 1)

ab
0
(2)4 4 p + p0 k k 0 f (p)f (p0 )f (k)f (k 0 )
0 (p + p ),

f i
0 0 0
gj
g0 0 j 0
(p)f i (p0 ) 0 (k) 0 (k 0 )
0
(33)
ff gg ii jj
0
0
[3 a ] [3 b ]
Nf2 (Nc 1)

ab
0
(2)4 4 p + p0 + k + k 0 f (p)f (p0 )f (k)f (k 0 )
0 (p + p ).
(34)
Here, [3 ] = 3 , and the refer to L or R chirality with the chiral substructures defined
as

LL LR
LL LR
,
=
,
(35)
=
RL RR
RL RR
in which color indices have been suppressed. It is clear from their definitions that LR =
RL , LL = RR , LR = RL , and LL = RR .
Written in terms of these chiral elements, the two diagrams of Fig. 4 correspond to the
following set of equations:

ab
ab
ab
ab
(q) = 1 + c1a LR
(q)I1 (q) + c2a LR
(q)I2 (q) + c3a LL
(q)I3 (q) ,
LL

ab
ab
ab
ab
(q) = c1a LL
(q)I1 (q) + c2a LL
(q)I2 (q) + c3a LR
(q)I3 (q) ,
LR

ab
ab
ab
(q) = c3a LL
(q)I3 (q) + c2a LR
(q)I2 (q) ,
LL

ab
ab
ab
(q) = c3a LR
(q)I3 (q) + c2a LL
(q)I2 (q) ,
(36)
LR
582
where no sums are implied on color indices a, b. The adjoint color vectors {ci } determine
the internal propagators and are found to be
c1 = (0, 0, 0, 1, 1, 1, 1, 0),
c2 = (1, 1, 1, 0, 0, 0, 0, 1),
c3 = (1, 1, 1, 0, 0, 0, 0, 1).
The integrals which result from these loops are:
Z

1
d4 p p+ p
(p+ )(p ) Z(p+ ) + Z(p ) ,
I1 (q 2 ) 2
4
2
2
(2) p+ p
0
Z
2
d4 p p+ p
(p+ )(p )Z(p+ )Z(p ),
I2 (q 2 ) 2
2 p2
(2)4 p+
0
Z
2
d4 p
(p+ )(p )F (p+ )F (p ),
I3 (q 2 ) 2
(2)4
0
(37)
(38)
where we have defined (p ) = p 12 q and
.
(39)
Nc (Nc 1)
The set of BetheSalpeterGorkov equations (36) can easily be solved for correlations
functions of all adjoint colors. For colors a = b = 4, . . . , 8, in direct analogy to similar
calculations of pions in the instanton vacuum [24], the q 2 = 0 pole arises due to a
cancellation in the denominator which is a direct consequence of the gap equation. The
calculation follows identically for a = 4, 5, 6, and 7, where we find

44
44 2
.
(40)
q 2 = LL
q =
LR
1 2 I1 (q 2 )2

After writing the gap equation (23) in these terms,

1 (0),
1 = I
we have
44
LR
where
Z1

2

1 Z
22
2 2
I1 q
2,
= 1 I1 (0) q
q 2
q

I1 (q 2 )
= 2
q 2 q 2 =0

Z
1
d4 p (p)2 12 (p)p0 (p) + 14 p2 0 (p)2
= 2
(2)4
p2 [p2 + (p)2 ]
0
2

1
2
0
2 (p) (p) p (p)
.
p2 [p2 + (p)2 ]2
(41)
(42)
(43)
The primes denote differentiation with respect to p. The value of Z will change with
density, since not only does it depend on 0 but the integrand will exhibit functional
dependence on . In vacuum numerical evaluation gives Z1 = 8.07 103 .
583
For the 88 and 88 solving the coupled equations becomes more involved, since all
diagrams in Fig. 4 are present. One finds
1 2 [I2 (q 2 ) + I3 (q 2 )]2
Z
= 2,
q
88 2
88
(q ) = LR
(q 2 ) =
LR
(44)
where the second line is obtain by making use of the SchwingerDysonGorkov equations,
Eqs. (19), in the limit of small q 2 . This completes the set of NambuGoldstone modes. One
can naturally compute the correlation function for the additional diquark correlators with
a = b = 1, 2, 3, and find a crucial sign difference in the combination of integrals:
1 [I2 (q 2 ) I3 (q 2 )]2
1
= 2
.
q + m2

11
11
q 2 = LR
q2 =
LR
(45)
The mass can be determined from an O(q 2 ) expansion of the denominator, however this
will not be done here as for our purposes it is enough to verify the a = 1, 2, and 3 modes
are indeed massive.
6. Gluon polarization operator

With the gapped quark propagators, conserved current interactions, and composite
modes defined in the previous sections, we now compute the leading modification to the
gluon polarization operator. All diagrams prove to be color diagonal, and so we write

2
ab
2
(46)
5ab
q = 5 q .
Our interest is the static limit, q 2 0, which may be considered an effective mass.
In the presence of a color-3, scalar diquark the gluons are divided into three classes.
Gluons of adjoint colors 1, 2, and 3 belong to the residual SU (2) gauge group and as
such remain massless. Gluons 4, 5, 6, and 7 couple one gapped quark (of fundamental
color 1 or 2) with the ungapped species and share a degenerate mass. Gluon 8, diagonal
in fundamental color, obtains a mass proportional to the previous four. One polarization
operator from each class will be explained here to avoid unnecessary repetition. All
possible contributions to order g 2 are diagramed in Figs. 5 and 6 and, depending on the
gluonic species, some of these diagrams vanish and others combine to cancel in the static
limit.
We begin with the case of gluons 47, considering corrections to ha4 a4 i. Diagrams (5b),
(5f), and those involving (6c) require pairs of Gorkov propagators and thus vanish since
the quarks of color 3 (to which these gluons couple) cannot propagate anomalously. There
are four graphs which constitute the contact term proportional to .
Diagram (5a) is a standard loop, where one of the quarks is gapped (color 1 in the case of
gluon 4) and the other not (color 3). After subtracting off the vacuum part of this diagram,
584
Fig. 5. Quark loop diagrams contributing to the gluon polarization operator.
Fig. 6. Diagrams coupling the NambuGoldstone modes to gluons. Their sum squared contributes to
the polarization operator.
which remains a concern of gluon renormalization and is not relevant to the Meissner mass,
we find
Z

(p)2
d4 p
2
2
.
(47)
0
=
g
N
q
5(5a)
f
(2)4 p2 [p2 + (p)2 ]

The remaining integral is finite due to an power-law cut-off in the the function (p) arising
from the finite size of instantons.
Additional diagrams arise from the modified SINT . Not only does this generate additional
current interactions (29), but the interaction (14) itself contains a contact contribution to
the gluon two-point function. This, diagram (5c), is the second variation of the action (14)
with respect to the fourth gluon field,

2 SINT
2

0
=
.
(48)
q
5(5c)
4
4
a (q)a (q) q 2 =0
Evaluated to order g 2 it is
5(5c)
1
q 0 = g 2 Nf
4
2
d4 p 12 p2 0 (p)2 + (p)p2 00 (p)

.
(2)4
p2 [p2 + (p)2 ]
Diagrams (5d) and (5e), constructed with the additional current piece, are
Z
1 2
d4 p (p)p0 (p)
(5d) 2
,
5 q 0 = g Nf
4
(2)4 p2 [p2 + (p)2 ]
Z
1 2
p2 0 (p)2
d4 p
2
.
5(5e)
q 0 = g Nf
8
(2)4 p2 [p2 + (p)2 ]
(49)
(50)
Eqs. (47), (49), and (50) comprise the microscopic equivalent of the Higgs contact term
(6), and their sum is

2
5(5)
q 0
585
= g 2 Nf
d4 p (p)2 14 (p)p0 (p) 14 (p)p2 00 (p)2

.
(2)4
p2 [p2 + (p)2 ]
The integral can be trivially rewritten and then manipulated to yield

Z
d4 p (p)2 12 (p)p0 (p)
(2)4
p2 [p2 + (p)2 ]

(p)p2 00 (p)
(p)p0 (p)
= Z1 20 .
+
4p2 [p2 + (p)2 ] 4p2 [p2 + (p)2 ]
(51)
(52)
The final equality is achieved by integrating the second and third terms by parts.
The NambuGoldstone modes couple to the gluons as in Fig. 6. The construction
[(6a)+(6b)] 44 (q 2 ) [(6a)+(6b)] supplies the q q /q 2 piece to ensure transversality:
Z

Z
d4 p 12 (p)2 [(p) p0 (p)]2
(6) 2
2
5 q 0 = g Nf q q 2
q
(2)4
p2 [p2 + (p)2 ]2

(p)2 12 (p)p0 (p) + 14 p2 0 (p)2 2
p2 [p2 + (p)2 ]
q
q
= g 2 Nf 20 Z1 2 .
(53)
q
We have now accounted for all contributions to gluons 4, 5, 6, and 7.
Analysis of the eighth gluon follows in a similar fashion, although with terms from
every diagram. A superficial difference lies in the factors arising from the elements 8
which lead to a polarization 4/3 times the previous result. More subtle is the combination
of diagrams (5a) and (5b). For a8 this sums to 4/3 times Eqs. (47), whereas for a1,2,3
they cancel one another. This cancellation is only manifest to O(40 ) here, due to the fact
that the diagrams are constructed with a pair of one-loop, resummed quark propagators
as determined in Section 3. This generates a (p)2 term in each integrand denominator,
in essence including higher-order terms in the perturbative expansion (in ) which violate
gauge invariance. After the vacuum pieces are subtracted, these diagrams contribute the
following to a8 :

q2 0
5(5a+5b)

Z
2(p)2
1
p2
d4 p
1 2
2
= g Nf 0
3
(2)4 [p2 + (p)2 ]2 [p2 + (p)2 ]2 p2

Z
d4 p
(p)2
4
4
+
O(
)
(54)
= g 2 Nf 20
0 .
3
(2)4 p2 [p2 + (p)2 ]
For gluons 1, 2, and 3, the sign of the second term is changed and the corresponding
integral is of order 40 , and thus this quantity vanishes to the order to which we know
quark propagators, 20 . The same follows for pairs of diagrams with similar construction,
such as (5e) and (5f).
By combining Eqs. (46), (51), (52), and (53) we arrive at the satisfyingly compact
expressions
586
5ab
a, b = 1, 2, 3,
0,
g 2 N 2 Z 1 ab q q ,
2
a,
b = 4, 5, 6, 7,
f 0
q 0 =
q2
q
q
1
4
g 2 Nf 2 Z ab
, a, b = 8.
0
3
q2
(55)
Transversality requires that the contact term be proportional to the wave function
renormalization of the NambuGoldstone modes and this result, a Ward Identity for color
superconductivity, is recovered here.
Finally, to determine a numerical value for the masses we must fix the coupling constant
g. Evaluating in the instanton vacuum, one finds the large finite action [15]
S0 =
8 2
' 12,
g2
or g ' 2.6 and the perturbative expansion parameter s = g 2 /4 = 0.54.

The gluon masses squared are thus
0,
a = 1, 2, 3,
2 2 1
2
Ma2 = 2g 0 Z ' (150 MeV) , a = 4, 5, 6, 7,
8 2 2 1
2
3 g 0 Z ' (175 MeV) , a = 8.
(56)
(57)
These masses apply to the vacuum, = 0.

In order to estimate the finite-density behavior of the Meissner mass, we can simply take
the values of 0 for a given from the results of Ref. [6]. As detailed in that paper, the
instanton form factor (11) becomes density dependent and thus (p) should be replaced
E ). However, the changes in Z arising from the finite-
by a complicated (p4 , |p|,
modifications of the form factors are minor compared compared to the changes in the gap
magnitude, 0 ; we numerically estimate this correction to be about 3%. For simplicity
we therefore considered only the scaling from 0 () (taken from previous work [6]).
Fig. 7. Scaling of M, Z1 , and 0 with quark chemical potential . Each quantity is shown relative
to its vacuum value. For reference, diquark condensation is only stable in quark matter at and above
' 300 MeV.
587
1/2
Each non-zero gluon mass is proportional to 0 Z

(see Eq. (57)) and therefore all
will scale identically with density. The resulting gluon mass M(), as well as that of
the renormalization constant Z ()1/2 , is shown in units of its vacuum value in Fig. 7.
Note that although Z rises with increasing quark density, this effect is not sufficient to
overcome the falling gap and the masses continuously decrease. The first point of physical
relevance would occur around ' 300 MeV, the common prediction for chiral restoration
to a color superconductor. Results for matter at lower densities correspond to an unstable
solution and are only of academic interest [6,2023].
7. Similarity with chiral symmetry breaking

Apart from the gauge coupling g the gluon masses (57) are determined by the
2 22 Z 1 . This quantity is the analog of the F 2 constant in the chiralcombination Fqq
0
broken phase, both in its physical meaning and algebraically.
In the chiral-broken phase the NambuGoldstone bosons are pions. The correlation
A = A , in the massless quark limit has the
function of the axial current, j5
5
transverse form

A
q q
B
AB
2 AB
2 .
(58)
j5 (q)j5(q) = 5 (q) = F
q
The transversality of 5 is the consequence of the conservation of the axial current; the
1/q 2 pole is due to the pion in the intermediate state. Were there gauge bosons coupled to
the quark axial current their mass would be equal to F multiplied by the corresponding
gauge coupling. In case of diquark (vs. quarkantiquark) condensation the relevant currents
are color ones, and we obtain a similar form for the correlation function of two color
currents, Eq. (55). One needs only to multiply Fqq by the gauge coupling to deduce the
gluon mass.
If instantons are dilute, the leading contribution to the Kronecker part of the polarization
operator arises from the 5(4a)
piece, Eq. (47). It is the only contribution which diverges
logarithmically if one neglects the momentum dependence of the gap (p). For the same
2 in the dilute limit comes from the pieces
reason the leading contribution to the Z1 and Fqq
not containing the derivatives 0 (p). One has therefore in the dilute limit:
Z
220
d4 p
R2
(p)2
2
'
'2
log
,
(59)
Fqq
(2)4 p2 [p2 + (p)2 ] 8 2
2
where 0 is the superconducting gap at zero momentum. It follows from the gap
equation (23) that 0 N/V = /R 2 . Similarly, the axial correlation function (58)
computed in Ref. [24] gives (in the same approximation)
Z 4
4Nc M02
R2
M(p)2
d p
2
'
log
,
(60)
F ' 4Nc
(2)4 [p2 + M(p)2 ]2
8 2
2
where M(p) is the dynamical quark mass (the chiral gap) whose value at zero momentum
is determined from a corresponding gap equation to be M0 N/V Nc .
588
We see, thus, that not only have the Meissner mass and the F constant analogous
meaning, but their algebraic structure is quite similar. One expects, therefore, that the
numerical value for the Meissner mass is of the order of F , and this expectation is
confirmed by an exact numerical calculation of the previous section.

We have analyzed the problem of spontaneous gauge symmetry breaking brought
about by a diquark condensate. Since the broken symmetry is continuous and gauged the
resulting NambuGoldstone modes do not remain in the spectra, instead mixing with the
longitudinal components of the gauge fields to produce massive gauge bosons. This, a
dynamical Higgs mechanism, can be called the color Meissner effect in the context of
color superconductivity.
To reveal the gauge boson masses mathematically one has to compute the polarization
operator, which ought to be transverse,

q q
2 ab
(q)
=
M
,
(61)
5ab
q2
where the massless pole 1/q 2 is the manifestation of the NambuGoldstone intermediate
state. The coefficient, M 2 gives the mass of the gauge boson. In this paper, we explicitly
solved this problem for the case of diquark condensation as induced by the instanton
background, the effective action from which we found necessary to modify in order to
maintain a conserved color current.
Through computing the gluon polarization operators to order g 2 we find the effective
gluon masses to be on the order of the diquark gap. The three gluons comprising the
residual SU(2) group remain massless and hence a quarkgluon medium would become
color-biased in such a phase. The analysis here was done in the limit of zero temperature
and, initially, vanishing chemical potential , though strictly speaking at zero the color
superconductor is only metastable with the ground state being the usual chiral-broken
phase. We then estimated finite-density dependence of the calculated quantities. At the
critical density, where one expects the phase transition to the color superconducting phase,
we deduce that the Meissner masses of the 4th, 5th, 6th, and 7th gluon are about 120 MeV
and the 8th gluon has a mass about 140 MeV. These quantities would be of physical
relevance should a low temperature, high density region become experimentally accessible.
At a chemical potential low enough to leave the instanton background approximately
unchanged ( < 0.6 GeV), the instanton effects analyzed here would still be present and
likely dominant.
In computing the coupling of the NambuGoldstone modes to gluons we have
established that their mixing is described by an effective theory in which the composite
diquark is replaced by a complex scalar in the fundamental representation of the gauge
group, a . The effective Lagrangian coincides with that of elementary scalar field
covariantly coupled to gluons, aside from an overall factor Z which is the wave function
renormalization of the composite scalar field. There are no a priori reasons for this factor
589
to be close to unity (as it is in the case of a weakly-coupled elementary Higgs field), and
indeed we find a substantial deviation from unity. Meanwhile, it is a crucial factor for the
estimate of the Meissner mass.
If electromagnetic interactions were taken into account, the gluon a8 will invariably
mix with the (massless) photon. This mixing, estimated from general arguments to be
rather small [25], reorganizes the fields into a massive new gluon and massless new
photon. Given that the gluonic sector alone can be recast as a YangMillsHiggs theory,
this additional result would complete the analogy to the electroweak sector of the Standard
Model.
Finally, we would like to comment on the phase transition between ordinary chiralbroken phase and the color superconductivity. The point is that all estimates existing
in the literature (see Ref. [6] and references therein) indicate that this phase transition
happens alarmingly early: taken literary, the claim is that the interior region of a heavy
nucleus is actually in a boiling state. However, those estimates generally neglect the
influence of the dense medium on the gluonic background fluctuations which induce the
color superconductivity itself. The Meissner mass of about 150 MeV found here is a large
quantity and, together with the Debye mass and other effects, will suppress instantons.
Therefore, by taking into account back-influence effects one has a chance to save ordinary
nuclear matter from a premature phase transition by moving that transition to higher
densities.
Although the instanton density is not expected to change significantly for any chemical
potential below the inverse instanton size (600 MeV), at and beyond this point perturbative
Meissner and screening masses will become increasingly important. At asymptotically
large densities perturbative gluon exchange is the source of diquark condensation [7,8,26
28], whereas in this work we have been concerned with the low density, nonperturbative
regime. Determining the behavior at a moderate density scale which would be an
interpolation between the two is necessary if one wishes to confidently consider signals
of color superconductivity in any potentially realizable situation, be it in some future
heavy-ion collider or a neutron star.
Acknowledgements
G.W.C. thanks D. Rischke for discussions and both the Leon Rosenfeld Fund and
USDOE grant DE-FG02-88ER40388 for support.
References
[1]
[2]
[3]
[4]
D. Diakonov, H. Forkel, M. Lutz, Phys. Lett. B 373 (1996) 147.

M. Alford, K. Rajagopal, F. Wilczek, Phys. Lett. B 422 (1998) 247.
R. Rapp, T. Schfer, E.V. Shuryak, M. Velkovsky, Phys. Rev. Lett. 81 (1998) 53.
D. Diakonov, V. Petrov, in: G. Zinoviev, V. Shelest (Eds.), Hadron Matter under Extreme
Conditions, Naukova Dumka, Kiew, 1986, p. 192.
590
[5] D. Diakonov, in: A. Di Giacomo, D. Diakonov (Eds.), Selected Topics in Nonperturbative QCD,
Varenna, 1995, hep-ph/9602375.
[6] G.W. Carter, D. Diakonov, Phys. Rev. D 60 (1999) 016004.
[7] D. Bailin, A. Love, Phys. Rep. 107 (1984) 325.
[8] D.T. Son, M.A. Stephanov, Phys. Rev. D 61 (2000) 074012.
[9] M. Alford, J. Berges, K. Rajagopal, Nucl. Phys. B 558 (1999) 219.
[10] T. Schfer, F. Wilczek, Phys. Rev. D 60 (1999) 074014.
[11] T. Schfer, E. Shuryak, Rev. Mod. Phys. 70 (1998) 323.
[12] G. t Hooft, Phys. Rev. D 14 (1976) 3432; Phys. Rev. D 18 (1978) 2199, Erratum.
[13] A. Hasenfratz, C. Nieter, Phys. Lett. B 439 (1998) 366.
[14] E. Shuryak, Nucl. Phys. B 203 (1982) 93.
[15] D. Diakonov, V. Petrov, Nucl. Phys. B 245 (1984) 259.
[16] J.W. Bos, J.H. Koch, H.W.C. Naus, Phys. Rev. C 44 (1991) 485.
[17] R.D. Bowler, M.C. Birse, Nucl. Phys. A 582 (1995) 655.
[18] R.S. Plant, M.C. Birse, Nucl. Phys. A 628 (1998) 607.
[19] W. Broniowski, hep-ph/9909438; hep-ph/9911204.
[20] J. Berges, K. Rajagopal, Nucl. Phys. B 538 (1999) 215.
[21] R. Rapp, T. Schfer, E.V. Shuryak, M. Velkovsky, Ann. Phys. 280 (2000) 35.
[22] T.M. Schwarz, S.P. Klevansky, G. Papp, Phys. Rev. C 60 (1999) 055205.
[23] B. Vanderheyden, A.D. Jackson, Phys. Rev. D 61 (2000) 076004.
[24] D. Diakonov, V. Petrov, Sov. Phys. JETP 62 (1985) 204; Sov. Phys. JETP 62 (1985) 431; Nucl.
Phys. B 272 (1986) 457.
[25] M. Alford, J. Berges, K. Rajagopal, Nucl. Phys. B 571 (2000) 269.
[26] D.T. Son, Phys. Rev. D 59 (1999) 094019.
[27] M. Rho, E. Shuryak, A. Wirzba, I. Zahed, hep-ph/0001104.
[28] D.H. Rischke, nucl-th/0001040.

Features in the primordial power spectrum

of double D-term inflation
Julien Lesgourgues
SISSAISAS, Via Beirut 2-4, 34013 Trieste, Italy
Received 17 December 1999; revised 17 April 2000; accepted 16 May 2000
Abstract
Recently, there has been some interest for building supersymmetric models of double inflation.
These models, realistic from a particle physics point of view, predict a broken-scale-invariant power
spectrum of primordial cosmological perturbations, that may explain eventual nontrivial features
in the present matter power spectrum. In previous works, the primordial spectrum was calculated
using analytic slow-roll approximations. However, these models involve a fast second-order phase
transition during inflation, with a stage of spinodal instability, and an interruption of slow-roll.
For our previous model of double D-term inflation, we simulate numerically the evolution of
quantum fluctuations, taking into account the spinodal modes, and we show that the semiclassical
approximation can be employed even during the transition, due to the presence of a second inflaton
field. The primordial power spectrum possesses a rich structure, and possibly, a non-Gaussian spike
on observable scales. 2000 Elsevier Science B.V. All rights reserved.
Keywords: Early universe; Inflation; Primordial fluctuations; Phase transitions
PACS: 98.80.Cq; 04.62.+v; 05.70.Fh
1. Introduction
The standard inflationary prediction [13] concerning the primordial power spectrum
of scalar metric perturbations (denoted in the longitudinal gauge) is a simple powerlaw: k 3 h||2 i k n1 . However, for more than ten years, there has been some interest for
models now called broken-scale-invariant (BSI), predicting deviations from a power-law.
These models generally involve, in addition to the usual inflaton field, other (effective)
fields, driving successive stages of inflation, or just triggering phase transitions [411].
From the beginning, BSI models have been motivated by observations. After decisive
experiments (like the APM redshift survey and COBE), it appeared that the standard
cold dark matter model (sCDM), even with a power-law primordial spectrum, was not
in agreement with observations. Adding the BSI primordial spectrum of double inflation
could improve the situation [12,13], but later it was shown that this model cannot account
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 0 1 - 1
594
J. Lesgourgues / Nuclear Physics B 582 (2000) 593626
for small-scale cosmic microwave background (CMB) anisotropy measurements [14].

Nevertheless, interest for BSI models revived recently, when some authors found possible
evidence for a feature in the present matter power spectrum, around 120 h1 Mpc [1519]:
this is still a controversial and open question, which future redshift surveys should answer
precisely; anyway, in [20], it was shown that this feature could be associated with the BSI
step-like spectrum of an inflationary model proposed by Starobinsky [21] (combined with
a cosmological constant), while in [18], similar conclusions were reached with a spike
in the primordial spectrum. Finally, there has been a recent attempt [22] to connect BSI
models with a possible deviation from gaussianity in COBE data [23].
The step-like spectrum of [21], based on a nonanalyticity in the inflaton potential,
intends to be an effective description of some underlying, more complicated and more
realistic model. As noticed in [24], this underlying model should certainly involve more
than one scalar field, and some rapid phenomenon, such as a phase transition, occurring
approximately 60 or 50 e-folds before the end of inflation. Such models were recently
proposed [2527] in the context of global supersymmetry or supergravity. 1 The main
advantage of supersymmetric inflationary models is that they do not require very small
parameters, due to the existence of flat directions in the tree-level potential, and of small
one-loop quantum corrections in the effective potential.
In these works [2527], the power spectrum was calculated using analytic formulas, for
modes exiting the Hubble radius during one of the two slow-roll stages. Indeed, when
inflation is driven by a single slow-rolling scalar field, a simple expression relates the
spectrum of scalar metric perturbations, k 3 h||2 i, to the field potential (and its derivatives)
at the time of Hubble crossing for each mode. As we will see later, in some situations,
the analytic results can be extended to multiple slow-roll inflation. However, in more
complicated situations, it is necessary to calculate the primordial spectrum from first
principles, i.e., to follow numerically the evolution of the mode functions of field and
metric perturbations: this amounts in integrating a system of second-order differential
equations. At the end of inflation, the mode functions of give the primordial spectrum.
Here, we want to perform such a numerical simulation, in order to check the analytic
results, and to obtain the shape of the primordial spectrum on intermediate scales, for
modes exiting the Hubble radius between the slow-roll stages. For these scales, and also
on larger scales, the calculation appears to be very interesting, and reveals some unexpected
features.
The supersymmetric double inflationary models of Refs. [2527] are, in the vocabulary
of inflation, double hybrid models, 2 involving two inflaton fields, and two trigger fields,
driving second-order phase transitions: one between the two inflationary stages, and one at
the end of the last stage. Since we will focus on scales crossing the Hubble length around
the intermediate phase transition, the second trigger field, the one ending inflation, will
1 Also, while this work was being completed, there has been an interesting proposal for generating features in
the primordial power spectrum through resonant fermion production during inflation [28].
2 Double hybrid or double supersymmetric inflation had been introduced earlier [29,30], not in the context of
BSI primordial spectra, but as a solution to the problem of initial conditions for hybrid inflation.
595
not be important. 3 So, our study involves three fields: the inflaton field A, in slow-roll
during both stages of inflation; the inflaton B, in slow-roll during the first stage only, and
then performing damped oscillations; finally, the trigger field C, stable in a local minimum
during the first stage, and then performing oscillations around the true minimum. In fact, C
is a complex Higgs field, charged under a U (1) gauge symmetry, and the phase transition
describes the spontaneous symmetry breaking of this U (1). The inflatons A and B are also
complex fields, but their phases are not affected by the evolution, and are fixed from the
beginning; so, we can treat them as real fields.
This fast second-order phase transition, triggered by the non-slow-rolling field C, and
taking place approximately 50 e-folds before the end of inflation, while the field A is still
in slow-roll, is a new situation for the calculation of the primordial power spectrum.
Indeed, in the context of inflation, second-order phase transitions have been first
considered by Kofmann and Linde [6] and Kofmann and Pogosyan [8], but for models
with two fields instead of three: an inflaton and a trigger field. Then, during a short
stage, corresponding to the beginning of the phase transition, the mass of the trigger field
becomes negative, and adiabatic/isocurvature perturbations are exponentially amplified.
This mechanism is often called spinodal instability. It results in the formation of a large
and narrow spike in the primordial spectrum, together with the appearance of topological
defects, which are diluted later if the second stage of inflation is sufficiently long [6]. These
models were recently studied in more details, in the context of supernatural inflation [33],
and natural hybrid inflation [34], and connected with primordial black hole production.
However, with two fields only, there is not as much freedom as in our model (in particular,
because the trigger field must be light enough for driving a second inflationary stage). Then,
the spike in the primordial spectrum produces enormous density fluctuations (/ 1),
allowed only on very small scales; so, the second stage of inflation cannot last more
than 30 e-folds. In our model, since we have a third field driving the second stage of
inflation, we will observe for natural parameter values a moderate spike, which exceeds
the approximately flat power spectrum only by a factor of order one. This is typically the
kind of feature that can improve current fits to the observed matter power spectrum [18],
and in contrast with previous models, it is worth calculating precisely the spike shape and
amplitude.
Second-order phase transitions during inflation have also been recently studied in
details by Boyanovsky, de Vega, Holman and collaborators (see for instance [3537]).
However, in these works, the Higgs field is in slow-roll, and drives inflation: it is
a very interesting situation, involving strongly nonlinear evolution, and requiring some
nonperturbative approach. The stage of spinodal instability leads to the emergence of
a stochastic homogeneous background from the coarse-graining of super-Hubble scales,
like in stochastic inflation [38]. We will see that in our model, such features are avoided,
3 However, the second phase transition will produce cosmic strings [31], which can be as important for the
formation of structure as the primordial power spectrum itself [31,32]. In current studies, these mechanisms are
completely separated, and here we consider the calculation of the primordial spectrum only. We are grateful to
Rachel Jeannerot for pointing out this fact.
596
and the usual perturbative semiclassical approach can be employed (again, because we
have an inflaton A supporting inflation during and after the transition).
Obviously, the reason for which double hybrid inflationary models did not get a lot
of attention before, is that without supersymmetric motivations, they appear as quite
nonminimal, with many free parameters (some of them being tuned to unnaturally small
values). On the other hand, if one plays with superpotentials and tries to build doubleinflationary models, double hybrid inflation arises as one of the simplest possibilities. One
can imagine many variants of double supersymmetric inflation: the potential can be of
the type of double F-term inflation [25], double D-term inflation [26], or mixed; special
supergravity effects can arise [27]; different gauge groups can be broken [31]; different
assumptions can be made about the field charges. In the following, we will consider only
our previous double D-term model, and neglect supergravity corrections, 4 but most of our
results would arise in the other variants, for some (natural) parameter values.
Which kind of novelty do we expect from this study?
First, since slow-roll conditions are abruptly violated during the transition (so
that the Hubble parameter H will depend as much on the kinetic energy of the
oscillating fields B, C, as on the potential energy of the inflaton field A), we should
devote a special attention to the evolution of super-Hubble wavelengths during the
intermediate stage. Previous studies [2527] assume an analytic expression for the
amplitude of large wavelength perturbations (those which exit the Hubble radius
P
during the first stage), of the type k 3 h||2 i V i (Vi /Vi0 )2 (see the next section
for definitions and references), where the sum runs over slow-rolling fields: in our
case, A and B. Strictly speaking, this formula was originally derived and employed in
the context of multiple slow-roll inflation [4]. We want to test this expression in our
case; we will see in fact in Section 3 that the violent slow-roll interruption induces a
significant deviation from the analytic prediction (with our choice of parameters, by
a factor 4.3 for k 3/2 h||2 i1/2 ).
Second, interesting physics could emerge from the description of the Higgs field
dynamics during the transition. At the very beginning of the phase transition, the
Higgs field has to roll away from an unstable equilibrium point. For the purpose of
the simulation, one may think that it is sufficient to put by hand an arbitrary small
value for the Higgs zero-mode, and then let this zero-mode roll down. Actually this
is completely misleading. The final result would depend strongly on the initial small
value (which controls the duration of the spinodal stage). In fact the Higgs dynamics
has to be studied carefully, taking into account, for the Higgs modulus |C|, the
exponential amplification of large wavelength modes due to spinodal instability, and
the emergence of a homogeneous background out of small quantum perturbations. We
will see that these features have a characteristic signature on the primordial spectrum:
a non-Gaussian spike.
4 These corrections are important when the fields take very large values, close to the Planck mass. In particular,
this may be the case at the beginning of inflation [27,39,40]. In our model, with the small gauge coupling constants
that we will consider, supergravity corrections are subdominant.
597
Let us now start working. The details of our model and its potential are given
in [26]. However, a numerical simulation of the transition would be impossible without
a continuous expression for the one-loop corrections around the critical value Bc , at which
symmetry breaking starts. This expression has been calculated for toy models of F-term
and D-term inflation in [41], and it is straightforward to generalize it to our case. The
complete effective potential is given in the appendix, together with some remarks on its
motivations from string theory. We choose a particular arbitrary set of parameters, also
given in the appendix.
In Section 2, we recall for non-experts the results usually expected in multiple slow-roll
inflation.
In Section 3, we treat the problem in the simplest possible way, i.e., neglecting the Higgs
quantum perturbations. Even at this level, we find that the amplitude of the large-scale
plateau exceeds the analytic prediction.
In Section 4, we discuss the Higgs dynamics at the very beginning of the transition,
around the unstable equilibrium point.
In Section 5, we go to the next level of complexity for the primordial spectrum
calculation, and take into account the Higgs longitudinal perturbations (still neglecting the
vacuum degeneracy, which is expected here to be irrelevant). We use the usual framework
of the semiclassical approximation, and carefully justify this approach.
2. Well-known results in multiple slow-roll inflation

Let us recall some well-know results about quantum perturbations in multiple slow-roll
inflation. The matter of this section can be found in other papers, but we included it for
clarifying the results and notations of the next sections. We follow essentially Polarski and
Starobinsky [42] (see also [43]), and we devote a particular attention to a subtle problem
(the definition of the Vi s), which is of particular importance in our case.
2.1. Semiclassical equations
In absence of a phase transition, there are some well-known techniques for the
calculation of the power spectrum. Usually, the slow-rolling fields i , i = 1, . . . , N , are
decomposed into a classical zero-mode plus quantum perturbations:
Z
d3 k
i (k, x, t).
(1)
i (x, t) = i (t) +
(2)3/2
The field perturbation i , and the metric perturbation (in the longitudinal gauge) can
be expanded over an N -dimensional basis of annihilation operators a j (k), j = 1, . . . , N ,
satisfying canonical commutation relations [a j (k), a j0 (k)] = jj 0 (k k0 ):
X
eikx ij (k, t) a j (k) + eikx ij (k, t) a j(k).
(2)
i (k, x, t) =
j =1,...,N
This expansion defines a set of mode functions ij and j (from now on, we omit
all arguments k and t). The basis is chosen in such way that at some initial time, the
598
perturbation i lies along a i . Commutation relations for the fields yield the Wronskian
normalization conditions for the mode functions, valid at any time:
X

ij ij ij ij = ia 3 ,
j =1,...,N

j j j j =
j =1,...,N
i
4ak 2m4P
2i
(3)
i=1,...,N
(we are using the reduced Planck mass: mP = (8G)1/2 = 2.4 1018 GeV). The
background equations of motion are:

V
1
1 X 2
= 0, with H 2 =
V
+
(4)
i + 3H i +
i
i
2
3m2P
i=1,...,N
(of course, even if not written explicitly, the potential and its derivatives are evaluated
P
at (1 , . . . , N )). They imply the useful relation H = 1 2 i 2i . The perturbation
2mP
equations of motion read:

2

X
k
2V
2V
i j 2 V j
+
=
4
j 0 j ,
ij + 3H ij +
ij
i
i j 0
a2
i2
j 0 =1,...,N
j + H j =
1
2m2P
i ij .
(5)
i=1,...,N
Initial conditions for the field mode functions inside the Hubble length (when k aH )
are given by WKB solutions of Eqs. (5) (with j neglected) normalized with (3):
R
a 1
ij = ij ei
2k
k
a
dt
(6)
Equivalent conditions for the metric mode functions are given at the next order in aH /k:
j = i
j
ei
2
mP (2k)3/2
k
a
dt
(7)
The primordial spectrum is defined as the power spectrum of , evaluated outside the
Hubble radius, during matter domination:
X
|j |2 .
(8)
k 3 h0|||2 |0i = k 3
j =1,...,N
In fact, in order to obtain the shape of the primordial spectrum, it is not necessary to follow
the evolution of until matter domination, but until the end of inflation. Indeed, at this
time, all the modes observable today are outside the Hubble radius, and are dominated by
the so-called growing adiabatic solution, which evolution is independent of k. In other
P
words, the primordial spectrum k 3 j |j |2 changes only by an overall normalization
factor between the end of inflation and matter domination.
599
2.2. Analytic solutions

We summarize here the analytic solutions for multiple slow-roll inflation with uncoupled
inflaton fields. It is useful to note that in any case (with or without slow-roll), and for each
value of j , two solutions of Eqs. (5) have an analytic expression outside the Hubble radius
(when k a H ):
!
!
Zt
Zt
ij
H
1
H
0
0
e
(9)
=
a dt + Cej .
Cj a dt Cj ,
j = Cj 1
i
a
a
a
0
The mode with coefficient Cj is the growing adiabatic mode, the one with Cej the decaying
adiabatic mode. Other solutions (2N 2 isocurvature modes for each j ) have no generic
e If one
expressions. There is also a freedom to add to j an arbitrary decaying mode D/a.
looks only for slowly-varying solutions, the system (5) simplifies in:
2

k
2V
V
1 X
+
=
2
,
H
=
(10)
i ij ,
3H ij +
ij
j
j
i
a2
i2
2m2P i
and admits only N slowly-varying independent solutions (for each j ). Now, during slowroll stages, we have:
3H i +
V
= 0,
i
H2 =
V
.
3m2P
In this case, the N slowly-varying solutions can be found analytically [42]:

P

Cj
ij
i 0 d i 0 j Vi 0
2H
dij ,
=
H
V
i
P

d
H
i 0 d i 0 j Vi 0
.
j = Cj 2 H
H
dt
V
(11)
(12)
The solution with coefficient Cj is the growing adiabatic mode (as can be seen by
comparing with Eq. (9)) and the modes with coefficient dij are the isocurvature modes.
For fixed j , only N 1 out of the N coefficients dij are independent, and the definition
is complete only once an arbitrary constraint has been chosen (for instance, d1j = 0). The
system (12) can be inverted in:
P
ij
Cj
1 X
Vi
i 0 d i 0 j Vi 0
.
(13)
ij ,
dij =
+
Cj = 2
2
V /i
2H i 2H
V
mP
i=1,...,N
A crucial point, to which we will come back later, is that the quantities Vi are not uniquely
defined. The condition under which the former solution is valid during slow-roll reads:
i, dVi /dt = (V /i ) i . So, the Vi can be found by integrating V /i over time or
directly over i , but there is still a freedom to add a constant term to each Vi . This freedom
is equivalent to an ambiguity in the splitting between adiabatic and isocurvature modes.
Indeed, we see from Eqs. (12) that adding constants to the Vi s is equivalent to changing the
definition of Cj and dij . So, to remove this freedom, we must give a more precise definition
of isocurvature modes during inflation. In fact, isocurvature modes can be clearly identified
600
at the end of inflation, from the fact that their contribution to j vanishes [42,44]. Using
this as a definition, it is possible to calculate the Vi s in a unique manner, provided that
the whole evolution of background fields is known, till the end of inflation (in contrast
with single-field inflation, for which Cj depends only on background quantities evaluated
at t = tk , the time of Hubble radius crossing).
The most widely-studied multiple inflationary models (for instance, those with a polynomial potential and no coupling terms) share the following property. When one field exits slow-roll, its time-derivative, which is decreasing, becomes small with respect to the
other ones ( i 0, Vi cst). So, the simplified background equations (11) and the solutions (12) hold continuously at any time. Simply, at each transition between two stages, the
perturbation equation for the field exiting slow-roll becomes irrelevant (the perturbation
vanishes), but the other solutions remain valid. In this case, we have a simple prescription
for defining the Vi s. 5 In the last stage of single-field inflation, driven by one inflaton 1 ,
only the adiabatic mode contributes to j , by definition. So, using Eq. (12), we see that we
must have V1 = V , and Vi6=1 = 0. This fixes the integration constants: at any time, we can
obtain the Vi by integrating backward on the field trajectory, expressed as (1 , . . . , N )
Ti (i ), from a point at the end of the last slow-roll stage Ti (i end ), back to an arbitrary
point:
Zi
Vi = 1i V Ti (i end ) +
i end

V
Ti (i ) d i .
i
(14)
So, at each transition, one potential Vi goes to zero, and in the last stage only the
growing adiabatic mode remains. The interesting quantity at the end, which is C 2
P 2
j Cj (since at the end of inflation the primordial spectrum defined in Eq. (8) is equal
to k 3 C(k)2 (H /H 2 )2 ), can
be immediately found by applying Eq. (13) at tk . Indeed, at this
time, ij ' ij H (tk )/ 2k 3 , and
C =
2
X
j =1,...,N
Cj2

2
Vj
H 2 (tk ) X
=
.
2k 3 m4P j V /j t =tk
(15)
On the other hand, for models with violent (waterfall) transitions, i.e., for which nonslow-rolling fields acquire large time-derivatives with respect to V , the solution of Eqs. (12)
applies separately to different slow-rolling stages. Between these stages, the background
and perturbation evolution can be quite complicated. This makes it very difficult to define
properly the Vi s, and to distinguish between the adiabatic mode and the isocurvature
modes (excepted, of course, during the last slow-rolling stage). So, the standard result
of Eq. (15) may be difficult to employ, due to the impossibility of defining the Vi s
analytically. The model considered in this paper is an illustration of this problem, which,
to our knowledge, had not been addressed previously.
5 I am very grateful to D. Polarski and A. Starobinsky for providing this prescription.
601
3. Spectrum neglecting Higgs perturbations

We will now try to calculate the primordial power spectrum numerically. For a pioneering work on this kind of numerical simulation, see Ref. [9], in which several cases of
multiple field inflation are treated.
The simplest way to study the phase transition is to neglect completely the Higgs
perturbations, and deal only with the inflaton perturbations (Aj , Bj )j =1,2 . In other
words, we consider C as an exactly homogeneous field: C = Cei , and we solve the

following equations:
+ V = 0, idem for B, C,
+ 3H A
A
A
2 + B 2 + C 2
A
,
3m2P H 2 = V +
2
2

2
2
2 V V B ,
j + k + V Aj = 4A
j + 3H A
A
j
j
j
2
2
A
A B
a
A
2

2
2
j + k + V Bj = 4B j 2 V j V Aj ,
j + 3H B
B
a 2 B 2
B
A B

1
A Aj + B Bj .
j + H j =
2m2P
(16)
Note that the small coupling term 2 V /A B arises only at the one-loop order, and when
|B| < Bc . At the beginning of the phase transition, C is sitting at the origin, which becomes
an unstable equilibrium point at the critical e-fold number N = Nc . So, we have to push
it away, and introduce an initial value at Nc , invoking quantum fluctuations as a physical
justification. In contrast with the results of Section 5, the simplified calculation of this
section does not depend much on this initial value, provided that it is small enough and
does not introduce a discontinuity. We take however for C the expectation value of the
coarse-grained quantum fluctuations, that will be calculated in the next section:
2 1/3

mc
2 1/2
H
C = |Cquantum|
.
(17)
N
After Nc , (B, C) rapidly evolve towards the equilibrium point (0, C eq ). In fact, it is useful
to distinguish five stages:
(i) first slow-roll stage: A and B are in slow-roll, C = 0.
(ii) spinodal stage: while C grows from the initial value given above to a critical
value C spinodal(B), the effective mass square m2c is negative: this is spinodal
instability. This stage will have a characteristic signature only in Section 5, when
C perturbations will be taken into account. In our model this stage lasts 0.15
e-folds.
(iii) quasi-static transition: B and C roll down to the minimum. This evolution is fast
with respect to the Universe expansion ( 0.8 e-folds in our model) and (B, C) are
far from being in slow-roll. However, we call this transition quasi-static, because
602
for the parameters chosen here (in particular ' 103 ), C remains close to the
valley of its local minima, 6 and performs small oscillations around the elliptical
trajectory C 2 = C 2eq (1 B 2 /B 2c ) on which V /C = 0.
(iv) second stage with oscillations: only A is in slow-roll, but the fields (B, C) are not
stabilized in their minimum and their oscillations affect the background evolution.
at any time, but, if we average over one
|A|
In our model we find that |C|
oscillation, |B| |A| during approximately 15 e-folds. This stage is inflationary

(|H | H 2 ), but differs from a usual slow-roll stage since the evolution of H is
driven by the oscillating field B, not the inflaton field A.
and we go back to ordinary
|A|
(v) second stage with slow-roll: at some point, |B|
single-field slow-roll inflation, during approximately 35 e-folds.
The slow-roll analytic prediction for the primordial spectrum is given by Eq. (15), and
at first (tree-level) order, the definition of (VA , VB ) resulting from Eq. (14) and from the
potential (Eq. (46)) is:
VA ' V ( Aend ) '
ZB
VB '
2 g 2 ( )2
gA
B
B A
,
2
2
2 gA + gB
1 2
B C 2 dB
2
(18)

B2
2
with C 2 = max C eq
1
,0
B 2c
2
2
2
(gA A + gB B )
during the first stage,
2
2
2(gA + gB )

2
= 1
2B 2C 2 1 B
during the transition,
eq
4
2B
0
during the second stage.
(19)
Note that at any time V = VA + VB . This is due to the fact that the definition (14) implies
R
V = VA + VB + VC , and if C remains in the valley of minima, VC = V
C dC ' 0.
Let us now compare the slow-roll predictions with the results of the numerical
simulation. We will first discuss in details the results for two modes: on with k < a(tc )H (tc )
and one with k > a(tc )H (tc ).
3.1. Large wavelength results
We integrate the equations for a mode that crosses the Hubble radius around N = 53,
while Nc = 50 (the origin for N has been chosen so that the last inflationary stage ends
at N = 0). The results for |1 | and |2 | are shown on the first plot of Fig. 1. Since the final
power spectrum will depend on the final value of |1 |, which is much bigger than |2 |
6 So, we do not deal with the case in which C undergoes a chaotic stage with large oscillations. This would
arise with 102 6 6 1. In this case, we do not have a robust justification for the expression of one-loop
corrections [41], and everything becomes more complicated.
603
Fig. 1. Evolution of the metric perturbations, when Higgs perturbations are neglected: (top) |1 |
and |2 | for a long wavelength mode; (middle left) splitting of |1 | in adiabatic and isocurvature
contributions for this mode; (middle right) same for |2 |; (bottom) |1 | and |2 | for a small
wavelength mode. The transition starts at Nc ' 50, while the last inflationary stage ends at N ' 0.
The long wavelength mode exits the Hubble radius at N = 53, the small wavelength mode at
N = 44.
604
during the last stage, it is important to understand the evolution of this quantity, and to
compare with slow-roll predictions.
In first approximation, the evolution of 1 can be reproduced analytically. Indeed, we
check that the perturbation B1 , though amplified exponentially during the transition,
remains a negligible source term for 1 and A1 , except during a short period (two e-folds)
during which it is comparable with other source terms. So, at first order, we can just neglect
B1 , and deduce the evolution of 1 and A1 from a system of two differential equations
instead of three. 7 Moreover, we look only for the slowly-varying solution, which obeys to:
A
A
1
.
(20)
2m2P
There is an exact solution (normalized to |A1 | = H / 2k 3 at t = tk ), that reproduces fairly

well the numerical solution:
3H A 1 +
2V
V
1 ,
A1 = 2
A2
A
H 1 =
Zt 2
H
A
dt,
exp
A1 =
3
m2P H
A(tk ) 2k
t
Zt 2
2
1
A
A
dt.
exp
1 =
2H
3
m
A(tk ) 2 2k
P
(21)
tk
2 (this is the
We must keep in mind that between tc and some time t1 , one has B 2 + C 2 > A
15 e-fold second stage with oscillations). On the other hand, during the second stage
2 , and Eqs. (21) read:
with slow-roll, we have 2m2P H = A
A
,
H
H
,
H2
Zt 2 2
H (tk ) V (tk )
B +C
exp
dt.
C1 =
V
m2P H
m2P 2k 3 A (tk )
A1 = C1
1 = C1
(22)
tk
Now, since the contribution to the last integral is made essentially for tc < t < t1 with
H ' (B 2 + C 2 )/(2m2P ), the last term with the exponential can be approximated by
H 2 (t1 )/H 2 (tk ), and then
C1 '
H (tk ) V (t1 )
.
m2P 2k 3 V
A (tk )
(23)
This approached result matches exactly the slow-roll prediction (15), because V (t1 ) is
nothing else but VA . In fact, we find that the numerical result (which takes into account the
2 + B 2 + C 2 , and the small correction due to the fact that B
exact relation 2m2P H = A
1
cannot be neglected during two e-folds) exceeds the slow-roll prediction by a factor 4.3.
7 Note that this simplification is appropriate for finding and A , but the exact adiabatic solution (9) does
1
1
not satisfy the reduced system (two equations). This shows that the solution for 1 and A1 must be a mixture
of adiabatic and isocurvature perturbations.
605
To conclude on the evolution of 1 , note that it is possible in principle to isolate

numerically the adiabatic and the isocurvature contribution. We subtract to 1 a solution
e in order to remove any
e We tune (C1 , Ce1 , D)
of the type of Eq. (9), and also, a solution D/a.
behavior proportional to 1/a or H /a, and to cancel 1 during the second slow-roll stage.
The result is unique and stands for the isocurvature mode. The second plot of Fig. 1 shows
the result of this decomposition. At the end of this formal exercise, 1 appears as the result
of an almost exact cancellation during the first stage between adiabatic and isocurvature
contributions, which are of opposite sign. We see that both modes are excited during the
transition, but the isocurvature mode decays like a 3 during the second stage and vanishes.
On the other hand, the solution for j = 2 differs completely from the standard slow-roll
prediction. We expect from Eqs. (15), (18) and (19) that C2 3C1 . It is in fact 103 times
smaller. Again, this result shows that the usual definition of (VA , VB ) is not valid during
and before the slow-roll interruption. This appears clearly when we separate numerically
the adiabatic and isocurvature contribution (third plot of Fig. 1). We see that during the
first stage, the isocurvature mode completely dominates.
In conclusion, even if C = (C12 + C22 )1/2 differs from the slow-roll prediction only by a
factor of order one, we have proved that the usual results of Eqs. (14) and (15) cannot be
employed in our model. The numerical results can be matched with Eqs. (13) and (15) by
tuning VA and VB during the first stage, but we do not know how to predict these values
analytically: this calculation seems to be difficult, and a numerical simulation is probably
unavoidable.
3.2. Small wavelength result
We now consider the modes that cross the Hubble radius well after the transition. For
the smallest wavelengths, this happens during the last single-field slow-roll stage, and the
usual results of Eqs. (15), (18) and (19) are automatically valid. In our model, these modes
correspond to very small wavelengths today, not observable on cosmic scales.
We are more interested in modes crossing the Hubble radius during the oscillatory
2 . The evolution of ( , ) for
period, 8 when the evolution of H is driven by B 2 A
1
2
such a mode, with Hubble crossing at N ' 44 (six e-folds after Nc ), is shown on the
last plot of Fig. 1. Since we are now in single-field inflation, 1 and 2 should be pure
adiabatic modes; we check that this is the case, by fitting (1 , 2 ) with the expression of
Eq. (9) (for N > 44). The coefficient C1 matches exactly the slow-roll analytic prediction,
and C2 C1 , as expected for a non-slow-rolling field perturbation.
This can be explained easily. During the transition, 1 remains a slowly-varying
solution, because the effective mass of A1 is dominated by k 2 /a 2 . So, at Hubble crossing,
1 is in the slowly-varying adiabatic mode, and C1 is exactly the same as in the slowroll prediction (i.e. Eq. (15) with i = 1). On the other hand, |2 | is strongly affected by the
8 Since the oscillations of B produce oscillations in the effective mass of C, one may expect that C modes
could undergo a stage of parametric resonance, that would show up in resonant metric amplification [48]. This is
not the case because the field C is very heavy, so that the relative amplitude of the mass oscillations is kept very
small.
606
Fig. 2. The primordial power spectrum of adiabatic fluctuations, k 3/2 C(k), or equivalently,
k 3/2 |(k)|, when the Higgs perturbations are neglected (with logarithmic scale and arbitrary units).
It is a step-like spectrum with superimposed oscillations on intermediate scales.
transition, because the effective mass of B2 becomes suddenly much larger than k 2 /a 2 . At
Hubble crossing, |2 | is in the decaying adiabatic mode, proportional to a 1 , and stabilize
with a value much smaller than |1 |, as expected for a non slow-rolling field perturbation
in single-field inflation.
3.3. Primordial power spectrum
We plot on Fig. 2 the power spectrum of adiabatic fluctuations, k 3/2 C(k), or equivalently
(up to a constant), k 3/2(k). It is a step-like spectrum with superimposed oscillations
on small scales, quite similar to the analytic spectrum of Starobinsky [21], but more
smooth (the amplitude of the first oscillations with respect to the amplitude of the step
is smaller). On intermediate and small scales, it is also close to the spectrum of [11] (from
double polynomial inflation), with the notable difference that on large scale we have an
approximately flat plateau, instead of a logarithmic dependence on k.
For our choice of primordial parameters, the step has got an amplitude p ' 10, and
spans over one decade in k space. As we saw before, this spectrum arises solely from
C1 , i.e., from metric perturbations calculated with initial condition of the type (A 6= 0,
B = 0).
4. Dynamics of the Higgs field

Until now, we have treated the Higgs field C as a simple classical homogeneous
background field, and put by hand an initial condition at N = Nc , in order to push the
field away from unstable equilibrium. We will now enter into more details concerning the
607
Higgs quantum fluctuations, in order to justify and calculate the initial condition at Nc (by
averaging over quantum fluctuations), and to prepare the work of the following section, in
which we will consider the effect of the Higgs quantum perturbations on the primordial
power spectrum. We rewrite the potential (Appendix, Eq. (46)) under the more suggestive
form:

2 Bc2 4
1
C .
(24)
V = V (A, B) + 2 B 2 Bc2 C 2 +
2
4
8Ceq
When B < Bc , the potential is a usual Mexican hat with respect to C.
4.1. Fluctuations during the first stage of inflation
We decompose the field C as in Eqs. (1) and (2), using now a three-dimensional operator
basis a j , j = 1, 2, 3. Since C is complex, we would need in principle a four-dimensional
basis, in order to quantify separately the real and imaginary part C1 and C2 . However,
in this section, the U (1) symmetry will still be preserved, and the mode functions are
identical for C1 and C2 . So, for concision, we introduce only one degree of freedom j = 3,
describing simultaneously both directions in the complex plane.
The Wronskian condition (3) applies also to the mode function Cj . During the first
stage of inflation, C = 0. So, in addition to Eqs. (16) (now taken for j = 1, 2, 3), we have
the following perturbation equation:
2

k
2V
+
(25)
Cj = 0,
Cj + 3H Cj +
a 2 C 2
with a WKB solution valid both within and outside the Hubble radius:
k2 2V
9
a 3/2 i R m dt
e
, with m2 2 +
H 2.
Cj = 3j
2
4
a
C
2m
(26)
During the first stage of inflation, at the first perturbative order considered here, 3
vanishes, and so do the nondiagonal mode functions (A3 , B3 , Cj 6=3 ). C3 is decoupled
from other perturbations and its evolution depends mainly on the mass 2 V /C 2 =
1 2
2
2
2 (B Bc ) that goes to zero at N = Nc .
4.2. Formation of inhomogeneities just after Nc
When N > Nc , the mass of C becomes negative, causing an exponential amplification
of the mode functions, known as spinodal instability. Let us restrict the analysis to the very
short stage after Nc , during which C3 remain small enough for the linear equation (25) to
be still valid. 9 As we can check a posteriori, this stage lasts 0.15 e-folds. If we rescale
C3 to a 3/2 C3 , we see that Eq. (25) can be rewritten:
9 This is the case if h0| |
d3 k C |2 |0i1/2 C ; then, the cubic term in V /C is negligible with respect

eq
k
(2 )3/2
to the linear term, and

Z
Z
2V
V
d3 k
d3 k
Ck '
(0)
Ck .
3/2
2
C
(2 )
C
(2 )3/2
608

+

k2 2V
9 2
H
+
= 0.
a 2 C 2 4
(27)
We will solve this equation under a few approximations, and compare our result with
an exact numerical simulation. Since we are interested in a short period after Nc , with
typically N Nc 6 0.2, it is appropriate to linearize the N -dependence of the effective
mass-term in Eq. (27):

d{ 2V /C 2 }
2V
2
2
= (N Nc ),

(28)
> 0,
C 2
dN
Nc

k2
k2
=
1 2(N Nc )
a 2 ac2
(in the particular model we are studying, can be easily found from the slow-roll condition
for B:
2 + g 2 )2
2 2 ln 2(gA
2 B V
A
B B
=
.
2
2
3H B
3H
16 2
Also, during this short stage, H is approximately constant and N Nc = H (t tc ). Under
these approximations, Eq. (27) reads:
2 =
2k H (t tk ) = 0 with
2k = 2 + 2
k2
ac2
tk = tc +
2k H

9 2
H
.
ac2 4
k2
(29)
(30)
The modes start growing exponentially at t = tk (for k = 0, t = tc 94 H /2 , but smaller

wavelengths start growing later). The correctly normalized solution of equations (29)
and (25) is given, up to an arbitrary complex phase, by:
r
(1)
(t tk )1/2 H1/3 2i3 k H 1/2(t tk )3/2 ,
(31)
C3 = a 3/2,
=
6
(1)
where H1/3 [z] is a Hankel function of the first kind. Indeed, for t < tk , this solution has an
asymptotic expression which is identical to Eq. (26). For t ' tk , one has:

1/3
3
2
3/2
,
(32)
C3 = a

30 23 k H 1/2
and for t > tk in very good approximation:
exp 2 k H 1/2 (t tk )3/2
C3 = a 3/2 p 3
2k H 1/2 (t tk )1/2

.
(33)
Is it possible to define an effective background from the coarse-graining of long wavelength

modes? The field C, coarse-grained over a patch of size l 2a/k, is a stochastic
classical 10 quantity. Its real and imaginary part obey to a Gaussian distribution (at least if
all modes are in the vacuum state), with variance hC 2 i1/2 computed from
10 The effective background can be considered as classical when large wavelengths modes have very large
expectation values. Then, observers measuring the value and momentum of C cannot feel the noncommutative
ZkcgZ
C
2
609
d3 k d3 k0
h0| C(k, x, t) C (k0 , x, t) |0i
(2)3
1
=
2 2
Zkcg
k 2 |C3 |2 dk,
(36)
where kcg is the coarse-graining cut-off. When t < tc , we know that scales outside and
around the Hubble radius have a k-independent amplitude (see Eq. (26)). So, hC 2 i scales
3 (intuitively, all modes have the same amplitude, but small wavelength have a bigger
like kcg
statistical weight, proportional to k 2 ). On the other hand, when t > tc , large wavelengths
start being amplified earlier; so, in the coarse-graining integral, there is a competition
between the term |C3 |2 and the term k 2 , and a scale will be privileged. More precisely, let
us use the asymptotic expression (33). The function k 2 |C3 |2 peaks around the scale 11

1/2
H
kmax = ac
,
(37)
2 N Nc
and only modes in the range 0.1kmax < k < 3kmax , which are within the Hubble radius,
contribute to the coarse-graining integral (36). A numerical simulation confirms this result
for 0.1 6 N Nc 6 0.2 (see Fig. 3). This means that just after Nc , the field C has an
inhomogeneous structure, and can be seen as effectively homogeneous only in regions of
typical size a/kmax .
Inside each patch, we can define an effective homogeneous background from the coarsegraining of large wavelength modes, and use the standard semiclassical approximation. The
constant phase in a given patch can be chosen randomly, while the squared modulus |C|
obeys a probability distribution that can be calculated exactly. Indeed, with any coarsegraining cut-off kcg 3kmax , we can compute the root mean square of the real and
imaginary part of C:

2 1/2
1 ac 3/2 20 H 1/4
C
=
exp 23 H 1/2(t tc )3/2 .
(38)
3/2
4
a
t tc
operator structure of the field, and the quantum field behaves as a classical stochastic field. More precisely, it is
know [45,46] that the modes can be approximated by classical stochastic quantities when their expectation values
are much bigger than the minimal expectation value set by Heisenberg uncertainty principle:

h9|y(k)
p (k) + c.c.|9i h9|[y(k),
p (k)]|9i = 1.
(34)
This condition on operators can be translated in terms of mode functions and read for the vacuum state:
i(C C C C).
|C C + C C|
(35)
For Eq. (26) (before Nc ) this means H mC , which does not hold for the non-slow-rolling field C; but when
the exponential amplification starts, the condition above is rapidly satisfied for spinodal modes.
11 To find this it is appropriate to use the approximation

3/2

2
k 2 |C3 |2 k 2 exp 4/3H 1/2 t tc k2 2
.
ac
610
Fig. 3. The left column shows the Higgs perturbation power spectrum: log |C3 |2 versus log k, for
different times chosen around the beginning of the transition, i.e. when the effective mass of C
becomes negative. From top to bottom, we have N Nc = 0.025, 0.075, 0.175. The vertical scale
is the same in the three plots. We show on the horizontal axis the particular values corresponding
to k = aH and k = a|m|, where m2 2 V /C 2 . We see that spinodal modes, with k 6 a|m|,
are amplified. On the right column, we plot the integrand of the coarse-graining integral k 2 |C3 |2 .
The vertical scale increases from top to bottom. It appears clearly that after a fraction of e-fold,
only modes with H < k/a < |m| contribute to this integral, which indicates the emergence of an
inhomogeneous background, with first-order homogeneity recovered on smaller scales (higher k).
On the last plot, we also show the scale kmax calculated analytically from Eq. (37).
Since C1 and C2 are Gaussian stochastic numbers with variance hC 2 i1/2 , then |C|2 obeys a
2 distribution and |C| has a mean value ' 3/2 hC 2 i1/2 and a variance ' 1/2 hC 2 i1/2
(the numerical factors can be found in tables; see, for instance, Eq. (26.4.34) in [47]). We
2 i1/2 is a solution of the zerosee explicitly, by comparing with Eqs. (32) and (33), that hC
2 1
2
1/2
(32 H )1/3 . Actually, the
mode equation with initial condition hC i (tc ) = 60 3
numerical simulation shows that the exact result exceeds this estimate by a factor 2.3, so
that the correct initial condition is
1/3
2 1/2
C
(tc ) = 0.12 2 H
.
(39)
So, we really have in each patch an effective homogeneous background, obeying for
N > 0.1 to the equation C + 3H C + V /C = 0, and with an initial 2 distribution of
probability.
611
On the other hand, on scales k 6 kmax , the background is essentially inhomogeneous,

and due to strong mode-mode coupling, the linear semiclassical approximation breaks soon
after Nc .
4.3. Later evolution and the homogeneous phase approximation
So, it is not possible to find an exact solution for the evolution of C during spinodal
instability and at later time. We should recall here that out-of-equilibrium phase transitions
and spinodal instability have been widely studied in condensed matter physics, and also in
the context of spinodal inflation, using either the large-N limit [35,36] (which is welldefined, but appropriate for the symmetry breaking of U (N), N ), or just a Hartree
Fock approximation [37] (which is more difficult to interpret, and cannot account so far
for the metric back-reaction). In such studies, the typical scale of homogeneity for the
coarse-grained background is larger than the Hubble radius (kmax < a(tc )H (tc )), while the
observable primordial spectrum arises from modes with k > a(tc )H (tc ). So, it is possible
to separate completely the modes contributing to the effective zero-mode assembly,
and those contributing to the primordial spectrum. In contrast, in our case, we need to
follow modes on scales k with k < a(tc )H (tc ) < kmax . So, at first sight, it seems that our
problem is much more complicated than spinodal inflation, due to the inhomogeneity of
the background. In fact, it is not, because we have an inflaton field A driving inflation even
during the transition, and we will see that a perturbative semiclassical approach can still be
employed. 12
The following observations will suggest an approximation under which we can continue
the primordial spectrum calculation:
since the causal horizon during the first inflationary stage is much bigger than the
Hubble radius, (A, B) are homogeneous on the largest scales observable today, Bc is
reached everywhere at the same time, and the phase transition is triggered coherently.
So, the modulus |C| should be approximately homogeneous, even on the largest
scales: we should be able to cast it into a zero-mode plus small perturbations;
the phase inhomogeneities should not be much relevant for the primordial spectrum
calculation. The existence of a negative mass, leading to exponential amplification,
only arise in the longitudinal direction |C|. The degree of freedom associated with
the complex phase is a Goldstone boson; it is eaten up by the gauge field, which
becomes massive. Starting from zero, this mass soon exceeds the Hubble parameter:
so, there cannot be large quantum fluctuations arising from this sector. The existence
of phase inhomogeneities is also linked to the formation of cosmic strings at the very
beginning of spontaneous symmetry breaking, separated at least by a characteristic
length > a/kmax . These strings could eventually contribute to the formation of
cosmological perturbations today, but as we said in the introduction, we do not include
such mechanism in this study.
12 Another difference with respect to spinodal inflation is that we will never coarse-grain or average modes
far outside the Hubble radius. This is just because the spinodal field is not slow-rolling, and then, modes are
exponentially amplified also on sub-Hubble scales.
612
So, for our purpose, it seems reasonable to neglect the phase inhomogeneities, forget
the vacuum degeneracy, and assume a homogeneous phase (x, t) = . In the following
section, we will recompute the primordial power spectrum, taking now into account the
Higgs longitudinal quantum perturbations, and we will find that these perturbations have
a characteristic signature on the primordial power spectrum.
5. Spectrum with Higgs longitudinal perturbations

We go on neglecting the vacuum degeneracy. We saw in Section 4.2 that just after
the transition, the modulus |C| has got a mean value 3/2 h|C|2 i1/2 and a variance
1/2 h|C|2 i1/2 that we calculated explicitly. Assuming a homogeneous phase is almost
equivalent to identifying this modulus with a real scalar field, canonically quantized, and
decomposed, following the usual semiclassical approximation, into a background field
C = 3/2 h|C|2 i1/2 , plus quantum fluctuations. The quantum mode functions should
be
matched with the quantities Cj that we already studied around Nc , divided by 2. With
these prescriptions, the mean value and the variance of the modulus previously studied
in the last section coincide exactly with those of the real scalar field introduced in this
paragraph. 13
This matching is the key point of our study. It is not exact, because, in a classical
stochastic sense, it is based only on the first two momenta of the 2 distribution: we
artificially gaussianize the fluctuations. However, the matching seems appropriate for
estimating the amplitude of the quantum fluctuations of the modulus |C|, even if any
information on a possible non-gaussianity is lost. Since in this approach we introduce a
(physically justified) zero-mode, we expect that the growth of spinodal modes will slowdown and terminate rapidly, and that the perturbative semiclassical approach will remain
valid. This will be justified explicitly in Section 5.3.
In summary, we simulate the usual background equations for (A, B, C, H ) and the
following perturbation equations:
2

2
k
2V
2 V V B ,
+
=
4
A
A
Aj + 3H Aj +
j
j
j
j
a 2 A2
A
A B
2
2
2
2
j + k + V Bj = 4B j 2 V j V Aj V Cj ,
j + 3H B
B
B
A B
B C
a 2 B 2
2

2
2
j + k + V Cj = 4C j 2 V j V Bj ,
j + 3H C
C
C
B C
a 2 C 2
1
A + B B + C C .
A
(40)
j + H j =
j
j
j
2
2mP
When we start following a mode Cj well inside the Hubble radius, we employ the initial
13 At least, when (N N ) > 0.1 (i.e., when h|C|2 i1/2 obeys the zero-mode equation, see Eq. (38)), and
c
(N Nc ) < 0.2 (i.e., when the equation for Cj remains linear, see Eq. (25)).
613
condition of Eq. (26) divided by a factor 2. At N = Nc , we put by hand a non zero-value
for C, given by Eq. (39) multiplied by a factor 3/2.

5.1. Large wavelength results
The evolution of (|1 |, |2 |, |3 |) for a mode crossing the Hubble radius at N ' 53
(i.e. during the oscillatory stage) is shown on the upper plot of Fig. 4. By subtracting
the previous solution (Fig. 1) to the new solution (Fig. 4), we check that the evolution of
|1 | is unchanged, except during a few e-folds after the beginning of the transition. The
2V
difference is due to the small (one-loop) coupling term A
B : when B1 is amplified, it
excites |1 |, but later on, this mode decays and does not leave an observable signature.
On the other hand, the negative effective mass of Cj and the large (tree-level) coupling
Fig. 4. We now take into account the Higgs longitudinal perturbations, and show the evolution of
the metric perturbations (|1 |, |2 |, |3 |), for (top) a long wavelength mode, crossing the Hubble
radius at N = 53, and (bottom) a small wavelength mode, crossing at N = 43.
614
between Bj and Cj generate a considerable amplification of |2 | and |3 |. At the

end, when the isocurvature modes and the decaying adiabatic mode can be neglected, the
remaining growing adiabatic mode is smaller than the one for |1 |, but only by one order
of magnitude.
So, on large scale, what is the observable difference between these results,
and those of
Section 3? First, since initial conditions for B2 are proportional to (a k)1 , we expect
the large-scale power spectrum for k 3/2 |2 | to be, as usual, approximately scale-invariant.
In other words, the contribution found for one mode will be the same for other largescale modes. So, the large-scale power spectrum will still be a flat plateau, but possibly
with a small deviation from gaussianity, caused by the small contribution of |2 |. As we
said before, our method does not allow a quantitative evaluation of this deviation, and
we leave this for further studies. Second, since initial conditions for C3 are independent
of the scale, and since the factor k 2 /a 2 is always subdominant in the effective mass of
C3 on large wavelengths, |3 | should be scale-independent on these wavelengths. So, the
contribution of k 3/2 |3 | to the power spectrum k 3/2h||2 i1/2 will be proportional to k 3/2 ,
and will peak for scales crossing the Hubble length during the transition. We therefore
expect an observable non-Gaussian spike, whose shape and amplitude will be calculated in
Section 5.4.
5.2. Small wavelength result
The evolution of (|1 |, |2 |, |3 |), for a mode crossing the Hubble radius at N ' 43
(i.e. during the oscillatory stage), is shown on the lower plot of Fig. 4. |1 | and |2 | are
identical to previous results of Section 3.2. |3 | has the same behavior as |2 |, but is even
smaller. So, for small wavelengths, including the Higgs modulus perturbation makes no
difference at all.
5.3. Consistency check of the semiclassical approximation
Before giving the primordial spectrum of perturbations, we will check here the
consistency of our approach, and the validity of the semiclassical approximation in our
case. Strictly speaking, the semiclassical approximation holds whenever for each field, the
KleinGordon equation

V
(x, t) = 0
g 1/2 g 1/2 g (x, t) +
(41)
can be cast into a background equation for the zero-mode, plus independent linear
equations for field and metric mode functions. The first term, with spacetime derivatives,
is linear provided that metric perturbations remain small:
2
Z

d3 k

|0i1/2 1.
(42)
h0|
3/2
(2)
The second term, with the potential derivative, is linear in two cases:
(i) obviously, when the perturbations are small with respect to the zero-mode:
615
Z
2

d3 k

h0|
|0i1/2 .
(2)3/2
Then, we can write:
V
2V
V
(x, t) '
()
+
()

d3 k
;
(2)3/2
(43)
(ii) also, when both the zero-mode and the perturbation expectation value are small
enough for nonquadratic terms in the potential to be neglected. Then, we have the
following linearization:
2V
V
(x, t) =
(x, t),
(44)
and 2 V / 2 is a constant.
We have seen that around N = Nc + 0.15, in the homogeneous phase approximation,
we can define a zero-mode and a set of perturbations. Since all these quantities are very
small, we are in the second case, and the semiclassical approximation is valid. But what
happens later? We know from Fig. 4 that one e-fold after Nc , the perturbations of C, B and
reach a maximum; if, at that time, the expectation values of the perturbations exceed the
homogeneous background fields (or, in the case of metric perturbation, exceed one), then
our approach is simply inconsistent.
To address this question, we first study the evolution of the power spectra of Cj , Bj
and j , for Nc < N < Nc + 1. The results for Cj are shown on three-dimensional plots,
in Fig. 5. This simulation is just the continuation of the one performed in Section 4.2, and
P
at low N we recognize the curves of Fig. 3. On the upper plot, we show j h|Cj |2 i, in
logarithmic scale, vs. k and N . We see that large scales (spinodal modes) are exponentially
amplified during the spinodal stage, and then perform coherent oscillations, driven by
P
the background evolution. On the lower plot, we show k 2 j h|Cj |2 i, in linear scale,
vs. k and N . We observe that the main contribution to the averaging integral (Eq. (36))
always arise from the same region in k space, centered around the approximately constant
wavenumber kmax . This observation is technically important, because it justifies the use of
finite limits of integration in the averaging integral. In particular, we see that we do not have
to worry with ultraviolet divergences. Provided that the upper limit of integration kcut-off is
chosen in a finite range (namely, around the largest k values shown of Fig. 5), then h|C|2 i
does not depend on this cut-off. Similar results are found for Bj and j .
Then, by integrating on k, we find the expectation values h|C|2 i1/2 , h|B|2 i1/2 , h||2 i1/2 ,
and plot them with respect to C, B and 1 (see Fig. 6). It appears that for N ' Nc + 0.20,
when C approaches the valley of minima (on the figure, we recognize this valley from
the small oscillations around it), i.e., when V /C becomes nonlinear, the quantum
perturbations are subdominant by one order of magnitude. When C reaches Ceq , the
quantum expectation value is smaller by three orders of magnitude. After, we know from
Fig. 4 that the field will become more and more homogeneous. For B, we see that the
perturbations are always subdominant (when B performs its first oscillation, it goes through
zero, but the perturbations are still quite small and we remain in the linear regime). For ,
616
Fig. 5. Before calculating the expectation value of quantum perturbations, and compare it with
background quantities, it is useful to follow the various power spectra, plotted here for C. The
upper plot, |C|2 (logarithmic scale) versus k and N , shows that spinodal modes are exponentially
amplified, and then oscillate coherently. The second plot, k 2 |C|2 (linear scale) versus k and N ,
shows that the main contribution to the averaging integral arises always from the same modes, with
k kmax ; this plot shows how to define the limits of integration in the averaging integral.
the maximal value is h||2 i1/2 103 . So, the semiclassical approach holds at any time
in the range Nc < N < Nc + 1, and therefore, until the end of inflation.
In conclusion, in this model, the semiclassical approximation can be employed during
the whole evolution, because there exists a time (N ' Nc + 0.15) at which the perturbations
have grown sufficiently to create an effective zero-mode (obeying the usual zero-mode
equation), but not too much, so that we are still in the linear region of the potential
derivative. It is at this time 14 that we have done the matching between the complex field
(with no zero-mode) and a real scalar field (with a zero-mode) standing for its modulus.
If such a time did not exist, complicated nonlinear approaches and further approximations
would have to be employed. This would be the case if the second inflaton field A was
14 This time can be clearly identified on the upper plot of Fig. 6: without the factor 3/2 (respectively 1/2)
in C (respectively h|C|2 i1/2 ), the curves would be tangent around N = Nc + 0.15.
617
Fig. 6. Consistency check of the semiclassical approximation. The expectation value of the quantum
perturbations h||2 i1/2 is generally much smaller than the background fields (than one in the
case of metric perturbations). When it is not the case, both quantities are very small and the linear
approximation is still valid. The three plots correspond to the case of C, B, and . The case of A is
not shown because we know from the beginning that this field remains extremely homogeneous. The
evolution is shown only until N = Nc + 1, which is the time of maximal inhomogeneity, as can be
seen on Fig. 4.
618
not supplying a constant potential energy, and supporting the Universe expansion. Indeed,
in that case, the evolution of the scale-factor and of the fields would become completely
stochastic, and it would be a hopeless challenge to keep track of modes crossing the Hubble
length before and during the transition.
5.4. Primordial spectrum including Higgs longitudinal perturbations
We show our final result on Fig. 7, for the parameters used previously (middle). On the
other plots, we show some primordial spectra corresponding to different values of . The
contribution of |1 | is unchanged with respect to Fig. 2, but we now have an additional
contribution from 2 and 3 on intermediate and large scales. Let us summarize the
characteristics of the power spectrum:
On small scales (large k), we have an approximately flat plateau, exactly identical
to the analytic predictions. Its amplitude can be found in our previous paper [26,
Eqs. (6) and (7)], as a function of the FayetIlliopoulos terms (A , B ), of the coupling
constants (gA , gB ) and of the duration of the second stage 1N (different values of
1N correspond to different choices for the inflaton fields initial values). The plateau
has got a tilt nS = 1 1/1N ' 0.98, like in usual (single-field) inflationary models,
when the potential derivative is provided by logarithmic one-loop corrections.
On intermediate scales (the value of these scales is given by 1N ), we find a spike, as
expected from Sections 5.1 and 5.2 (since k 3/2 |3 | is proportional to k 3/2 for small k,
and strongly suppressed for large k). This spike may be non-Gaussian, for reasons
that we already explained. Clearly, it is beyond the scope of this paper to quantify
the non-gaussianity, because the semiclassical approximation is not appropriate for
extracting the statistics of the modes. The amplitude and width of this spike with
respect to large scales depend essentially on . Indeed, is proportional to the Higgs
mass, and then, controls the duration of the transition: for low values, the transition
is long, exponential amplification is enhanced, and the spike is higher. Its width is
roughly of one decade in k space (but, since k 3/2 |3 | is proportional to k 3/2 on
large scales, small spikes are automatically more narrow). The shape of the spike
around its maximum is slightly nontrivial, reflecting the first oscillations of the field B
around zero during the transition. It is reasonable to state that if the primordial power
spectrum really possesses a spike (in the range in which this is still an open possibility,
i.e., 0.1 102 h Mpc1 6 k 6 5 102 h Mpc1 ), then its amplitude is already
constrained by observations to be smaller than 10 (here, we define the amplitude
as the ratio between the maximum value of k 3/2|| and the one, for instance, of the
large-scale plateau). This corresponds to the condition > 103 .
On large scales (small k), we also have an approximately flat plateau, but as we saw
in Section 3.1, we do not know how to predict analytically the amplitude and tilt
of this plateau, due to the violent slow-roll interruption during the transition. For
our choice of parameters (A , B , gA , gB , 1N ), there is a ratio of 4.3 between the
amplitude obtained numerically for k 3/2||, and the analytic prediction (Eq. (13)
of [26], or Eq. (23) of this paper). This ratio is found to be approximately independent
619
Fig. 7. The primordial power spectrum of adiabatic fluctuations, k 3/2 C(k), or equivalently, k 3/2 (k),
when the Higgs longitudinal perturbations are taken into account (in logarithmic scale and arbitrary
units). Together with the total power spectrum, we plot the three contributions from j , j = 1, 2, 3.
For scales crossing the Hubble length during the phase transition, 3 produces a spike (possibly
non-Gaussian). On the first three plots, the parameters (A , B , gA , gB ) are fixed to the values of the
appendix, and = 2 103 (top), 103 (middle), 0.5 103 (bottom).
620
Fig. 8. The primordial power spectrum of adiabatic fluctuations, for the same parameters as on Fig. 7
(upper plot), excepted lower values of A , B . On the right, the small scale plateau is unchanged,
because we kept A B fixed. On the left, we get less power on large and intermediate scales.
of . The large scale tilt is very close to one. Finally, metric perturbations on these
scales are composed of a (Gaussian) component 1 , and of a (possibly non-Gaussian)
component 2 , generated, like 3 , by the exponential amplification of perturbations
during the transition. So, the relative contribution of 2 depends, like the spike, on .
For > 103 , we find that 2 is subdominant, as can be seen in the figures, but not
negligible (for = 103 , |1 |/|2 | = 3.4). We also investigate the dependence of
the large-scale plateau on A , B . Since small-scale amplitude depends on A B
(see Eq. (7) of [26]), we vary A + B while keeping A B fixed. We find that the
amplitude on large and intermediate scales increases with A + B . On Fig. 8, A + B
has been divided by 2.8, and other parameters are the same as in Fig. 7 (upper plot).
In this case, the ratio between the two plateaus is 1.5.
6. Conclusion
We calculated numerically the primordial power spectrum for a particular model of
double supersymmetric inflation, called double D-term inflation [26]. Following the usual
semiclassical approach, we integrated for each independent mode function the equations
of motion of field and scalar metric perturbations. Some technical difficulties (attached to
any model of double hybrid inflation) arise between the two inflationary stages. Indeed, the
first stage ends with a rapid second-order phase transition, taking place during inflation, and
corresponding to the spontaneous symmetry breaking of a U (1) gauge symmetry, triggered
by a Higgs field C. This transition is followed by another inflationary stage, equivalent to
usual hybrid inflation. We were especially interested in the behavior of modes exiting the
Hubble radius during the transition, since they are expected to produce nontrivial features
in the primordial power spectrum.
621
After spontaneous symmetry breaking, it is impossible to follow exactly the mode

functions of the complex field C. Indeed, on the largest observable wavelengths, the
complex phase of C is completely inhomogeneous, at least at the beginning of the phase
transition; so, on these scales, the usual perturbative semiclassical approach is broken by
mode-mode coupling. However, we argued that only longitudinal Higgs perturbations,
i.e., perturbations of |C|, can contribute to the primordial power spectrum. Indeed, these
perturbations undergo a stage of spinodal instability, during which large wavelength modes
are exponentially amplified, while the perturbations of the complex phase and of the gauge
field (which combine into a massive gauge field, following the usual Higgs mechanism),
are those of a non-slow-rolling field with large positive mass square.
So, we performed a calculation in which longitudinal Higgs perturbations are taken into
account, with the modulus |C| treated as an ordinary real scalar field, described by the
semiclassical equations. We showed that at the beginning of the transition, a zero-mode
for |C| emerges from quantum fluctuations of Cs real and imaginary parts. Since the final
result depends crucially on this initial zero-mode, we calculated it precisely in Section 4.2.
Then, by comparing the background fields and the averaged quantum perturbations, we
showed that the semiclassical approach holds at any time. However, during one e-fold or so
after the beginning of the transition, it is not far from its limits of validity (with typically a
factor 103 between background quantities and perturbations). This fact is a first hint that the
contribution to the power spectrum resulting from longitudinal Higgs perturbations could
be significantly non-Gaussian. The second hints comes from the fact that at the beginning
of the transition, we identified a modulus with a real scalar field. In this process, we
matched carefully the mean value and expectation value of both quantities, but information
on higher momenta was lost. It is beyond the scope of this paper to evaluate the nongaussianity.
Our final power spectrum, shown on Fig. 7 for different parameter values, can be divided
in three regions:
(i) a small-scale Gaussian plateau, whose amplitude and tilt (nS ' 0.98) can be
calculated analytically;
(ii) a possibly non-Gaussian spike, with an amplitude depending on the superpotential
coupling parameter . For 103 , the spike induces variations at most of order
one in the primordial spectrum;
(iii) a large-scale plateau, for which we do not have an exact analytic prediction. For
103 , deviations from gaussianity, if any, are expected to be small.
For a precise comparison of our model with CMB and large-scale-structure (LSS) data,
one needs in principle, not only the primordial power spectrum of scalar adiabatic
perturbations, but also the spectrum of primordial gravitational waves, and the contribution
of cosmic strings formed during the two phase transitions (the one considered here, and
the one occurring at the end of inflation). However, as we said in [26], the tensor-toscalar ratio T /S (for the first temperature anisotropy multipoles) depends essentially on
the gauge coupling constants (as in single-field D-term inflation). Both gA and gB should
be at most of order 101 (if not, inflation takes place with inflaton values of the order of the
Planck mass, and supersymmetry/supergravity theories are not expected to be valid; see the
622
appendix for the consistency of this requirement with string theories). With this order of
magnitude for the coupling constants, one can show easily that T /S is negligible in double
D-term inflation. On the other hand, cosmic strings can contribute significantly to the
matter power spectrum and to CMB anisotropies, especially for D-term inflation, as shown
in [31] (see also [32]). Since, in current studies, perturbations from strings and from the
primordial spectrum are just added in quadrature, it is perfectly reasonable, at least in a first
step, to calculate only the primordial spectrum. The ratio of strings-to-inflation large-scale
temperature anisotropies, RSI = C5S /C5I , is difficult to estimate; theoretical predictions give
a ratio around 3 [31] or 4 [32], but the authors of this last reference argue that there are
many uncertainties in these calculations, so that smaller values can still be considered
seriously. With a mixture of inflationary and string perturbations, our model could still
be interesting: the features would appear nevertheless, but smoothed by the cosmic string
contribution.
In this work, we did not compare our results with CMB and LSS observations. Indeed,
current data cannot distinguish with precision the kind of small features that we obtain, at
least if they are located on scales 0.1 102 h Mpc1 6 k 6 5 102 h Mpc1 . However,
as we said in the introduction, comparisons with observations of typical BSI spectra
(with a step or a spike) have already been performed. The conclusions are currently quite
encouraging (see for instance [18,20]) and provide strong motivations for studying BSI
inflationary models, until a final statement is made by future redshift surveys, such as the
2-degree Field project 15 or the Sloan Digital Sky Survey, 16 combined with forthcoming
balloon and satellite CMB experiments.
An important qualitative aspect of our results is the possible deviation from gaussianity
on intermediate and large scales. This may be related with the evidence for a nongaussianity in the COBE data [32], but further work is needed on this point, since, on
the one hand, we do not have a precise theoretical prediction of non-gaussianity in this
model, and on the other hand, the conclusion of [32] involves large uncertainties.
Finally, we should stress that experimental evidence for a BSI primordial power
spectrum would be a very positive breakthrough for cosmology. On the one hand, the
introduction of one or two additional inflationary parameters, associated with the shape
of the spike or the step, would not compromise cosmological parameter extraction, as
indicated by [49], because the effect of these parameters would be orthogonal to the
others. On the other hand, experimental data would encode more information, and provide
additional exciting constraints on inflationary models, fundamental parameters, and highenergy physics.
Acknowledgements
This work was initiated after illuminating discussions with David Polarski and Alexei
Starobinsky. I would also like to thank Bruce Bassett, Rachel Jeannerot, Stephane Lavignac
15 http://meteor.anu.edu.au/colless/2dF
16 http://www.astro.princeton.edu/BBOOK
623
and Andrei Linde for providing useful comments, and Antonio Masiero for stimulating
discussions. This work is supported by the European Community under TMR network
contract No. FMRX-CT96-0090.
Appendix
The model. Double D-term inflation [26] is based on the superpotential W =
e + B
e , and on the definition of two gauge groups (U (1)A , U (1)B ), with
eB
e+ B
eA
e+ A
A
corresponding gauge coupling constants (gA , gB ) and FayetIliopoulos terms (A , B ). We
e (other choices
e B,
e (1, 0) for A
e , and (1,1) for B
choose the charges to be (0,0) for A,
e
can also lead to successful models). During inflation, we need only to follow the fields A,
e
e
e
e
B and B . Moreover, the complex phases of A and B remain constant during inflation
(except for very peculiar initial conditions that we do not consider here). So, we deal with
four canonically normalized real fields (A, B, C1 , C2 ) defined as:

e,
e .
B 2 e
C1 + iC2 C 2 B
B ,
(45)
A 2 A
The scalar potential reads:

2
2
g2
g2
1
1
1
V = A A C 2 + 2 B 2 C 2 + B B C 2
2
2
4
2
2
2
X
mi
1
+
ai m4i ln
.
64 2
2
(46)
i=1,...,9
The last term is the one-loop correction. Following [41], (ai , mi ) are taken from Table 1.
There,q(>) (respectively (<) ) means when |B| > Bc (respectively when |B| 6 Bc ), and
2 + g 2 )/ 2 . In the table, we have used the notations:
Bc = 2(gA
A
B B
Table 1
i
ai
m2i
5(>)
(<)
1 2 A2 + g 2 1 hCi2
2
2
A A
1 2 A2 g 2 1 hCi2
2
2
A A
1 2 hCi2
2
1 2 (2B 2 + hCi2 )
2
1 2 B 2 g2 g2
A A
B B
2
2 + g2 1 2 B 2
2 gA
A
B
B
2
6(>)
(<)
2 + g 2 )hCi2
(gA
B
1 2 A2
2
1 2B2 +
4
1 2B2 +
4
1 ( 2 + 2g 2 + 2g 2 )hCi2 +
A
B
4
1 ( 2 + 2g 2 + 2g 2 )hCi2
A
B
4
624
v
u

u 2 g 2 A + g 2 B 1 2 B 2
A
B
2
t
,
hCi
2 + g2 )
(gA
B

2

2
2
1
16
+ 2gB2 hCi2 12 2 gA
+ gB2 hCi4 .
2 B 2 + 2 + 2gA
(47)
(48)
The one-loop corrections are automatically continuous in |B| = Bc . The renormalization

scale must be chosen around Bc and, as in [41], we optimize this choice
numerically by requiring the continuity of the potential derivative.
Choice of parameters. The dependence of the primordial spectrum on the parameters
is discussed in Section 5.4. Before this section, we choose some arbitrary parameters
(A , B , gA , gB ) inside the allowed region defined in [26], for which the power spectrum
has got the same order of magnitude as the one indicated by COBE, and the amplitude of
the step (between the large-scale and the small-scale plateau) is of order one:
p
p
A = 3 103 mP ,
B = 4.2 103 mP ,
gA = gB = 102 .
(49)
The value of is completely irrelevant for our study (it is relevant only at the very end
of inflation). On the other hand, the choice of is crucial. Until Section 5.3, we take
= 103 . As mentioned in Section 3, Footnote 6, values of greater than 102 would
render the problem very complicated; we would not have a robust justification for the
expression of one-loop corrections, and the transition would consist in chaotic oscillations
in (B, C) space.
String motivations. Double D-term inflation is a consistent model from the point
of view of supersymmetry, when the FayetIliopoulos terms are put by hand from the
beginning. Here, we briefly address the issue of consistency with string theory. It turns out
that in heterotic string theory, one can always redefine the gauge groups in order to end
up with only one anomalous U (1), and a single associated FayetIliopoulos term. So, in
this framework, it is impossible to obtain two terms at low energy. However, it is well
known that even the simplest models of single D-term inflation are hardly compatible with
heterotic string theory [50,51], mainly because the FayetIliopoulos term generated by the
GreenSchwartz mechanism is typically too large to account for COBE data. It has been
noted by Halyo [52] that the situation is much better in type I string orbifolds, or type IIB
orientifolds. Then, the FayetIliopoulos terms are not generated at the one-loop order [53],
but at tree level, and depend on the vacuum expectation value of some moduli [54,55]
corresponding to the blowing-up modes of the underlying orbifold; since there is some
freedom in fixing the moduli, the value of required by COBE does not appear as unnatural
in this framework (also, in these theories, there is no unification of all coupling constants
as in heterotic string theory, and one can envisage scenarios where the coupling constant g
of the anomalous gauge symmetries is slightly lowered with respect to the standard model
coupling constant [52,56], so that the inflaton takes values significantly smaller than the
Planck mass during inflation, and supergravity, or even global supersymmetry can be
employed). But, nicely, another unusual feature in type I string orbifolds and orientifolds of
type IIB is the coexistence of several anomalous U (1)s, with associated FayetIliopoulos
625
terms [54,55]. So, our model may naturally arise from these theories. We are very grateful
to Stephane Lavignac for having kindly provided information on these points.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
S.W. Hawking, Phys. Lett. B 115 (1982) 295.

A.A. Starobinsky, Phys. Lett. B 117 (1982) 175.
A.H. Guth, S.-Y. Pi, Phys. Rev. Lett. 49 (1982) 1110.
A.A. Starobinsky, Pisma Zh. Eksp. Teor. Fiz. 42 (1985) 124 [JETP Lett. 42 (1985) 152].
L.A. Kofmann, A.D. Linde, A.A. Starobinsky, Phys. Lett. B 157 (1985) 361.
L.A. Kofmann, A.D. Linde, Nucl. Phys. B 282 (1987) 555.
J. Silk, M.S. Turner, Phys. Rev. D 35 (1987) 419.
L.A. Kofmann, D.Y. Pogosyan, Phys. Lett. B 214 (1988) 508.
D.S. Salopek, J.R. Bond, J.M. Bardeen, Phys. Rev. D 40 (1989) 1753.
D. Polarski, A.A. Starobinsky, Nucl. Phys. B 385 (1992) 623.
D. Polarski, Phys. Rev. D 49 (1994) 6319.
S. Gottlber, J.P. Muecket, A. Starobinsky, Astrophys. J. 434 (1994) 417.
P. Peter, D. Polarski, A.A. Starobinsky, Phys. Rev. D 50 (1994) 4827.
J. Lesgourgues, D. Polarski, Phys. Rev. D 56 (1997) 6425.
J. Einasto et al., Nature 385 (1997) 139.
J. Einasto, M. Einasto, P. Frisch, S. Gottlber et al., Mon. Not. R. Astron. Soc. 289 (1997) 801;
Mon. Not. R. Astron. Soc. 289 (1997) 813.
J. Adams, G. Ross, S. Sarkar, Nucl. Phys. B 503 (1997) 405.
J. Silk, E. Gawiser, in: Proc. of Nobel Symposium, Particle Physics and the Universe, 1998,
Phys. Scripta, astro-ph/9812200.
T. Broadhurst, A.H. Jaffe, astro-ph/9904348.
J. Lesgourgues, D. Polarski, A.A. Starobinsky, Mon. Not. R. Astron. Soc. 297 (1998) 769.
A.A. Starobinsky, JETP Lett. 55 (1992) 489.
L. Wang, M. Kamionkowski, Phys. Rev. D 61 (2000) 3504.
P.G. Ferreira, J. Magueijo, K.M. Grski, Astrophys. J. 503 (1998) L1.
A.A. Starobinsky, Grav. Cosmol. 4 (1998) 88.
M. Sakellariadou, N. Tetradis, hep-ph/9806461.
J. Lesgourgues, Phys. Lett. B 452 (1999) 15.
T. Kanazawa, M. Kawasaki, N. Sugiyama, T. Yanagida, Phys. Rev. D 61 (2000) 3517.
D.J.H. Chung, E.W. Kolb, A. Riotto, I.I. Tkachev, hep-ph/9910437.
C. Panagiotakopoulos, N. Tetradis, Phys. Rev. D 59 (1999) 083502.
G. Lazarides, N. Tetradis, Phys. Rev. D 58 (1998) 123502.
R. Jeannerot, Phys. Rev. D 56 (1997) 6205.
C. Contaldi, M. Hindmarsh, J. Magueijo, Phys. Rev. Lett. 82 (1999) 2034.
L. Randall, M. Soljacic, A.H. Guth, Nucl. Phys. B 472 (1996) 377.
J. Garcia-Bellido, A. Linde, D. Wands, Phys. Rev. D 54 (1996) 6040.
D. Boyanovsky, D. Cormier, H.J. de Vega, R. Holman, S.P. Kumar, in: N. Snchez, A. Zichichi
(Eds.), Proc. 6th Erice Chalonge School on Astrofundamental Physics, Kluwer, 1998, hepph/9801453.
D. Boyanovsky, H.J. de Vega, R. Holman, in: Lectures given at the ESF Network Workshop
Topological Defects and the Nonequilibrium Dynamics of Symmetry Breaking Phase Transitions, Les Houches, France, February 1999, pp. 1626, hep-ph/9903534.
D. Cormier, R. Holman, Phys. Rev. D 60 (1999) 041301.
A.A. Starobinsky, in: H.J. de Vega, N. Sanchez (Eds.), Current Topics in Field Theory, Quantum
Gravity, and Strings, Springer, Berlin, 1986.
626
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
A. Linde, A. Riotto, Phys. Rev. D 56 (1997) 1841.

G. Lazarides, N. Tetradis, Phys. Rev. D 58 (1998) 123502.
R. Jeannerot, J. Lesgourgues, accepted in Phys. Rev. D, hep-ph/9905208.
D. Polarski, A. Starobinsky, Phys. Rev. D 50 (1994) 6123.
J. Garcia-Bellido, D. Wands, Phys. Rev. D 53 (1996) 5437.
A.A. Starobinsky, J. Yokoyama, gr-qc/9502002.
D. Polarski, A. Starobinsky, Class. Quant. Grav. 13 (1996) 377.
J. Lesgourgues, D. Polarski, A. Starobinsky, Nucl. Phys. B 497 (1997) 479.
M. Abramowitz, I.A. Stegun (Eds.), Handbook of Mathematical Functions, Dower, New York,
1965.
B.A. Bassett, F. Tamburini, D.I. Kaiser, R. Maartens, Nucl. Phys. B 561 (1999) 188.
J. Lesgourgues, S. Prunet, D. Polarski, Mon. Not. R. Astron. Soc. 303 (1999) 45.
D. Lyth, A. Riotto, Phys. Rep. 314 (1999) 1.
J. Espinosa, A. Riotto, G. Ross, Nucl. Phys. B 531 (1998) 461.
E. Halyo, Phys. Lett. B 454 (1999) 223.
E. Poppitz, Nucl. Phys. B 542 (1999) 31.
L.E. Ibanez, R. Rabadan, A.M. Uranga, Nucl. Phys. B 542 (1999) 112.
Z. Lalak, S. Lavignac, H.P. Nilles, Nucl. Phys. B 559 (1999) 48.
L.E. Ibanez, C. Munoz, S. Rigolin, Nucl. Phys. B 553 (1999) 43.

Solution of Dirac equation around a spinning

black hole
Banibrata Mukhopadhyay 1 , Sandip K. Chakrabarti 2
Theoretical Astrophysics Group, S.N. Bose National Centre For Basic Sciences, JD Block, Salt Lake, Sector-III,
Calcutta-700091, India
Abstract
Chandrasekhar separated the Dirac equation for spinning and massive particles in Kerr geometry
in radial and angular parts. Chakrabarti solved the angular equation and found the corresponding
eigenvalues for different Kerr parameters. The radial equations were solved asymptotically by
Chandrasekhar. In the present paper, we use the WKB approximation to solve the spatially complete
radial equation and calculate analytical expressions of radial wave functions for a set of Kerr
and wave parameters. From these solutions we obtain local values of reflection and transmission
coefficients. 2000 Elsevier Science B.V. All rights reserved.
PACS: 04.20.-q; 04.70.-s; 04.70.Dy; 95.30.Sf
Keywords: Black holes; Kerr geometry; Dirac waves
1. Introduction
One of the most important solutions of Einsteins equation is that of the spacetime
around and inside an isolated black hole. The spacetime at a large distance is flat
and Minkowskian where usual quantum mechanics is applicable, while the spacetime
closer to the singularity is so curved that no satisfactory quantum field theory could be
developed as yet. An intermediate situation arises when a weak perturbation (due to, say,
gravitational, electromagnetic or Dirac waves) originating from infinity impinges on a
black hole, interacting with it. The resulting wave is partially transmitted into the black
hole through the horizon and partially scatters off to infinity. In the linearized (test field)
approximation this problem has been attacked in the past by several authors [14]. The
master equations of Teukolsky [2] which govern these linear perturbations for integral
spin (e.g., gravitational and electromagnetic) fields were solved numerically by Press and
1 bm@boson.bose.res.in
2 chakraba@boson.bose.res.in
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 1 0 - 2
628
B. Mukhopadhyay, S.K. Chakrabarti / Nuclear Physics B 582 (2000) 627645
Teukolsky [5] and Teukolsky and Press [6]. While the equations governing the massive
Dirac particles were separated by Chandrasekhar [3]. So far, only the angular eigenfunction
and eigenvalue (which happens to be the separation constant) have been obtained [7].
Particularly interesting is the fact that whereas gravitational and electromagnetic radiations
were found to be amplified in some range of incoming frequencies, Chandrasekhar [4]
predicted that no such amplifications should take place for Dirac waves because of the very
nature of the potential experienced by the incoming fields. However, these later conclusions
were drawn using asymptotic solutions and no attempt has so far been made to determine
the nature of the radial wave functions, both incoming and outgoing, for the Dirac wave
perturbations. He also speculated that one needs to look into the problem for negative
eigenvalues (- ) where one might come across super-radiance for Dirac waves.
In the present paper, we revisit this important problem to study the nature of the radial
wave functions as a function of the Kerr parameter, rest mass and frequency of incoming
particle. We also verify that super-radiance is indeed absent for the Dirac field. Unlike
the works of Press and Teukolsky [5] and Teukolsky and Press [6] where numerical
(shooting) methods were used to solve the master equations governing gravitational and
electromagnetic waves, we use an approximate analytical method for the massive Dirac
wave. The details of the method would be presented below.
The plan of the paper is as follows: in the next section, we present the equation governing
the Dirac waves (waves for half-integral massive spin particles) as they were separated
into radial and angular coordinates. We then briefly present the nature of the angular
eigenvalues and eigenfunctions. In Section 3, we present our method of solution and
present the spatially complete radial wave functions. Finally, in Section 4, we draw our
conclusions.
2. The Dirac equation in Kerr geometry

Chandrasekhar [3] separated the Dirac equation in Kerr geometry into radial (R) and
angular (S) wave functions. Below, we present these equations from Chandrasekhar [4]
using the same choice of units: we choose h = 1 = G = c.
The equations governing the radial wave-functions R 1 corresponding to spin 12 ,
2
respectively, are given by:
1/2 D0 R1/2 = - + imp r)1/2R+1/2 ,
1/2 D 1/2 R 1 = (- imp r)R1/2 ,
0
+2
(1a)
(1b)
where, the operators Dn and Dn are given by,

(r M)
iK
+ 2n
,
(r M)
iK
+ 2n
,
Dn = r
Dn = r +
and
(2a)
(2b)
629
= r 2 + a 2 2Mr,
(3a)
K = (r + a ) + am.
(3b)
Here, a is the Kerr parameter, n is an integer or half-integer, is the frequency of incident

wave, M is the mass of the black hole, mp is the rest mass of the Dirac particle, - is the
eigenvalue of the Dirac equation and m is the azimuthal quantum number.
The equations governing the angular wave-functions S 1 corresponding to spin 12 ,
2
respectively, are given by:
L 1 S+ 1 = (- amp cos )S 1 ,
(4a)
L1 S 1 = +(- + amp cos )S+ 1 ,
(4b)
where, the operators Ln and Ln are given by,

Ln = + Q + n cot ,
(5a)
= Q + n cot
(5b)
Q = a sin + m cosec .
(6)
Ln
and
Note that both the radial and the angular sets of equations, i.e., Eqs. 1(a), 1(b) and
Eqs. 4(a), 4(b) are coupled equations. Combining Eqs. 4(a), 4(b), one obtains the angular
eigenvalue equations for the spin- 21 particles as [7]

amp sin
L1 + - 2 a 2 m2p cos2 S 1 = 0.
(7)
L 1 L1 +
2
2
- + amp cos 2
2
There are exact solutions of this equation for the eigenvalues - and the eigenfunctions
S 1 when = mp / = 1 in terms of the orbital quantum number l and azimuthal quantum
2
number m. These solutions are [7]:

1 2
y2
-2 = l +
,
(8)
+ a (p + 2m) + a 2 2 1
2
2(l + 1) + a x
and
1
2
Slm = 1 Ylm
2
ay
1 Yl+1m ,
2(l + 1) + a x 2
(9)
where
p = F (l, l);
x = F (l + 1, l + 1);
y = F (l, l + 1)
and

1/2

l2 1m0|l1m
F (l1 , l2 ) = (2l2 + 1)(2l1 + 1)

l2 1 12 0l1 12 + (1)l2 l l2 1m0l1 m l2 1 12 0l1 12 + (1)l2 l 2 l2 1 12 1l1 21

(10)
with h | are the usual ClebshGordon coefficients. For other values of one has
to use perturbation theories. Solutions upto sixth order using perturbation parameter a is
630
- are required to solve the radial equations which

given in Chakrabarti [7]. The eigenvalues
we do now.
This radial equations (1a), (1b) are in coupled form. One can decouple them and express
the equation either in terms of spin up or spin down wave functions R 1 but the expression
2
loses its transparency. It is thus advisable to use the approach of Chandrasekhar [4] by
changing the basis and independent variable r to

2Mr+ + am/
r
2Mr + am/
r
log
1
log
1
(11)
r = r +
r+ r
r+
r+ r
r
(for r > r+ ),
d
d
= 2 ,
dr dr
2 = r 2 + 2 ,
(12)
2 = a 2 + am/,
(13)
to transform the set of coupled Eqs. 1(a), 1(b) into two independent one dimensional wave
equations given by:

1/2
d
i P+ 1 = 2 (- imp r)P 1 ,
(14)
2
2
dr

1/2
d
+ i P 1 = 2 (- + imp r)P+ 1 .
(15)
2
2
dr
Here, D0 = ( drd + i ) and D0 = ( drd i ) were used and wave functions were
redefined as R 1 = P 1 and 1/2R+ 1 = P+ 1 .
2
2
2
2
We now define a new variable,
= tan1 (mp r/- ),
(16)
which yields
cos = q
- 2 + m2p r 2
and
and
mp r
sin = q
- 2 + m2p r 2
q
(- imp r) = exp(i) - 2 + m2p r 2 ,
so the coupled equations take the form

1/2
mp r
1/2
d
,
i P+ 1 = 2 - 2 + m2p r 2
P 1 exp i tan1
2
2
dr
and

1/2
mp r
1/2
d
.
+ i P 1 = 2 - 2 + m2p r 2
P+ 1 exp i tan1
2
2
dr
Then, defining,
P+ 1 = + 1
2

1
1 mp r
exp i tan
2
-
(17)
(18a)
(18b)
(19a)
and

mp r
1
P 1 = 1 exp + i tan1
,
2
2
2
-
we obtain,
d+ 1
2
dr
631
(19b)

1/2
1
- mp
1/2 2
i 1 + 2
1
1 =
- + m2p r 2
+2
2
2 - 2 + m2p r 2
2
(20a)
and

1/2
- mp
1/2
1
+ i 1 + 2
+ 1 .
(20b)
1 = 2 - 2 + m2p r 2
2
2
2
2
2
dr
2 - + mp r
-m
m r
1
tan1 ( -p ) so that dr = (1 + 2 2p - 2 1 2 2 )dr , the
Further choosing r = r + 2
+mp r
above equations become,

d
i + 1 = W 1 ,
(21a)
2
2
dr
d 1
2
and

d
+ i 1 = W + 1 ,
2
2
dr
(21b)
where
W=
1/2(- 2 + m2p r 2 )3/2

.
2 (- 2 + m2p r 2 ) + - mp /2
(22)
Now letting Z = + 1 1 we can combine the differential equations to give,

2
2

d
W Z+ = i Z ,
(23a)
dr
and

d
+ W Z = i Z+ .
(23b)
dr
From these equations, we readily obtain a pair of independent one-dimensional wave
equations,

2
d
2
+
(24)
Z = V Z ,
dr 2
where
1/2 (- 2 + m2p r 2 )3/2
dW
= 2 2
V = W 2
dr
[ (- + m2p r 2 ) + - mp /2 ]2
1/2 2

3/2
- + m2p r 2
(r M)(- 2 + m2p r 2 ) + 3m2p r
3/2 (- 2 + m2p r 2 )5/2
2 2
[ (- + m2p r 2 ) + - mp /2 ]3

2r(- 2 + m2p r 2 ) + 2m2p 2 r + - mp (r M)/ .
(25)
632
One important point to note: the transformation of spatial coordinate r to r (and r ) is

taken not only for mathematical simplicity but also for a physical significance. When r
is chosen as the radial coordinate, the decoupled equations for independent waves show
diverging behaviour. However, by transforming those in terms of r (and r ) we obtain
well behaved functions. The horizon is shifted from r = r+ to r = unless 6 s =
am/(2Mr+ ) (Eq. (11)). In this connection, it is customary to define c where 2 = 0 (Eq.
(13)). Thus, c = m/a. If 6 s , the region is expected to be super-radiant [4] because
for integral spin particles for 6 s there exhibit super-radiation.
3. Solution of the radial equation

Out of the total physical parameter space, in one region (region I) the total energy of the
particle is always greater than the height of potential barrier and in the other region (region
II) the energy is less than of the maximum height of the potential barrier. In region II, the
wave hits the wall of barrier and tunnels through it. One has to treat these two cases a little
differently.
The usual WKB approximation [8] is used to obtain the zeroth order solution. We
improve the solution by properly incorporating the inner and outer boundary conditions.
After establishing the general solution, we present here the solution of Eq. (24) for three
sets of parameters as illustrative examples. For those examples the choice of parameters is
made in such a way that there is a significant interaction between the particle and the black
hole, i.e., when the Compton wavelength of the incoming wave is of the same order as the
radius of the outer horizon of the Kerr black hole. So,
h
G[M + M 2 a 2 ]
.
(26)
2
c
mp c
We choose as before G = h = c = 1, so
mp
.
[M + M 2 a 2 ]
(27)
Similarly, the frequency of the incoming particle (or wave) should be of the same order as
the inverse of the light crossing time of the radius of the black hole, i.e.,
c3
.
G[M + M 2 a 2 ]
Using the same units as before, we can write,
p

1
.
mp M + M 2 a 2
(28)
(29)
In principle, however, one can choose any values of and mp for a particular black hole
and the corresponding solution is possible.
One can easily check from Eq. (25) that for r (i.e., r ) V m2p . So we
expand the total parameter space in terms of the frequency of the particle (or wave), and
the rest mass of the particle, mp . It is clear that in half of the parameter space spanned
633
Fig. 1. Contours of constant wmax = max( 12 ddk

) are shown to indicate that generally w 1 and
k r
therefore the WKB approximation is valid in most of the physical region. Labels indicate values of
wmax .
by mp where, < mp , particles are released at finite distance with so little energy
that they cannot escape to infinity. In this case, the total energy 2 of the incoming
particle at a large distance is less than the potential energy of the system. We will not
discuss solutions in this region. The rest of the parameter space ( > mp ) is divided into
two regions I: E > Vm and II: E < Vm , where E is the total energy of the incoming
particle and Vm is the maximum of the potential. In region I, the wave is locally sinusoidal
because the wave number k is real for the entire range of r . In region II, on the other hand,
the wave is decaying in some region when E < V , i.e., where the wave hits the potential
barrier and in the rest of the region, the wave is propagating. We shall show solutions in
these two regions separately. In region-I whatever be the physical parameters, the energy
of the particle is always greater than the potential energy and the WKB approximation is
generally valid in the whole range (i.e., k12 ddkr 1). In cases of region-II, the energy of the
particle is always less than the maximum height of potential barrier. Thus, at two points
(where k = 0) the total energy matches the potential energy and in the neighbourhood of
those two points the WKB approximate method is not valid. They have to be dealt with
separately. In Fig. 1, we show contours of constant wmax = max( k12 ddkr ) for a given set
(, mp ) of parameters. The labels show the actual values of wmax . Clearly, in most of the
parameter regions the WKB approximation is safely valid for any value of r . One has to
634
employ a different method (such as using Airy functions, see below) to find solutions in
those regions where wmax attains a large value which indicates the non-validity of WKB
method.
3.1. Solutions of region I
We rewrite Eq. (24) as,
d 2 Z+
+ ( 2 V+ )Z+ = 0.
dr2
(30)
This is nothing but the Schrdinger equation with total energy of the wave 2 . This can be
solved by regular WKB method.
Let
Z
q
2
u(r ) = k(r )dr + constant,
k(r ) = V+ ,
k is the wavenumber of the incoming wave and u as the Eikonal. The solution of the
Eq. (30) is,
A
A+
Z+ = exp(iu) + exp(iu)
k
k
(31)
A2+ + A2 = k.
(32)
with
The motivation of Eq. (32) is to impose the WKB method at the each space point so that
sum of the transmission and reflection coefficients are same at each location. In this case
2 > V+ all along and also 1k ddkr k, so the WKB approximation is valid in the whole
region.
It is clear that a standard WKB solution where A+ and A are kept constant throughout,
can not be accurate in whole range of r , since the physical inner boundary condition
on the horizon must be that the reflected component is negligible there (since there the
potential barrier height goes down to zero). Thus the WKB approximation requires a slight
modification in which a spatial dependence of A is allowed. On the other hand, at a large
distance, where the WKB is strictly valid, A+ and A should tend to be constants, and
hence their difference is also a constant:
A+ A = c.
(33)
Here, one can choose also the sum of A+ and A are constant instead of difference as
Eq. (33), but the final result will not be affected. Here, c is determined from the WKB
solution at a large distance. For simplicity we choose A s are real. This along with Eq. (32)
gives,
p
c
2k(r) c2
.
(34)
A (r) = +
2
2
This spatial variation, strictly valid at large distances only, should not be extendable to the
horizon without correcting for the inner boundary condition. These values are to be shifted
635
by, say, Ah respectively, so that on the horizon one obtains the physical R and T . We first
correct the reflection coefficient on the horizon as follows: Let Ah be the value of A on
the horizon (see Eq. (34)),
p
c
2k(r+ ) c2
.
Ah = +
2
2
It is appropriate to use A = A Ah rather than A since A vanishes at r = r+ .
Incorporating these conditions, the solution (31) becomes,
A
A+
Z+ = exp(iu) + exp(iu)
q
q
(35)
with the usual normalization condition

A2+ + A2 = q
(36)
where A+ = A+ A+h .
Determination of A+h is done by enforcing R obtained from Eq. (37a), which is shown
below, is the same as that obtained by the actual WKB method. The q is used to compute
the transmission coefficient T from Eq. (36). In this way, normalization of R + T = 1 is
assured.
q is found to
The normalization factor q k as r and the condition q1 ddq
r
be satisfied whenever 1k ddkr k is satisfied. This is the essence of our modification of the
WKB. In a true WKB, A are constants and the normalization is with respect to a (almost)
constant k. However, we are using it as if the WKB is instantaneously valid everywhere.
Our method may therefore be called Instantaneous WKB approximation or IWKB for
short. Using the new notations, the instantaneous values (i.e., local values) of the reflection
and transmission coefficients are given by (see Eq. (35)),
A2
,
q
A2
T = +.
q
R=
(37a)
(37b)
Whatever may be the value of the physical parameters, 1k ddkr k is satisfied in whole range
of r for region I.
The variation of reflection and transmission coefficients would be well understood if we
imagine the potential barrier consists of a large number of steps. From simple quantum
mechanics, in between each two steps, we can calculate the reflection and transmission
coefficients [9]. Clearly these reflection and transmission coefficients at different junctions
will be different. This is discussed in detail below. To be concrete, we choose one set of
parameters from region I. Here, the total energy of the incoming particle is greater than the
potential barrier height for all values of r . We use, Kerr parameter, a = 0.5; mass of the
black hole, M = 1; Mass of the particle, mp = 0.8; orbital angular momentum quantum
number, l = 1/2; azimuthal quantum number, m = 1/2;
frequency of the incoming wave,
= 0.8. The derived parameters are, r+ = M + M 2 a 2
= 1.86603; c = 1; s =
0.066987; 2 = 0.0625. For these parameters, the eigenvalue is - = 0.92 [7].
636
Fig. 2. Behaviour of (a) V+ (solid curve) and V (dashed curve), (b) V+ (solid curve), k (dashed
curve), total energy E (short-dashed curve), (c) local transmission (T , solid curve) and reflection (R,
dashed curve) coefficients as functions of r . The parameters are a = 0.5, M = 1, mp = 0.8, l = 1/2,
m = 1/2, = 0.8.
Here it is clear that is in between c and s and 2 < 0, r+ > ||. So we are in a
strictly non super-radiant regime since here, > s [4].
From Eq. (24) we observe that there are two wave equations for two potentials V+ and
V . The nature of the potentials is shown in Fig. 2(a). It is clear from the Fig. 2(a) that
the potentials V are well behaved. They are monotonically decreasing as the particle
approaches the black hole, and the total energy chosen in this case ( 2 ) is always higher
than V . For concreteness, we study the equation with potential V+ . A similar procedure
(IWKB method) as explained above can be adopted using the potential V to compute Z
and its form would be
A0+ A0+h
A0 A0h
p
p
exp(iu0 )
exp(iu0 ).
(350 )
Z =
0
0
q
q
Note the occurrence of the negative sign in front of the reflected wave. This is to satisfy
the asymptotic property of the wave functions.
637
Fig. 3. (a) Steps (solid) approximating a potential (dotted), thus reducing the problem to that
of a quantum mechanics. The parameters are same as in Fig. 2. (b) Comparison of variation of
instantaneous reflection coefficient R and transmission coefficient T with the radial coordinate r
using analytical WKB method (solid) and step-potential method (dotted). The parameters are same
as Fig. 2.
In Fig. 2(b), we show the nature of V+ (solid curve), k (dashed curve) and E(= 2 )
(short-dashed curve). In the solutions (Eqs. (35) and (350 )) the first term corresponds to the
incident wave and the second term corresponds to the reflected wave.
In Fig. 2(c), the variation of reflection and transmission coefficients are shown. It is seen
that as matter comes close to the black hole, the barrier height goes down. As a result, the
penetration probability increases, causing the rise of the transmission coefficient.
Local values of the reflection and transmission coefficients could also be calculated
using the well known quantum mechanical approach. First one has to replace the potentials
(as shown in Fig. 2(a)) by a collection of step functions as shown in Fig. 3(a). The standard
junction conditions of the type
Z+,n = Z+,n+1 ,
(38a)
where
Z+,n = An exp[ikn r,n ] + Bn exp[ikn r,n ]
and
where

dZ+
dZ+
=
,
dr n
dr n+1
(38b)

dZ+
= ikn An exp(ikn r,n ) ikn Bn exp(ikn r,n )
dr n
at each of the n steps were used to connect solutions at successive steps. From the simple
quantum mechanical calculation we obtain the reflection and transmission coefficients at
the each junctions. Clearly at different junctions, i.e., at different radii this reflection and
638
transmission coefficients will be different. As before, we use the inner boundary condition,
to be R 0 at r . In reality we use as many steps as possible to follow accurately
the shape of the potential. Smaller step sizes were used whenever k varies faster. Fig. 3(b)
shows the comparison of the instantaneous reflection and transmission coefficients in both
the methods. The agreement shows that the WKB can be used at each point quite successfully.
It is to be noted, that, strictly speaking, the terms reflection and transmission
coefficients are traditionally defined with respect to the asymptotic values. The spatial
dependence that we show are just the dependence of the instantaneous values. This is
consistent with the spirit of IWKB approximation that we are using.
The radial wave functions R+ 1 and R 1 which are of spin up and spin down particles
2
2
respectively of the original Dirac equation are given below,
a+ cos(u ) + a cos(u + )
Re R 1 1/2 =
2
2 k
0 cos(u0 + )
a 0 cos(u0 ) a
,
(39a)
+ +
2 k0
a+ sin(u ) a sin(u + )
Im R 1 1/2 =
2
2 k
0 sin(u0 + )
a 0 sin(u0 ) + a
,
(39b)
+ +
2 k0
a+ cos(u + ) + a cos(u )
Re R 1 =
2
2 k
0 cos(u0 )
a 0 cos(u0 + ) a
,
(39c)
+
2 k0
a+ sin(u + ) a sin(u )
Im R 1 =
2
2 k
0 sin(u0 )
a 0 sin(u0 + ) + a
.
(39d)
+
2 k0
a0
a0
Here, a+ = (A+ A+h )/ q/k and a = (A Ah )/ q/k. +0 and 0 are the transk
k
mitted and reflected amplitudes, respectively, for the wave of corresponding potential V .
In Fig. 4(a)(d) we show the nature of these wavefunctions. The eikonals used in plotting
these functions (see Eqs. (39a)(39d)) have beenRcalculated
by approximating V in terms
p
2 V dr . This was done since V
of a polynomial and using the definition u(r ) =
is not directly integrable. Note that the amplitude as well as wavelength remain constants
in regions where k is also constant.
3.2. Solutions of region II
Here we study the solution of a region where for any set of physical parameters, the total
energy of the incoming particle is less than the maximum height of the potential barrier.
So the WKB approximation (more precisely, our IWKB approximation) is not valid in
639
Fig. 4. Nature of real and imaginary parts of radial wave functions for Case 1.
the whole range of r . In the regions where the WKB is not valid, the solutions will be
the linear combination of Airy functions because the potential is a linear function of r in
those intervals. At the junctions one has to match the solutions including Airy functions
with the solution obtained by WKB method.
In the region where the WKB approximation is valid, local values of reflection and
transmission coefficients and the wave function can be calculated easily by following the
same method described in previous sub-section (solutions of region I) and the solution will
be same as Eqs. (35), (350 ). In other regions, the equation reduces to
d 2 Z+
xZ+ = 0,
dr2
(40)
where x = 1/3(r p), is chosen to be positive and p is the critical point where the
total energy and potential energy are matching.
640
Let Z+ (x) = x 1/2Y (x) and considering region x > 0 the equation (40) reduces to

dY
1
d2 Y
x3 +
Y (x) = 0.
(41)
x2 2 + x
dx
dx
4
By making another transformation
2
= x 3/2
3
we obtain
(42)

dY
1
2
+
+
Y ( ) = 0,
d 2
d
9
2d
2Y
(43)
this is the modified Bessel equation. The solution of this equation is I+ 1 ( ) and I 1 ( ).
3
3
So the solution of Eq. (40) will be

(44)
Z+ (x) = x 1/2 C1 I+ 1 ( ) + C2 I 1 ( ) .
3
When x < 0 the corresponding equation is,

2
dY
1
2d Y
2
+
+
Y ( ) = 0,
d 2
d
9
which is the Bessel equation. The corresponding solution is

Z+ (x) = |x|1/2 D1 J+ 1 ( ) + D2 J 1 ( ) ,
3
(45)
(46)
where J 1 and I 1 are the Bessel functions and the modified Bessel functions of order
3
3
1/3, respectively.
The Airy functions are defined as

1
(47)
Ai(x) = x 1/2 I 1 ( ) I+ 1 ( ) , x > 0,
3
3
3

1
(48)
Ai(x) = |x|1/2 J 1 ( ) + J+ 1 ( ) , x < 0,
3
3
3

1 1
(49)
Bi(x) = x 2 I 1 ( ) + I+ 1 ( ) , x > 0,
3
3
3

1
(50)
Bi(x) = |x|1/2 J 1 ( ) J+ 1 ( ) , x < 0.
3
3
3
In terms of Airy functions, the solutions (44) and (46) can be written as
3
3
(C2 + C1 ) Bi(x) for x > 0,
(51)
Z+ = (C2 C1 ) Ai(x) +
2
2
3
3
(D2 D1 ) Bi(x) for x < 0.
(52)
Z+ = (D2 + D1 ) Ai(x) +
2
2
By matching boundary conditions it is easy to show that the solution corresponding x > 0
and that corresponding x < 0 are continuous when C1 = D1 and C2 = D2 .
As an example of solutions from this region, we choose: a = 0.95, M =p1, mp = 0.17,
l = 1/2, m = 1/2, and = 0.21. The black hole horizon is at r+ = M + (M 2 a 2 )
=
641
Fig. 5. Plots are same as in Fig. 2. The parameters are a = 0.95, M = 1, mp = 0.17, l = 1/2,
m = 1/2, = 0.21.
1.31225, c = 0.526316, s = 0.180987, 2 = 1.356 and - = 0.93 [7]. It is clear that

the values of c , s and 2 indicate the region is non super-radiant. In Fig. 5(a), we show
the nature of V+ and V , however, while solving, we use the equation containing V+ (Eq.
(24)). Unlike the case in the previous subsection, here 2 is no longer greater than V at all
radii. As a result, k 2 may attain negative values in some region. In Fig. 5(b), nature of V+
(solid curve), parameter k (dashed curve) and energy E (short-dashed curve) are shown.
Here, WKB approximation can be applied in regions other than r 4 to 1 and 1 to 7
where k is close to zero and the condition 1k ddkr k is not satisfied.
In the region r = 7 to 1 around the turning point r = 4.45475 the solutions turns out
to be [10],
Z+ = 1.087526 Ai(x) + 0.788968 Bi(x).
(53)
642
Similarly, the solution from r 1 to 4, i.e., around the turning point r = 2.8053
turns out to be [10],
Z+ = [1.328096 Ai(x) + 0.774426 Bi(x)].
(54)
It is to be noted that in the region r 1 to 1, even though the potential energy dominates
over the total energy,
method is still valid. Here the solution will take
WKB approximation
the form exp(u)/ k and exp(+u)/ k. Asymptotic values of the instantaneous reflection
and the transmission coefficients (which are traditionally known as the reflection and
transmission coefficients) are obtained from the WKB approximation. This yields the
integral constant c as in previous case.
From Eqs. (37a), (37b) local reflection and transmission coefficients are calculated,
behaviours of which are shown in Fig. 5(c). The constants Ah and A+h are calculated
as before. Note the decaying nature of the reflection coefficient inside the potential barrier.
3.3. Solutions in the super-radiant region
In this region, the potential diverges at r = ||. Here, the barrier height goes up to infinity
and then the potential changes sign so that its nature changes from repulsive to attractive
and vice versa. This is because 6 s (which is the condition for super-radiance) and
r+ < || [see Eqs. (11) and (13)]. Unlike the previous two cases, the relation between r
and r is not single valued. Here, at both r = r+ and r = , the value of r = . With the
decrement of r, r is decreased initially up to r = ||. Subsequently, r starts to rise and at
the black hole horizon it diverges. Obviously, in this case particles hit the barrier and we
can solve the equation following the same methods as explained in the previous cases, i.e.,
the solutions are the same as Eqs. (35) and (350 ) for the region where the WKB method is
valid and Eqs. (51) and (52) where the WKB method is not valid.
For illustrative example, here, we choose: a = 0.95, M = 1, mp = 0.105, l =
1/2,
m = 1/2, and = 0.105. The black hole horizon is located at r+ = M +
M 2 a2
= 1.31225, and c = 0.526316, s = 0.180987, 2 = 3.62 and - = 0.97 [7].
Chandrasekhar showed [4] that for integral spin particles this region exhibits superradiance and conjectured that for half-integral spins the super-radiance may be absent.
We investigate here if this conjecture is valid.
The behaviour of potentials V+ and V are shown in Fig. 6(a). It is clear that at r = ||
the potential diverges and the nature of the potential is changed from repulsive to attractive
(for V ) and vice versa (for V+ ). Here, we will treat the equation with V as the potential
(it is equally easy to do the problem with V+ ). We first divide our computations into two
parts. In the repulsive part of the potential (i.e., when V > 0), particles come from infinity
and most of them reflect back from the infinitely high barrier. In the attractive part of
the potential (i.e., when V+ < 0), particle radiates outwards in the r coordinate (actually,
particle goes towards the horizon but due to multivalueness of the radial coordinate r (with
respect to r) the horizon is mapped to infinity).
In Fig. 6(b), nature of V , k and E are shown. The WKB approximation (more precisely
IWKB approximation) method is valid from infinity to r 40 since, otherwise, 1k ddkr k
643
Fig. 6. Behaviour of (a) V+ (solid curve) and V (dashed curve), (b) V (solid curve), k for
region where potential is positive (krep , dashed curve), k for region where potential is negative
(katt , short-dashed curve), total energy E (dotted curve), (c) local transmission (T , solid curve) and
reflection (R, dashed curve) coefficients as functions of r . The parameters are a = 0.95, M = 1,
mp = 0.105, l = 1/2, m = 1/2, = 0.105.
is not satisfied. In those other regions one has to apply a different method (which was
also explained in last sub-section) to find solutions. The local values of the reflection
and transmission coefficients and the wave function of the particle are calculated as in
the previous cases. Since the matter which tunnels through the infinitely high barrier face
infinitely strong attractive field, the possibility of extraction of energy would be zero. In
Fig. 6(c), the variations of local transmission and reflection coefficients are shown. The
net transmission of the wave through the horizon is non-negative all along and therefore
super-radiation is absent, although is less than s . We believe that the non-existence
of super-radiation is due to (r ||)3 variation of the potential near the singular point.
644
Because of the existence of attractive field, the extraction of energy is difficult, so the net
transmission of the wave through the horizon from is always positive. This argument is
valid for any set of parameters where 6 s .
4. Conclusion
In this paper, we studied scattering of massive, spin-half particles from a Kerr black
hole, particularly the nature of the radial wave functions and the reflection and transmission
coefficients. Our main motivation was to give a general analytical expression of the solution
which can be useful for further study. We showed few illustrative cases as examples.
We verified that these analytical solutions were indeed correct by explicitly solving the
same set of equations using quantum mechanical step-potential approach as described in
Section 3. We classified the entire parameter space in terms of the physical and unphysical
regions and the physical region was further classified into two regions, depending on
whether the particle hits the potential barrier or not. Again, the region where particle hits
on the barrier, is divided into two parts, one is super-radiant region and other is non-superradiant region. We chose one illustrative example in each of the regions. We emphasize that
the most interesting region to study would be close to mp . However, we pointed out
(Fig. 1) that for mp 6 0.35, the WKB solutions cannot be trusted, and other methods (such
as those using Airy functions) must be employed.
We used the well known WKB approximation method as well as the step-potential
method of quantum mechanics to obtain the spatial dependence of the coefficients of
the wave function. This in turn, allowed us to determine the reflection and transmission
coefficients and the nature of wave functions. The usual WKB method with constant
coefficients and (almost) constant wave number k is successfully applied even when
the coefficients and wave number are not constant everywhere. Solution from this
instantaneous WKB (or IWKB) method agrees fully with that obtained from a purely
quantum mechanical method where the potential is replaced by a collection of steps. Our
method of obtaining solutions should be valid for any black hole geometry which are
asymptotically flat so that radial waves could be used at a large distance. This way we
ensure that the analytical solution is close to the exact solution. In region II, in some
regions, the WKB method cannot be applied and hence Airy function approach or our
step-potential approach could be used.
In the literature, reflection and transmission coefficients are defined at a single point.
These definitions are meaningful only if the potential varies in a small region while studies
are made from a large distance from it. In the present case, the potential changes over a
large distance and we are studying in these regions as well. Although we used the words
reflection and transmission coefficients, in this paper very loosely, our definitions are
very rigorous and well defined. These quantities are simply the instantaneous values and in
our belief more physical. The problem at hand is very similar to the problem of reflection
and transmission of acoustic waves from a struck string of non-constant density where
reflection and transmission occurs at each point.
645
Among other things, we verify Chandrasekhars conjecture [4] based on the asymptotic
solution, that for spin- 21 particle the phenomenon of super-radiance is absent. We believe
that this is due to the very way the potential develops the singularity at r = ||. Here
V (r ) (r ||)3 , which results an attractive potential in some region very close to
the black hole. In contrast, V (r ) (r ||)4 when electromagnetic and gravitational
waves are scattered off the black hole [4] does not create an attractive part in the potential
and possibly exhibit the phenomenon of super-radiance. It is noted that all the cases where
potential diverge at r = (i.e., so called super-radiation cases) arise for 6 s with the
negative values of azimuthal quantum number (here, m = 1/2) and the positive Kerr
parameter, a. For positive values of m and positive values of a, potential does not diverge
at any point for all values of . If we change the spin orientation of the black hole (negative
values of a) and take positive m again divergence of the potential will arise. Thus, it seems
that the cases with opposite sign of a and m are physically more interesting.
It is seen that for different physical parameters the solutions are different. The waves
scattered off are distinctly different in different parameter regions. In a way, therefore,
black holes can act as a mass spectrograph! Another interesting application of our method
would be to study interactions of Hawking radiations in regions just outside the horizon.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
S. Teukolsky, Phys. Rev. Lett. 29 (1972) 1114.

S. Teukolsky, Astrophys. J. 185 (1973) 635.
S. Chandrasekhar, Proc. R. Soc. London A 349 (1976) 571.
S. Chandrasekhar, The Mathematical Theory of Black Holes, Clarendon Press, London, 1983.
W.H. Press, S. Teukolsky, Astrophys. J. 185 (1973) 649.
S. Teukolsky, W.H. Press, Astrophys. J. 193 (1974) 443.
S.K. Chakrabarti, Proc. R. Soc. London A 391 (1984) 27.
J. Mathews, R.L. Walker, Mathematical Methods of Physics, 2nd edn., The Benjamin
Cummings Publishing Company, California, 1970.
[9] A.S. Davydov, Quantum Mechanics, 2nd edn., Pergamon Press, Oxford, New York, 1976.
[10] M. Abramowitz, I.A. Stegun (Eds.), Handbook of Mathematical Functions, National Bureau of
Standards, Washington, 1972.

Constructing classical and quantum superconformal

algebras on the boundary of AdS3
Jrgen Rasmussen a,b,1
a Department of Theoretical Physics, Tata Institute of Fundamental Research, Homi Bhabha Road, Colaba,
Mumbai 400 005, India

b Physics Department, University of Lethbridge, Lethbridge, Alberta, Canada T1K 3M4 2
Received 6 March 2000; revised 27 April 2000; accepted 22 May 2000
Abstract
Motivated by recent progress on the correspondence between string theory on anti-de Sitter
space and conformal field theory, we address the question of constructing spacetime N extended
superconformal algebras on the boundary of AdS3 . Based on a free field realization of an affine
SL(2|N/2) current superalgebra residing on the world sheet, we construct explicitly the Virasoro
generators and the N supercurrents, N is even. The resulting superconformal algebra has an affine
SL(N/2) U (1) current algebra as an internal subalgebra. Though we do not complete the general
superalgebra, we outline the underlying construction and present supporting evidence for its validity.
Particular attention is paid to its BRST invariance. In the classical limit where the free field realization
may be substituted by a differential operator realization, we discuss further classes of generators
needed in the closure of the algebra. We find sets of half-integer spin fields, and for N > 6 these
include generators of negative weights. An interesting property of the construction is that for N 6= 2 it
treats the supercurrents in an asymmetric way. Thus, we are witnessing a new class of superconformal
algebras not obtainable by conventional Hamiltonian reduction. The complete classical algebra is
provided in the case N = 4 and is of a new and asymmetric form. 2000 Elsevier Science B.V. All
rights reserved.
PACS: 11.25.Hf; 11.25.-w
Keywords: Superconformal field theory; AdS/CFT correspondence; String theory; Free field realization
1. Introduction
Recently, Maldacena proposed a duality between type IIB string theory and (super-)
conformal field theory (CFT) on the boundary of anti-de Sitter space (AdS) [1], further
elaborated on in Refs. [2,3]. This remarkable conjecture has induced a tremendous activity
1 rasmussj@cs.uleth.ca
2 Address after 10th of February 2000.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 2 2 - 9
650
J. Rasmussen / Nuclear Physics B 582 (2000) 649676
in theoretical high energy physics. In the cases involving AdS3 , see, e.g., Refs. [410] and
the review [11], the corresponding spacetime CFT is two-dimensional and thus possesses
many well-known properties.
The boundary of AdS3 enjoys conformal symmetry, as recognized by Brown and
Henneaux [12]. In the work [6], Giveon et al. have constructed explicitly the generators
of the spacetime conformal algebra from the string theory on AdS3 . The construction
starts from the world sheet SL(2) current algebra. In terms of the Wakimoto free field
realization [13] of that, they have provided a general expression for the Virasoro generators
and computed the central charge.
In Ref. [14] Ito has succeeded in constructing superconformal algebras (SCAs) on the
boundary of AdS3 and again the construction starts from a world sheet current algebra.
However, in order to obtain an extended conformal symmetry in spacetime one needs
to consider higher world sheet current (super-)algebras than SL(2). Thus, Ito has found
that the appropriate Lie superalgebras leading to N = 1, 2 and 4 superconformal algebras
are osp(1|2), sl(2|1) and sl(2|2), respectively. The explicit constructions are based on
generalized Wakimoto free field realizations of the associated affine current superalgebras
[1517]. A related approach to construct N = 1, 2 and 4 SCAs is discussed in [18] in which
also one- and two-point functions and some unitary representations are considered. In Ref.
[19] spacetime N = 3 superconformal theories are studied.
The objective of the present paper is to take a first step in the direction of classifying
the SCAs that may be induced by string theory on AdS3 , thus generalizing the work by
Ito [14]. A main property of an appropriate world sheet current (super-)algebra is that
the bosonic part may be decomposed as G = SL(2) G0 . This is necessary as we want
the free fields in the Wakimoto realization of the embedded SL(2) to be considered as
coordinates on AdS3 . As discussed in Ref. [6], a purely bosonic world sheet algebra with
such a decomposition leads immediately to an affine Lie algebra on the boundary of AdS3 .
The present paper is devoted to discussing the class of superconformal algebras that may
be constructed starting from affine SL(2|N/2) current superalgebras having as bosonic part
SL(2) SL(N/2) U (1). By construction, N is even.
We construct explicitly the Virasoro generators, the N supercurrents, and the generators
of an internal SL(N/2) U (1) KacMoody algebra. For N = 2, the SL(N/2) and U (1)
current algebras collapse to a single U (1) current algebra. This conventional N = 2
superconformal algebra has already been obtained by Ito [14]. For higher N we turn to
a classical limit in which the generators may be substituted by first-order linear differential
operators. The resulting classical SCAs are centerless and are shown to include classes of
primary generators of half-integer weights which are all smaller than 2. Some weights
are negative for N > 6. SCAs based on free field realizations and with generically
nonvanishing central charges are denoted quantum SCAs as opposed to such classical
SCAs. Thus, our use of the notion quantum is not in the quantum group sense of
q-deformations.
A new and important property of the construction is that for N 6= 2 it treats the
supercurrents asymmetrically. This is illustrated in the case N = 4 where the classical
SCA is completed and found to be of a new and asymmetric form. Thus, it is not included
651
in the standard classification of N = 4 SCAs [2024]. In particular, it deviates essentially

from the small N = 4 SCA which has otherwise been announced to be the result [14] of
a construction similar to the one employed in the present paper. This is argued not to be
the correct result. The full quantum N = 4 SCA with generic central charge is sketched in
the present paper, while proof and further discussion will appear elsewhere [25]. There we
shall also show that the small N = 4 SCA may be obtained by replacing the original world
sheet SL(2|2) current superalgebra by the related SL(2|2)/U (1) current superalgebra.
A complete classification along the lines indicated is reached when the SCAs induced
by any world sheet current superalgebra with SL(2) G0 decomposable bosonic part have
been constructed. We anticipate that the techniques employed in the present paper may be
enhanced to cover the general case and hope to come back elsewhere with a discussion on
this generalization.
As pointed out in Ref. [6], BRST invariance of the construction of the spacetime
conformal algebra is equivalent to requiring the Virasoro generators to be primary fields
of weight one with respect to the world sheet energy-momentum tensor, ensuring that
the integrated fields commute with the world sheet Virasoro algebra. This carries over
to the superconformal case, and we shall verify that the generators of our SCAs meet
the requirement of being primary of weight one with respect to the world sheet current
superalgebra Sugawara tensor.
The algebras constructed in Ref. [18] are simpler than the ones by Ito [14] as they are
based on smaller Lie superalgebras. For example, the N = 4 SCA is constructed from
an sl(2|1) Lie superalgebra. However, the central charges are essentially fixed, and the
algebras are in general not ensured to be BRST invariant. 3
The remaining part of this paper is organized as follows. In Section 2 we review
the construction of the Virasoro algebra and the immediate extension to an affine Lie
algebra [6].
In Section 3 we introduce our notation for Lie superalgebras and their associated current
superalgebras, and review the free field realizations of the latter obtained in Ref. [17].
In Section 4 we provide our explicit construction of the supercurrents, the Virasoro
generators, and the generators of the internal SL(N/2) U (1) KacMoody algebra.
In Section 5 BRST invariance of the construction is addressed.
In Section 6 we discuss the classical N extended SCAs and write down the explicit result
for N = 4.
Section 7 contains concluding remarks, while details on the Lie superalgebra sl(2|M)
are given in Appendix A.
2. Virasoro algebra
The standard Wakimoto free-field realization of the affine SL(2) current algebra with
level k [13] is
3 We thank O. Andreev for pointing out that the N = 4 SCA in Ref. [18] is nevertheless BRST invariant.
652
E = ,
H = 2 +
k + 2 ,
F = 2 +
k + 2 + k . (1)
Here and throughout the paper, normal ordering is implicit. The operator product
expansions (OPEs) of the ghost fields , and the bosonic scalar field are:
1
,
(z)(w) = 2 ln(z w),
(2)
(z) (w) =
zw
where regular terms have been omitted. In Ref. [6] it is shown that the world sheet SL(2)
current algebra with level k induces the Virasoro algebra

c 3
n n n+m,0
(3)
[Ln , Lm ] = (n m)Ln+m +
12
on the boundary of AdS3 . The generators are given by
I
dz
Ln (z),
Ln = a+ (n) n+1 E + a3 (n) n H + a (n) n1 F
(4)
Ln =
2i
with constants 4
a+ (n) = 12 (1 n)n,
a3 (n) = 12 (1 n)(1 + n),
a (n) = 12 n(1 + n).
(5)
The central charge is found to be

c = 6k p,
(6)
where p is the integer winding number:

I
dz
.
(7)
p=
2i
BRST invariance requires Ln to be conformal primary of weight 1 with respect to the
world sheet Virasoro generator
1
1
2 .
(8)
T = +
2
2 k + 2
This property is readily verified.
The Virasoro algebra is immediately extended to an affine Lie algebra if the world sheet
affine SL(2) current algebra is replaced by an affine G = SL(2) G0 current algebra, where
G0 is a Lie group. Indeed, let Ja denote the currents of the affine G0 current algebra with
central extension k 0 , and define the generators
I
dz n
(z)Ja (z).
(9)
Ia; n =
2i
From the defining OPE
a,b k 0
fa,b c Jc (w)
,
(10)
+
2
(z w)
zw
where a,b and fa,b c are the CartanKilling form and the structure constants, respectively,
of the underlying Lie algebra g0 , one finds
Ja (z)Jb (w) =
[Ln , Ib; m ] = mIb; n+m ,
[Ia; n , Ib; m ] = fa,b c Ic; n+m + npk 0 a,b n+m,0 .
(11)
4 Note that the conventions used here differ slightly from the ones used in Ref. [6], which is also reflected in a
sign discrepancy in the central charge (6).
653
Summation over properly repeated indices is implicit. The central extension k 0 p of

the spacetime affine Lie algebra generated by {Ia; n } is seen to be the one, k 0 , of the
world sheet G0 current algebra multiplied by the winding number p of the embedded sl(2)
subalgebra. As the currents Ja are spin one primary fields, the construction is readily seen
to be BRST invariant due to the decomposition SL(2) G0 .
3. Affine current superalgebra

3.1. Lie superalgebra
Let g0 and g1 denote the even and odd parts, respectively, of the Lie superalgebra
g of rank r, see Ref. [26] and references therein. = 0 1 is the set of roots
of g where 0 (1 ) is the set of even (odd) roots. The set of positive roots > 0
is + = 0+ 1+ . A choice of simple roots is written {i }i=1,...,r . A distinguished
representation is characterized by exactly one simple root being odd. Related to the
triangular decomposition
g = g h g+ ,
(12)
the raising and lowering operators are denoted E , J g+ and F , J g , respectively,

with + , while Hi , Ji h are the Cartan generators. Generic Lie superalgebra
elements are denoted Ja and satisfy
[Ja , Jb } = fa,b c Jc
(13)
where [, } is an anticommutator if both arguments are fermionic, and otherwise

a commutator. The Jacobi identities read

(14)
Ja , [Jb , Jc } = [Ja , Jb }, Jc + (1)p(Ja )p(Jb ) Jb , [Ja , Jc } ,
where the parity p(Ja ) is 1 (0) for Ja an odd (even) generator. The CartanKilling form
a,b ,
2h a,b = str(adJa adJb ),
(15)
and the Cartan matrix Aij = j (Hi ) are related as i,j = Aij j ,j . h is the dual Coxeter
number. The Weyl vector
X
X
, 1 = 12
,
(16)
= 0 1 , 0 = 12
0+
1+
satisfies i = i2 /2.
For each positive even or odd root > 0 we introduce a supertriangular coordinate
denoted by x or , respectively, where is Grassmann odd. In terms of the matrix
X
X
x f,a b
f,a b
(17)
Cab (x, ) =
0+
1+
654
one may then realize the Lie superalgebra in terms of differential operators [17]:
Ja (x, , , ) =
Va (x, ) +
r
X
Pa (x, )j ,
(18)
j =1
>0
where is the weight of the representation, and j are the labels defined by
Hj |i = (Hj )|i = j |i.
(19)
is differentiation with respect to x or depending on the parity of , whereas V and

P are finite dimensional polynomials:

0

0
0
0
Vi (x, ) = C(x, ) i ,
V (x, ) = B C(x, ) ,
X
00
0
0
(x, ) =
eC(x,) B C(x, ) 00 ,
V
00 >0
Pj (x, ) = 0,
Pi (x, ) = i ,

j
j
P (x, ) = eC(x,) ,
B(u) is the generating function for the Bernoulli numbers Bn :

X Bn
1 eu 1 X
1
u
=
un ,
=
un .
B(u)
=
B(u) = u
e 1
n!
u
(n + 1)!
n>0
(20)
(21)
n>0
The formal power series expansions (20) all truncate and become polynomials due to the
nilpotency of the matrix C (17). For later use, let us also introduce the notation V+ + for
the first of the polynomials in (20):

+

(22)
V+ + (x, ) = B C(x, ) + = B C+ + (x, ) ,
where C+ + is the submatrix of C (17) with both row and column indices positive (even or
odd) roots. V+ + is immediately seen to be invertible
1
1 X
n
1
C+ + (x, ) .
= B C+ + (x, )
=
(23)
V+ + (x, )
(n + 1)!
n>0
Most Lie superalgebras with even subalgebra g0 = sl(2) g0 have the property that the
embedding of sl(2) in g carried by g1 is a spin 1/2 representation. 5 This means that the
space of odd roots may be divided into two parts:
1 = 1 1+ ,
(24)
where the roots 1 are characterized by

1
sl(2)
= ,
2
2
sl(2)
(25)
and we have the correspondence

1+ = sl(2) + 1 ,
(26)
5 This is true for all basic Lie superalgebras with even subalgebra g0 = sl(2) g0 except osp(3|2M) where the
embedding is a spin 1 representation, see, e.g. [27].
655
sl(2) is the positive root associated to the embedded sl(2). In particular, the division (24)
is present in the case of our main interest, namely the Lie superalgebra sl(2|M) which is
considered in Section 4 and further in Appendix A.
3.2. Free-field realization
Associated to a Lie superalgebra is an affine Lie superalgebra characterized by the
central extension k, and associated to an affine Lie superalgebra is an affine current
superalgebra whose generators are conformal spin-one primary fields and have the mutual
operator product expansions: 6
Ja (z)Jb (w) =
a,b k
fa,b c Jc (w)
.
+
(z w)2
zw
(27)
We use the same notation J, E, F, H for the currents as for the algebra generators.
Hopefully, this will not lead to misunderstandings. The associated Sugawara construction
T=
1
a,b Ja Jb
2(k + h )
(28)
generates the Virasoro algebra with central charge

c=
k sdim(g)
.
k + h
(29)
The standard free-field construction [1517] consists in introducing for every positive
even root 0+ , a pair of free bosonic ghost fields ( , ) of conformal weights (1,0)
satisfying the OPE
0
(z) (w) =

.
zw
The corresponding energy-momentum tensor is

X

T =
(30)
(31)
0+
with central charge

c = 20+ = dim g0 r.
(32)
For every positive odd root 1+ one introduces a pair of free fermionic ghost fields
(b , c ) of conformal weights (1, 0) satisfying the OPE
0
b (z)c (w) =

.
zw
(33)
6 We note that the extension of the Virasoro algebra to include an affine Lie algebra (11) discussed in Section 2
may readily be generalized to an affine Lie superalgebra simply by substituting the Lie group G0 with a Lie
supergroup; the only change being that the commutator [Ia; n , Ib; m ] becomes an anticommutator for Ja and Jb
both fermionic.
656
The corresponding energy-momentum tensor is

X
c b
Tbc =
(34)
1+
with central charge

cbc = 21+ = dim g1 .
(35)
For every Cartan index i = 1, . . . , r one introduces a free scalar boson i with contraction
i (z)j (w) = i,j ln(z w).
(36)
The corresponding energy-momentum tensor

1
1
2
T =
2
k + h
has central charge
(37)
h sdim(g)
,
(38)
k + h
where the superdimension sdim(g) of the Lie superalgebra g is defined as the difference
dim(g0 ) dim(g1 ). In obtaining (38) we have used Freudenthalde Vries (super-)strange
formula
h
sdim(g).
(39)
2 =
12
The total free field realization of the Sugawara energy-momentum tensor is T = T +
Tbc + T and has indeed central charge (29).
The generalized Wakimoto free-field realization of the affine current superalgebra is
obtained by the substitution
x (z),
i k + h i (z),
x (z),
c = r
b (z),
c (z),
(40)
in the differential operator realization {Ja (x, , , )} (18), (20), and a subsequent addition
of anomalous terms linear in or c:
X
X

Va (z), c(z) (z) +
Va (z), c(z) b (z)
Ja (z) =
0+
1+
r
X

j
Pa (z), c(z) j (z)
+ k + h
j =1
+ Jaanom

(z), c(z), (z), c(z) .
Anomalous terms are only added to the lowering generators F (z):
0 for a = i, > 0,
P 0 0 0 (z)F 0 ( (z), c(z))

,
+
Jaanom (z), c(z), (z), c(z) =
P
0 (z)F
c
+
1
0
, 0 ( (z), c(z))
+
for a = < 0,
and are given by
(41)
(42)
X
F0 , 0 ( , c) = k
+
V+ + ( , c)
1
0 ,
V+ + ( , c)
, 0+ , +
V+ + ( , c)
1+ , ,+
X
F1 , 0 ( , c) = k
+
V+ + ( , c)
0 ,

V ( , c) V
( , c)
0
V ( , c) V
( , c),
0
V+ + ( , c)
1
1
1
V+ + ( , c)
1+ , 0+ , +
V ( , c) V
( , c)
0
1
,0+ , ,+
1
V+ + ( , c)
0+ , 1+ , +
657
V+ + ( , c)
V ( , c) V
( , c)
0
1
V ( , c) V
( , c)
0
1
V ( , c) V
( , c).
0
(43)
, 1+ , +
In particular, for a simple root the anomalous term F, 0 is a constant independent of

and c:

(44)
Fi , 0 ( , c) = 12 i , 0 2k + h i ,i Aii .
This concludes the explicit free field realization of general affine current superalgebras
obtained in Ref. [17], where the polynomials V , P and (V+ + )1 are given in (20) and (23).
4. Generators of the superconformal algebra
In this section we shall construct the Virasoro generators and the supercurrents of the
SCA in spacetime that may be induced by the affine SL(2|N/2) current superalgebra
on the world sheet. Some steps towards constructing the complete set of SCA generators
are also taken. For simplicity, we consider the underlying Lie superalgebra sl(2|N/2)
in the distinguished representation, see Appendix A. Let us introduce the abbreviations
= 1 , E1 = E1 , etc., for the objects related to the embedded sl(2). Hopefully, no
misunderstandings will arise, as we are also using to represent a general bosonic ghost
field argument in the polynomials V and P , though in general we will leave out the
arguments.
Using the explicit polynomials
V11 ( , c) = 1,
V11 ( , c) = 2 ,
1
V
( , c) = 2 ,
1
it is straightforward to verify that the Virasoro algebra (3) is generated by

I
dz
Ln (z),
Ln = a+ (n) n+1 E1 + a3 (n) n H1 + a (n) n1 F1 ,
Ln =
2i
(45)
(46)
658
and has central charge

c = 6k1 p1 ,
(47)
k1
= , and
= 1 ,1 k is the level of
p1 is the winding number (7) for the ghost field
the embedded sl(2) or the level in the direction 1 .
As a preparation for constructing the algebra generators, let us introduce the generators
I
dz n
(z)Ja (z)
(48)
Ja; n =
2i
and consider the OPE

1 n
a+ (n) n+1 f1 ,a c + a3 (n) n f1,a c + a (n) n1 f1 ,a c Jc
Ln (z)Ja (w) =
zw
o
na3 (n) n2 Va1 2 E1 + H1 F1
n
1
a+ (n) n+1 (z)1 ,a + a3 (n) n (z)1,a
+
(z w)2
+ a (n) n1 (z)1 ,a na3 (n) n2 (z) 2 (z)V1 (z)
o
(z) Va1 (w) .

+ (z)V1 (z) V
(49)
1
1
Here and in the following summations over repeated indices are implicit. Summations over
root indices are meant to be over all positive roots if not otherwise indicated. Actually,
this restriction is not necessary as the supertriangular coordinates are defined for positive
roots only, i.e., exists only for > 0. Now, due to the structure of the root space (see
Appendix A) we immediately obtain
[Ln , J; 0] = 0
for 0 \ {1 }.
(50)
Likewise, it follows that

[Ln , J ; 0 ] = 12 (n 1) n (n + 1)J ; n n J + ; n+1 ,

[Ln , J + ; 0 ] = 12 (n + 1) n1 nJ ;n1 (n 1) J + ;n .
(51)
The idea is to use the right hand sides in the construction of the supercurrents. To this end
let us consider the general setting where a primary field of weight h,

(52)
[Ln , m+ ] = (h 1)n m n+m+ ,
may be obtained as a commutator of the form

I
Lm , B = b(m; ; h)m+ ,
(53)
is a possible noninteger shift in the modes, whereas b is some function. From the Jacobi
identities this function satisfies the recursion relation

(n m)b(n + m) = (h 1)n m b(m) (h 1)m n b(n)
(54)
allowing the simple solution
b(m) = b0 + b1 m,
(1 h)b0 = b1 .
(55)
659
This is precisely of the form we have encountered in (51). Since we want to construct
supercurrents of weight 3/2 we should choose Z + 12 , and we define the generators
I
dz
G ; n+1/2 (z),
G ; n+1/2 =
2i
(56)
G ; n+1/2 = (n + 1) n J n n+1 J + .
Of course, it still remains to verify that the supercurrents G are indeed primary:

[Ln , G ; m+1/2 ] = 12 n m 12 G ; n+m+1/2 .
(57)
However, that follows immediately from the simple computation of the OPE Ln G ; m+1/2 ,
and we have thus constructed half of the supercurrents. Note that both commutators in (51)
lead to the same supercurrent as we have
(
2
[Ln , J ; 0 ],
(58)
G ; n+1/2 = n1
2
n+2 [Ln+1 , J + ; 0 ].
This means that G may be represented as a commutator for all integer modes n.
Let us now turn to the construction of the supercurrents G. For a = 1
it
follows from (49) that a situation like (58) occurs provided
1
1
V
+ = V .
(59)
1
This important relation is proven in Appendix A where also some identities involving V
are derived. We are led to define the supercurrents G by

I
dz
G ; n1/2 =
G ; n1/2 (z),
2i
G ; n1/2 = (n 1) n J + n n1 J +

1
2
n(n 1) n2 V
E1 + H1 F1
V1
+ n(n 1) n2 2 V1 + V1 V
1
= (n 1) n J + n n1 J +
n(n 1) n2
b k 1 + h /2
1

V1 ,

1
(60)
where
= 2 V1 + V1 V
.
1
1
(61)
As indicated in (60), one may show that the upper root index is always an odd (and
positive) root. The analogue to (58) reads:
(
2
[Ln , J ; 0 ],
(62)
G ; n1/2 = n+1
2
n2 [Ln1 , J + ; 0 ].
660
As in the case of the supercurrents G, there is a free overall scaling. However, in order to
produce the conventional prefactor of plus one multiplying the Virasoro generator in the
anticommutator {G, G} (see (73)), the relative factor is fixed. 7
Before proving that G is a primary field of weight 3/2,

(63)
Ln , G ; m1/2 = 12 n m + 12 G ; n+m1/2 ,
let us observe the following property of the construction. Consider the Jacobi identity

(64)
J; 0 , [Ln , J ; 0 ] + Ln , [J ; 0 , J; 0 ] = J ; 0 , [Ln , J; 0 ] = 0,
where 0 \ {1 } is any even (positive or negative) root different from 1 . From the
construction of G and G it follows that 8
G ; n+1/2 = [J ; 0 , G ; n+1/2 ] for 0 \ {1 },
G ; n1/2 = [J ; 0 , G ; n1/2 ]
for 0 \ {1 }.
(65)
Here we have used that
f , = f , = 1
for 0 \ {1 }.
(66)
Besides providing information on the underlying algebraic structure of our construction,

the translational property (65) may be used to reduce considerations for general supercurrents to similar ones for the supercurrents G2 ; n+1/2 and in particular G2 ; n1/2 , 2 is
the only fermionic simple root, see Appendix A. Thus, as a first application we shall prove
that G is primary. From the Jacobi identities we find

Ln , G ; m1/2 = J2 ; 0 , Ln , G2 ; m1/2
(67)
leaving us with the task of proving that G2 is primary. To that end we work out
1
2
=
= 12 c + C,
V
2
1
(68)
where c and C are the fermionic ghost fields associated to the odd roots 2 and 1 + 2 ,
respectively, and the supercurrent becomes
G 2 ; m1/2 = (m 1) m J2 + m m1 J1 2

1
m(m 1) m2 V
2 E1 + H1 F1 12 .
2
(69)
Now, one may compute the OPE Ln G 2 ; m1/2 and reduce the result using
V11 +2 = 12 c,
to the desired commutator

Ln , G2 ; m1/2 =
V11 +2 = C,
1
1
2n m + 2
1 +2
1
V
= 12 V
1
2
G2 ; n+m1/2 .
(70)
(71)
This concludes the proof of (63) that G is primary of weight 3/2.

7 A more commonly used convention is a prefactor of plus two. However, we have found it natural to define G
and G without introducing any powers of 2. To comply with the standard convention is straightforward, though.
8 Subtleties for n = 1, 2 are immediately resolved by the alternative commutator representations (58) and
(62).
661
4.1. Affine SL(N/2)U (1) current subalgebra

In order to derive the entire set of generators of the SCA, one should first consider the
anticommutators {G , G }, {G , G } and {G , G }. It is readily seen that
{G ;n+1/2 , G ;m+1/2 } = 0,
(72)
whereas a rather cumbersome but essentially straightforward computation reveals that

G ; n+1/2 , G ; m1/2 = , Ln+m + (n m + 1)K ; ; n+m
+ 16 cn(n + 1)n+m,0 , ,
where the current K is defined by
I
dz
K ; ; n (z),
K ; ; n =
2i
1
n
c
K ; ; n = n n1 V
( J + J ) f , Jc

+ 12 , n1 n 2 E1 + H1 F1 H1

1
+ n n1 V+ V V
.
(73)
(74)
There are several ways of representing K ; ; n of which the following two turn out to
be useful:
(
1
({G ; n+1/2, G ; 1/2} , Ln ),
(75)
K ; ; n = n+1
1
n1 ({G ; 1/2 , G ; n1/2 } , Ln ).
In particular, they may be used in a straightforward verification that the current K is
primary of weight 1:

(76)
Ln , K ; ; m = mK ; ; n+m .
In Section 6 we shall provide evidence from considering the classical counterpart, that
{K ; ; n } generate an affine SL(N/2) U (1) current subalgebra. The number of
generators is, accordingly:
1 2

= (N/2)2 = dim sl(N/2) + 1.
(77)
+
A novel feature of our construction is its asymmetry in the two sets of supercurrents {G}
and {G}, originating in (72) and

G ; n1/2 , G ; m1/2 6= 0 for 6= , n 6= m, n + m 6= 1.
(78)
A proof at the classical level is presented in Section 6, however it is obvious that a result
as (78) at the classical level remains true at the quantum level. The right hand side of (78)
involves new fields to be introduced in Section 6.
4.2. Underlying Lie superalgebra
Here we shall express the underlying Lie superalgebra in terms of selected modes of the
SCA generators. From the Virasoro generator we have:
E1 = L1 ,
H1 = 2L0 ,
F1 = L1 ,
(79)
662
while the supercurrents allow us to write:

J = G ; 1/2,
J = G ; 1/2 ,
J + = G ;1/2 ,
J + = G ; 1/2.
(80)
As we have
K ; ; 0 = 12 , H1 f , c Jc ,

u
X
0
Hu +1 + Jv u ,
{J2 u , J(2 v ) } = u,v 2H2
(81)
i=u0
where Jv u is defined only for v 6= u (see Appendix A), we find that the remaining
(N/2)2 Lie superalgebra generators are given by
Jv u = K2 u ; (2 v ); 0
for u 6= v,
H2 = K2 ; 2 ; 0 L0 ,
Hi = K2 i1 ; (2 i1 ); 0 K2 i2 ; (2 i2 ); 0
for i = 3, . . . , N/2 + 1.
(82)
4.3. N = 2 superconformal algebra

For N = 2 the only positive -root is 2 (i.e. 1
+ = {2 }) and we have the four
generators:
Ln = a+ (n) n+1 E1 + a3 (n) n H1 + a (n) n1 F1 ,
Gn+1/2 = (n + 1) n J2 n n+1 J ,
G n1/2 = (n 1) n J2 + n n1 J

1
n(n 1) n2 V
2 E1 + H1 F1 12 ,
2
1
( J J2 ) 12 n (H1 + 2H2 )
Kn = n n1 V
2

+ 12 n n1 2 E1 + H1 F1 ,
contribution 12 (k + 1/2)n n1
where = 1 + 2 . Note that the

integration as
I
dz n1
= npn,0 = 0.
n
2i
The N = 2 SCA becomes
(83)
to Kn vanishes upon
c
[Ln , Lm ] = (n m)Ln+m + (n3 n)n+m,0 ,
12

[Ln , Am ] = h(A) 1 n m An+m , A {G, G, K},

{Gn+1/2 , Gm+1/2 } = Gn1/2 , Gm1/2 = 0,

Gn+1/2 , Gm1/2 = Ln+m + (n m + 1)Kn+m + 16 cn(n + 1)n+m,0 ,

Kn , Gm1/2 = 12 Gn+m1/2 ,
[Kn , Gm+1/2 ] = 12 Gn+m+1/2 ,
[Kn , Km ] =
1
12 cnn+m,0 ,
(84)
663
and closure is seen to be ensured by the four generators (83). This result has already been
obtained by Ito [14], though his construction is based on a slightly different but equivalent
free field realization of the associated affine SL(2|1) current superalgebra.
5. BRST invariance
Before turning to the classical SCA let us discuss an important property of our
construction. As pointed out in Ref. [6], BRST invariance of the construction of the space
time conformal algebra from a world sheet SL(2) current algebra requires the Virasoro
generators to be primary fields of weight one with respect to the world sheet energymomentum tensor. This ensures that the integrated fields commute with the world sheet
Virasoro algebra. This requirement carries over to the superconformal case, where all
(super-)currents (the Virasoro generators L, the supercurrents G and G, the affine Lie
algebra generators K, etc.) are primary of weight one with respect to the Sugawara energymomentum tensor of the world sheet SL(2|N/2) current superalgebra. A naive inspection
immediately tells that the four types of currents considered so far have weight one, so
all we need to verify is that they are primary. This amounts to verifying that third- and
higher-order poles in the OPEs with the Sugawara tensor T all vanish. From the free-field
realization of T it follows that no higher-order poles than third order appears. Using that
the affine currents J are primary fields, the BRST invariance of the supercurrents G is
readily confirmed as V1 = 0. The BRST invariance of the Virasoro generators L follows
from (45), whereas the invariance of the supercurrents G is ensured provided
1
n2 1
V +
0 = n(n 1) n1 V
+ n(n 1)
1
1
n1 1
n(n 1)V
V1 + (n 1) n2 V11 (n 2) n3 V
n
1

1
n
n1
n2
V1
V1 V .
n(n 1) V1 +

(85)
The three lines vanish separately due to (59), (45) and (131), respectively. Likewise, BRST
invariance of the affine Lie algebra generators K amounts to verifying
0 = n n1 f , c Vc1

+ 12 , n (n + 1) n V11 + n n1 V11

1
(n 1) n2 V
n n1 V11 .
1
(86)
Here we have used that V1 = 0, and the identity (86) follows from (45) and (133).
One may take a more general point of view observing that the (anti-)commutator of two
BRST invariant fields commutes with the world sheet Virasoro generators. This follows
from the Jacobi identities. Thus, having established that all but one field appearing on the
right hand side of a (anti-)commutator (of two BRST invariant fields) are BRST invariant,
is sufficient to conclude that the final field is likewise BRST invariant. A trivial example is
the alternative deduction that K is BRST invariant following from the anticommutator (73).
664
6. Classical superconformal algebra

Here we shall distinguish between classical and quantum SCAs. Our use of the notion
quantum is not in the quantum group sense of q-deformations but rather as opposed to
classical as described in the following. Let us recall the situation for free field realizations
of affine current superalgebras discussed in Section 3. In that case one may start with
a first-order linear differential operator realization of the underlying Lie superalgebra. The
free field realization of the associated current superalgebra is then obtained by substituting
with (normal ordered products of) free fields (40) and subsequently adding quantum
corrections, anomalous terms or normal ordering terms (41). We shall denote the
differential operator realization a classical limit or version of the associated quantum
free field realization. A classical algebra thus defined has vanishing central extensions. 9
A similar situation may be expected in the present case. Thus, there should exist
a classical counterpart of the full SCA which allows a differential operator realization
(and accordingly has vanishing central extensions). Based on this assumption our program
is to first work out the classical SCA for then to perform the appropriate substitutions
and additions of anomalous terms in order to obtain the full (quantum) SCA. It should be
stressed that to each mode of the generators of the quantum SCA, there is an associated
differential operator. This results in an infinite dimensional algebra of differential operators
contrary to the situation described above where the classical algebra is a standard (finite
dimensional) Lie superalgebra.
Having identified the differential operator
X
Y a (x, )Ja (x, , , )
(87)
A(x, , , ) =
a
as a generator of the classical SCA, the corresponding quantum generator

I
dz
A(z)
(88)
A=
2i
is obtained by performing the substitutions (40) in A(x, , , ) and adding appropriate
anomalous terms linear in derivatives of the spin 0 ghost fields in order to produce A(z):
X
X

Y a (z), c(z) Ja (z) +
X (z), c(z) (z)
A(z) =
a
0+
X0 (z), c(z) c (z).
(89)
1+
So with the ansatz (87), A(z) is linear in the affine currents Ja (z) with spin 0 ghost
field dependent coefficients. Note that in the expression (89) some anomalous terms are
hidden in the definition of Ja (z), cf. (41).
9 It should be stressed that a nonvanishing central charge of a classical Virasoro algebra may well exist when
classical is defined to denote single contractions only, as 2 (z) 2 (w) = 12/(z w)4 . Here we have used the
convention (z)(w) = 2 ln(z w). However, terms like 2 are excluded in our differential operator realization
picture employed in the present paper. Note that BRST invariance is used as an implicit guideline as 2 has
weight 2 with respect to the world sheet energy-momentum tensor.
665
The question of BRST invariance of A (88) may be addressed even without explicit
knowledge on the anomalous term. This follows from the fact that any term of the form
P 0
P
Z ( , c) +
Z ( , c) c is primary of weight one. Thus, in order to establish
that A is BRST invariant it suffices to consider the term linear in the affine (super-)currents.
In the following we shall accordingly define a classical differential operator to be BRST
invariant when its naively quantized form linear in the affine (super-)currents is BRST
invariant.
Before continuing our program let us briefly justify it. It has turned out to be an
immense technical task to complete the derivation of the SCA for general N . Even at
the classical level, the computations are rather involved. A study of the centerless classical
SCA seems therefore a natural first project to concentrate on, and one from which one
may get structural insight into the full quantum SCA. In the following we shall present
some essential steps in the direction of deriving the classical SCA. In the presumably most
interesting case of N = 4, the classical SCA is completed. The full quantum N = 4 with
generic central charge will be presented elsewhere [25].
6.1. Algebra generators
In the remaining part of this section all fields A are represented by their classical
differential operator analogues A(x, , , ). To be explicit, let us summarize our findings
for the classical generators:
Ln = a+ (n)x n+1 E1 + a3 (n)x n H1 + a (n)x n1 F1 ,
G ; n+1/2 = (n + 1)x n J nx n+1 J + ,
G ; n1/2 = (n 1)x n J + nx n1 J +

1
2
n(n 1)x n2 V
x E1 + xH1 F1 ,
1
n
c
K ; ; n = nx n1 V
(xJ + J ) x f , Jc

+ 12 , x n1 nx 2 E1 + (n 1)xH1 nF1 ,
h(L, G, G, K) = (2, 3/2, 3/2, 1),
(90)
Ja (E, H or F ) denotes the differential operator Ja (x, , , ) given in (18) and (20), while
1
V
is the polynomial in the supertriangular coordinates x and given in (20). Here and
in the following x may denote either x 1 or a general triangular coordinate argument,
though it should be clear from the context which it is. h(Am ) indicates that Am is primary
of weight h:

(91)
[Ln , Am ] = (h(A) 1)n m An+m .
The generators respect among others the anticommutators:

{G ; n+1/2 , G ; m+1/2} = 0,

G ; n+1/2 , G ; m1/2 = , Ln+m + (n m + 1)K ; ; n+m .
(92)
We shall now discuss the subalgebra generated by {K}. One finds straightforwardly
[K ; ; n , K ; ; m ] = , K ; ; n+m , K ; ; n+m .
(93)
666
In order to show explicitly that this has the affine structure

SL(N/2) U (1)
(94)
we introduce the following notation. From Appendix A we know that any root 1
+
may be represented as 2 u for some u = 1, . . . , N/2, so abbreviate K by
Ku; v; n = K2 u ; (2v ); n .
(95)
Note also that v u > 0 for u > v. Define now:

ei; n = Ki+1; i+1; n Ki; i; n ,
H
ei; n = Ki+1; i; n ,
E
ei; n = Ki; i+1; n ,

F
(96)
where i = 1, . . . , N/2 1 by construction. One may then show that these correspond to
the Chevalley generators of an (centerless) affine SL(N/2) Lie algebra. In general, the
currents Ku; v; n correspond to raising operators for u > v, and to lowering operators for
u < v. Furthermore, the generator
Un =
N/2
X
Ku; u; n
(97)
u=1
ei,m
is seen to commute with all ladder operators Ku; v6=u; m , with the Cartan generators H
and with Um itself. Thus, U generates a (centerless) U (1) current algebra and we have the
decomposition (94).
Let us return to the anticommutator {G , G } and prove the classical counterpart
of the assertion (78). The anticommutator may be computed directly or obtained as
a special case of a much more general consideration: introduce the operator

1
1
G ,..., ; m1+k/2 = V
(m 2 + k)x m J + mx m1 J +
...V

k
k1
+
..
.

+ (m 2 + k)x m J + mx m1 J + V1 . . . V1
1
1
k
2
1
1
m
V
.
.
.
V
E
(m
+
k
1)(m
+
k
2)x
1
+ (m + k 2)mx m1 H1

m(m 1)x m2 F1
(98)
which is seen to reduce to G for k = 1. G ,..., ; m is bosonic (fermionic)

k
1
for k even (odd). In the latter notation the mode m is meant to be integer or halfinteger depending on the parity of the generator, i.e. depending on k being even or odd,
respectively. Note that G ,..., is antisymmetric in its root indices. J is defined not
k
1
1
, and is only written to the left of the V -monomial for convenience
to act on V
...V

j+1
of notation. Thus, within G ,..., one has

1
J V1
j
j+1
. . . V1
k
= (1)
kj
V1 . . . V1 J .
j+1
667
This resembles normal ordering needed in the free field realization and may be regarded as
a normal ordering of the differential operator. It is to be employed throughout this section.
One may now work out the (anti-)commutator

G ,..., ; n , G ,..., ; m
k
l
1
1

(99)
= (k 2)m (l 2)n G ,..., , ,..., ; n+m .
k
1
|1
|
generate a subalgebra of dimension 2| | 1 =
We observe that {G ,..., }k=1
1
k
2N/2 1 and that 2N/21 of the generators are fermionic. Note that the commutator for
k = l = 2 vanishes identically. One may also show that (classically) the generators are
primary:

Ln , G ,..., ; m = (1 k/2)n m G ,..., ; n+m , h = 2 k/2. (100)
k
Thus, including L as a generator of the subalgebra it has dimension 2| | and equal

numbers of bosonic and fermionic generators. BRST invariance is readily verified, either
directly or as a consequence of the recursive relation (99) and the general approach of
Section 5.
We note that the generator (98) may be written in the following compact form:
G ,..., ; m1+k/2
1
k
( k
X

= x m2
(m + k 2)x 2 J + mxJ +
j
j =1
1
V
(m + k 1)(m + k 2)x E1 (m + k 2)mxH1 + m(m 1)F1

2
1
1
,
V
...V

1
(101)
1
1 V ij .
V
i
(102)
1
As G ,..., is a first-order differential operator, /V
is meant to act only on the
k
explicitly written products of V s within G ,..., itself. A special situation occurs for
k
1
k = 2 since we then have
G , ; m = mS , ; m ,
1
2
1
2
( 2
X

x 2 J + xJ +
S , ;m = x m2
1
j =1
+ (m 1)F1
2
1 (m + 1)x E1 mxH1
V

V 1 V 1 ,
1
(103)
668
and S , is readily seen to have weight 0. G , may be interpreted as the

1
2
1
2
derivative of the scalar S , .
1
2
The list of generators presented hitherto is by no means exhaustive. Let us consider
generators which may be obtained by the adjoint action of {G } on G ,..., . Firstly,
1
k
we find the (anti-)commutator
[G ; n+1/2 , G ,..., ; m }
1
k
= m (k 2)n ; ,..., ; n+m+1/2
k
k2
k3
k
X
(1)j 1 , G
j
j =1
,
1 ,...,[
j ,...,k ; n+m+1/2
(104)
where ; ,..., corresponds to the special case l = 1 in the following general

k
1
expression:
,..., ; ,..., ; m
l
l
X
= x m(kl)/21
(1)i1

m (k l)/2 J m + (k l)/2 xJ+
i
i=1
)
b
1
1
x
f , Jc
V
...V
1
1 . . .
1 . . .
1
i
j
V V
V
V
k
1
j =1
j
1
i
l
( k
X

l
x m(kl)/21
x 2 J + xJ +
+
j
j V 1
kl2
j
j =1

m + (k l)/2 x 2 E1 m + (k l)/2 1 xH1
)
1
1
+ m (k l)/2 F1
V
(105)
...V
.
1 . . .
1
V
V
k
1
k
X
These generators are only defined for certain integer pairs (l, k) to be discussed
below. A hat over an object indicates that the object is left out. We observe that
,..., ; ,..., is bosonic (fermionic) for l + k even (odd), and that it is
1
l
1
k
antisymmetric in the positive root indices and in the negative root indices, separately. Using
the polynomial relations listed at the end of Appendix A, one may show that satisfies
the recursion relation

G ; n+1/2 , ,..., ; ,..., ; m
1
k2
kl2
k
X
(1)lj +1 ,
j
j =1
1 ,...,l ; 1 ,...,j ,...,k ; n+m+1/2

.
k l 2 ,1 ,...,l ; 1 ,...,k ; n+m+1/2
In addition, ,..., ; ,..., may be shown to be primary of weight
1
l
1
k

h ,..., ; ,..., = 1 (k l)/2.
(106)
(107)
669
As for G ,..., (98), BRST invariance of ,..., ; ,..., is verified straightfor1

k
1
l
1
k
wardly, either directly or indirectly. There is a slight subtlety in (106) for k = l = 1, which
is easily resolved, though, as (105) may be interpreted as G . One then has
[K ; ; n , G ; m+1/2 ] = , G ; n+m+1/2 12 , G ; n+m+1/2 .
(108)
In the definition of ,..., ; ,..., we obviously have k l 2 6= 0, and as the

l
expression is obtained by the adjoint action of G on G ,..., , 1 6 l < k 2, is seen

1
k
to be a valid domain. It is relevant for the discussion below on N = 4 that besides (l, k) =
(1, 1) which corresponds to the K generators, also (l, k) {(2, 1), (1, 2), (2, 2), (3, 2)} may
be reached by this adjoint action. It turns out that three of these four generators may be
expressed in terms of simpler generators. First we note that up to permutations in the root
indices, , , ; , is the only nonvanishing generator of type (3, 2) (provided
2
6= i and 1 6= 2 , of course), while , ; is the only nonvanishing generator

of type (2, 1) (provided 6= ). It is easily shown that they satisfy

, , ; , ; m+1/2 = , , , , G ; m+1/2
1
2
3
1
2
3
1
2
2
3
2
2
1
1
, , , , G ; m+1/2
3
1
1
2
3
2
1
1
2
+ , , , , G ; m+1/2 ,
2
, ; ; m+1/2 = , G , G .
1
(109)
Secondly, we observe that , ; , is the only nonvanishing generator of type

(2, 2) and that it may be reduced as
, ; , ; m = , K2 ; ; m + , K2 ; ; m
1
+ , K1 ; ; m , K1 ; ; m.
(110)
For l > k + 1, k = 1, 2, the generator ,..., ; ,..., is readily seen to vanish.

1
l
1
k
Relevant for the discussion on the N = 4 SCA in the following, is the result

K ; ; n , G ; m1/2 = 12 , G ; n+m1/2 , G ; n+m1/2
+ n ; , ; n+m1/2 ,
(111)
which one may work out explicitly.

Finally, let us add a comment on the closure of the algebra. Based on the results
obtained so far one should expect that the SCA is generated by a BRST invariant set
e
} where 0 6 l, k and l 6 k + 1. The notation implies for example

{
1 ,...,
l ; 1 ,..., k
e
e = L. Not all
e = G and
=G
, while for l = k = 0 we have
1 ,...,k
1 ,...,k
e need be present as independent fields as illustrated by the reductions (109) and

the fields
e generates
(110). At present we do not have a complete proof of the assumption that {}
the SCA, though it is easy to prove (using the polynomial relations of Appendix A) that
the following 8 BRST noninvariant building blocks
1
1
x n+ J ,
V
...V

1
k+
1
1
V
x n+ J + ,
...V

1
+
k+
670

1
1
1
nc
c
[
V
. . . V
...V
x f , Jc ,

j
1
1
x n J + ,
V
...V

1
V
1
. . . V
nH
H1 ,
1
1
V
x nE E1 ,
...V

+
k
1
1
V
x n J ,
...V

kc
1
V
kE
1
. . . V
x nF F1
(112)
are close under (anti-)commutations.

Despite the fact that closure of the BRST invariant algebra is still not ensured, nor are all
the BRST invariant generators at hand, above we have presented substantial evidence that
our construction does produce a finitely generated and BRST invariant N extended SCA.
Below we shall demonstrate that this is indeed the case for N = 4. We intend to come
bask elsewhere with further discussion on the BRST invariant SCA and its quantization as
described above.
As |1
| = N/2, we observe that for N = 6, 8, . . . the SCA contains primary generators
of negative weight causing problems for the unitarity of the associated CFT, and in
particular for its applications to string theory. From an algebraic point of view, however,
we see no severe obstacles arising from the appearance of negative weights, and believe
that further studies of these algebras are warranted. Their explicit realizations and linearity
in the affine currents seem to make such investigations feasible.
6.2. Classical N = 4 superconformal algebra
We recall that sl(2|2) has precisely two positive (fermionic) -roots, see Appendix A:
1
+ = {2 , 2+3 2 + 3 },
2 = 2 1 ,
3 = 1 2 .
(113)
Using the results on (classical) SCA obtained above for general N we may almost
immediately complete the (classical) N = 4 SCA. We find that closure is ensured by the
following 12 BRST invariant generators which may be characterized as:
Virasoro generator: L,
supercurrents:
G2 , G2+3 , G2 , G2+3 ,
e = K ; ,
affine SL(2):
E
2+3
h = 2,
h = 3/2,
e = K2+3 ; 2+3 K2 ; 2 ,
H
e = K ; ,
F
affine U (1):
U = K2 ; 2 + K2+3 ; 2+3 ,
h = 1,
h = 1,
fermions:
2 = 2+3 ; 2+3 ,2 ,
2+3 = 2 ; 2 ,2+3 ,
h = 1/2,
S,
(nSn = G2 ,2+3 ; n
h=0
h = 1).
scalar:
2+3
The nontrivial (anti-)commutators are

[Ln , Am ] = (h(A) 1)n m An+m ,
(114)
671
{G ; n+1/2 , G ; m+1/2 } = 0,

G ; n+1/2, G ; m1/2 = , Ln+m + (n m + 1)K ; ; n+m ,

G2 ; n1/2 , G2+3 ; m1/2 = (n m)(n + m 1)Sn+m1 ,

em = 2E
en , F
em = 2F
en+m ,
en+m ,
en , E
H
H

em = H
en+m ,
en , F
E

en , G2 ; m+1/2 = G2+3 ; n+m+1/2 ,
E

en , G2+3 ; m+1/2 = G2 ; n+m+1/2 ,
F

en , G ; m+1/2 = G ; n+m+1/2 ,
H
2
2

en , G2+3 ; m+1/2 = G2+3 ; n+m+1/2 ,
H

en , G2+3 ; m1/2 = G2 ; n+m1/2 n2 ; n+m1/2 ,
E

en , G2 ; m1/2 = G2 ; n+m1/2 + n2 ; n+m1/2 ,
H

en , G2+3 ; m1/2 = G2+3 ; n+m1/2 n2+3 ; n+m1/2 ,
H

en , G2 ; m1/2 = G2+3 ; n+m1/2 n2+3 ; n+m1/2 ,
F

Un , G2 ; m1/2 = n2 ; n+m1/2 ,

Un , G2+3 ; m1/2 = n2+3 ; n+m1/2 ,
[Sn , G2 ; m+1/2 ] = 2+3 ; n+m+1/2 ,
[Sn , G2+3 ; m+1/2 ] = 2 ; n+m+1/2 ,
{G2+3 ; n+1/2 , 2+3 ; m1/2} = Un+m ,
{G2 ; n+1/2 , 2 ; m1/2 } = Un+m ,

G2 ; n1/2 , 2+3 ; m1/2 = (n + m 1)Sn+m1 ,

G2+3 ; n1/2 , 2 ; m1/2 = (n + m 1)Sn+m1 ,

en , 2+3 ; m1/2 = 2 ; n+m1/2 ,
E

en , 2 ; m1/2 = 2+3 ; n+m1/2 ,
F

en , 2 ; m1/2 = 2 ; n+m1/2 ,
H

en , 2+3 ; m1/2 = 2+3 ; n+m1/2 ,
(115)
H
Am denotes any of the 12 BRST invariant generators listed in (114). We observe that only
the derivative of the scalar S appears on the right hand sides of (115). Thus, the zero mode
of S decouples from the algebra. One may verify explicitly that the Jacobi identities are
satisfied.
Finally, we note that this centerless N = 4 SCA is of a new and asymmetric form.
In particular, it deviates essentially from the small N = 4 SCA announced in Ref. [14]
to be obtained by a similar construction. Even though the free field realization of the
associated affine SL(2|2) current superalgebra used in Ref. [14] is slightly different from
ours, one may show that the result in Ref. [14] for the N = 4 SCA is incorrect. One way of
reaching this conclusion is to consider the analogue to our (78) and specialize to the case
n = 0. In the notation of Ref. [14], this corresponds to considering the anticommutator
H dz
1
2
1
j1 +2 (still in the notation of
{G1/2 , Gm1/2 } in which case G1/2 reduces to 2i
672
Ref. [14]). We find that the anticommutator for generic m is nonvanishing in agreement
with our result but contrary to the definition of the small N = 4 SCA. Nevertheless, the
small N = 4 SCA can be obtained by a construction equivalent to the one employed above.
One simply has to replace the original world sheet SL(2|2) current superalgebra by an
SL(2|2)/U (1) current superalgebra, whereby the resulting spacetime SCA reduces to the
standard small N = 4 SCA. This will be discussed further in Ref. [25]. There we shall also
work out the full quantum N = 4 SCA for generic central charge. For completeness, we
include here the resulting central extensions of the classical N = 4 SCA:

c 3
n n n+m,0 ,
[Ln , Lm ] = (n m)Ln+m +
12

G ; n+1/2 , G ; m1/2 = , Ln+m + (n m + 1)K ; ; n+m
+ 16 cn(n + 1)n+m,0 , ,
en+m + 1 cnn+m,0 ,
=H
em
en , F
E
6

em = 1 cnn+m,0 .
en , H
H
3
(116)
All other (anti-)commutators remain unchanged.
7. Conclusion
In the present paper a new class of two-dimensional N extended SCAs has been
discussed. The algebras are induced by free field realizations of affine SL(2|N/2) current
superalgebras, where N is even. In the framework of string theory on AdS3 the affine
SL(2|N/2) current superalgebra resides on the world sheet providing a spacetime SCA
on the boundary of AdS3 . The construction generalizes recent work by Ito [14]. The
Virasoro generators, the N supercurrents, and the generators of an internal SL(N/2)U (1)
KacMoody algebra have all been constructed explicitly. Reducing the considerations to
a classical centerless limit has provided additional insight into the structure of the full
SCA. BRST invariance has also been addressed. The classical N = 4 SCA is complete
and of a new type. In particular, it differs from the small N = 4 SCA. It also illustrates
the new and important property of the general construction that it treats the supercurrents
asymmetrically.
The results presented here offer (stringy) representations of superconformal algebras
which are linear in the currents. This suggests that they may be useful when discussing
representation theoretical questions, and in the computation of correlation functions. Many
other applications may be envisaged.
Several classes of Lie supergroups enjoy decompositions of the bosonic part G =
SL(2) G0 as in the case of SL(2|N/2). Based on their associated current superalgebras,
we anticipate that other classes of SCAs may be constructed along the lines employed in
the present paper. This is currently being investigated.
In the classification of CFT with extended symmetries, the construction of SCAs in the
present paper presents an alternative to conventional Hamiltonian reduction and otherwise
constructed nonlinearly extended SCAs [2733]. Whole new classes of extended Virasoro
673
algebras seem to be the result of it. There are strong indications that we are even able to
produce new and purely bosonic (and linear) extensions of the Virasoro algebra. These will
be the subject of a forthcoming publication.
Acknowledgements
The author is grateful to K. Olsen and J.L. Petersen for comments at early stages of this
work, and thanks The Niels Bohr Institute, where parts of this work were done, for its kind
hospitality.
Appendix A. Lie superalgebra sl(2|M)

The root space of the Lie superalgebra sl(2|M) in the distinguished representation may
be realized in terms of an orthonormal two-dimensional basis {1 , 2 } and an orthonormal
M-dimensional basis {u }u=1,...,M with metrics
0 = ,0 ,
u u0 = u,u0 ,
u = 0.
(117)
The 12 (M + 1)(M + 2) positive roots are then represented as

1+
+ = {1 u | u = 1, . . . , M},
0+ = {1 2 } {u v | u < v},

1
+ = {2 u | u = 1, . . . , M},
(118)
where the M + 1 simple roots i are

1 = 1 2 ,
2 = 2 1 ,
u+2 = u u+1 .
(119)
The associated ladder operators E , F , and the Cartan generators Hi admit a standard
oscillator realization (see, e.g. [34]):
E1 2 = a1 a2 ,
E u = a bu ,
Eu v = bu bv ,
F1 2 = a2 a1 ,
F u = bu a ,
Fu v = bvbu ,
H1 = a1 a1
H2 = a2 a2
Hu+2 = bu bu
()
a2 a2 ,
+ b1 b1 ,
(120)
bu+1
bu+1 ,
()
where a and bu are fermionic and bosonic oscillators, respectively, satisfying

() ()

bu , bv = u,v ,
bu , a = 0.
a , a0 = ,0 ,
(121)
It is not possible to design a root string from u v < 0 to 1 implying that

Vu1v = 0.
(122)
This fact is used in deriving (50). However, root strings from to 1 do exist. They are
of the form
+ + (2 u ) + 1 = 1 ,
+ + (1 u ) = 1 ,
where root strings from + are obtained by inserting an additional 1 :
(123)
674
+ + + 1 + + (2 u ) + 1 = 1 ,
+ + + (2 u ) + 1 + 1 = 1 ,
+ + + 1 + + (1 u ) = 1 ,
+ + + (1 u ) + 1 = 1 ,
(124)
Possible additions of positive even roots v 0 v are indicated by . Let us consider the
polynomials (20)
X
1
1
1
n
(C)
(125)
V =
n!

n>0
and compare the two polynomials in order to derive the relation (59). This we do by
1
1
n+1 in V 1 . There are two cases, as such
considering n!1 (C)n in V
and (n+1)! (C)
+

u
u
1
2
or c
for some u, and we will discuss them separately. In
terms involve either c
the first case the relevant root strings are the lower one in (123) and the two lower ones in
(124). Their differences are characterized by the structure constants
f(2 u ),(1 u ) 1 = 1
f(1 v ),1
(2 v )
and
(126)
f(2 u ),(1 u )
= 1,
f(1 u ),1 u fj,1
= 1. (127)
1
V
,
the situations (127) occur

Now, each time the situation (126) occurs in n!1 (C)n in
1
1
(C)n+1 in V
.
This
is
in accordance with (59).
n times and once, respectively, in (n+1)!
+

u
2
leads to the characterizations
A similar analysis of the terms involving c
f(2 u ),2 u j fj,1 1 = 1,
f(1 v ),1
(2 v )
f(1 u ),2 u
and
(128)
f(2 u ),2 u fj,1

1
f1 ,1 f1,1
= 1,
= 2.
(129)
1
Each time the situation (128) occurs in n!1 (C)n in V
, the situations (129) occur n 1
1
1
n+1
in V + . However, as the last situation
times and once, respectively, in (n+1)! (C)
contributes with a factor 2, the terms involving c2 u also agree with the relation (59),
which is thereby proven.

A.1. Polynomial relations
Using that the polynomials V enter in a differential operator realization of a Lie
superalgebra leads to the polynomial relations
fa,b c Vc = Va Vb (1)p(a)p(b)Vb Va ,
(130)
as discussed in Ref. [17]. The parity p(a) of the index a is defined as 1 for a an odd root
and 0, otherwise. Particularly useful are the following relations:
1
V1 V
= 0,
1
1
V1 V
= V ,
V 1 = V
V
1
V + V
= , ,
(131)
1
V V
= , ,
1
1
1
V + V
= V V ,
c
f , Vc1
1
f , c Vc V
f + , c Jc
f + , + c Jc
f , + c Jc
f , c fc, d Jd
f , c fc,+ d Jd
f , c fc, d Jd
f , c fc,+ d Jd
f , c fc,1 d Jd
f , c fc,1 d Jd
f , c f , d fc,d e Je
675
1
V V
= 0,
(132)
= , ,
1
= , V
,
(133)
= , E1 ,
= , H1 + f , c Jc ,
= , F1 ,
(134)
= , J , J ,
= , J + ,
= , J , J ,
= , J + ,
= , E1 ,
f , c fc,1 d Jd = 0,
= , F1 ,
(135)
= , f , Jc , f , Jc .
c
(136)
In deriving some of these relations we have made use of the explicit oscillator realization
(120).
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231.

S. Gubser, I. Klebanov, A. Polyakov, Phys. Lett. B 428 (1998) 105.
J. Maldacena, A. Strominger, J. High Energy Phys. 12 (1998) 005.
J. de Boer, Nucl. Phys. B 548 (1999) 139.
A. Giveon, D. Kutasov, N. Seiberg, Adv. Theor. Math. Phys. 2 (1998) 733.
D. Kutasov, F. Larsen, R.G. Leigh, Nucl. Phys. B 550 (1999) 183.
J. de Boer, H. Ooguri, H. Robins, J. Tannenhauser, J. High Energy Phys. 12 (1998) 026.
M. Yu, B. Zhang, Nucl. Phys. B 551 (1999) 425.
D. Kutasov, N. Seiberg, J. High Energy Phys. 04 (1999) 008.
O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Phys. Rept. 323 (2000) 183.
J.D. Brown, M. Henneaux, Commun. Math. Phys. 104 (1986) 207.
M. Wakimoto, Commun. Math. Phys. 92 (1986) 605.
K. Ito, Phys. Lett. B 449 (1999) 48.
M. Bershadsky, H. Ooguri, Phys. Lett. B 229 (1989) 374.
K. Ito, Phys. Lett. B 259 (1991) 73; Int. J. Mod. Phys. A 7 (1992) 4885.
J. Rasmussen, Nucl. Phys. B 510 (1998) 688.
O. Andreev, Nucl. Phys. B 552 (1999) 169; Nucl. Phys. B 561 (1999) 413.
S. Yamaguchi, Y. Ishimoto, K. Sugiyama, J. High Energy Phys. 02 (1999) 026.
M. Ademollo, L. Brink, A. DAdda, R. DAuria, E. Napolitano, S. Sciuto, E. Del Guidice,
P. Di Vecchia, S. Ferrara, F. Gliozzi, R. Musto, R. Pettorino, Phys. Lett. B 62 (1976) 105; Nucl.
Phys. B 114 (1976) 297.
[21] K. Schoutens, Phys. Lett. B 194 (1987) 75; Nucl. Phys. B 295 (1988) 634.
[22] A. Sevrin, W. Troost, A. Van Proyen, Phys. Lett. B 208 (1988) 447.
[23] A. Ali, A. Kumar, Mod. Phys. Lett. A 8 (1993) 1527.
676
[24] A. Ali, Classification of two-dimensional N = 4 superconformal symmetries, hep-th/9906096.

[25] J. Rasmussen, Comments on N = 4 superconformal algebras, hep-th/0003035.
[26] V.G. Kac, Adv. Math. 26 (1977) 8; in: Lect. Notes in Math., Vol. 676, Springer, Berlin, 1978,
p. 597; Adv. Math. 30 (1978) 85.
[27] K. Ito, J.O. Madsen, J.L. Petersen, Nucl. Phys. B 398 (1993) 425; Phys. Lett. B 318 (1993) 315.
[28] V. Knizhnik, Theor. Math. Phys. 66 (1986) 68.
[29] M. Bershadsky, Phys. Lett. B 174 (1986) 285.
[30] K. Schoutens, Nucl. Phys. B 314 (1989) 519.
[31] P. Bowcock, Nucl. Phys. B 381 (1992) 415.
[32] E.S. Fradkin, V.Ya. Linetsky, Phys. Lett. B 282 (1992) 352.
[33] A. Sevrin, K. Thielemans, W. Troost, Nucl. Phys. B 407 (1993) 459.
[34] D.-S. Tang, J. Math. Phys. 25 (1984) 2966.

Smooth gauge strings and

D > 2 lattice YangMills theories
Andrey Yu. Dubin
ITEP, B. Cheremushkinskaya 25, Moscow 117259, Russia
Received 7 March 2000; accepted 28 April 2000
Abstract
Employing the nonabelian duality transformation [A. Dubin, hep-th/9910264], I derive the gauge
string form of certain D > 3 lattice YangMills (YMD ) theories in the strong coupling (SC) phase.
With the judicious choice of the actions, in D > 3 our construction generalizes the GrossTaylor
stringy reformulation of the continuous YM2 on a 2d manifold. Using the EguchiKawai model
e for connected YM-flux
as an example, we develop the algorithm to determine the weights w[M]
e
worldsheets M immersed into the 2d skeleton of a D > 3 base-lattice. Owing to the invariance of
e under a continuous group of area-preserving worldsheet homeomorphisms, the set of weights
w[M]
e can be used to define the theory of the smooth YM-fluxes which unambiguously refers to
{w[M]}
a particular continuous YMD system. I argue that the latter YMD models (with a finite ultraviolet
cut-off) for sufficiently large coupling constant(s) are reproduced, to all orders in 1/N, by the smooth
gauge string thus associated. The asserted YMD /String duality allows to make a concrete prediction
for the bare string tension 0 which implies that (in the large N SC regime) the continuous YMD
systems exhibit confinement for D > 2. The resulting pattern is qualitatively consistent (in the
extreme D = 4 SC limit) with the Wittens proposal motivated by the AdS/CFT correspondence.
PACS: 11.15.Ha; 11.15.Pg; 11.15.Me; 12.38.Aw
Keywords: Lattice; YangMills; Duality; String
1. Introduction
The quest for a string representation of the D = 4 continuous YangMills (YMD ) gauge
theory shaped, to a large extent, many branches of the contemporary mathematical physics.
Currently, there are two (qualitatively overlaping) candidates for the stringy systems
conjecturaly dual to YMD . The first one, due to Polyakov, puts forward certain ansatz [1]
for the world-sheet action which is to ensure the invariance of the Wilson loop averages
dubin@vxitep.itep.ru; tel.: 7-095-129-9674; fax: 7-095-883-9601
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 6 2 - 5
678
A.Yu. Dubin / Nuclear Physics B 582 (2000) 677715
hWC i with respect to the zig-zag backtrackings of the contour C. The complementary
approach has recently sprung to life after daring conjecture of Maldacena [2] (further
elaborated in [3,4]) about the so-called AdF/CFT correspondence concerning the N = 4
SUSY YM4 . Being motivated by the latter correspondence, Witten has made a speculative
proposal [5] advocating that the ordinary nonsupersymmetric YM4 exhibits confinement
at least when considered for large N in the specific strong coupling (SC) regime. Namely,
employing certain SUSY YMD , D > D, properly broken to a nonsupersymmetric YMD
system, one is to fix an effective finite ultraviolet (UV) cut-off while keeping the relevant
YMD coupling constant(s) sufficiently large.
At present, the support of the conjectured duality-mappings is fairly limited to a few
indirect though reasonably compelling arguments. On the other hand, final justification of
that or another stringy representation calls either to reproduce the YMD loop-equations in
stringy terms or to provide with an explicit transformation of the gauge theory into a sort
of string theory. It is our goal to make a step in the second direction approaching (with new
tools) the old challenge: the exact reformulation of the continuous D > 3 YMD theory in
terms of the microscopic colourelectric YM-fluxes. More specifically, we focus on the
pattern of the smooth gauge string inherent in the presumable large N SC expansion (as
opposed to the standard weak coupling (WC) series) valid in the SC regime similar to
that of the Wittens proposal [5]. Complementary, the considered generic strongly coupled
continuous YMD models in D = 4 might be viewed as the prototypes of the effective
low-energy YM4 theory. The latter is supposed to result via the Wilsonian renormgroup
(RG) flow of the effective actions (starting from the standard YM4 in the WC phase) up
to the confinement scale. In this perspective, the built in UV cut-off in D = 4 is to be
qualitatively identified with a physical scale YM4 that is of order of the lowest glueball
mass.
The short-cut way to a continuous model of smooth YM-fluxes is suggested by the twodimensional analysis. Here, the pattern of the flux-theory [15,16] proposed by Gross and
Taylor is encoded not only in the continuous YM2 systems on a 2d manifold. It is also
inherent in the associated RG invariant 2d lattice gauge models [23,24] introduced via the
plaquette-factor
X F (R)
e
R (U ), eF (R) = dim R e ({bk },N,{Cp (R)}) ,
(1.1)
Z bk U =
R
where dim R, R (U ), and Cp (R), p = 1, . . . , N, stand respectively for the dimension,

character and pth order Casimir operator associated to a given SU(N) irreducible
representation (irrep) R (while {bk N 0 } denotes a set of the dimensionless coupling
constants). The key observation is that, in the D > 3 lattice YMD systems (1.1), the
pattern of the appropriately constructed flux-theory as well refers (owing to a subtle D > 3
descendant of the 2d RG invariance) to the properly associated continuous YMD models.
To support this assertion, I first reformulate the strongly coupled D > 3 lattice YMD
models (1.1) in terms of the gauge string which does appropriately extend the D = 2
GrossTaylor stringy pattern [15,16] into higher dimensions. This lattice theory of the
YM-flux is endowed with certain continuous (rather than discrete as one might expect
679
from the lattice formulation) group of the area-preserving homeomorphisms. In turn, the
latter symmetry ensures that the considered D > 2 pattern of the lattice flux-theory can be
employed to unambiguously define the associated D > 2 smooth gauge string invariant
under the area-preserving diffeomorphisms. The remarkable thing is that, in the latter
continuous D > 3 flux-theory, one can identify (see Section 8) such SC conglomerates of
the (piecewise) smooth flux-worldsheets which are in one-to-one correspondence with the
judiciously associated varieties of the WC Feynman diagrams on the side of the properly
specified continuous YMD model. Therefore, the proposed smooth gauge string provides
with the concrete realization of the old expectation [10] (for a recent discussion see [5]) that
certain nonperturbative effects close up the windows of the Feynman diagrams trading
the latter for the string worldsheets.
More specifically, building on the nonabelian duality transformation [29] recently
proposed by the author, we show that in the D > 2 lattice systems (1.1) the free energy
and the Wilson (multi)loop observables can be rewritten in terms of the following statistics
e )] of a given connected worldsheet M(T
e ) (with the
of strings. The lattice weight w[M(T
support on a subspace, represented by certain 2d cell-complex T, of the 2d skeleton of the
D-dimensional base-lattice)

22hb

e ) b k ,
e )] = exp A
2 0 bk
J M(T
N
(1.2)
w[M(T
is composed of the three different blocks which altogether conspire so that the contribution
of the strings with any backtrackings (i.e., foldings) is zero. In Eq. (1.2), the first factor
e The
is the exponent of the NambuGoto term proportional to the total area A of M.
2
D-independent bare string tension 0 = 0 (to be defined by Eq. (1.6) below) is
measured in the units of the UV cut-off squared (which, in D > 3, regularizes the
transverse string fluctuations). Next, there appears the standard t Hooft factor, where h
and b are respectively the genus and the number of the boundary contours Ck , k = 1, . . . , b,
e
of M.
e )|{bk }] (being equal to
Finally, given a particular model (1.1), the third term J [M(T
e
unity for a nonselfintersecting surface M) is sensitive only to the topology (but not to the
e In particular, the dependence of J [M(T
e )|{bk }] on the
geometry) of selfintersections of M.
coupling constants {bk } is collected from the elementary weights assigned to the admissible
e T . As
movable singularities (to be specified later on) of the associated map : M
a result, the last term (similarly to the remaining ones) is invariant under the required
continuous group of the area-preserving homeomorphisms so that the pattern (1.2) directly
e
applicable to a generic smooth worldsheet M.
Next, consider the D > 3 continuous flux-theory defined as the statistics of the smooth
e which are postulated to be endowed with the weights (1.2) corresponding
worldsheets M
e into the Euclidean space RD . In compliance with the above
to the smooth mappings of M
asserted SC/WC correspondence, a given specification (1.1) the smooth gauge string refers
to the following unique continuous YMD model. The local Lagrangian of the latter is to
be reconstructed as the D-dimensional pull-back
a
a
; , = 1, 2} {F
; , = 1, . . . , D}
L2 (F ) LD (F ): {F
(1.3)
680
a F
of the D = 2 Lagrangian (composed of the O(D)-invariant combinations of the F
tensor)
D = 2: LD (F ) = D4
X X [tr(F k )]pk
,
gr ({bk })
(1.4)
n>2 rYn
associated to such continuous YM2 theory [23,25] that its partition function on a 2d disc
(of unit area with the free boundary conditions) is equal to the plaquette-factor (1.1).
(In D > 3, within each trace of (1.4), the involved (F )ij -factors are prescribed to be
totally symmetrized to exclude the [F , F ]-commutator dependent terms identically
vanishing in D = 2.) To separate out the kinematical D-dependent rescaling of the
coupling constants, we have introduced the parameter . The latter is to be identified with
the effective UV cut-off, for the considered strongly coupled YMD systems, predetermined
by the necessary regularization of the D > 3 YM-flux transverse fluctuations. Remark also
that the N - and -dependent coupling constants gr ({bk }) gr ({bk }, N, ) are canonically
P
labelled by the S(n) irreps r Yn (parametrized by the partitions of n: nk=1 kpk = n).
In sum, our proposal is that the particular theory of the smooth gauge string,
thus induced through (1.1), reproduces (to all orders in 1/N ) the corresponding
local continuous
D > 3 YMD model (1.4) for sufficiently large coupling constants
Pn
2
p
k
k=1
O(N 0 )}. (The latter constants are measured in the units of 1/N and
{gr N
akin to Eq. (9.1).) Note that in D = 2, after proper identification of the coupling constants,
this correspondence is justified by the matching with the 2d pattern [15,16] of Gross and
Taylor. One may also expect that the asserted YMD /string correspondence is unique to
the extent that the stringy degrees of freedom are directly identified with the worldsheets
of the microscopic, conserved YM-flux immersed (by mappings with certain admissible
singularities to be specified after Eq. (1.11) below) into a D-dimensional base-space.
The above constraint on {gr } (selecting the SC regime) accounts for the fact that in
D > 3 the proposed YMD /String duality is limited by the stability of the YM-flux. In
particular, the physical string tension ph (entering, e.g., the asymptotics of the Wilson
loop averages) must be positive. As we will see, despite the extra J [ ]-factor in Eq. (1.2),
the bare and the physical string tensions (defined within the 1/N expansion) are
conventionally related so that the latter condition in the large N limit reads

(1.5)
gr bk , . . . , : ph = 0 b k D 2 > 0,
i.e., the bare part 0 = 2 0 should be larger than the D > 3 entropy contribution ent =
D 2 due to the transverse string fluctuations. Among our results, the central one is the
explicit {bk }-dependence of 0 = 2 0 ({bk }) entering the worldsheet weights (1.2). It is
uniquely reconstructed

(1.6)
bk , N, {Cp (R)} = n 0 bk 1 + O N 1 , 0 = 2 0 O N 0 ,
from the formal 1/N asymptotics (with R Yn(N) , n O(N 0 )) of the admissible function
( ) defining the associated lattice model (1.1). For example, the smooth gauge string
induced from the HeatKernal action [21]
X X

Z g 2 U =
dim R R (U ) exp g 2 C2 (R)/2 ,
681
(1.7)
n>0 RY (N)
n
being presumably dual to the standard YMD theory with g 2 = 4D g 2 1/N , refers to
the following bare string tension (with g 2 1/N )
2
)/4g 2
LD (F ) = tr(F
0 = g 2 ND2 /2.
(1.8)
Summarizing, the predicted pattern (1.6) of 0 implies that the continuous D > 3
YMD systems are confining at least in the large N SC regime belonging to the domain
(1.5). Although in this regime the standard WC series are expected to fail, the latter
SC phenomenon suggests a mechanism of confinement for YM4 in the WC phase (see
conclusions). Remark also that the physical string tension in Eq. (1.5) generically is not
adjusted to be infinitely less than the squared UV cut-off (which in D = 4 matches
with the above identification YM4 ). Actually, in the considered YMD models the
introduced smooth flux-worldsheets might become unstable degrees of freedom even prior
to the saturation of (1.5) owing to the presumable large N phase transition(s) (generalizing
the D = 2 situation [14]).
Finally, the distinguished regime of the smooth gauge string is the extreme large N SC
limit where 0 ({bk }) 2 . As in Eq. (1.5) the entropy-constant D is {bk }-independent,
in this limit the leading order of the Wilson loop averages hWC i is given by the minimal
area contribution that allows for a number of nontrivial predictions. First of all, the physical
string tension in this regime merges with the bare one, ph 0 , determined by Eq. (1.6).
(We will show in Section 9 that the Wittens SC asymptotics [5] of ph in D = 4 is
semiquantitatively consistent with the latter prediction). Also, provided the minimal area
Smin (C) has a support on a 2d manifold (rather than on a 2d cell-complex), in this limit the
pattern of the D > 3 averages hWC i is reduced to the one in the corresponding continuous
D = 2 YM2 theory (1.4).
1.1. The D > 2 pattern of the gauge string weights
1.1.1. The D = 2 case
eM
Recall that, in the 2d framework initiated by Gross and Taylor, the partition function X
of a given continuous YM2 theory on a 2d manifold M of the area A (without boundaries)
is rewritten as the sum
Z
f
e M,
eM = df (1)Pf N exp[0 nf A], f : M
(1.9)
X
|Cf |
e
M
e of M specified by the
over the topologically distinct branched covering spaces M
mappings f . The latter maps locally satisfy the definition [27,30] of the immersion
e M everywhere on M except for a set of isolated points where the singularities,
M
corresponding to a branch-point or/and to a collapsed (to a point) subsurface connecting
e as the associated Riemann
a few sheets, are allowed. To visualize the pattern of M
surface (without foldings but with the singularities to be viewed as certain collapsed
682
e of the total Euler

2d subsurfaces) identified with a particular string worldsheet M
characteristic f , I propose the cutting-gluing rules which readily generalize to the D > 3
case. As for |Cf |, Pf , and nf , in Eq. (1.9) they denote respectively the symmetry factor
(i.e., the number of distinct automorphisms of f : f = f ), the parity, and the degree
(i.e., the number of the covering sheets) of the map f . The sum (1.9) includes all admissible
eM ] is supposed to be
disconnected contributions in such a way that the free energy ln[X
e
deduced [15,16] from (1.9) constraining the worldsheets M to be connected.
To make contact with the T = M, A = nf A case of (1.2), observe first that the
summation (1.9) over the maps f effectively enumerates the admissible selfintersections of
e In particular, the sum over the singularities of f implies the one over the nonmovable
M.
and (the positions of) the movable branch points parametrized by the corresponding
Pn
cyclic decomposition {p}:
k=1 kpk = nf of nf . These two types of the points are
respectively assigned with the nontrivial weights w {p} and w{p} ({bk }) (included into the
measure df of (0.5)) which are {bk }-dependent only in the movable case. (The w{p} -,
e )|{bk }] of (1.2).)
w {p} -factors, together with (1)Pf /|Cf |, are collected into J [M(T
Therefore, the summation df implies in particular the multiple integrals
{p} Z
Y [w{p} ({bk })]m{p} mY
{p}
[m{p} ]!
(k)
d 2 X{p}
(1.10)
k=1 M
(k)
on M of (a given number m{p} of) the movable branch points
over all positions X{p}
R
(k)
that introduces additional dependence on the area A = M d 2 X{p} of M. (Nonmovable
singularities of the f -map can be placed anywhere on M but do not carry any area- or {bk }dependent factors.) Finally, the bare string tension 0 = 2 0 ({bk }) (defined via Eq. (1.6))
enters the exponent of (1.9) as the no-fold variant of the NambuGoto term. Our attention is
mostly confined to the option (1.8) corresponding to the D > 2 SU(N) HeatKernal lattice
gauge theory (1.7). In this case, 0 = 2 g 2 N/2 and there are only the simple transposition
branch points (i.e., the ones connecting a pair of the sheets) weighted by w2 = g 2 N .
Finally, the D = 2 representation (1.9) of the continuous YM2 theories (1.4) is valid
also in the corresponding 2d lattice gauge systems (1.1) on a discretized surface M.
The origin of this two-fold interpretation of (1.9) resides in the RG invariance of the
latter lattice models. Complementary, owing to the symmetry of the continuous YM2
models (1.4) under the area-preserving diffeomorphisms, the RG invariance of (1.1) results
e under the continuous group of the areain the invariance of the lattice weights w[M]
preserving worldsheet homeomorphisms. These homeomorphisms continuously (rather
than discretely) translate the positions of the singularities of the map f everywhere on
M including interiors of the plaquettes.
1.1.2. Extension to the D > 3 case
A given D > 3 lattice YMD system (1.1) (having some lattice LD as the base-space
B) can be equally viewed as the YM model (1.1) defined on the 2d skeleton TD of
LD represented by the associated 2d cell-complex. As we will discuss in Section 7, this
683
reformulation allows to reveal certain D > 3 descendant of the 2d RG invariance of (1.1)

that in turn foreshadows the existence of the D > 3 extension of the pattern (1.9).
eLD of the YMD system (1.1) defined on B = LD .
Consider the partition function X
Combining the nonabelian duality transformation [29] with the methods of algebraic
topology [30], we derive that the pattern (1.9) remains to be valid in D > 3 with the
only modification. It concerns the structure of the relevant mappings (to be summed over)
e of the YM-flux. The D = 2 maps
specifying the admissible topology of the worldsheets M
f are superseded by the mappings
e T TD ,
: M
(1.11)
e onto a given subspace T (represented by a 2d cell-complex) of the 2d

of a 2d surface M
skeleton TD of the D-dimensional base-lattice LD . Akin to the D = 2 case, the maps
e T anywhere except for
locally comply with the definition [27,30] of the immersion M
a countable set of points on T where certain singularities are allowed. The relevant (for the
smooth implementation of the gauge string) ones are either of the same type as in the 2d
f -mappings (1.9), i.e., the branch points or/and the collapsed subsurfaces, or of the type
corresponding to the homotopy retraction (see, e.g., Appendix C) of the latter irregularities.
In particular, the included into d-measure {bk }-dependent weights w{p} ({bk }) (of the
movable branch points) in D > 3 are equal to their counterparts in the 2d case (1.10)
associated to the same continuous YM2 theory (1.4). (Akin to Eq. (1.9), one can argue that
eLD ] is provided by the restriction of the sum for partition function
the free energy ln[X
e
e
XLD to the one over the connected worldsheets M.)
The D > 3 nature of the mappings (1.11) is reflected by the fact that, owing to possible
e ), the tangent-space T may be represented
higher-dimensional selfintersections of M(T
by the 2d cell-complex T = k Ek rather than by a 2d surface. (Recall that any 2d cellcomplex, after cutting out along the links shared by more than two plaquettes, becomes
[21,30] a disjoint union of 2d surfaces Ek of the areas Ak with certain boundaries.
Conversely, the set {Ek } can be combined back into T = k Ek according to the incidence
numbers [21] corresponding to the entries of the associated incidence-matrix.) Therefore,
e ) (via our cuttingglueing algorithm developed for the canonical
the construction of M(T
branched coverings (1.9)) entails the appropriate generalization of the notion of the
P (k)
e ) can be viewed
Riemann surface (with the total area A = k n Ak ). Alternatively, M(T
as the specific generalization of the branched covering space (wrapped around T ) suitable
for the analysis of the gauge string.
e with
Next, as it is shown in Section 8, the total contribution of the worldsheets M
arbitrary backtrackings (that, generalizing the D = 2 case, bound any zero 3-volume)
e in D > 2 arises when
e )|{bk }] = 1 pattern (1.2) of w[M]
vanishes. The simplest J [M(T
e
e )
the genus h connected string worldsheet M(T ) is represented by an embedding T = M(T
e
e
(i.e., M(T ) does not selfintersect). Nevertheless, the set of all possible J [M(T )|{bk }] = 1
embedding-weights (1.2) does not discriminate between those distinct models (1.1)(1.4)
which provide with one and the same bare string tension (1.6). To distinguish between the
different models, the specification of the remaining immersion-weights (assigned to the
selfintersecting string worldsheets) is indispensable. In this way, one reconstructs the data
684
encoded in a given continuous YM2 theory (1.4) that originally served as the bridge (1.3)
between the gauge string and the corresponding continuous YMD model.
e )] are invariant under
Owing to the asserted pattern, the D > 3 lattice weights w[M(T
certain continuous group of the area-preserving homeomorphisms extending the D = 2
e )]} can be unambiguously used to introduce the statistics
ones. As a result, the set {w[M(T
e ), with the latter homeomorphisms
of the (piecewise) smooth YM flux-worldsheets M(T
being traded for the corresponding diffeomorphisms. As for the sum over the worldsheets,
it is specified by the one running over the -mappings (1.11). This time, one is to consider
the piecewise smooth immersions (with the admissible singularities which, in D > 4,
can be restricted to the ones listed after Eq. (1.11)) of the 2d manifolds M into the De with the support on
dimensional spacetime B = RD that results in the worldsheet M
T RD . In turn, it constitutes the proper class of the 2d cell-complexes T which can play
the role of the tangent-spaces for the considered (piecewise) smooth maps .
As we will discuss in Section 8, the D > 3 dynamics of the smooth gauge string in
certain sense is dramatically simpler compared to its lattice counterpart. First of all, it is
convenient to make certain redefinition of the bare string tension (that amounts to the substitution 0 = /2 (1 1/N 2 )/2 in the case of (1.7)) to get rid of a redundant subset
of the movable singularities present in the SU(N) mappings (1.11) even for a 1-sheet cove )
ering of T . Given this modification, consider the subset of the smooth worldsheets M(T
(without boundaries) which are constrained to be strictly nonselfintersecting in D > 5 and
allowed to selfintersect at an arbitrary union of isolated points in D = 4. (Actually the
latter condition can weakened e.g. to include non(self)intersecting boundary contour(s)).
e )|{bk }] = 1 weight-pattern (1.2). The point is
They are assigned with the simplest J [M(T
e parametrized by
that in D > 4 the latter subset is dense in the set of all worldsheets M
the (piecewise) smooth mappings (1.11) into RD , provided the latter redefinition of 0 . In
particular, it justifies (at least in D > 4) the validity of the simple relation (1.5). As for
the more complicated pattern (1.2) of the weights, in D > 4 it is observable for example
within such Wilson loop averages hWC i where the corresponding minimal surface Smin (C)
selfintersects (or when the boundary contour C has some zig-zag backtrackings).
2. Outline of the further content

To make the analysis of the D > 2 lattice gauge string more concise, we employ the
twisted EguchiKawai (TEK) representation [11,12] of the large N infinite-lattice SU(N)
eLD of a lattice
gauge systems. Recall that in the limit N the partition function (PF) X
D
2
YMD theory like (1.1) in a D = 2p volume L = N can be reproduced (at least within
both the SC and the WC series)
!LD
Z Y
D(D1)/2
D
Y

2
+
+
eLD = lim
dU
Z g t U U U U
,
(2.1)
lim X
N
=1
=1
eD of the associated D-matrix SU(N) TEK model with the reduced

through the PF X
spacetime dependence (where t = exp [i2/N 2/D ] ZN ). Therefore, in the N SC
685
phase, the free energy of (2.1) yields the generating functional for the YMD string-weights
assigned to the worldsheets corresponding to the 2d spheres immersed into LD . On the
side of the TEK model, the surfaces are wrapped around the EK base-lattice TEK which
has the topology of the 2d-skeleton of the D-dimensional cube with periodic boundary
conditions. Thus, TEK is homeomorphic to the 2d cell-complex visualized as the union
D(D1)/2
TEK = =1
E E = l ,
E ,
(2.2)
of the D(D 1)/2 mutually intertwined 2-tora E sharing D uncontractible cycles (i.e.,
compactified links of the D-cube) l in common.
To proceed further, let us first label the SU(N) irreps summed up in each -species of
the Z-factor (1.1) entering (2.1) by
,
R Yn(N)
n+ =
D(D1)/2
X
n .
(2.3)
=1
It is convenient to rewrite the SU(N) TEK PF (2.1) in the following form

X
X
X
eD =
t n+ eS({R }) B({R }) =
t n+ B({n }),
X
{n } {R
(2.4)
{n }
(N)
Yn }
introducing the elementary master-integrals B({R }) (which are then composed into
B({n }))
Z Y
Y
Y
dU
R (U );
eS({R }) =
eF (R ) ,
(2.5)
B({R }) =
{}
{}
{}
where U U U U+ U+ , F (R) is defined by Eq. (1.1), and we have used the identity
(N)
. In the SC phase, we are concerned below, the
R (tV ) = t n(R) R (V ), t ZN , R Yn(R)
twist-factor in Eq. (2.4) is irrelevant [11,12] in the limit N so that the TEK model is
reduced to the original EguchiKawai one. In this regime, the t = 1 correspondence (2.1)
is supposed to be valid in any (not necessarily even) D > 2.
Next, making use of the nonabelian duality transformation [29], in Section 3 we rewrite
the TEK master-integral B({n }) as the weighted sum of the Trn+ -characters (i.e., traces)
X

eS({R }) Trn+ D An+ ({R }) ,
(2.6)
B({n }) =
(N)
{R Yn }
P
of certain master-elements An+ ({R }) = S(n+ ) a( |{R }) . Being defined by
Eqs. (3.34), (3.35), they take values in the tensor representation of the S(n+ ) algebra
(with n+ being given by Eq. (2.3)). The latter is deduced by linearity from the canonical
representation [19,20] for S(n)-group elements
{i n }
D( ){j n } = j1 (1) j2 (2) jn(n) ,
: k (k), k = 1, . . . , n,
(2.7)
where ji denotes the N -dimensional Kronecker delta function.

To relate (2.6) with the stringy pattern like (1.9), in Section 4 we represent this equation
in the form suitable for the algebraic definition of the topological data. First of all,
686
one observes (see Appendix B) that the Trn+ -trace in Eq. (2.6) is simply related to the
associated character of the regular S(n+ )-representation. As a result, B({n }) can be
rewritten as the delta-function on the S(n+ )-algebra (that selects the contribution of the
weight of the S(n+ ) unity-permutation 1 [n+ ] )

(2.8)
B({n }) = Trn+ D A n+ = n+ (1)
n An+
+
that already played the important role in the D = 2 analysis [15,16]. Next, employing the
(1)
SchurWeyl duality, both the operator n+ (defined by the U (N) variant of Eq. (3.9), see
Section 3) and
X
eS({R }) An+ ({R })
(2.9)
A n+ =
(N)
{R Yn }
are reformulated entirely in terms of the symmetric group elements (i.e., no SU(N)
irreps R are left). The resulting expression (4.8) is suitable to specify the mappings
(1.11). The D > 3 nature of (1.11) is reflected by the outer-product structure of A n+
which is represented as certain combination of the S(n )-blocks embedded to act in the
common enveloping space of the S(n+ ) = S(n ) algebra (with {}, {}). The
technique, dealing with such compositions, is naturally inherited from the nonabelian
duality transformation [29].
The remaining material is organized as following. In Section 5, we rederive by our
methods the BaezTaylor reformulation [31] of the GrossTaylor 2d pattern (1.9) for
the HeatKernal model (1.7). We also briefly sketch how our method generalizes for
any admissible model (1.1). Being more compact than the GrossTaylor one, the D = 2
representation of the BaezTaylor type has the structure where the full stringy pattern is
not entirely manifest. To circumvent this problem we formulate a simple prescription how
to transform the latter into the former.
The D > 3 generalization of the 2d BaezTaylor representation [31] is derived in
Section 6 for the EguchiKawai models (2.1) which fully justifies the announced pattern
(1.11) of the D > 3 mappings. (In particular, we make certain conjecture concerning
the concise topological reinterpretation, generalizing the D = 2 one [22], of the D > 3
sums like (1.9) in the formal topological limit when ( ) 0.) As well as in the D = 2
case, to relate the obtained D > 3 representation with the manifest stringy pattern (1.2),
a natural extension of the D = 2 prescription (formulated in Section 5) is suggested. In
Section 7, we discuss the structure of the asserted continuous group of the area-preserving
e are endowed with) and
homeomorphisms (the generic lattice gauge string weights w[M]
make a brief comparison with the earlier large N variants [79] of the Wilsons SC
expansion [6] devoid of the latter invariance. In Section 8, we discuss the major qualitative
features of the gauge string accentuating the similarities and differences between the
continuous flux-theory and the conventional paradigm of the D > 3 fundamental
strings. In the last section, we put forward a speculative proposal for the mechanism of
confinement in the standard weakly-coupled continuous gauge theory (1.8) at large N .
Also a preliminary contact with the two existing stringy proposals [1,5] is made. Finally,
the appendices contain technical pieces of some derivations used in the main text.
687
eD
3. The dual representation of X
To derive (2.6), we apply the nonabelian duality transformation to the partition function
of the SU(N) TEK model (1.1)(2.1) following the general algorithm formulated in [29]
for a generic SU(N) D-matrix system. To begin with, the SU(N) character is to be
represented in the form (see, e.g., [22,29]) reminiscent of the one of Eq. (2.6). Let CR
denote the canonical Young idempotent [19,20] proportional, PR = dR CR , to the Young
projector
dR X
R ( ), R Yn ,
(3.1)
PR =
n!
S(n)
where R ( ), dR are the character and the dimension associated to the S(n) irrep R (while
[PR , ] = 0, S(n)). Then, R (U ) assumes (akin to (2.6)) the form of the trace:
R (U ) = Trn [D(CR )U n ], R Yn(N) , where

Trn D( )U n =
N
X
i1 i2 ...in =1
Ui1 (1) Ui2 (2) Uin (n) ,
(3.2)
and the tensor D( ) is defined in Eq. (2.7).

Altogether, it implies that the master-integral (2.5) can be rewritten in the synthetic form
Z
D

Y
dU ,
(3.3)
B({R }) = Tr4n+ D 4n+ ({R }) D U U+
=1
where the S(4n+ )-algebra valued tensor D(4n+ ({R })) (to be defined by Eq. (3.19)
below) multiplies the complementary tensor given by the ordered direct product of the
4n+ elementary N N matrices (U )ij , (U+ )kl :
D
O
n

(U )n U+ ,
D U U+
2n+ =
=1
D
X
n .
(3.4)
=1
Given 4n+ ({R }), we first derive the intermediate S(4n+ ) representation

B({R }) = Tr4n+ D J4n+ ({R })
(3.5)
which in Section 3.3 will be transformed into the final S(n+ ) form of Eq. (2.6). To
determine the operator J4n+ S(4n+ ), in (3.3) one is to substitute the dual form [29]
of the SU(N) measure, i.e., to represent the result of the D different U -integrations as an
S(4n+ )-tensor akin to D(4n+ ({R })).
Before we explain the latter procedure, let us make the pattern of the element
P
4n+ ({R }) = S(4n+ ) ( |{R }) more transparent relating the explicit form of
the S(4n+ )-group tensor D( ) with the one of (3.4). To this aim, we introduce first a
particular S(4n+ ) permutation {n } via the mapping m (m), m = 1, . . . , 4n+ . Then,
generalizing the tensor representation (2.7), the tensor D({n } ) stands for
i
i(n1 ) k(n1 +1)

ln +1
1
1
jn
j(1)
1
k(2n1 )
l2n
k(4n+ 2nD )
l4n
+ 2nD
k(4n+ nD )
l4n
+ nD
k(4n+ )
l4n
(3.6)
688
where, to each individual U -, U+ -factor in the product (3.4), we associate one copy of the
Kronecker delta-function. In this way, both D({n } ) and the block (3.4) are defined to act
on one and the same S(4n+ ) space (to be manifestly constructed, see Eqs. (3.13), (3.14)
below). Complementary, the pattern of the trace (3.5) of J4n+ ({R }) (defined through the
structures like (3.6)) naturally generalizes the above convention (3.2) to the case of (3.4).
3.1. The dual form of the SU(N) measure
On a given base-lattice, a generic multilink integral (like in (2.1)) evidently can be
p ...q
expressed [21] in terms of the 1-link integrals M G (n, m)j11...lmm :
Z
Z
p
p
q
q
{p n }
{q m }
(3.7)
(U )j11 (U )jnn (U + )l11 (U + )lmm dU D(U ){j n } D(U + ){l m } dU
composed of the N N matrices (U )jkk , (U + )lkk in the (anti)fundamental representation
of the considered Lie group G. As we will show in a moment, the SU(N) TEK partition
function (PF) (2.1) is invariant under the substitution of the SU(N) link-variables by
the U (N) = [SU(N) U (1)]/ZN ones. In the U (N) case, where the dual [29] form of
M U (N) (n, m) in terms of the S(n)-valued tensors (2.7) reads
X
{q n }
p ...q
{p n }
D( 1 (1)
){j n } D( ){l n }
(3.8)
M U (N) (n, m)j11...lmm = [n, m]
n
p
S(n)
that renders manifest the previously known interrelation [16,21] between (3.7) and the
(1)
symmetric groups structures. The operator n S(n) belongs to the (U (N) option of
the) family [29]
X
dR (n! dim R/dR )m CR , m Z,
(3.9)
(m)
n =
(N)
RYn
where CR = PR /dR is defined by Eq. (3.1). In Eq. (3.9), dR and dim R are respectively
the dimension of the S(n)-irrep and chiral U (N)-irrep both described by the same Young
(N)
tableau Yn containing not more than N rows. (In Eqs. (4.4), (4.8) below, we will consider
(m)
the SU(N) variant of n where the sum is traded for the one over the SU(N) irreps.)
As for the possibility to substitute the SU(N) TEK link-variables by the U (N) =
[SU(N) U (1)]/ZN ones, the pattern of Eq. (2.1) ensures that the TEK action is invariant
under the D copies of the extended transformations [U (1)]D : U t U , where t
U (1) rather than taking value in the center-subgroup ZN of SU(N). As a result, the
eD . As
nondiagonal moments M SU(N) (n, m), n 6= m, do not contribute into the TEK PF X
for the remaining diagonal integrals M SU(N) (n, n), the latter extended invariance justifies
[29] the required substitution: M SU(N) (n, n) = M U (N) (n, n), n Z>0 . Moreover, in the
context of the large N SC expansion, in Eq. (3.9) one can insert the SU(N) variant
. (The difference can be traced back to the contributions of the SU(N) stringof (1)
n
junctions which are supposed to be irrelevant to all orders in 1/N .)
Actually, the derivation of (3.5) calls for the alternative S(2n) reformulation [29] of the
S(n) S(n) formula (3.8) that will require the explicit form of an S(2n)-basis. For this
purpose, we first recall that each individual matrix U+ or U can be viewed [19,20] as the
689
operator acting on the associated elementary N -dimensional subspace |i ()i according

P
j
+
to the pattern: U |i i = N
j =1 Ui |j i and similarly for U |i+ i. Complementary, for a
given S(n), the operator (2.7) acts as the corresponding permutation of the elementary
subspaces |ik i
{j n }
(3.10)
|i1 i|i2 i |in i = |i 1 (1) i|i 1 (2) i |i 1 (n) i D( ){i n } |j in ,
P
where in the r.h.s. the summation {jk } is implied. As a given S(4n+ )-basis is constructed
as the outer product of the 4n+ building blocks |i ()i (ordered according to a particular
prescription), the elementary subspaces |ik i of Eq. (3.10) are represented by |i ()i.
Returning to the S(2n)-reformulation of Eq. (3.8), in the basis |I2n i = |In(+) i |In() i
()
(with |In i = |i >n ) it reads
Z
{j 2n }
jn+1 ...j2n
j ...j
(1)
1 [n] {i 2n } ,
(3.11)
dU D(U )i11...inn D(U + )in+1
...i2n = D 2n (2n) n
where 1 [n] denotes the unity-permutation of the S(n) group, n S(n) is defined by
P
Eq. (3.9), while (2n) = S(n) ( 1 ), i.e.,
X
j1 ...jn
jn+1 ...j2n
{j 2n }
D 1 i ...in D( )in+1
(3.12)
D( (2n)){i 2n } =
...i2n S(2n).
(1)
S(n)
To restore the -labels, n n , observe first that the ordering of the {U }-factors in
Eq. (3.4) is associated to the following basis
D

O
I4n(+) =
|I2n() i,
(+) ()
|I2n() i = In() In() ,
(3.13)
=1
O ()
() D1
I
I
n() =
n() ,
()
n()
I
|i ()in() ,
n() = |i ()i
(3.14)
6=
P
()
n (used in Eq. (3.11)) matches with |I () i.
where 2n+ = D
=1 n , and |In i = |i i
n()
Therefore, for a given link , the left and the right S(n )-subblocks of (2n ) in Eq. (3.12)
(+)
()
i and on |In()
i. The same convention is used for the S(n )act respectively on |In()
(1)
subblocks in the direct product (n 1 [n ] ) entering Eq. (3.11).
The remaining S(2n )-operator 2n , being considered in the alternatively ordered basis
|I2n() i for each |I2n() i-subsector,

n

(3.15)
|I2n() i I2n() = |i+ ()i |i ()i
n() ), takes the simple form of the outer product of
(with |i ()in() = D1
6= |i ()i
the 2-cycle permutations c2 C(2)
2n = (c2 )n S(2n ),
c2 : {12} {21},
(3.16)
where each c2 S(2) acts on the elementary sector |i+ ()i |i ()i. It completes the
embedding of the S(2n ) operators (3.11), representing the individual 1-link integrals, to
act in the common enveloping S(4n+ )-space. (It is noteworthy [29] that the three S(2n )
subblocks of the inner-product in the r.h.s. of Eq. (3.11) commute with each other.)
690
3.2. The dual form of the TEK action

Let us now turn to the derivation of the master-element 4n+ ({R }) entering the
synthetic representation (3.3) of the master-integral (2.5). For this purpose, we first specify
an alternative, more suitable S(4n+ ) basis
|I4n(+) i =
D(D1)/2
O
|I4n() i,
(3.17)
|I4n() i = (|i+ ()i|i+ ()i|i ()i|i ()i)n ,
(3.18)
=1
where in Eq. (3.18) the product of the four elementary blocks |i ()i is associated to the
elementary -holonomy U entering Eq. (2.5). As it is derived in Appendix A, in this
basis one obtains
4n+ ({R }) =
D(D1)/2
O
P4n (R ) 4n ;
n+ =
n ,
(3.19)
{}
=1
where each of the operators P4n (R ), 4n S(4n ) is supposed to act on the

corresponding |I4n() i subspace of |I4n(+) i. Given (3.18), 4n assumes the simple form
of the outer product
4n = (c4 )n S(4n ),
c4 : {1234} {4123},
(3.20)
with each individual 4-cycle permutation c4 C(4) acting on the elementary plaquette
subspace
|i+ ()i|i+ ()i|i ()i|i ()i
(3.21)
ordered in accordance with the original pattern (2.5) of the plaquette-labels of U .

As for P4n (R ), making use of the alternative basis |I4n() i |I4n() i of the
S(4n )-subspace in (3.18):
|I4n() i = |i+ ()in |i+ ()in |i ()in |i ()in ,
(3.22)
we employ (see Appendix A) the following representation

q

q

P4n (R ) = 1 [n ] 1 [n ] CR dR CR dR ,
(3.23)
where 1 [n ] denotes the S(n )-unity. More explicitly, the four (ordered) S(n )-factors
in Eq. (3.23) are postulated to act on the corresponding four (ordered) n -dimensional
subspaces (3.22) of |I4n i. Altogether, we have formulated the required (for the derivation
of Eq. (3.5)) embedding of the S(4n ) operators 4n , P4n (R ) into the enveloping
S(4n+ )-space.
3.3. B({R }) as the Trn+ -character
Combining together Eqs. (3.19) and (3.11) in compliance with the pattern of Eq. (3.3),
we arrive at the dual representation of the master-integral (2.5) as the S(4n+ ) character
(3.5) with the master-element

D(D1)/2
J4n+ ({R }) = =1
691

4n D
=1 12n ({R }) ,
(3.24)
(3.25)
12n ({R }) = 2n (2n ) K2n ({R }) S(2n ).
N
For later convenience, we have introduced the S(n ) S(n ) combination that in the
|I2n() i basis (3.13) reads
q

O
D1
C
d
(3.26)
K2n ({R }) = (1)
R
R
,
n
6=
(1)
where n
(+)
()
acts onto |In() i, while each of the CR factors acts on the |In() i-subspace
()
i.
of |In()
Next, let us complete the duality transformation trading the intermediate S(4n+ )
representation (3.5) for its final S(n+ ) pattern (2.6). To this aim, it is convenient to
start with the alternative following alternative form of the master-element (3.24). As it
is demonstrated in Appendix A, the dual representation (3.5) does not alter when in the
element (3.24), (3.25) one makes the substitution
D
2 D
D
=1 2n =1 (2n() ) = =1 1[2n ] = 1[4n+ ] ,
(3.27)
so that (inside the Tr4n+ -character) in Eq. (3.25) all the operators 2n() can be omitted.
3.3.1. The D = 2 case
Next, it is appropriate to proceed with the simplest D = 2 case of (3.5) corresponding
to the well studied continuous SU(N) gauge theory on a 2-torus. It will provide not only
with a cross-check of our D > 2 formalism but also with the motivation for the announced
reduction, S(4n+ ) S(n+ ), of the enveloping space. In D = 2, the this reduction is
encoded in the basic property (following from Eqs. (A.1), (A.3) in Appendix A) of the 4n
operator (3.20)
n

n
e
e = U U+ U U+ ,
(3.28)
= Tr4n D(4n )U
, U
Trn U
while U = U U U+ U+ . Making the substitution U , U+ , one
obtains
!
"
!#
"
!#
2
D
D
O
Y
Y

(+)
()
(+)
()
4n

= Trn D
, (3.29)

Tr4n D
(+)
=1
=1
()
()
=1
where in the l.h.s. the operators S(n) (combined into the outer product) act on the
associated |In() i subspaces of the S(4n) basis (3.13). As for the ordering inside the two
inner -products in the r.h.s. of Eq. (3.29), in both products it complies with the ordering
of the |i+ ()i (or, equally, |i ()i) elementary blocks in Eq. (3.21).
Let us apply the identity (3.29) to the D = 2 option of (3.5), (3.24). Combining
Eq. (3.29) with the orthonormality PR1 PR2 = R1 ,R2 PR1 of PR = dR CR and taking into
account the standard relation Trn [CR ] = dim RR ( )/dR (between the projected Trn trace and the canonical S(n) character R , see, e.g. [29]), one easily obtains
2 X

R ([1 , 2 ])
dR
1
1
,
(3.30)
=
B(R) =
dim R
n!
dR
dim R
{ S(n)}
692
where R Yn(N) , and [1 , 2 ] conventionally stands for (1 2 11 21 ). We have also used

that the block in the curly brackets of (3.30) is equal to unity according to the identity
derived in [16]. Together with Eq. (2.4), the expression (3.30) for B(R) precisely matches
with the (genus one) result of [13,23] derived by the combinatorial method of [24].
3.3.2. The D > 3 case
Returning to the generic D > 3 Tr4n+ -character (2.5) of the master-element (3.24), the
S(4n+ ) S(n+ ) reduction of the enveloping space is performed with the help of the
following generalization of the D = 2 identity (3.29)
!!#
"
!
D(D1)/2
D
O
O

(+)
()
4n

Tr4n+ D
"
=1
= Trn+ D
D
Y
=1
(+) 1 [ n+ ]
=1
!
D
Y
() 1 [ n+ ]
=1
!!#
(3.31)
Let us simply explain the meaning of the above pattern, while for more details see
Appendix A. In the l.h.s. of (3.31), the outer -product is defined in the same way
as in Eq. (3.24), and the operators () S(n ) (composed into the outer -product)
()
i subspace of the S(4n+ ) basis (3.13). As for the r.h.s. of
act on the associated |In()
(3.31), we first construct the following S(n+ ) basis. To begin with, one is to introduce
the N -dimensional spaces |i()i, i = 1, . . . , N, parametrized by the plaquette label =
P
1, . . . , D(D 1)/2. Then, the S(n+ ) operators (recall that n+ = {} n ) can be viewed
as acting on
|In(+) i =
D(D1)/2
O
|In() i,
n
|In() i = |i()i ,
(3.32)
=1
according to the same rule (3.10) that has been already used for the S(4n+ ) operators.
Given this convention, each operator () is postulated to act on the associated S(n )
subspace |In() i of |In(+) i,
D1
O

n
In() i =
|i()i ,
6=
n =
D1
X
n ,
(3.33)
6=
where the ordering of the |i()i blocks matches with the one in Eq. (3.32). As for 1 [n+ /n ] ,
in Eq. (3.31) it denotes the unity permutation on the S(n+ n ) subspace of (3.32)
complementary to (3.33). In the D = 2 case, where n+ = n1 = n2 , the general Eq. (3.31)
readily reduces to its degenerate variant (3.29).
To complete the construction (3.31), one should specify the ordering inside the two
inner -products of its r.h. side. We defer this task till the end of the section and now
apply the identity (3.31) to the Tr4n+ character (3.5) of the master-element (3.24). For this
purpose, all what we need is the proper identification of the composed into J4n+ ({R })
permutations with () . To this aim, let denote by and the permutations which
693
enter the definition (3.1) of the relevant operators CR = PR /dR S(n ) (combined
S(n ) being given by Eq. (3.9)) assigned with
into K2n of Eq. (3.26), with (1)
n
the associated labels {}, {}. Complementary, let stands for the permutations
()
1
entering the definition (3.12) of (2n ). Identifying (D1
6= ) ,
(+) (without numb summation over ), after some routine machinery one derives for the
master-element
An+ ({R }) =
D
Y
=1 R
(N)
n Yn
F=
O
{}
!
CR
Y
{}
dR2
n ! dim R
1 [ n+ ]
n
F ({ }; {R }),
(3.34)
S(n )
Y
{}
1 CR
1 [ n+ ]
n
!
,
(3.35)
so that its trace (2.6) determines the B({R })-block (2.5) of the TEK partition function
(PF).
Altogether, it establishes the exact duality transformation of the TEK PF (2.1) which is
one of the main results of the paper. In certain sense (modulo the explicit presence of the
-twists), it provides with the D > 3 generalization of the representation [13,23] for the
PF of the continuous gauge theory on a 2d manifold. Also, it can be compared with the
considerably simpler pattern of the PF of the judiciously constructed solvable D-matrix
models [28] where B({R }) depends nontrivially only on the associated generalized
LittlewoodRichardson coefficients. (On the contrary, the pattern (3.24) encodes the
general KlebschGordan coefficients.)
3.3.3. The ordering inside the inner -products

Finally, let us discuss the ordering inside the two inner -products of the r.h. side of
Eq. (3.31). As it is shown in Appendix A, this ordering is entirely predetermined by the
following characteristics G() of a particular -label called its cardinality. Consider the
D(D 1)/2 dimensional vector M({}) defined so that its components Mk ({}) are in
one-to-one correspondence with the D(D 1)/2 labels (where 1 6 < 6 D). The
latter parametrize the th plaquette-holonomies U entering the characters in Eq. (2.5).
Let the th component of M({}) is equal to the first link-label of the plaquettelabel: M ({}) = . Then, the cardinality (ranging from 0 to D 1) is postulated to
PD(D1)/2
[, Mk ({})], i.e., the number of times the particular -label
be G() = k=1
enters the entries of the vector M({}). In Eq. (2.5), it is always possible to arrange
for a nondegenerate cardinality assignment {G()} when G(1 ) 6= G(2 ) if 1 6= 2 .
()
Then, given a nondegenerate set {G()}, the ( 1 [n+ /n ] ) factors in the Eq. (3.31)
are ordered (from the left to the right) according to the successively decreasing G()assignments of their -labels.
694
4. SchurWeyl transformation of B({n })

eD into the D > 2
To transform the dual representation (2.6)(3.34) of the TEK PF X
stringy representation like (1.9), one is rewrite B({n }) entirely in terms of the symmetric
groups variables which are suitable for the algebraic definition of the topological data
associated to the mappings (1.11). For this purpose, we employ the following two useful
identities (derived in Appendices B and D) which reflect the SchurWeyl complementarity
of the Lie and the symmetric groups. The first one trades the ubiquitous operator (3.9) for
the product of the two elements of the associated S(n) algebra
X
(N)
(N n n )m , Pn(N) =
PR ,
(4.1)
(m)
n = Pn
(N)
RYn
(N)
(N)
(N)
where the projector (Pn )2 = Pn , Pn = 1 if n < N (that independently appeared

within the method of [31]) reduces the S(n) Yn -variety of irreps to the Yn(N) -one of either
U (N) or SU(N). As for the second element
X
(1/N)nK[ ] , [n , ] = 0, S(n),
(4.2)
n =
S(n)
(belonging the center of the S(n)-algebra), it is defined [16] by the equation

n
X
R ((N n n )m ) dR m
m
,
n K[ ] =
(k 1)pk ,
(dim R) =
dR
n!
(4.3)
k=1
P
where m Z, and the factor K[ ] = nk=1 pk in Eq. (4.2) denotes the total number of various k-cycles in the cyclic decomposition of the conjugacy class [ ] = [1p1 , 2p2 , . . . , npn ],
Pn
k=1 kpk = n.
Generalizing (4.1), the second key-identity deals with the similar sum weighted this time
by the factor e (which defines a generic model (1.1))

X
n! dim R m
e (...,{Cp (R)}) dR
CR = Pn(N) (N n n )m Qn ( ).
(4.4)
d
R
(N)
RYn
In the HeatKernal case (1.7) where ( ) = C2 (R)/2N , the S(n)-algebra valued

operator Qn ( ) reads

X
n2
,
(4.5)
exp Tb2(n) , Tb2(n) =
Qn ( ) = exp n 2
2
N
N
(n)
T2
where T2(n) T2 denotes the S(n) conjugacy class [1n2 21 ] of the simple transposition,
and = g 2 N . In a generic admissible model (1.1), as it is demonstrated in Appendix E,
the operator Qn ( ) generalizes to
X

X
(n)
(n)
{p} {bk }, n, N Tb{p}
=
{p} ,
(4.6)
, Tb{p}
Qn ( ) = exp
{p}
(n)
{p} T{p}
695
(n)
where Tb{p}
denotes the sum of the S(n) permutations belonging to a particular conjugacy
P
(n)
class T{p} labelled by the partition {p} of n: ni=1 kpk = n,. As for the weight {p} ( ), it
assumes the form (which, in particular, results in the required asymptotics (1.6))
M{p}
{p} ({bk }, n, N)
N
Pn
k=1 (k1)pk
X X

s{p} bk , m, l N 2l nm ,
(4.7)
m=0 l>[m/2]
where m, l Z>0 , and [m/2] = m/2 or (m + 1)/2 depending on whether m is even or

odd (while s{p} ( ) O(N 0 )). The specific pattern of {p} ( ) implies that in Eq. (1.1)
the function ( ) satisfies certain conditions (see Appendix E) that ensure the consistent
stringy interpretation of (4.7) to be discussed in Section 6 (for Eq. (4.5)) and in Appendix E
(for Eqs. (4.6), (4.7)).
Combining (4.1), (4.4) with the delta-function reformulation (2.8) of the Trn+ character
(derived in Appendix B), we arrive at the explicit S(n ) representation for the building
block B({n }) of the TEK partition function (2.4). This central, for the present discussion
of the lattice gauge string, expression reads
!
!
O Qn n X Y
Y 1 (1)
1[ n+ ]
n 1[ n+ ] , (4.8)
n+ n+
n
n
n !
{}
{ } {}
{}
where, to all orders in 1/N , one can safely use the SU(N) variant of the representation (4.1)
(m)
of n . It is noteworthy that the [Z2 ]D(D1)/2 invariance (with respect to R R )
of the sums in (2.1) defining the plaquette-factor (1.1) results in the invariance of Eq. (4.8)
under the simultaneous permutations , (), (), S(D) of the link-labels
, = 1, . . . , D, in the two ordered inner products.
5. The stringy form of B(n) in D = 2
To begin with, in the D = 2 case (where n+ = n12 = n1 = n2 ) all the involved into (4.8)
S(n ) operators act in one and the same S(n)-space that matches with the reduced formula
(3.29). The resulting amplitude in the HeatKernal case (1.7) reads (with B(n) 1)
2
B(n) =
2 (n n 2 )
N
n!
b(n)
n Pn(N) e N T2 (N n n )12+1 [1 , 2 ]
(5.1)
(n)
{ },T2 S(n)
that is in complete agreement with the genus-one result [31] of Baez and Taylor derived by
a different method. It provides with the more compact reformulation of the GrossTaylor
stringy pattern (1.9), although the transformation (to be summarized by Eqs. (5.7), (5.8)
below) relating the two representations is not entirely manifest. Let us proceed recasting
(5.1) into the form almost equivalent to the one of (1.9).
To make contact with the pattern of the f -mapping of Eq. (1.9), first it is convenient to
decompose
X
(n) b(n)
Pn(N) T{p}
(5.2)
T{p} ,
Pn(N) =
(n)
T{p} S(n)
696
where Tb{p} is defined in Eq. (4.6). Then, expanding all the exponents except e 2 n = e 0 n ,
one is to rewrite (5.1) as
(n)
B(n) =
X N 2(t +s)i
P (N) (T{p} ) Kn (i, s, t),
|Cf ({p})| n
(5.3)
e
i,s,t >0 T{p} S(n) f M
Kn (i, s, t) = e 2 n
()i+s+t (1)i ns (n2 n)t

,
i!s!t!
2s+t
(5.4)
where in the first exponent of (5.1) one is to decompose n2 = n/2 + n(n 1)/2, and for
(n)
in the rest of the section. As for the symmetry factor,
simplicity we denote T{p} T{p}
X
e
f M({p},n,i)
1
=
|Cf ({p})|
X
{ S(n)}

1
b{p} Tb2 i [1 , 2 ] ,
n T
n!
(5.5)
it emerges when one reformulates the r.h.s. of (5.5) as the sum over certain maps (1.9)
(to be explicitly constructed below). The latter can be viewed as the topological mappings
(i.e., immersions without singularities)

e f 1 (qs ) = M {qs },
(5.6)
f :f M
e {f 1 (qs )} onto the base-space torus M with i + 1 deleted points {qs }.
of the space M
These maps define [30] the admissible (by the data in the r.h.s. of (5.5)) branched covering
e of M which can be visualized as the Riemann surfaces M
e M({p},
e
spaces M
n, i) (to
be identified with the worldsheets of the YM-flux) with n-sheets and i + 1 branch points
located at {qs }.
As for the sums (5.3) over the nonnegative integers i, t, s, in addition to the number
of the movable simple branch-points (where two sheets are identified), they refer to the
e
extra movable singularities of the map (1.9). Namely, one is to attach to M({p},
n, i)
the t collapsed to a point microscopic tubes (connecting two sheets) and the s collapsed
to a point handles (glued to a single sheet). Altogether, it results in the worldsheet
e
M({p},
n, i|t, s).
The key-observation [16] is that the factor |Cf ({p})| is equal to the number of
e
distinct automorphisms of the branched covering space M({p},
n, i|t, s) in question.
(N)
the 1/N factor
Complementary, in the absence of the Pn -twist (i.e., when T{p} 1),
enters (5.3) in the power equal to the G = 1 option of the RiemannHurwitz formula h = n
(2G 2) + 2(t + s) + i calculating the overall genus of the corresponding Riemann surface
e 1,
n, i|t, s). Also, in what follows, we assume that the contribution of the movable
M(
collapsed handles is reabsorbed into the redefinition 0 = /2 (1 1/N 2 )/2 of the
SU(N) bare string tension which eliminates the corresponding singularities of the map
(1.11). (In the case (4.6) of the generic model (1.1), the modified tension is given by the
(n)
n = 1 restriction of {p} ({bk }, n, N) associated to T{p} = 1 [n] .)
From the the general expression (4.6), it is clear that the pattern (1.9) emerges in a
generic model (1.1) as well. In particular, owing to the pattern of Eq. (4.7), the large
N asymptotics ln[Qn ( )] = n 0 ({bk })(1 + O(1/N)) is consistently provided by the
the leading l = 0 term in Eq. (4.7) describes the
Tb{p} = 1 term of (4.6). For T{p} 6= 1,
697
branch-point canonically parametrized (see, e.g. [22]) by T{p} . The latter point decreases
P
the associated Euler character by k (k 1)pk which matches (akin to the pattern (4.3) of
b{p} . As it is discussed in Appendix E
n ) with power of the 1/N factor assigned in (4.7) to T
(where the earlier results [17,22] are summarized and reformulated), the l > 1 terms can
be reinterpreted as the movable subsurfaces (of various topologies) collapsed to a point.
(N)
As for the factor Pn (T{p} ), inherited from the decomposition (5.2) of the projector
(N)
Pn , its dependence on N is not particularly suitable for a manifest 1/N expansion
like (1.9). It calls for a nontrivial resummation which would reproduce the the welldefined large N SC series (1.9) (obtained by Gross and Taylor without resort to (5.1)).
(N)
The latter pattern effectively eliminates Pn (T{p} ) at the expense of working with the
S(n+ ) S(n ) double-representation, B(n) B({n }), that refers to the two coupled
sectors of the opposite worldsheet orientation. The prescription to reconstruct B({n })
from the amplitude like (4.1) is quite simple: all the involved S(n)-structures are traded for
their S(n+ ) S(n ) counterparts
n ( ) n+ n ( ),
+ +n
N n n N n
+ +n ((n+ )2 +(n )2 2n+ n )/N 2 )
Qn Qn+ ,n = e 2 (n
+ , (5.7)
n+ ,n ,
+)
b(n
e N (T2
+Tb2(n
(5.8)
where Eq. (5.8) implies the HeatKernal case (1.7) and can be generalized to a generic
model (1.1). (The definition and interpretation of n+ ,n and other ingredients in
Eqs. (5.7), (5.8) can be found in [16].)
5.1. Construction of the branched covering spaces
In the remaining subsections, we discuss the major issues related to the effective
enumeration of the mappings (5.6) and their automorphisms. Let us proceed with an
explicit algorithm which, given the symmetric group data in the r.h.s. of (5.5), reconstructs
e
the topology of the Riemann surfaces M({p},
n, i) in the l.h.s. of (5.5). As the elements
Tb{p} , (Tb2 )i are associated [18] to the branch points (BPs), it is convenient to start with the
simpler case of the topological covering spaces (without BPs singularities) removing the
latter elements. The full branched covering spaces (BCSs) can be reproduced reintroducing
the BPs onto the corresponding covering spaces (CSs).
e of a given 2d surface M (the 2-torus in what follows), it
As for a particular n-sheet CS M
can be composed with the help of the cuttinggluing rules borrowed from the constructive
topology. To begin with, cut a 2-torus M along the two uncontractible cycles () trading
the latter for the pairs of edges () (), = , . It makes M into a rectangular
H with the boundary edge-path represented as ()() 1 () 1 (). Then, consider
e = H n (where n = {1, . . . , n}) of H by n copies
the trivial covering H
of this rectangular with the boundary edge-paths given by (k )(k ) 1 (k ) 1 (k ),
k = 1, . . . , n. Perform the set of the pairwise reidentifications of the involved edges
(k ) = ( (k) ),
: k (k), k n ,
(5.9)
where 1 , 2 are supposed to satisfy the n -constraint (5.5) (with the excluded
b{p} (Tb2 )i ).
contribution of T
698
Evidently, the two sets (5.9) of the reidentifications can be concisely represented as
the two closed (i.e., without branch end-points) branch cuts $ of the Riemann surface
e }) with n sheets. According to the pattern (5.9), each connected component of
M({
e }) has the topology of 2-torus. In compliance with (1.9), it matches with the
M({
N -independence of the argument of the n -function (5.1) taking place after the exclusion
of Pn(N) exp[Tb2(n) /N].
b{p} (Tb2 )i factor of (5.5)), recall that
To reintroduce the branch points (encoded in the T
each admissible BP is the end-point qk of the associated branch cut $ (k) which should be
e }). To implement $ (k) , we first
included additionally to the closed cuts $ of the CS M({
(k)
(k)
e
cut M({ }) along the support of $ . (Both $ and $ are all supposed to terminate
at a common base-point p = () () of M.) Then, on the left and on the right sides
of each cut $ (k) , the resulting two copies of the n new edges of the sheets are reidentified
according to the prescription (5.9). The only modification is that, instead of , one is
to substitute the appropriate permutations {p} T{p} and (s) T2 entering respectively
Tb{p} and the sth Tb2 -factor in the inner product (Tb2 )i . It completes the construction of the
e
admissible Riemann surfaces M({p},
n, i) entering the l.h.s. of (5.5).
5.2. The homomorphism of 1 (M {qs }|p) into S(n)
To enumerate the equivalence classes of BCSs and justify the asserted interpretation of
|Cf ({p})|, one is to employ the relation to the following group homomorphism [18] where
the first homotopy group 1 (M {qs }}|p)
: 1 (M {qs }|p) S(n),
(5.10)
is mapped into S(n). (In Eq. (5.10), M {qs } denotes the base-surface M with the i + 1
excluded points qs , associated to the branch points, and with the base-point p.) Given an
BCS encoded in the n -function (5.5), choose a set n = {1, 2, . . ., n} to label the n sheets
(at the base-point p). Consider the lift [22,30] of the closed paths in M {qs } (defining
e {f 1 (qs )}. Then, the (equivalence classes of
1 (M {qs }|p)) into the covering space M
the) paths induce the permutations of the labels which determine the corresponding S(n)
operators acting on n .
To make (5.10) explicit, let us first specify the pattern of the first homotopy group.
Consider a topological space T {q1 , . . . , qm } which, for our later purposes, is allowed
to be a (CW) 2d cell-complex T (not necessarily reduced to a 2d surface) with m deleted
points qk . Recall that in this case the group 1 (T {q1 , . . . , qm }) (in what follows we will
everywhere omit the specification of the base-point p) can be represented as the following
abstract group [30]. The generators of the latter group are associated to the homotopy
equivalence classes (HEC) of the uncontractible closed paths based at a given point p
(supposed to be distinct from the set {qs }). In the case at hand, additionally to the generators
r , r = 1, . . . , P (corresponding to P HECs of the uncontractible cycles of T ), there are
extra m generators (s) , s = 1, . . . , m, which refer to the HECs of closed paths encircling a
single deleted point qs . Finally, there exists a constructive algorithm to find the complete set
of K Z>1 relations {Fl ({r , (s) }) = 1}l=1,...,K that completes the intermediate mapping
of the 1 (T {q1 , . . . , qm }) generators into the abstract group.
699
Returning to the case of a genus g 2d surface T = Mg , there is a single relation (with

P = 2g, [i , k ] i k i1 k1 ) defining 1 (Mg {qs })
!
g
m
Y
Y

(s)
(s)
[j , g+j ]
=
= 1.
(5.11)
F r ,
j =1
s=1
Comparing (5.11) with the pattern (5.1)(5.5), one deduces the explicit form of the
homomorphism (5.10) (with the identification g = 1, m = i + 1)

: ( ) = , (1) = {p} ,
( (s) ) = (s1),
s > 2,
(5.12)
b{p} and the (s 1)th Tb2 -factor

T{p}
T2 which enter respectively T
(in the inner product (Tb2 )i ).
where {p}
and (s1)
5.3. The symmetry factor

Given the homomorphisms (5.12), one can readily enumerate the equivalence classes
e
M({p}, n, i) of the BCSs employing the following notion of equivalence [15,16] of two
homomorphisms 1 and 2 . The latter are postulated to belong to the same equivalence
class if there exists some S(n) so that

(5.13)
1 ( ) = 2 ( ) 1, 1 M {q1 , . . . , qm } , S(n).
Then, the basic theorem [18,30] of the topological coverings ensures that the inequivalent
homomorphisms (5.12) are in one-to-one correspondence with the associated homeomorphically distinct branched covering spaces.
Finally, let : f = f , denotes a particular automorphism of the branched covering
e Being restricted to the n-set n = f 1 (p)
(p 6= qs , s), the group of the
space M.
automorphisms is isomorphic [30] to (the conjugacy class of) the S(n)-subgroup Cf ({p})
that induces conjugations (5.13) leaving all the images 1 ( ) invariant: 1 ( ) = 2 ( ),
, () Cf ({p}). In turn, it justifies [16] the required interpretation of |Cf ({p})|.
6. The stringy form of B({n }) in D > 3
To rewrite the D > 3 amplitude (4.8) in the form generalizing the D = 2 stringy pattern
(n )
(5.3)(5.5), we first expand each factor Qn and select the i th power (Tb2 )i (where
(n )
the k th Tb -factor in the latter product is supposed to be defined via Eq. (4.5) in
2
(k )
(n
terms of T2 , k = 1, . . . , i ). Complementary, akin to Eq. (4.6) one is to

decompose
X
X
(m )
(m )
(n )
n T{p}
{p } ,
(6.1)
n =
(n )
S(n )
}
T{p
(n )
}
{p } T{p
b{p } (defined by Eq. (4.6)).

b(n ) T
and separate the contribution of a given T
{p }
700
Given the above expansions, one can prove (see below) that the associated building block
of (4.8)
X
{ }

(n )
O
Y 1

(Tb )i Y
b{p } 1 n+
n+ Tb{p+ }
1 [ n+ ]
T
Tb{p } 2
[ n ]
n
n !
{}
{}
e },{n },{i })
M({p
{}
1
|C ({p })|
(6.2)
can be rewritten as the sum over the relevant D > 2 mappings (1.11) to be reconstructed
in the next subsection. Extending the D = 2 theorem due to Gross and Taylor, each term
of the latter sum is weighted by the inverse number |C ({p })| of distinct automorphisms
e }, {n }, {i }).
e M({p
associated to a given M
e can be viewed as the
As we will demonstrate, the involved into (6.2) spaces M
generalized Riemann surfaces to be identified with the worldsheets of the YM-flux
wrapped around the EK 2d cell-complex (2.2). Combining all the pieces together, the
stringy reinterpretation of (4.8) essentially follows the steps discussed in the D = 2 case.
Namely, leaving aside the Pn(N)
(T{p } ) factors (4.1), the rest of the ingredients of (4.8)
readily fit in the consistent D > 3 extension (1.11) of the GrossTaylor stringy pattern
(1.9). Indeed, by the same token as in the D = 2 analysis, the involved (movable or
nonmovable) branch points, microscopic tubes, and handles are weighted by the 1/N e .
and n -dependent factors according to their contribution to the Euler character of M
The latter local matching is completed by Eq. (6.2) together with the following global
matching. To see it, let us remove temporarily the branch points and the collapsed
subsurfaces via the substitution
(m )
n N n m ,
Qn exp[n /2],
(6.3)
into Eq. (4.8). Then, the remaining n -dependent factor and the overall power (1/N)
(inherited from Eq. (4.8))
=
D(D1)/2
X
=1
D
X
=1
n + n+ = 0,
2n+ =
X
{}
n = 2
n ,
(6.4)
{}
match with the area and the 2-tora topology (revealed below) of each connected component
e corresponding to the deformation (6.3).
of M
By the same token as in the D = 2 case, the manifest stringy representation of the
full amplitude (4.8) (including the Pn(N)
(T{p } )-factors) calls for a resummation into
the appropriate large N SC series. Presumably, it eliminates these factors trading each
remaining in (4.8) S(n )-operator for its S(n+

) S(n ) descendant. It is suggestive that
the short cut way for the latter reformulation is provided by the direct extension of the
D = 2 prescription (5.7), (5.8): one is to substitute the labels n, n+ , n by their properly
associated -dependent counterparts n , n+

, n (where n , {}, {}, +, labels the
relevant S(n )-structures). Note also that, similarly to the D = 2 case, a generic D > 3
model (1.1) complies with the pattern (1.9) as well.
701
In conclusion, we remark that the amplitude (4.8) exhibits purely topological assignment
(see Section 6.2 for more details) of the S(n )-structures after the formal deformation
(m )
{n (N n n )m }, {Qn 1}. In particular, consider the powers m = 1 in

(m )
which the relevant S(n )-twists n enter the n factors (4.1): {m = 1}, {m = 1},
m+ = 1. The latter are equal to the weights in the famous formula computing the Euler
characteristic of a given cell-complex T : 2 2GT = np nl + ns = np mp + nl ml + ns ms ,
where np , nl , ns are the total numbers of respectively plaquettes, links, and sites of T . (The
earlier heuristic arguments (in the case of the single chiral sector), consistent with the above
pattern of the n assignment, can be found in [26]).
The topological nature of the considered deformation of the amplitudes like (6.2) can
be made transparent by the conjecture building on the D = 2 observation [22]. One may
eD yields the generating
expect that, after this deformation, the TEK partition function X
functional for the orbifold Euler characters of the Hurwitz-like spaces associated to the
e (of
following set. The latter includes all the generalized branched covering spaces M
the EK 2d cell-complex (2.2)) corresponding in Eq. (6.2) to the considered deformation
(N)
{Pn 1}. (For a close but somewhat distinct earlier conjecture, see [26].) Similar
proposal can be made for the (weaker) topological deformation {Qn 1} of the PF
eD reintroducing the Pn(N)
factors.
X
6.1. The generalized covering spaces of TEK {qs }

Given the data in the argument of the n+ -function in the l.h.s of Eq. (6.2), our aim is
e (i.e., the mappings (1.11))
eEK M
to reconstruct the associated topological spaces M
which are summed over in the r.h.s. of (6.2). This problem, being inverse to what is
usually considered in the framework of algebraic topology [30], will be resolved through
the sequence of steps which appropriately generalizes the D = 2 analysis of the previous
eEK do not coincide with the canonical branched
section. It is noteworthy that the spaces M
covering spaces (BCSs) of the TEK 2d cell-complex TEK . To say the least, a generic
canonical BCS of TEK is again a 2d cell-complex (with a number of branch points) but
not a 2d surface like in the asserted mappings (1.11). As we will see, the basic amplitude
(6.2) indeed complies with (1.11) and refers to the novel class of the associated to TEK
eEK to be called the generalized branched covering spaces (GBCSs) of TEK .
spaces M
Similarly to the D = 2 analysis, we start with the simpler case of the generalized
covering spaces (GCSs) of TEK corresponding to the deformation (6.3). Being specified by
the immersions (1.11) without the (branch points and collapsed subsurfaces) singularities,
the GCSs can be reconstructed generalizing the D = 2 surgery-construction of the
covering spaces. Take D(D 1)/2 -rectangulars H representing the decompactified
plaquettes of the EK base-lattice (2.2). Then, one is to begin with trivial (n -sheet)
e = H n of H with the edge paths q (q )(q ) 1 (q ) 1 (q ),
coverings H
q = 1, . . . , n . To reproduce the effect of the -permutations in (6.2), first let us denote
P
by {(k ), k = 1, . . . , n = D1
6= n } the -set of the (q ) edges (collected from the
(D 1) different -plaquettes) ordered in accordance with the pattern (3.14) of the S(n )
()
i. Similarly, we introduce the sets {(k ), k = 1, . . . , n }.
basis |In()
702
To reconstruct the associated to (4.8) GCS of TEK , at each particular -link one is to
perform the pairwise -identifications of (k ), (k ) according to the prescription (5.9).
In this way, we have constructed appropriate conglomerate of the generalized Riemann
e }) (with the total number D of the closed branch cuts) which are wrapped
surfaces M({
around TEK in compliance with the mapping (1.11). Evaluating the Euler characteristic of
e }) in accordance with (5.9), each connected component of the latter surface reveals
M({
the topology of a 2-tora. In turn, it matches with the overall power (6.4) of the 1/N factor
(m )
which is inherited via (6.3) from the product (4.8) of the n factors.
e }) of TEK , the branch points (encoded in (6.2) through the
Given the GCSs M({
(n ) i
b
b
) ) are reintroduced in essentially the same way as we did for
operators T{p } (T2
the D = 2 case. The only subtlety is that, to make the employed cuttinggluing rules
well-defined, it is convenient to view the EK cell-complex TEK as the homotopy retract
0 (possessing by construction the same fundamental
of a less singular 2d complex TEK
0
group 1 (TEK ) = 1 (TEK )). We refer the reader to Appendix C for the details, and now
proceed with the D > 3 generalization of the D = 2 GrossTaylor theorem concerning
the interpretation of the symmetry factor entering the D > 2 stringy amplitudes like (1.9),
(6.2).
6.2. The homomorphism of 1 (TEK {qs }) into S(n+ )
To effectively enumerate the constructed GBCSs of TEK and their automorphisms, we
first reconstruct what kind of homomorphism like (5.10) is encoded in the D > 3 pattern
(6.2). It will provide with the precise mapping of the fundamental group 1 (TEK {qs })
(of TEK with a number of deleted points {qs } associated to the branch points) into the
enveloping S(n+ ) group.
Observe first that the abstract group representation of 1 (TEK ) is defined, according to
(2.2), in terms of the D generators corresponding to the uncontractible cycles (i.e.,
the compactified -links) of TEK . Having excluded from TEK a set of points {qs }, it is
()
convenient to recollect them into the three varieties of the -subsets {qs } = {qk },
{}, {}, +, k = 1, . . . , b , belonging respectively to the interior of the -plaquette,
to the interior of the -link, and to the single site of TEK . The set of the 1 (TEK
()
{qs }) generators includes (additionally to the subset { } inherited from 1 (TEK )) k
associated to (the equivalence classes of) the closed paths encircling a single deleted point
()
qk . The representation of 1 (TEK {qs }) is then completed by the D(D 1)/2 relations
!
b
Y
Y
()
k = 1, = 1, . . . , D(D 1)/2,
(6.5)
[ , ]
=,,,+ k =1
each of which can be viewed as the -copy of (5.11).

Comparing the argument of the n+ -function in the l.h.s. of Eq. (6.2) with the pattern
of (6.5), it is straightforward to write down the precise form of the relevant D > 2
homomorphism generalizing (5.10):
: ( ) = , (1 ) = {p } , (k ) =
()
()
(k
1)
k > 2,
(6.6)

(k
where {p } Tn ,
1)
(n
703
()
T2 . Note that the three -varieties of the generators 1

(n )
()
match with the three species of Tbn in Eq. (6.2), while k represents the k th Tb2 (n )
factor in the product (Tb2 )i . In particular, Eq. (6.6) implies that in Eq. (6.5) b =
i + 1 while b+ = b = 1, .
6.2.1. The symmetry factor

To deduce the announced interpretation of |C ({p })| (entering the r.h.s. of Eq. (6.2)),
let us start with the following observation. The summation in the l.h.s of (6.2) can be
viewed as the sum over the associated homomorphisms (6.6) constrained by the condition:
(k 1)
(n )
T2 . Therefore, one is to identify the proper equivalence
{p } T{p } and
classes of the homomorphisms (6.6) to parametrize uniquely the topologically inequivalent
e constructed in Section 6.1.
spaces M
We assert that the two homomorphisms (6.6), 1 and 2 , are equivalent (i.e., the
e are homeomorphic) if and only if there
corresponding generalized branched coverings M
exists some S(n ) so that
1 ( ) = 2 ( )1 ,
0
1 (TEK
{qs }),
S(n ),
(6.7)
where the basis (3.32) for S(n+ ) is implied. For a preliminary orientation, one observes
that the conjugations (6.7) are induced by the S(n ) permutations of the sheets (of
e ) separately within each of the D(D 1)/2 distinct n -subsets which covers (see
M
0 . (Complementary, S(n ) is
Section 6.1) a given -tora E combined into TEK
the largest subgroup of S(n+ ) providing with the conjugations leaving the argument of
the n+ -function in Eq. (6.2) invariant.) We refer to Appendix D for the justification of
the notion (6.7) of the equivalence and now simply deduce its consequences. Consider
e , and let C ({p }) is the group of the inequivalent
a particular branched covering M
e : = . Take the restriction of a given to the n+ space
automorphisms of M
of { 1 (p)}). Akin to the D = 2 case, one shows that C ({p }) is isomorphic to the
conjugacy class (with respect to (6.7)) of the following subgroup of S(n ). The latter
is associated to such subset of conjugations (6.7) that leave all 1 ( ) invariant: 1 ( ) =
2 ( ), , () C ({p }). In sum, there are exactly ( n !)/|C ({p })| distinct
e that justifies the basic
homomorphisms {k } associated to the same topology-type of M
identity (6.2).
7. The area-preserving homeomorphisms

On a lattice, our D > 3 stringy proposal can be viewed as the confluence of the Wilsons
string-like reformulation [6] of the (finite N ) SC series with the power of the large N
expansion. The feature, which sharply distinguishes the D > 3 gauge string from the earlier
e
D > 3 proposals [79] in this direction, is the asserted invariance of the weights w[M]
under certain continuous group of the area-preserving worldsheet homeomorphisms. In
D > 3, the latter symmetry is encoded in the following D > 3 descendant inherited from
the D = 2 renormgroup (RG) invariance of the YM2 systems (1.1).
704
Recall first that a D > 3 YMD theory (1.1) (having some lattice LD as the base-space)
can be equally viewed as the YM theory being defined on the 2d skeleton TD of LD
represented by the associated 2d cell-complex. Consider the partition function (PF) of
(1.1) on TD = k Ek composed from the associated 2d surfaces Ek (of the areas Ak and
with certain boundaries) according to the corresponding incidence-matrix [21]. Akin to the
D = 2 case, this PF is invariant under subdivisions of TD (preserving the total areas Ak )
so that Ek can be made into the associated smooth 2d manifolds Mk (with boundaries)
e D = k Mk (homeomorphic to TD ). Therefore, refining
combined into a 2d cell-complex T
the discretization, the YMD lattice theory (1.1) can be adjusted to merge with the following
continuous system. To implement the latter, we first put the associated to (1.1) continuous
YM2 theory (1.4) (keeping free boundary conditions) on each Mk . Then, one is to average
over the gauge fields, living on the boundaries of {Mk }, in compliance with the associated
e D incidence-matrix.
to T
Similarly to the D = 2 case (see Section 1.1.1), on the side of the proposed D > 3 gauge
string, the above relation to the continuous YM2 system ensures the required invariance
of the lattice string weights (entering the amplitudes defined by the D > 3 extension
(1.11) of (1.9)). Indeed, on the one hand, both the parity P and the symmetry factor |C |
e ) and on the one of the corresponding
depend only on the topology of the worldsheet M(T
tangent-space T = k Ek . (Compare it with [79] where the singular lattice geometry of
T is built into the curvature-defects.) On the other hand, the sum over the mappings
e ). Indeed, the multiple
also supports the required invariance of the set of relevant M(T
integrals (rather than discrete sums as in [79]) over all admissible positions of the movable
branch points (and certain collapsed subsurfaces) span the entire interior of the discretized
R
2d surfaces Ek combined into T . The remaining discrete contributions into d refer
to the purely topological assignment of the nonmovable singularities anywhere in the
interiors of the appropriate subspaces of T . Altogether, the relevant group of the worldsheet
homeomorphisms continuously translates the positions of the admissible singularities of
the map (1.11) within the interiors of the associated subspaces of T .
e associated to the YM system (1.1)
To be more specific, we compare the weights w[M]
defined on a generic base-lattice FEK homeomorphic to the elementary EK 2d cellcomplex (2.2). Let the total area (measured in the dimensionless units) of a given torus E
is equal to A . Combining the above general arguments with the analysis of Sections 3 and
4, the general weights on FEK can be deduced from the basic TEK amplitude (4.8) trading
in each Qn -factor (4.5) the coupling constant for A . In particular, the (n )m -,
-twist assignment is indeed purely topological: the corresponding conglomerates of
the nonmovable branch points (and collapsed subsurfaces) can be placed anywhere (but
without summation over positions) in the interior of the associated macroscopic -, -cell
of FEK .
705
8. Smooth gauge strings vs. lattice ones

It is clear that the method, we have developed in Sections 36 on the example of the
TEK model (2.1), can be extended for the YMD theories (1.1) defined on a generic regular
subspace T of the 2d skeleton of the D-dimensional base-lattice LD . We defer the analysis
of the generic weights (1.2) for a separate paper and now simply stress those implications of
our present results which are T -independent and common for both the lattice and smooth
realizations of the gauge string. In this way, one actually decodes all the major peculiar
features of the continuous theory of the smooth YM-fluxes which are novel compared to
the conventional paradigm of the D > 3 fundamental strings.
To begin with, one observes that the relevant smooth mapping (see, e.g., Eq. (1.11) for
e without boundaries) are allowed to have singularities which usually
the worldsheets M
are not included into the D > 3 string-pattern (where the maps (1.11) are restricted to be
e is
sheer immersions). Complementary, certain class of the selfintersecting worldsheets M
e )|{bk }] 6= 1 that does not have a direct
endowed in Eq. (1.2) with the extra factors J [M(T
counterpart in the known D > 3 string theories. In turn, it is the extra J [ ]-weights which
encode the data sufficient to reconstruct the associated continuous YMD model (1.3), (1.4).
As for a single D > 3 Wilson loop average, the major novel ingredient is due to the various
movable branch points and the collapsed 2d subsurfaces (assigned with the {bk }-dependent
weights (1.10)). Their positions can be viewed as the zero modes associated to the minimal
surface contribution (provided the latter properly selfintersects) resulting in the extra areadependence in the preexponent of hWC i. This pattern generalizes the one of the D = 2
loop-averages [7,15,16].
e (and,
Next, let us demonstrate that the total contribution of the worldsheets M
consequently, of the tangent-spaces T ) with the backtrackings vanishes which matches
with the absence of foldings in the D = 2 GrossTaylor representation (1.9). To make this
feature manifest, we first recall that the relevant for (6.2) maps (1.11) (associated to the
e wrapped around
SU(N) TEK model) do not include any foldings of the worldsheets M
TEK . On the other hand, one could equally start with the large N U (N) TEK model
where the proposed technique would result in all kinds of foldings covering various
conglomerates of the plaquettes (with possibly different -orientation). The comparison
of these two complementary large N patterns justifies the above assertion.
On the side of the gauge theory defined on the standard cubic lattice LD , the TEK
e To see it, let us call
foldings are associated to the more general backtrackings of M.
e
worldsheets M(T ) (or tangent-spaces T ) without any backtrackings regular. Then, to any
regular tangent T (r) one can associate the variety of tangent-spaces creating all kinds of
the backtracking (bounding a zero 3-volume) with the support not necessarily belonging
the original regular tangent T (r) (in contradistinction to the foldings discussed in [7]).
The irrelevance of the backtrackings is intimately related to the invariance (in gauge
theories) of the Wilson loop averages hWC i under the zig-zag backtrackings of the
boundary contour C that is in sharp contrast with the situation in the conventional Nambu
Goto string and most of its existing generalizations. (Recently Polyakov [1] advocated
the latter invariance as the crucial feature of the strings dual to gauge theories.) In the
706
limit N , the zig-zag symmetry of hWC i can be made manifest confronting as

previously the two large N formulations of the open gauge string: the U (N) one with
e obtained
a given backtracking contour C should be compared with the SU(N) one with C
contracting zig-zags of C.
8.1. Correspondence with the WC Feynman diagrams
Let us reveal the WC/SC correspondence between the continuous YMD models (1.4)
in the WC phase and the associated smooth gauge string in the SC regime. According
e )] can be
to Section 7, once the backtrackings are absent, the relevant weights w[M(T
derived by putting the YM2 system (1.4) onto a given 2d cell-complex (e.g., 2d surface)
T (r) = k Mk where the 2d surfaces Mk are piecewise smooth. In the large N SC regime,
the latter YM system is represented by the conglomerates of the worldsheets properly
wrapped around T (r) . On the other hand, in the large N WC regime the WC perturbation
theory represents our system through the set the Feynman diagrams. Employing the
standard path integral representation of the propagator of the free particle (in a curved
space), the diagrams are visualized as the fishnet (of the gluonic trajectories) appropriately
wrapped around the same 2d complex T (r) . The crucial observation is that, when T (r) is
viewed as being embedded into RD , the latter fishnet can be reinterpreted as the specific
contribution of the WC perturbation theory in the D-dimensional continuous YMD model
(1.4) (uniquely associated to the chosen YM2 via Eq. (1.3)). To select this contribution on
the YMD side, not only the gluonic trajectories in RD should be constrained to have the
spacetime support on T (r) but also each colour a-component of the associated gluonic
a (z), z T (r) , as the Lorentz (or O(D)) tensor should belong to the
strength-tensor F
tangent space of T (r) at z. (To circumvent gauge-fixing, tricky to explicitly match between
the T (r) - and standard formulations, one is to introduce an infinitesimally small mass term
for the gauge field and then perform the above comparison.)
8.2. Suppression of the selfintersections
At this step, it is appropriate to discuss a number of crucial simplifications inherent
in the dynamics of the considered D > 3 continuous flux-theory compared to its
e (resulting
lattice counterpart. Consider first the subset of smooth closed 2d surfaces M
D
from the smooth immersions of a 2d manifold M into R which alternatively can be
represented by the maps (1.11) with T RD with selfintersections on submanifolds of
the dimensionality d > dcr = (4 D). The latter subset is of measure zero [27] in the
set of all smooth immersions M RD (with arbitrary selfintersections). Complementary,
the smooth immersions M RD , D > 4, are dense [27] in the space of all piecewise
smooth immersions M RD . In particular, in D > 5 the subset of -maps resulting
e is the open, dense
in embeddings (i.e., in closed nonselfintersecting 2d manifolds M)
subspace of the space of all piecewise smooth immersions M RD . (Another example,
in D > 3 the backtrackings (being viewed as two-dimensional selfintersections) are of
measure zero, i.e., unstable.) Next, suppose that the redefinition (discussed in the very
707
end of Section 1 and after Eq. (5.6)) of the bare SU(N) string tension 0 is performed
that eliminates certain types of the collapsed 2d subsurfaces originally built into (1.11).
Then, by virtue of the Whitney immersion theorem [27], the singularities of the generic
(piecewise smooth) mapping (1.11) in D > 4 can be restricted to the ones listed after
Eq. (1.11) with the exclusion (in the HeatKernal case (1.7)) of the movable collapsed
subsurfaces attached to a single sheet of the covering. As for the remaining admissible
collapsed subsurfaces and the branch-points, being necessarily attached to nontrivial
selfintersections of the worldsheet, they therefore correspond in D > 5 to measuree (induced by the
zero limiting points of the dense subspace of the 2d manifolds M
embeddings (1.11)) without boundaries. As for D = 4, the stable (i.e., of nonzero measure)
e can occur only at a set of isolated points.
selfintersections of a closed smooth surface M
The advantage of the gauge string is that, as it is clear from the previous sections,
such zero-dimensional selfintersections are not weighted by any extra factors so that the
e )|{bk }] = 1, b = 0, reduction
corresponding worldsheets are still assigned with the J [M(T
of the weight (1.2). Altogether, it justifies the assertion made in the very end of Section 1.
e )|{bk }] 6= 1 pattern is observable in D > 3, recall first the basic
To see how the J [M(T
theorem [27] on the stability of the smooth immersions: the stability is equivalent to the
local stability. In particular, it implies that the unstable in the case of a closed 2d surface
selfintersections can not be stabilized (in the bulk) introducing some selfintersecting
boundary contour(s). To be more specific, consider the framework of the quasiclassical
expansion for hWC i where the weight of the minimal surface enters as the isolated, discrete
contribution disparate from the continuum of the string fluctuations. As a result of the
e )|{bk }] 6= 1 factor may
above theorem, in D > 4 the only place where a nontrivial J [M(T
be observable (beyond the considered SU(N) redefinition of 0 ) for a nonbacktracking
contour C seems to be the contribution of the selfintersecting minimal surface. The
simpliest option, where the weights (1.10) of the movable branch points can be measured,
is to take contours winding a number of times around a nonselfintersecting loop C. On
the other hand, let C any non-backtracking loop associated to some nonselfintersecting
e )|{bk }] = 1,
minimal-area surface(s). In this case, we expect that the simpliest J [M(T
b = 1 pattern of (1.2) (with the above redefinition of 0 and with the worldsheets represente M RD ) in D > 4 is sufficient to reproduce the
ed by the smooth strict immersions M:
correct result for the contribution of the genus h smooth worldsheets to the average hWC i.
9. Conclusions
We propose the correspondence between the smooth gauge string (1.2), induced from the
lattice models (1.1), and the associated via (1.3) continuous YMD theory (1.4) (with a finite
UV cut-off) in the large N SC phase. This duality implies the concrete prediction (1.6) for
the bare string tension 0 = 2 0 as the function of the coupling constants entering the
YMD Lagrangian (1.4). In particular, it readily allows for a number of nontrivial large N
predictions in the extreme SC limit where 0 merges with the leading asymptotics of the
physical string tension ph . More generally, the correspondence asserts that the generic
708
continuous YMD model (1.4) is confining in D > 3 at least in the large N SC regime
accessible by our approach.
In turn, it suggests the mechanism of confinement in the standard weakly coupled (WC)
D = 4 continuous gauge theory (1.8) at large N . Consider the effect of the Wilsonian
renormgroup flow on the original YMD theory in the WC phase at the UV scale (where
the large N SC expansion, in terms of the proposed microscopic gauge strings, fails).
The idea is that the latter YMD theory, at the sufficiently low-energy scale, may be
superseded by such effective strongly coupled YMD system where the (effective) gauge
string representation is already valid. As the effective YMD system is quasilocal, we assert
that in the effective gauge string the NambuGoto term in (1.2) is traded for the whole
operator expansion (OPE) running in terms of the extrinsic and intrinsic curvatures of the
worldsheet. (Complementary, the weights of the movable branch points are modified in
such a way that the integration, over the positions of the latter points, results in exactly the
same pattern of the worldsheet OPE as for the descendant of the NambuGoto term.)
Finally, let us make contact with the two alternative stringy proposals [1,5]. As for [5],
Witten argues that in continuous YM4 theory (with a finite UV cut-off ) the physical
string tension ph in the extreme large N SC limit g 2 N , g 2 O(1/N), scales as
ph g 2 N2 . On the other hand, Eq. (1.6) predicts similar large N SC O(N g 2 )-scaling
of the bare string tension 0 (where g 2 g 2 ({bk }, N) O(N 1 0 )) in the following
subclass of the YMD systems. The latter are defined by the subset of the actions (1.4)
with the coefficients gr constrained by
1 n+ (r) 2Pn p
k=1 k ,
N
(9.1)
g 2 N : [gr ]1 O 2n+4 N 2 g
where (r) > 0 (and there is at least one irrep r for which (r) = 0). According to the
structure of the (abelianized) BornInfeld action, the Wittens prescription [5] is supposed
to induce the YMD action with the local part belonging to the variety (9.1) (with possible
addition of the commutator-terms which does not alter the conclusions). In sum, our
prediction (1.6) for the string tension in the extreme large N SC limit semiquantitatively
matches with the D = 4 pattern of [5] motivated by the AdS/CFT correspondence.
As for the Polyakovs D = 4 ansatz [1], it is aimed at a stringy reformulation of the
continuous YM4 theory (1.8) in the large N WC regime. Assuming that the general pattern
of the ansatz is applicable to the SC phase as well, an indirect comparison might be
possible. In particular, the advocated by Polyakov invariance of the worldsheet action under
the extended group of the diffeomorphisms (with the singularities due to the zero Jacobian)
matches with the two key features of the gauge string: the vanishing contributions of the
e with the backtrackings and the invariance of w[M]
e under the group of the
worldsheets M
area-preserving diffeomorphisms specified in the Section 7.
Acknowledgements
This project was started when the author was the NATO/NSERC Fellow at University of
British Columbia, and I would like to thank all the staff and especially Gordon Semenoff
for hospitality.
709
Appendix A. Reconstruction of 4n+ ({R })

To derive the S(4n+ ) operator 4n+ ({R }) of Eq. (3.3), we first rewrite the characters
of the original expression (2.5) in terms of certain S(4n ) operator D(4n (R ))

(A.1)
R (U U U+ U+ ) = Trn D 4n (R ) D2 U U+ ,
where D2 ({U U+ }) is the D = 2 option of (3.4). In the |I4n() i-basis (3.22),
D(4n (R )) reads explicitly
X
S(n )
R ( )
n !
2n+1
3n+1
l1(3n+1) ln(4n) l1n+1 ln2n ln+1
l2n3n l2n+1
l3n4n
(A.2)
which can be represented in the concise form of the inner-product

4n (R ) = 4n Pe4n (R ),
e4n (R) = 1 3 CR ,
P
[n]
(A.3)
where 4n is defined by Eq. (3.20). Each of the four (ordered) S(n )-operators in
e4n (R ) are postulated to act on the corresponding
the outer product composition of P
four (ordered) S(n )-subspaces (3.22) of the space |I4n() i. Finally, in the S(4n+ )
basis (3.17), one evidently obtains 4n+ ({R }) = {} 4n (R ) that matches with
Eq. (3.19) modulo a slight deviation of Pe4n (R ) from P4n (R ) (defined by
Eq. (3.23)).
The possibility to substitute in Eq. (3.3) the operator Pe4n by P4n is ensured by the
following feature of the pattern (A.1). Owing to the basic commutativity [D( ), U n ] = 0,
S(n), any of the unity operators 1 [n] (the operator Pe4n is composed of) can
be substituted by the CR -factor properly weighted according to the multiplication rule
(CR )2 = CR /dR . Remark also that in Eq. (A.3) the relative order of the factors 4n and
Pe4n is immaterial since [4n , 4k=1 k ] = 0, k S(n).
Next, let us prove that, in the dual representation (3.24) of (2.5), the substitution (3.27)
(of D
=1 2n by its square) does not change the character (3.5). The latter symmetry
can be traced back to the fact that the multiple integral (2.5), represented by the S(4n+ )
element (3.24), is a real-valued function. To take advantage of this fact, observe that 2m =
(2n 2n ), m = 2n, is the operator which represents the complex conjugation of the
n ] entering (2.5):
e
characters R (U ) = Tr4n [D(4n )U

n
n
e
e ,
= Tr4n D 4n 2m U
(A.4)
Tr4n D(4n ) U
e is introduced in Eq. (3.28).
where 4n 4n (R ) is defined in Eq. (A.3), and U
As the master-integral (2.5) is invariant under the simultaneous transformation (A.4) of all
the involved characters while 2m = D
=1 2n , we arrive at the required invariance
under (3.27).
As for Eq. (A.4), it can be deduced by linearity from the following more elementary
identity. To formulate the latter, let us first introduce the representation of the basic traces
(the characters (A.4) are composed of)

n
1
n
e
e ,
4n U
= Tr4n D 4n
(A.5)
tr (U U U+ U+ )n Tr4n D(4n 4n )U
710
where 4n = (cn 1 3
[n] ) in the S(4n) basis (3.22), and cn is the n-cycle permutation. Then,
the required identity reads (with m = 2n)

n
n
e
e .
= Tr4n D 4n 4n 2m U
(A.6)
Tr4n D(4n 4n ) U
For its justification, we first introduce the tensor representation of the relevant complex
conjugation: Trn [D( ) nk=1 Uk ] = Trn [D( 1 ) nk=1 Uk+ ] where the orderings of the Uk
factors in the l.h.s. is the same as that of Uk+ in the r.h. side. The key-observation is that,
n
e
, the substitution Uk Uk+ can be performed as the conjugation
when nk=1 Uk+ = U
n
n
1
1
e
e
: 2m
U
2m = V
where
with respect to (2n 2n ) = 2m = 2m
+
+
e
V = U U U U . This is because, owing to Eq. (3.16), 2n interchanges (either
upper or lower) indices between the U n and (U + )n blocks of U n (U + )n . As in

1
(while 4n commutes with both 2m and 4n ),
Eq. (A.5) 4n can be substituted by 4n
1
, n = 2m. The last identity,

all what remains to be proved is that 4n 2m = 2m 4n = 4n
owing to the specific patterns (3.16) and (3.20) of 2n and 4n , directly follows from its
reduced m = 2 variant. This completes the justification of (A.6).
Finally, let us sketch the proof of the basis formula (3.31). To make sure that the
()
contraction of the indices is the same in the both sides of (3.31), the following
decomposition of the index-structure is helpful. Building on the prescription (3.6), we
rewrite the S(n ) operators D(() ) in the form
{j n }
{j n }...{j n }
(A.7)
D () {i n } D () {i n }...{i n } , , . . . , 6= ,
where each of the D 1 n -subsets of the indices (6= ) acts on the associated S(n )
subspace (3.32) of the enveloping space S(n+ ). Let consider any n -subset, associated
()
. A direct inspection then reveals that, in both
to () , as the composite block-index j
()
()
sides of Eq. (3.31), for a given , the four block-indices j , j (corresponding to
either = or = ) are c4 -cyclically contracted according to the pattern (3.19), (3.20)
of .
As for the ordering of the inner -products in the r.h.s. of (3.31), it should be deduced
from the particular ordering of the |i ()i-blocks composed into the elementary subspace
(3.21) used to define the D(D 1)/2 operators 4n . To justify the prescription stated in
Section 3, one is to combine the above analysis with the specified pattern of the ordering
in the elementary D = 2 case (3.29).
Appendix B. Tensor representation vs. regular one

Let us first derive the identity (2.8). Actually, this equation is nothing but the
transformation of the trace of the tensor D( )-representation (2.7) into that of the canonical
regular representation XREG [19,20] (both associated to a given S(n)-algebra). Recall that
XREG is defined [19,20] on the vector space 2 = {i ; i S(n)} by the homomorphism
S(n) Mi End(2) of the S(n)-group into the group of the endomorphisms of
2: i j = k 1kij (M(i ))kj (Mi )kj = 1kij . Here i j = q , and 1kij = 1 or 0
711
depending on whether q = k or q 6= k. Defined in this way, the matrices Mi satisfy the

same relations Mi Mj = Mq as the original S(n) group-elements i .
To make contact with the tensor representation (2.7), recall [19,20] first that
X
dR R ( ) = n! n ( ),
REG (PR ) = dR R ( ),
(B.1)
REG ( ) =
RYn
where tr[M( )] REG ( ) and PR = dR CR . As for n ( ), on the S(n)-group it reduces to

the standard Kronecker -function: n ( ) = [, 1 [n] ] (with 1 [n] being the trivial unitypermutation of S(n)). By linearity, it is then extended to the S(n)-algebra. Finally, rewriting
the V = 1 reduction of the second Frobenius formula (see, e.g., [22,29]) in terms of
REG (CR )
X
X
R ( )R (V )|V =1 =
dim R REG (CR ),
(B.2)
Trn [D( )] =
RYn(N)
RYn(N)
(1)
and employing the definition (3.9) of n (where the sum runs over the chiral U (N) irreps
(N)
R), we arrive at the basic identity (2.8). (Omitting the Pn projector, the latter identity
was discussed in [26].)
In conclusion, let us briefly sketch the derivation of Eqs. (4.1), (4.4). As for (4.1), in
its l.h.s. we use first (4.3) to substitute dR (n! dim R/dR )m by R ((N n n )m ). Replacing
P
CR = S(n) R ( 1 )/n!, we then combine the two resulting characters into one

1 X X
dR R (N n n )m Pn(N) 1 D( ).
n!
(B.3)
S(n) RYn
As [n , ] = 0, S(n), for the derivation of (B.3) we used the identity R ( ) =

R ( )R ( )/dR , S(n), if [, ] = 0 (which directly follows from the possibility
[19,20] to expand any such , in the center of S(n), in terms of the Young projectors =
P
(N)
RYn R PR ). We have used also that R (Pn ) = R ( ) or 0 depending on whether
P
or not R Yn(N) . Applying the completeness condition S(n) n ( 1 ) D( ) = D()
(where is any element of the S(n)-algebra) we arrive at the announced result (4.1).
Concerning Eq. (4.4), first one is to represent (see, e.g. [15,16]): C2 (R) = (nN +
2R (Tb2(n) )/dr n2 /N), R Yn , which is a particular case of the general relation (E.1).
(n)
(n)
Expanding R (Tb2 )/dr (where Tb2 is defined by Eq. (4.5)) into the preexponent, the
derivation of (4.4) is given by a minor modification of the steps developed in the context
of (4.1).
Appendix C. Reintroducing the branch points

e }) of TEK (resulting after the
Given the generalized covering spaces (GCSs) M({
deformation (6.3) of (4.8)), the generalized branched covering spaces (GBCSs) of TEK
e })
(encoded in the full expression (6.2)) can be reconstructed reintroducing onto M({
the relevant branch points (BPs). Alternatively, the GBCSs of TEK are reinterpreted as the
GCSs of TEK {qs } where the deleted (from TEK ) set of points {qs } is associated to the
712

(k
corresponding BP permutations {p } , in compliance with the homomorphism (6.6).

Then, a particular GBCS of TEK is reproduced from the associated GBC (of TEK {qs })
closing the temporarily deleted points { 1 (qs )}. At a given qs , the closure [18,22] is
performed as the mapping of the n(s) -set 1 (qs )
= (s) onto set of points matching
with the number of cycles in the cyclic decomposition of the associated to qs permutation
(6.6).
To make the above procedure well-defined, I propose to view TEK as the homotopy
0
with the same fundamental group:
retract of a less singular 2d cell-complex TEK
0
1 (TEK ) = 1 (TEK ). As the proposed retraction preserves the basic homomorphism (6.6),
the GBCSs of TEK can be consistently treated as the limiting case of the corresponding
0 .
GBCSs of TEK
First, let us thicken each -link into an infinitesimally thing cylinder Z = 6= E
shared by the D 1 corresponding 2-tora E . (To match with the canonical construction
[18,22] of the branch points, (for a given ) all E , 6= , can be always adjusted to
have the same orientation when restricted to Z .) Complementary, the intersection Z + =
Z = E of the D different cylinders Z (or, equivalently, of all the D(D 1)/2
-tora E ) is thicken into an infinitesimal disc Z + . (Again, all the cylinders Z can be
adjusted to have the same orientation when restricted to Z + .)
Choose a base-point p 6= qs , s, (common for all the equivalence classes of the paths
represented in the constraint (6.5)) in the interior Z+ of Z + . We require that the location
()
of the set {qs } = {qk } complies with

()
qk E (Z Z ) ,
qk (Z Z + ),
()
(+)
qk+ Z+ ,
(C.1)
where Z is the interior of the closed space Z , {}, {}, +. Also, the introduced in
0 ) are supposed to satisfy:
Section 6 closed branch cuts $ (inherited from the GCS of TEK
()
$ Z , while {qk } $ = 0, , k .
e }) of T 0 , one is to consider the
To reintroduce the branch points onto the GCSs M({
EK
0 .
additional (to $ ) branch cuts $(s) outgoing from the corresponding points qs of TEK
(Combining all the cuts together, one is supposed to arrive at a network = {$(s), $ }
whose global consistency is ensured by the n+ -constraint (6.2).) To implement $(s) , we
e }) along the 1 -image of $(s) (starting at qs and terminating either
cut the GCS M({
at some other point qk or, possibly, at an auxiliary vertex of the network ). Denote (sk )
e }))
and (sk ) (with k = 1, . . . , n(s) ) the resulting n(s) boundaries of the sheets (of M({
0
respectively on the left and on the right sides of $(s) . To obtain the GCS of TEK {qs }
corresponding to the homomorphism (6.6), one is to perform the pairwise identifications
of these boundaries according to the developed prescription (5.9). Matching with (6.2),
(k )
one simply substitutes in (5.9): s so that s is either {p } or depending on
the image of the homomorphism (6.6) assigned to a given branch point qs constrained by
(C.1).
713
Appendix D. Counting the generalized coverings

e in (6.2) are parametrized by the equivalence
The proof, that inequivalent spaces M
classes (6.7) of the homomorphisms (6.6), appropriately generalizes the analogous proof
[18,30] for the canonical coverings employed in [15,16,22]. First, one is to demonstrate that
e entering (6.2) are uniquely parametrized by certain
the relevant inequivalent spaces M
0 {q }|p) subgroups. Then, one proves the
S(n ) conjugacy classes of the 1 (TEK
s
one-to-one correspondence between the latter classes of the subgroups and the equivalence
classes (6.7) of the homomorphisms (6.6). Let us outline the above proofs with the
emphasis on the subtleties novel compared to [18,30].
To begin with, we observe that (by the same token [30] as in the canonical case) any
0 {q }|p) can be viewed as the image
subgroup of 1 (TEK
s
1

e (qs ) p , p 1 (p)
(D.1)
1 M
= n+ ,
e T 0 , reconstructed in Section 6.1 and Appeninduced by the -mapping (1.11): M
EK
e yields some branched covering
dix C. The consistency of (D.1) implies that a given M
(with n+ -sheets) above the single site s of TEK which is regularized, s Z+ (see
0 so that p Z . (In conEq. (C.1) of Appendix C), into an infinitesimal disc Z+ of TEK
+
tradistinction to the canonical case, the point p is not allowed to leave Z+ .) Perform a shift
of the base-point p 1 (p), located on a j th sheet of the branched covering of Z+ , to
e { 1 (qs )}. It induces [30]
p 0 1 (p) on some other kth sheet along the path M
0
the associated conjugation of the 1 (TEK {qs }|p) elements (and, hence, its subgroups)
0 {q }|p) of the path .
with respect to the element given by the image () 1 (TEK
s
Next, one notes that (similarly to the canonical construction [30]) the shift acts on
(n )
the n+ -set 1 (p)
= n+ as the simple S(n+ )-transposition (()) T2 + of the
two corresponding entries: j k, j, k, = 1, . . . , n+ . As a result, induces the S(n+ )
conjugation (6.7) of (6.6) with = (()) that can be visualized as the permutation
e . The latter does not necessarily leaves intact the
of the two associated sheets of M
e . The cuttinggluing technique of Section 6.1 implies that interchanges
topology of M
of the sheets encoded in the conjugations with S(n+ )/ S(n ) (i.e., between the
0 ) should be excluded when collecting
n -sheet coverings of different -tora E of TEK
e .
an equivalence class of maps (6.7) corresponding to a given topology of M
0 {q }|p) subgroups induced
Consider the S(n ) conjugacy classes of the 1 (TEK
s
by the combination of the elements () which are the images of those of the shifts that
0 . Taking into
connect the sheets within the n -sheet covering of a given torus E of TEK
account the above discussion, one readily modifies the canonical construction [30] to prove
that the considered classes of the subgroups are indeed in the 1-to-1 correspondence with
e { 1 (qs )} of T 0 {qs } (and, therefore,
the inequivalent generalized covering spaces M
EK
0
e
with inequivalent GBCSs M of TEK ). In particular, the cuttinggluing rules of Section 6.1
e to the corresponding classes of the 1 (T 0 {qs }|p)
ensure that the mapping from M
EK
subgroups is onto.
Finally, let us reveal the second 1-to-1 correspondence of the former classes of the
subgroups with the equivalence classes (6.7) of the homomorphisms (6.6). From above, it
714
0 {q }|p) subgroups can be related by the

is clear that the -images (6.6) of these 1 (TEK
s
S(n ) permutations corresponding to the interchanges of the sheets separately within
0 .
each of the D(D 1)/2 n -sheet coverings associated to particular E -tora of TEK
As a result, akin to the D = 2 case [18,30], the latter permutations are represented by the
required conjugations (6.7) with S(n ). Conversely, equivalent homomorphisms
0 {q }|p) subgroups (according to
determine equivalent S(n ) classes of the 1 (TEK
s
the discussed above construction of = (())). Summarizing, it completes the proof of
the two asserted correspondences.
Appendix E. The higher Casimirs actions

To derive the generalized form (4.6), (4.7) of Qn ( ) in (4.4), first one is to begin with
the (SchurWeyl) duality [17,22] between the Casimir eigenvalues
Cq (R)
=
N q1
X
T{p} S(n)
aq (N, n, T{p} )
R (Tb{p} )
,
dR
R Yn(N) ,
(E.1)
(n)
b{p} is defined by Eq. (4.6),
T
and the symmetric group characters {R (Tb{p} )} (where Tb{p}
while 1 6 q 6 N ). The coefficients aq ( ) are defined by the formula [17,22] which
can be deduced from Eq. (4.7) via the substitution: {p} (. . . , n, N) aq (N, n, T{p} ),
s{p} ({br }, m, l) sq (T{p} , m, l), M{p} M0 (T{p} ). In particular, (keeping n O(N 0 ))
the leading term of the formal 1/N expansion of the pth order Casimir operator reads:
Cq (R) = N q1 (n + O(N 1 )), R Yn(N) , which corresponds in (E.1) to the trivial unity
permutation T{p} = 1 [n] . As for the remaining T{p} 6= 1 [n] contributions, the branch point
interpretation of the l = 0 term in expression like (4.7) is discussed in the end of Section 5
(in fact sq (T{p} , m, 0) = 0 for m > 2). To interpret a given l > 1 term, first one is to resum
the powers nm in terms of the numbers of inequivalent subdivisions of n objects into
P
two subsets containing t and (n t) objects: nm = m
t =1 fm,n n!/t!(n t)!. As a result,
similarly to [17,22] the latter contributions include (in addition to the T{p} branch point)
the extra collapsed to a point 2d subsurfaces. Being cut out, these surfaces can be viewed
as having genus g = l [t/2] and 2[t/2] holes.
Finally, to select the admissible class of the functions (defining a given model (1.1)),
one is to require that in the associated large N SC series (1.9) the factor N 22h matches
e Then, to ensure Eq. (4.7), the admissible
with the genus h of the associated worldsheet M.
polynomial pattern of the function should belong to the variety (with M Z>1 )
M
X X

2l Y

Cqk (R)
s br , {qk }, l N
,
br , N, Cq (R) =
N qk 1
M
M,{qk } l>[
(E.2)
k=1
where the (properly weighted by the 1/N factors) l > 1 terms encode the additional,
compared to those encoded in (E.1), collapsed subsurfaces connecting a few sheets
according to the same rules as for (E.1). In particular, the pattern (E.1), (E.2) ensures
715
the existence of the proper asymptotics (1.6) of ( ) defining the bare string tension
0 ({br }) (to which only linear in Cq (R) terms in (E.2) contribute).
Finally, with the help of some elementary identities [15,16] from the theory of R (Tb{p} )
characters, the generic function (E.2) results in the generic pattern (4.6) of the generalized
operator Qn ( ). One can argue [17] that the required pattern (4.7) of the N -dependence
takes place provided, in the standard F -representation (1.4), the 1/N scaling of the
coefficients gr is constrained to yield the conventional t Hooft pattern of the 1/N
topological expansion.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
A. Polyakov, Nucl. Phys. Proc. Suppl. 68 (1998) 1.

J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231; Phys. Rev. Lett. 80 (1998) 4859.
S. Gubster, I. Klebanov, A. Polyakov, Phys. Lett. B 428 (1998) 105.
K. Wilson, Phys. Rev. D 10 (1974) 2445.
I. Kostov, Nucl. Phys. B 265 (1986) 223; Nucl. Phys. B 415 (1993) 29.
V. Kazakov, Sov. Phys. JETP 58 (1983) 1096.
K. O Brien, J. Zuber, Nucl. Phys. B 253 (1985) 621.
G. t Hooft, Nucl. Phys. B 72 (1974) 461.
T. Eguchi, H. Kawai, Phys. Rev. Lett. 48 (1982) 1063.
A. Gonzalez-Arroyo, M. Okawa, Phys. Rev. D 27 (1983) 2397.
B. Rusakov, Mod. Phys. Lett. A 5 (1990) 693.
M. Douglas, V. Kazakov, Phys. Lett. B 319 (1993) 219.
D. Gross, Nucl. Phys. B 400 (1993) 161.
D. Gross, W. Taylor, Nucl. Phys. B 400 (1993) 181; Nucl. Phys. B 403 (1993) 395.
O. Ganor, J. Sonnenschein, S. Yankielowicz, Nucl. Phys. B 434 (1995) 139.
C. Ezell, Trans. Am. Math. Soc. 243 (1978).
Wu-Ki Tung, Group Symmetry in Physics, WC, 1985.
W. Fulton, J. Harris, Representation Theory, Springer-Verlag, 1991.
J. Drouffe, J. Zuber, Phys. Rep. 102 (1983) 1.
S. Cordas, G. Moore, S. Ramgoolam, Lectures on 2D YangMills theory, equivariant
cohomology and topological field theories, hep-th/9411210.
E. Witten, Commun. Math. Phys. 141 (1991) 153.
A. Migdal, Sov. Phys. JETP 42 (1975) 413.
M. Douglas, K. Li, M. Staudacher, Nucl. Phys. B 420 (1994) 118.
S. Ramgoolam, Int. J. Mod. Phys. A 11 (1996) 3885.
M. Golubitsky, V. Guilemin, Stable Mappings and Their Singularities, Springer-Verlag, 1973.
A. Dubin, hep-th/9910088, submitted to Phys. Rev. D.
A. Dubin, hep-th/9910264, submitted to Nucl. Phys. B.
W. Massey, A Basic Course in Algebraic Topology, Springer-Verlag, 1991.
J. Baez, W. Taylor, Nucl. Phys. B 426 (1994) 53.

Dimensional reduction, magnetic flux strings,

and domain walls
C.D. Fosco a,1 , A. Lpez a,1 , F.A. Schaposnik b,2
a Centro Atmico Bariloche, 8400 Bariloche, Argentina
b Departamento de Fsica, Universidad Nacional de La Plata C.C. 67, 1900 La Plata, Argentina
Received 22 December 1999; accepted 22 May 2000
Abstract
We study some consequences of dimensionally reducing systems with massless fermions and
Abelian gauge fields from 3 + 1 to 2 + 1 dimensions. We first consider fermions in the presence of an
external Abelian gauge field. In the reduced theory, obtained by compactifying one of the coordinates
a la KaluzaKlein, magnetic flux strings are mapped into domain wall defects. Fermionic zero
modes, localized around the flux strings of the 3 + 1 dimensional theory, become also zero modes in
the reduced theory, via the Callan and Harvey mechanism, and are concentrated around the domain
wall defects. We also study a dynamical model: massless QED4 , with fermions confined to a plane,
deriving the effective action that describes the planar system. 2000 Elsevier Science B.V. All
rights reserved.
PACS: 11.10.Kk; 11.27.+d; 11.10.-z
Keywords: Domain wall; Dimensional reduction
1. Introduction
Topological defects in quantum field theory (an excellent reference on the subject is
[1]) play an important role in the description of many interesting phenomena, both in
high energy and condensed matter physics applications. Being part of the non-perturbative
spectrum, the understanding of their properties demands the study of topological classes
of the field configuration space, and the realization of the associated topological invariants
in the model under consideration. For some systems containing fermions, fermionic zero
modes arise whenever the topological charge is different from zero. This effect has many
important consequences for the study of the low momentum effective theory. Moreover,
1 CONICET.
2 Investigador CICBA, Argentina.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 1 5 - 1
C.D. Fosco et al. / Nuclear Physics B 582 (2000) 716732
717
since they are gapless excitations, they will strongly affect the response functions of the
model [2].
In this paper we are concerned with the relationship between topological defects and
zero modes in theories related by dimensional reduction. In particular, for the concrete
example of fermions in 3 + 1 dimensions coupled to an Abelian gauge field, we shall derive
the low-momentum effective action for the dimensionally reduced theory. Being originally
a massless theory, the momentum scale will be fixed by the inverse of the compactification
length. Also, as it will be explained, quantum corrections coming from all the Kaluza
Klein modes have to be included, to be consistent with large gauge invariance.
We will first consider the case of massless Dirac fermions in the presence of an external
Abelian field. We then go on to consider a dynamical model: massless QED in 3 + 1
dimensions. The reduction is here introduced for the fermions, assumed to be confined to
a planar spatial region. Being the only sources of the gauge field, this induces a reduction
also for the gauge field, regardless of the fact that it is apriori unconfined. For this system
we derive the effective theory, showing that there is room for the existence of stripe
defects, i.e., localized zero modes that appear due to changes in the sign of the fermionic
mass. This is just a manifestation of the well known Callan and Harvey mechanism [3].
When there is translation invariance along one of the spatial directions, the dynamics of
the defects is governed by a sine-Gordon-like action. Moreover, the soliton charge is linked
to the number of zero modes.
The organization of this paper is as follows: in Section 2, we consider fermions in the
presence of an external (non-dynamical) gauge field, in 3 + 1 dimensions. To take into
account quantum effects, Section 3 deals with the evaluation of the effective action using a
derivative expansion approximation for the fermionic determinant. In Section 4, we derive
the low-momentum effective theory for the interacting case of massless QED4 . Finally,
Section 5 is devoted to the conclusions.
2. Non-interacting case
By non-interacting we refer to a model containing a fermionic field in the presence
of an external (i.e., non-dynamical) gauge field A . The Euclidean action for the present
case is simply defined by
Z
/
(1)
SF = d4 x D
where
D
/ = D ,
D = + ieA .
(2)
We shall adopt the convention that indices from the beginning of the Greek alphabet always
run over the values 0, 1, 2, 3 while those from the end belong to the set 0, 1, 2 and are
reserved for the coordinates of the reduced spacetime. The third spatial coordinate, x3 ,
is compactified: 0 6 x3 6 L, and shall be sometimes also denoted s, to emphasize its
special role, as opposed to the uncompactified coordinates, x .
718
We shall first be concerned with configurations corresponding to magnetic flux strings in

3 + 1 dimensions, with the strings entirely contained in the reduced hyperplane x3 = 0. 3
These configurations are here considered regardless of the mechanism which may create
them. To have these magnetic field structures, special gauge field configurations are
required. Beforehand, we will use part of the gauge freedom to restrict the external gauge
field configurations to the ones satisfying the condition:
3 A3 = 0,
(3)
which can always be fulfilled, by performing a regular gauge transformation connected to

the identity. With this choice, A3 (x, s) A 3 (x),
1
A 3 (x)
L
ZL
ds As (x, s).
(4)
Besides considering static magnetic flux strings, we would also like to include a small
gauge field configuration in order to find the effective action. The strings, being topological
objects, should be treated non-perturbatively, while the small part is assumed to be
intrinsically perturbative.
In 3 + 1 dimensions, the magnetic field components Bi (i = 1, 2, 3) are, as usual, defined
by Bi = ij k j Ak . As the flux strings lay on the x3 = 0 plane, the third normal component
of the magnetic field vanishes: B3 = 0. Moreover, we can generate the two non-vanishing
components B1 , B2 using a gauge field such that A1 = A2 = 0, and A 3 6= 0. With this
choice,
Bi (x) = ij j A 3 (x).
(5)
It is evident that, from the point of view of the reduced, planar theory, A 3 is a scalar field
under transformations mapping the plane onto itself. It is convenient to use the definition
A 3 (x) = (x), with x (x0 , x1 , x2 ), to make explicit the fact that A 3 behaves like a scalar.
In the case of a flux string along the x1 axis, the magnetic field is:
B1 = (x3 )(x2 ),
(6)
where is a constant. 4
We need to write now a gauge field configuration leading to the magnetic field
of (6). There are of course many possibilities but, interestingly enough, if we want to
dimensionally reduce the theory, this freedom is substantially reduced. The reason is that
we do not want a gauge field configuration having a nontrivial x3 dependence far away
from the x3 = 0 hyperplane. The natural choice is then a gauge field strongly concentrated
on x3 = 0, and this determines the gauge field configuration to be of the form:
(7)
A3 (x) = (x3 ) sign(x2 ),
2
3 This is rather different to the more standard situation of a flux tube pointing in the x direction, and thus
3
corresponding to a magnetic vortex in 2 + 1 dimensions.
4 This -like configuration is, of course, an idealization. Realistic configurations will be, although highly
concentrated, regular functions.
719
which implies, for the 2 + 1 dimensional scalar field
sign(x2 ).
(8)
2L
The constant is, of course, related to the (quantized) total flux of the string, thus =
2k/e, where k is an integer, counting the number of elementary fluxes.
We note the emergence of a step-like configuration for the scalar field. This scalar
field, as we shall see next, plays the role of the mass for 2 + 1 dimensional fermions.
Hence, invoking the Callan and Harvey mechanism [3], we see that there are zero modes
concentrated on the walls, which indeed correspond to the dimensionally reduced version
of the zero modes around the flux strings. It is worth noting that the scalar field profile is
discontinuous because of the assumed -like profile for the magnetic field. It is convenient
to use instead a regular profile,
(x) =
h (x2 ),
2L
where h (x2 ) is a smoothed version of the step function, for example,
(x) =
h (x2 ) = tanh(x2 /),
(9)
(10)
where is a length which essentially measures the width of the stripe.

To discuss issues related to the fermionic zero modes, like the relationship between
the number of modes in 3 + 1 and 2 + 1 dimensions, a more detailed study of the
fermionic action (1) is required. The full gauge field configuration (perturbative plus nonperturbative) is assumed to be s-independent. This simplifying assumption allows us to
take advantage of the fact that the gauge field A 3 is independent of s = x3 to Fourier
expand the fermionic fields:
+
1 X in x3
e
n (x),
(x, x3 ) =
L n=
+
1 X in x3
n (x),
e
(x,
x3 ) =
L n=
(11)
with n = 2n/L . We then obtain for SF a representation as a series of decoupled 2 + 1

dimensional actions
SF =
+
X
SF,n .
(12)
n=
The explicit form for the action SF,n corresponding to each Fourier mode is:
Z

SF,n = d3 x n (x) D + i3 (n + e) n (x).
We assume the Dirac matrices are in the representation:

0 3
0 1
,
1 =
,
0 =
3 0
1 0
(13)
720

2 =
0 2
2 0

,
3 =
0 iI
iI 0

,
(14)
where j , with j = 1, 2, 3, denote the usual Pauli matrices, and I is the 2 2 identity
matrix. Writing each four-component fermionic field n in terms of two 2-component
(a)
fermions n (a = 1, 2), we see that
Z

d + n + e)n(1) + n(2) (x)(/
d n e)n(2)(x) .
(15)
SF,n = d3 x n(1) (/
We have introduced the notation d/ to refer to the Dirac operator in 2 + 1 dimensions, acting
on 2-component fermions. More explicitly,
d/ = D
(16)
with 0 3 , and j (j = 1, 2) again denoting the usual Pauli matrices. The covariant
derivative depends on a gauge field which, in our approximation, is a function of x only.
It is important to realize that all the fermionic modes should be kept if large gauge
invariance is to be preserved, since large gauge invariance amounts to shifts in , and this
field is directly coupled to the fermionic field.
It is now easy to study the Callan and Harvey fermionic zero modes of the reduced
theory, assuming that in the original model in 3 + 1 dimensions the background produced
a non-vanishing, quantized, magnetic flux. We first realize that the decomposition (12)
shows that there is an infinite number of 2 + 1 dimensional Dirac fermions, each Fourier
mode consisting of two flavours, distinguished by the sign of their mass terms. A Callan
and Harvey zero mode for the Fourier component n shall appear whenever n + e(x)
crosses zero more or less sharply. Thus, the configuration (8) (in its smooth version (9))
shows that there is at least a zero mode for the n = 0 component, but, if there is a flux
string with charge k, the scalar field profile will cross as many values of n , implying that
there are, indeed, k zero modes in the reduced theory. To see this, we take into account the
expression (9) for the domain wall configuration. The equation determining the locii of the
zero modes may be written as
h (x2 ) =
2
integer,
k
(17)
where k is the (fixed) number of elementary fluxes. In Fig. 1 we show, as an example, the
case of a stripe configuration h with k = 9. The horizontal lines correspond to different
ns, and a zero mode is produced whenever one of these lines intersects the curve.
It is interesting to realize that a proper understanding of the relation between the height
of the stripe and the number of zero modes requires the use of the smoothed version of
the steplike scalar field configuration. This should be evident from Fig. 1, if one imagines
the curve to be deformed to a steplike configuration.
We remark that the zero modes produced in this way are massless Dirac fermions in 1+1
dimensions, because each flavour produces a given chirality, and the signs are opposite.
721
Fig. 1. A stripe configuration of charge 9.
3. Derivative expansion for the fermionic determinant

We shall now consider the evaluation of the fermionic determinant, as a functional of a
and , in a derivative expansion approximation. The field a denotes the perturbative part
of the gauge field.
To avoid any risk of confusion, we shall, from now on, use two-component fermions
only. The effective action F [, a ] is defined as the result of functionally integrating
out the fermionic fields. Using the Fourier decomposition for the x3 = s coordinate, we see
that
Z +
Y
D n(1) Dn(1) D n(2) Dn(2)
eF [,a ] =
n=
exp
+
X
d
n(1) (/
+ n + e)n(1) + n(2) (/
d
n e)n(2)
)
(18)
n=
Thus F may be written in terms of fermionic determinants in 2 + 1 dimensions:

F [, a ] =
=
+
X

n=
+
X
d n e)
ln det(/
d + n + e) + ln det(/
2

Tr ln /
d + (n + e)2 .
(19)
n=
Large gauge transformations wind up a number of times around the periodic coordinate, s. Then they correspond to constant shifts in the scalar field :
2n
,
(20)
(x) (x) +
eL
where n is an integer. It is now clear, from the general expression (19), that all the Fourier
modes must be kept, if invariance under (20) is to be maintained.
722
In the derivative expansion technique, which we shall use in order to evaluate F , the
leading, zero derivative term, can usually be treated exactly. This term may be thought of
as depending basically on the constant component of the fields. The following terms in the
derivative expansion depend of course also on the fluctuating part, which is assumed to be
small in comparison with the constant part.
The piece of the effective action depending on the constant part is usually regarded as
an effective potential. In our case, there is no point in keeping a constant part for a
since, if present, it could be gauged away (because the planar system is assumed to have
a trivial topology). From (20), it is evident that to keep a constant component for is, in
turn, crucial. We thus decompose into two pieces,
(x) = 0 + (x),
(21)
=
where the constant 0 = h(x)i denotes the (x ) spacetime average of (x), and (x)
(x) 0 . Then we factorize a constant field determinant,
a ]
F [, a ] = F [0 , 0] + 1F [,
where
F [0 , 0]
d3 x Veff () =
+
X
(22)
2

Tr ln /
+ (n + e0 )2
(23)
n=
and
1 [,
a ] =
+
X

Tr ln 1 + e(/
+ n + e0 )1 (i/
a + )
n=
+
X

Tr ln 1 + e(/
n e0 )1 (i/
a )
.
(24)
n=
We have introduced the notation Veff under the integral symbol in (23), to emphasize the
property that it will play the role of an effective potential for . At the end of the evaluation,
we shall follow the common practice of replacing 0 by in Veff . This approximation is
justified within the derivative expansion technique, as long as the hypothesis leading to that
expansion are valid [4,5].
Large gauge invariance means that this potential is a periodic function of . The constant
field determinant is evaluated by taking the functional and Dirac traces in momentum
space:
+
X
F [0 , 0] =
2

Tr ln /
+ (n + e0 )2
n=
= 2
d3 x
+ Z
X
n=

d3 p 2
ln p + (n + e0 )2 .
3
(2)
(25)
To evaluate the sum over n, we follow techniques of standard application in finite

temperature quantum field theory [6]. We first take advantage of the fact that the series
runs from n = to n = +, to write
+ Z
X
d3 p
(2)3
n=

ln p2 + (n + e0 )2 + ln p2 + (n e0 )2 ,
F [0 , 0] =
d3 x
723
(26)
which is easily rearranged as

Z
+ Z
X
d3 p
(2)3
n=
2

ln n + (p + ie0 )2 + ln n2 + (p ie0 )2 .
F [0 , 0] =
d x
(27)
Each one of the terms in (27) may be evaluated by translating known results about the free
energy for a system of free bosons [6], what yields,
Z
Z

d3 p
3
ln 1 eL(p+ie0 ) + ln 1 eL(pie0 ,
(28)
F [0 , 0] = d x
3
(2)
where we have ignored the zero point contribution, since it is a -independent constant.
We shall, however, fix the ambiguity in the renormalization of this infinite constant by
demanding that the Veff vanishes at its minima.
The integral over p can be performed, and the result can be presented as a series,
Z
X

4
1
cos(neL0 ) (1)n ,
(29)
F [0 , 0] = d3 x 2 3
4
L n= n
or:
Z
d x Veff =
3
X
4
14 2
1
d x 2 3
cos(neL0 )
.
L n= n4
5L3
3
(30)
a].
Let us next consider the remaining part of the effective action, denoted 1F [,
Besides a dependence on 0 , it will also depend on (the fluctuating part of ) and a .
To be consistent with the derivative expansion, and also because both and a are
proportional to the same coupling constant e, the expansion must treat both fields as a
single entity. 5 To this end, we have found illuminating to define a new vector field ,
which summarizes the information on the a and configurations, through the relations
= ea ,
= e(x).
(31)
Of course, this involves the assumption that the Lorentz gauge a = 0 has been adopted
for the gauge field. The number of independent components for is three, since this field
is, in principle, not constrained by any gauge invariance requirement, and this matches
the number of components for the transverse a (two) plus the scalar fields (one). Using
a) may
trivial properties of the -matrices in 2 + 1 dimensions, we observe that 1F (,
be written as
5 Indeed, they originate in different components of the same 3 + 1 dimensional gauge field A .
724
1F [,
a ] =
X

/ ) + .
Tr ln 1 + (/
+ n + e0 )1 (/
(32)
n=
Because of its explicit Lorentz covariance, this expression could be regarded as a

consistency check for the procedure of introducing as a Lorentz vector.
Taking into account the fact that (32) is explicitly even under parity transformations, we
may write

(33)
1F [, a ] = 1even F
where the even label refers to the behaviour under parity transformations, and F =
a . The particular dependence of 1F on its arguments is due to the fact that:

1
(34)
= e 2 F .
We calculate the leading term in a derivative expansion for 1F , by noting that,

expanding up to second order in derivatives the 2 + 1 dimensional object

+ n + e0 )1 (/
/ ) ,
(35)
[ ] = Tr ln 1 + (/
we obtain
Z

e2
d3 x ( F )( F ) .
[ ] =
24|n + e0 |
Then we easily see that

1F [, a ] = [ ] + ( )
Z
X

1
e2
( )2 + (F )2 .
d3 x
=
12
|n + e0 |
n=
(36)
(37)
Namely, it corresponds to a local Maxwell term for a and a local kinetic term for .
In summary, the effective action induced by the integration of the fermionic degrees of
freedom is, to second order in a derivative expansion:

Z
X
4
14 2
1
cos(neL)
F [, a ] = d3 x
2 L3 n= n4
5L3
Z
X

1
e2
( )2 + (F )2 ,
(38)
d3 x
+
12
|n + e|
n=
where we have replaced 0 by in the contributions from the effective potential
(30) and the subleading term in the derivative expansion of the fermionic determinant
(37). As already mentioned, this procedure is justified within the derivative expansion
approximation.
4. Massless QED4
In previous sections, A was just a background for the fermion fields. Here, we shall
incorporate a Maxwell action so that A becomes dynamical. We then start from QED4
725
with the fermionic field constrained to the region 0 6 x3 6 L. The Abelian gauge field
is, in principle, defined on an unconstrained region. However, its dynamics is determined
entirely by the sources, which are, indeed, confined to 0 6 x3 6 L. When the momenta
involved in the processes are small in comparison with the scale L1 , an effective theory
can be obtained for the relevant degrees of freedom, which describe the dynamics on the
plane x3 = 0. This effective model contains Dirac fermions in interaction with a dynamical
Abelian gauge field, and with an (also dynamical) scalar field, in 2 + 1 dimensions. The
spectrum of this theory also contains stripes: (Dirac) fermionic zero modes localized on
linear defects.
Our starting point shall be to consider a model defined in terms of S, the action for
massless QED in four Euclidean dimensions
S = SF + SG ,
(39)
where SF and SG denote the fermionic and gauge field parts of the action, respectively.
They are given by
Z
D(x),
/
D
/ = / + ieA
/
(40)
SF = d4 x (x)
and
Z
SG =
1
d4 x F F ,
4
F = A A .
(41)
We assume that the gauge field components, A , are defined over the full four-dimensional
Euclidean spacetime, so that the integration region in (41) is unbounded.
The fermionic field is, in turn, defined on a region of small width L in the third
spatial dimension, namely: 0 6 x3 6 L, to simulate the physical situation of fermions
in a quasi planar system. There are, of course, many different choices for the boundary
conditions of the fermionic field at x3 = 0 and x3 = L, all of them compatible with current
conservation. In principle, the most natural one would be to impose the vanishing of the
normal component of the current on the borders, namely, j3 = 0 at x3 = 0, L. Although
this condition is natural, it is not an easy one to deal with from the point of view of
the calculation, because it explicitly breaks translation invariance along x3 . To avoid this
technical inconvenience, we prefer to use the simpler assumption that the fermionic field
is L-periodic in x3 . This allows us to Fourier expand in that coordinate. From the point
of view of the effective physics in the 0 6 x3 6 L region (the dimensionally reduced
theory), both choices should lead to qualitatively similar results. The important point is, as
we shall see, that the existence of a finite dimension allows for a description in terms of an
infinite number of fermionic modes in the dimensionally reduced spacetime. The precise
nature of the boundary conditions will of course affect the details of this phenomenon, like
the spacing between the fermionic modes, but not the gross features and properties of the
system.
The properties and objects we shall be concerned with can all be obtained, in principle,
from the knowledge of Z[j ; ,
], the generating functional of Greens functions
726
Z
Z[j ; ,
] =
"
DA D D exp S +
ZL
3
d x
ds j A +
+
#
(42)
with S as defined in (39). As we are interested exclusively in the phenomena localized

are
on the x3 = 0 plane, both the external source j , and the fermionic current ie
confined to the 0 6 s 6 L region.
It is evident that, from the point of view of the description of the physics on the x3 = 0
plane, the action for the gauge field yields more information than what is actually needed in
our situation, since we are not going to consider processes with sources outside the region
0 6 x3 6 L. It is possible, as we shall see, to obtain a 2 + 1 dimensional action describing
precisely the planar dynamics, and containing fermion and gauge fields with support on the
x3 = 0 plane. In a sense, we are going to integrate out gauge field modes, corresponding
to excitations lying outside the plane.
As a first step, let us see how to pass to a description where the relevant gauge field
dynamics is determined by an effective 2 + 1 dimensional gauge field action. We may
J
, the part of the action which involves the gauge field and its source, as follows:
write SG
Z
J
(43)
SG [A, J ] = SG [A] + Sgf [A] + d4 x J (x)A (x),
where we have included a gauge fixing term, Sgf [A], which, for the sake of simplicity, we
take to be of the Feynman type. 6
Z
1
Sgf = d4 x ( A)2 .
(44)
2
J stands for the full current, namely, the external source for the gauge field j plus
the fermionic current ie(x)

(x). We then integrate out the gauge field, obtaining a
non-local currentcurrent interaction:
Z
J

[A, J ] = exp Snl [J ] ,
(45)
DA exp SG
where
Z
1
(46)
d4 x d4 y J (x)K (x y)J (y)
2
with K (x y) denoting the (Feynman gauge) gauge field propagator:
Z
d4 k ik(xy)
e
.
(47)
K (x y) =
(2)4
k2
Taking now into account the fact that the current J vanishes when either x3 < 0 or x3 > L,
we see that (46) may be written more precisely as follows:
Snl [J ] =
1
Snl [J ] =
2
ZL
3
d x
Z
ds1
ZL
ds2 J (x, s1 )K (x y, s1 s2 )J (y, s2 ). (48)
d y
0
6 The following derivation can, of course, also be implemented for different gauge fixings, but the calculations
become more involved.
727
In the L 0 limit, the current may be taken to be approximately x3 independent, at least

from the effective theory point of view, since a dependence on x3 in this length scale would
correspond to a momentum component comparable to the scale L1 . Thus, we can replace
in (48) J (x, s) by its average J (x), defined by
1
J (x) =
L
ZL
ds J (x, s).
(49)
On the other hand, regarding the propagator, as 0 6 s1,2 6 L, we may use the
approximation
K (x y, s1 s2 ) ' K (x y)
(50)
where
K (x y) = K (x y, 0).
(51)
Therefore, (46) reduces to a non local currentcurrent interaction Snl in 2 +1 dimensions

Z
Z
L2
d3 x d3 y J (x)K (x y)J (y).
(52)
Snl [J ] =
2
To obtain a more explicit form for K (x y), we may write it in terms of its Fourier
representation
K (x y) =
d3 k ik(xy)
e
(2)3
Z+
dk3
.
2 k 2
(53)
Integrating over k3 , we easily see that the effective three-dimensional kernel for the
currents J (x) is
Z
d3 k ik(xy)
e
(54)
K (x y) =
(2)3
k2
while for the x3 component we have the scalar K s :
Z
d3 k ik(xy) 1
e
K s (x y) =
.
(2)3
k2
(55)
Notice that from now on k denotes the three dimensional momentum vector. K s is treated
separately, because A3 behaves in fact as a scalar field under spacetime coordinate
transformations in the x3 = 0 hyperplane.
It is important to realize that the 2 + 1 dimensional current J is conserved:
J = 0.
(56)
This follows from the fact that the current is approximately x3 -independent, in particular:
3 J3 = 0, plus the usual continuity equation J = 0. Then we may actually replace
in (54) by its transverse part:
k k
.
k2
(57)
728
Summarizing, the result of integrating out the gauge field may, in the small-L limit, be
represented in terms of an effective 2 + 1 dimensional action
Z
Z
L2
3
(x y)J (y)
d x d3 y J (x)K
Snl [J ] =
2
Z
Z
L2
3
(58)
d x d3 y Js (x)K s (x y)Js (y)
+
2
as in (54), but with instead of .
with K
This effective action Snl [J ] for the currents, can be equivalently rewritten as arising
from the integration of a gauge field a , and a scalar field , both defined in the three
dimensional spacetime x3 = 0. In other words, we can express the above action as the
result of functionally integrating out (in 2 + 1 dimensions) a gauge field a and a scalar
field , with the following action:
Z
Z
1

0
1
(x y)a (y)
d3 x d3 y a (x)K
J; a , =
Snl
2
Z
Z
1
d3 x d3 y (x)K s1 (x y)(y)
+
2
Z
Z
3 0
(59)
+ d x J (x)a (x) + d3 x Js0 (x)(x),
where J0 (x) = LJ (x). Note that the inclusion of this factor L in the definition of the
current reduces by one its mass dimensions, as it corresponds to the transition from
fermionic fields in 3 + 1 dimensions to 2 + 1 dimensions. The bosonic fields, a and ,
have the same mass dimension than their 3 + 1 dimensional counterparts, because they are
equipped with unusual, non-local kinetic terms.
Because of current conservation, we may also write a gauge invariant form for
0 [J; a , ], by introducing the longitudinal part of a into the game, obtaining:
Snl
Z
Z
1

1
0
F + d3 x J0 (x)a (x)
d3 x F
J; a , =
Snl
4
2
Z
Z
1
1
3 0

(60)
d3 x
+ d x Js (x)(x) +
2
2
where F = a a . This gauge invariant form should, of course, be gauge fixed in
order to recover a regular action.
We have thus re-derived a known result: the electromagnetic interaction due to a
Maxwell action, if restricted to charges living on a planar section of space, can be
reproduced by introducing a non-local gauge invariant action in 2 + 1 dimensions, like
in [11,12]. This is of course different to the usual, KaluzaKlein like dimensional
reduction prescription and in fact is closer to an alternative prescription proposed in [14].
To understand the basic differences between the two prescriptions, let us concentrate in the
case of a scalar field in 3 + 1 dimensions with action
Z
1
(61)
d4 x
S (3+1) =
2
729
with propagator K 3+1 satisfying in coordinate space the usual Green function equation
2 K (3+1) (x) = (x)
(62)
so that in coordinate and momentum (Minkowski) space one has

4 2 i
,
(x 0 )2 (x 1 )2 (x 2 )2 (x 3 )2 i0
1
.
K (3+1)(k) =
(k0 )2 (k1 )2 (k2 )2 (k3 )2 + i0
K (3+1) (x) =
(63)
(64)
The usual KaluzaKlein dimensional reduction consists in this case in dropping k32 in
(2+1)
in the reduced space obeys the analogous to (62),
Eq. (64). Then, the propagator KKK
(2+1)
2 KKK (x) = (x)

so that the reduced scalar theory corresponds to an action
Z
1
(2+1)
d3 x .
SKK =
2
(65)
(66)
The alternative prescription which leads to the reduction presented here corresponds to
dropping x 3 in (63). To see this, note that if one puts x3 = 0, the propagator K (2+1) does
not satisfy anymore (65) but instead, as can be easily seen,
p
2 K (2+1)(x) = (x),
(67)
so that the dynamics of the reduced scalar theory is governed by the action
Z
1
1
,
(68)
d3 x
S (2+1) =
2
2
which is precisely the non-local action for the scalar field in (60). An analogous result,
coinciding again with that in (60) can be found for a vector field like a .
Let us note that in the case in which the dimensional reductions is justified by the fact
that the physical system is (quasi) planar, the alternative prescription which consists in
dropping x 3 , effectively compels the system to evolve in a plane keeping the nature of the
interaction unchanged: for example, if in 3 + 1 dimensions electrons interact via a 1/r
potential, they still interact through this potential in 2 + 1 but with the third coordinate
constrained to be zero. In contrast, the usual KaluzaKlein prescription not only reduces
the space but also changes drastically the nature of the interaction: in the example above,
the electron interaction would change from 1/r to log r. If planar electrons are supposed
to be subjected to 3 + 1 interactions even when they are compelled to move in the plane,
the alternative prescriptions then appears to be more physical than the usual KaluzaKlein
one.
The prize one pays when using the alternative prescription is the non-locality of the
reduced action. It is interesting to note that, if looked at from the point of view of
Minkowski spacetime, this non-locality is due to a branch cut singularity starting at zero
momentum. Amusingly enough, this kind of singularity is entirely analogous to the one
that appears when one considers the vacuum polarization function for massless fermions
730
in 2 + 1 dimensions. In that case, the branch cut singularity is of course due to the
possibility of pair-creating fermions on-shell for any gauge field configuration with a nonzero momentum. The situation is, however, different to the dimensional reduction of our
example, since the singularity here corresponds to modes that have been integrated out,
and are in principle out of the spectrum of physical states. Thus the model is to be regarded
as an open system, since the unitarity relations corresponding to a closed system would in
turn require the introduction of asymptotic massless fermionic excitations into the game,
like in the case of (unitary) massless QED in 2 + 1 dimensions.
The reduction of the fermionic part of the action proceeds as in Section 2, while the
evaluation of the fermionic determinant has already been presented in Section 3.

So far we have obtained the effective action for the gauge fields (60), after reducing the
original gauge action from 3 + 1 to 2 + 1 dimensions, and the one for the fermionic degrees
of freedom (Eq. (38)), after integrating out the fermions. The final result is:
"
#
Z
X
4
14 2
1
3
cos(neL)
Seff [, a ] = d x
2 L3 n= n4
5L3
Z
X

1
e2
( )2 + (F )2
d3 x
+
12
|n + e|
n=
Z
Z
1
1
1
1
3
F +
.
(69)
d x F
d3 x
+
2
4
2
2
We see that the dynamics for the field that provides an induced mass for the fermions,
, is given by a non-local interaction term, and a series of sine-Gordon like term. It is also
through a series that is a functional of that this last field couples to a .
As mentioned above, non-local effective theories for the electrodynamics of particles
moving on a plane were discussed in [1113]. In particular, a Lagrangian containing a nonlocal Maxwell term plus a coupling to matter current confined to a plane and coinciding
with the first line in Lagrangian (60) was obtained in [12] where it was proven that this
effective 2 + 1 non-local Lagrangian was equivalent to one in which the current term is
replaced by a ChernSimons term. This last Lagrangian describes precisely the bosonized
version of a 2 + 1 massless fermion theory [11] and, as shown in [13], this is not a
coincidence. Indeed, the functional approach to bosonization [15,16] for 2 + 1 free Dirac
fermions with mass m, leads to a bosonization recipe which is exact for the matter current:
1
(70)
J (x) a .
4
Concerning the bosonic action, it cannot be written explicitly, except for certain particular
limits. In particular, in the vanishing mass limit one can compute exactly the bosonic action
[13] obtaining a result which coincides with the first line in Eq. (60) (if one uses (70) to
bosonize the matter current). It is worth emphasizing here that we have not started from a
731
2 + 1 model but rather from massless QED4 , and that the reduction mechanism explained
above brings into play, besides the usual a field, also an scalar field which has a nontrivial, sine-Gordon like dynamics and it is moreover coupled to a . The necessity for
the existence of this field may be traced back of course to the fact that the compactified
dimension allows for a certain non-trivial structure in that direction, reflected in the large
gauge transformations of which the scalar field is the subject.
To finish this discussion we look at the propagation of fermions in a (a , ) background.
In a first approximation, the dynamics of these background fields could be regarded
as entirely determined by the effective action (69) only, and as a next step one should
introduce the reaction of the fermionic fields on the bosons. Looking back at the action
(18) for the fermions in 2 + 1, we see that for each species n(1) and n(2) , there will be a
fermionic zero mode every time n + e passes through zero, i.e., when e = 2n/L.
These zero modes are produced because of the Callan and Harvey mechanism, since there
is a change of sign for the mass term of a 2 + 1 dimensional fermion. Since we have two
2-component fermions, and their mass terms have opposite signs, there shall be two chiral
fermions, of opposite chiralities, localized around each defect. This is of course equivalent
to saying that we shall have a massless Dirac fermionic zero mode around each defect.
The existence of these gapless zero modes has of course relevance for the calculation of
transport properties.
If we restrict ourselves to configurations having an approximate invariance under
translations along one of the spatial coordinates, the effective action for the field will of
course be the one of a two dimensional field theory with a potential periodic in the field, and
with an unusual kinetic term. Despite this last property, the periodic nature of the potential
implies that there is room for the existence of a non-trivial topological charge Q, defined
as for the case of the sine-Gordon model. If the system is invariant under translations in x2 ,
then
Z+

eL
eL
dx1 1 (x) =
(+) () .
(71)
Q=
2
2
A configuration of definite charge Q = q, corresponds to the field starting in one of the

minima at x1 = and covering q periods of the potential before finishing at another
minimum of the potential, at x1 = +. We may assume, for the sake of clarity, that
the effective potential is identical to the first component of the series. Then, there will
be exactly q fermionic zero modes, since they appear each time the potential passes
through a maximum. This discussion shows that there is an interesting relation between
the topological solitonic charge and the number of fermionic zero modes.
Acknowledgements
C.D.F. and A.L. are supported by CONICET and Instituto Balseiro, Argentina. This
work is supported in part by grants from ANPCYT (PICT 97/0053, 97/2285 and 98/0303924), CONICET, and by Fundacin Antorchas, Argentina.
732
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
R. Rajaraman, Solitons and Instantons, North Holland, Amsterdam, 1982.

C.D. Fosco, A. Lpez, Nucl. Phys. B 538 (1999) 685.
C. Callan, J.A. Harvey, Nucl. Phys. B 250 (1985) 427.
I.J.R. Aitchison, C.M. Fraser, Phys. Rev. D 31 (1985) 2605.
G. Grignani, G. Semenoff, P. Sodano, Phys. Rev. D 53 (1996) 7157.
J. Kapusta, Finite-Temperature Field Theory, Cambridge Univ. Press, 1993, Appendix A.3, and
references therein.
G. Grignani, G. Semenoff, P. Sodano, O. Tirkkonen, Nucl. Phys. B 473 (1996) 143.
C.D. Fosco, G.L. Rossini, F.A. Schaposnik, Phys. Rev. Lett. 79 (1997) 1980; Phys. Rev. D 56
(1997) 6547; Phys. Rev. D 59 (1999) 085012.
C.D. Fosco, E. Fradkin, A. Lpez, Phys. Lett. B 451 (1999) 31.
A.R. Kavalov, I.K. Kostov, A.K. Sedrakyan, Phys. Lett. B 175 (1986) 331.
E.C. Marino, Phys. Lett. B 263 (1991) 634.
E.C. Marino, Nucl. Phys. B 408 (1993) 551.
D.G. Barci, C.D. Fosco, L.E. Oxman, Phys. Lett. B 375 (1996) 267.
J.D. Edelstein, C. Nez, F.A. Schaposnik, J.J. Giambiagi, Mod. Phys. Lett. A 11 (1995) 1037.
E. Fradkin, F.A. Schaposnik, Phys. Lett. B 338 (1994) 254.
F.A. Schaposnik, Phys. Lett. B 356 (1995) 39.

Seminaive-dimensional renormalization
M. Pernici
INFN, Sezione di Milano, Via Celoria 16, I-20133 Milano, Italy
Received 5 January 2000; revised 20 March 2000; accepted 3 May 2000
Abstract
We propose a treatment of 5 in dimensional regularization which is based on an algebraically
consistent extension of the BreitenlohnerMaisont HooftVeltman (BMHV) scheme; we define
the corresponding minimal renormalization scheme and show its equivalence with a non-minimal
BMHV scheme.
The restoration of the chiral Ward identities requires the introduction of considerably fewer finite
counterterms than in the BMHV scheme. This scheme is the same as the minimal naive dimensional
renormalization in the case of diagrams not involving fermionic traces with an odd number of 5 ,
but unlike the latter it is a consistent scheme.
As a simple example we apply our minimal subtraction scheme to the Yukawa model at two loops
in presence of external gauge fields. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.10.G; 11.15
Keywords: Dimensional regularization; Chiral theories
1. Introduction
Dimensional regularization [15] is the most powerful scheme for making perturbative
computations in quantum field theory. It is particularly effective in absence of chiral
symmetries, since minimal subtraction [6] preserves the vectorial gauge symmetries.
In [1] it has been shown that defining 5 as in four dimensions the axial anomaly
is reproduced. In [7] it has been proven that this definition of 5 leads to a consistent
regularization scheme (BMHV) at all orders in perturbation theory.
The restoration of the chiral Ward identities in the BMHV scheme has been studied
in [8].
In the BMHV scheme many of the computational advantages of the dimensional
regularization scheme are lost, since the restoration of the chiral Ward identities requires
the introduction of non-invariant counterterms for practically any diagram containing 5 .
Work supported in part by M.U.R.S.T.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 6 8 - 6
734
M. Pernici / Nuclear Physics B 582 (2000) 733755
Due to this difficulty, comparatively few quantities have been computed in this scheme.
Among them, there are some one-loop processes in the standard model [9,10], two-loop
anomalous dimensions of four-fermion operators in QCD [11], the three-loop anomalous
dimension of axial currents in QCD [12]. The AdlerBardeen non-renormalization
theorem [13] has been verified at two loops in this scheme in [14,15].
Systematic computations of the bare action in the BMHV scheme have been made
in non-abelian gauge theories with chiral fermions [16], where the finite one-loop
counterterms necessary to restore the SlavnovTaylor identities have been computed; in
the simplest Yukawa model at two loop in the minimal scheme [17]; in a general class of
Yukawa models at two loops, where the finite two-loop counterterms necessary to restore
the rigid Ward identities have been computed, together with some of those necessary to
recover the local Ward identities in presence of external gauge fields [18].
Most of the computations in chiral theories are made in some version of the naivedimensional regularization scheme (NDR), first considered in [19] and [20]; other
proposals of NDR have been made in [21,22]. The common feature of all these versions
of NDR is that in fermionic loops with an even number of 5 s, these are considered to be
anticommuting with all the gamma matrices, thus avoiding the spurious anomalies present
in the BMHV scheme. However all these regularization schemes have some algebraic
inconsistency. To avoid these algebraic inconsistencies it has been suggested that the cyclic
property of the trace be upheld [2325], but in this case bosonic symmetry is not respected,
and no consistency proof for these renormalization schemes is available.
Reviews on various proposals for dimensional regularization in chiral theories can be
found in [26,27].
In this paper we introduce a consistent regularization scheme, in which to the usual
and 5 formal objects of the BMHV scheme two new 5 -like objects 1 and 2 are added;
the former is not anticommuting with all the gamma matrices, as 5 , while the latter is
anticommuting; they are defined in such a way that traces with an even number of 1 or an
odd number of 2 vanish.
We will call this scheme seminaive-dimensional regularization (SNDR); due to its
algebraic consistency in d = 4 dimension, algebraic manipulations on the bare
quantities can be done unambiguously.
We define minimal subtraction (MS) in this scheme in such a way that, after taking the
limit 0 on the minimally subtracted Green functions, there exists a homomorphism
from these abstract subtracted Green functions to the usual renormalized Green functions,
which preserves the algebraic properties on convergent diagrams, like the skeleton
expansion.
We prove the equivalence of the MSSNDR scheme with a non-minimal BMHV
scheme.
The proof of the consistency of the MSSNDR scheme is the main result of this paper.
We do not deal with the practical problem of restoring the Ward identities, apart from an
explicit example and some general considerations, showing that in this scheme there are
considerably fewer spurious anomalies than in the MSBMHV scheme.
735
As an example of the practical convenience of our scheme with respect to the BMHV
scheme, we treat in some detail the two-loop renormalization of the Yukawa model in
presence of external gauge currents, which we have worked out in the BMHV scheme
in [18]; while in the latter scheme minimal subtraction does not preserve the (anomalous)
chiral Ward identities, so that finite counterterms must be added to restore them for every
relevant term, in the SNDR scheme, after performing minimal subtraction only two finite
two-loop counterterms are needed to preserve the anomalous Ward identities, with the
anomaly appearing in the Bardeen form [28]; the AdlerBardeen theorem [13] is verified.
We discuss briefly the case of chiral gauge theories and we compare SNDR with
previously proposed NDR schemes.
2. Seminaive-dimensional regularization and minimal subtraction

2.1. BreitenlohnerMaisont HooftVeltman scheme
We recall how gamma matrices are treated in the t Hooft and Veltman [1] dimensional
regularization scheme as elaborated by Breitenlohner and Maison (BMHV) [7]. We use
Euclidean space conventions.
In the BMHV scheme one considers the Lorentz covariants I , , , p , etc. as
formal objects, satisfying the usual tensorial rules. is the Kronecker delta in d = 4
dimensions; a formal rule for summed indices is given:
= ,
p = p ,
= d.
(1)
The Lorentz indices of these formal covariants do not take a specific value.
The gamma matrices satisfy the relation
{ , } = 2 I,
(2)
where I is the identity,

I 2 = I,
I = I = .
(3)
The trace is cyclic and satisfies

tr I = 4.
(4)
In the BMHV scheme additional (d 4)-dimensional or evanescent tensors , p

and are introduced; the Kronecker delta in the (d 4)-dimensional space is ,
satisfying
= ,
= = ,
.
p p ,
(5)
The Kronecker delta in four dimensions in , satisfying

,
p p ,
(6)
736
The Levi-Civita antisymmetric tensor has no evanescent component:

= 0
(7)
and satisfies
1 2 3 4 1 2 3 4 =
sign()54i=1 i (i) .
(8)
S4
The matrix 5 is defined by

1
,
4!
which implies
5 =
(9)
{ 5 , } = 2 5 ,
= I,
(10)
[ , ] = 0.
5 5
(11)
After having performed the Dirac algebra with these rules, reduced the number of LeviCivita tensors, eliminated traces, reduced products of gamma matrices to the corresponding
antisymmetric products and eliminated summed indices, the BMHV elements are reduced
to normal form. The Feynman graphs are expressed in terms of meromorphic functions
of , which are the coefficients of these normal form terms.
The minimal subtraction on a l-loop 1PI Feynman graph renormalized up to order l 1
consists in subtracting local terms, polynomial in momenta and masses, which are the
poles of the Laurent series in of the meromorphic functions defining such a graph, both
for hatted and non-hatted tensor structures in normal form.
The bare action is a formal object, from which the Feynman graphs are constructed
following the usual rules. In the minimal subtraction scheme the bare action l-loop
counterterms are the poles of the l-loop 1PI Green functions, computed using the Feynman
rules derived from the (l 1)-loop bare action.
After making the minimal subtraction, the limit for 0 is taken in the entire functions
which are the coefficients of the normal form terms, and all the hatted tensors are set to
zero; finally the formal sums, traces and the normal form covariants are identified with the
usual four-dimensional operations and objects, respectively.
In Euclidean space the reflection symmetry takes the place of hermiticity [29]. Reflection
symmetry is an antilinear involution, with
(x)
= 1 (x 0 ),
0 )1 ,
(x) = (x
= ,
x 01
x 1 ,
x 0
= ,
5
(12)
(13)
=
=
for 6= 1. , and are invariant under reflection
where
symmetry. Considering a multiplet of fermions, the following fermionic bilinears are
reflection symmetric:
Z

H1 + H2 + H3 5 + iH4 5 + iH5 A + H6 5 B
x

+ iH7 V + iH8 5 A + iH9 V + H10 5 A ,
(14)
737
where the scalar A, the pseudoscalar B, the vector V and the pseudovector A are real
fields, with
A(x) = A(x 0 ),
B(x) = B(x 0 ),
V (x) = V0 (x 0 ),
A (x) = A0 (x 0 )
and Hi are hermitian matrices commuting with the gamma matrices.

A non-minimal subtraction consists in performing, after the minimal subtraction, the
subtraction of finite local counterterms; to avoid ambiguities, one must characterize which
are the normal form terms on which the finite subtractions are made.
In the case of a reflection symmetric theory, a convenient choice of fermionic bilinears
finite counterterms are the non-hatted terms in (14). The advantage of this choice is that in
the case of non-chiral theories the d-dimensional Lorentz symmetry of the bare action is
preserved.
2.2. Extension of the BMHV algebra
Add to the BMHV Dirac algebra the objects and 1 satisfying the following defining
relations, assumed to be valid for arbitrary :
{, } = 2 1 ,
(15)
= I,
(16)
I = ,
(17)
[1 , ] = 0,
(18)
[1 , ] = 0,
(19)
tr = 4 ,
(20)
tr 12
= 0.
(21)
We assume that the trace has the property (4) and that it is cyclic on this enlarged algebra.
Under reflection symmetry, = and 1 = 1 .
Define
2 1 .
(22)
Consider first the case 6= 0.

Let us prove that
I = ,
(23)
{1 , } = 2 1 ,
(24)
{2 , } = 0,
(25)
1 I = I 1 = 1 ,
(26)
1 2 = 2 1 = 0,
12
+ 22
= I,
13
(27)
= 1 ,
23
= 2 .
(28)
738
Eq. (23) is obvious. From (15) one gets

1
1
1
1 = {, } = + .
2
2
2
(29)
Using (15), (18), (22) and (23) one gets the relations (24), (25) and (26). From (19), (24)
and (25) one gets easily

2
1 , = 22 , = 0,
1 2 = (d 8)1 2 ,
(30)
[1 , 2 ] = 0.
From (16), (22) and (30) one has

d = 2 = 12 + 22 + {1 , 2 }

= d 12 + 22 + (d 8){1 , 2 },
so that
{1 , 2 } = 0.
(31)
Since 6= 0, we get {1 , 2 } = 0, which together with (19) implies (27).

Finally (28), which follows trivially from (16) and (27), says that 12 and 22 are
orthogonal projectors commuting with all the elements of this algebra.
Let us prove the following relations for the trace:
tr 1 = tr 2 = 0,
tr 12 1
tr 22 1
(32)
r = 0,
(33)
r = tr 1 r ,
(34)
tr 2 1 r = 0,
(35)
tr 1 1 r = tr 1 r
(36)
for arbitrary choice of indices 1 , . . . , r , r > 1. One has

d tr 1 = tr 1 = tr 1 = (8 d) tr 1 ,
d tr 2 = tr 2 = tr 2 = d tr 2
(37)
from which (32) follows. To prove the remaining relations (33)(36), observe that if for all
X = X ,
(38)
then, for indices 1 , . . . , r all different

d tr X 1 r = tr X 1 r
= tr X 1 r = (1)r (2r d) tr X 1 r ,
which for non-integer d has solution only for
(39)
tr X 1 r = 0.
739
(40)
Using this observation and the fact that 12 , 1 5 and 22 commute with all the gamma
matrices, while 2 anticommutes with all of them, one gets relations (33)(36) for indices
1 , . . . , r all different. In particular one gets
tr 2 5 = 0.
(41)
The general case for (33)(36) follows from this particular case since, simplifying
gamma matrices with equal indices, one is reduced to the relations
tr 12 = tr 2 = 0,
tr 22 = 4,
tr 1 5 = 4,
(42)
which follow trivially from the relations (16)(21) and from (32), (41).
Using (32)(36) all the traces can be computed, observing that the traces with more than
two 1 or two 2 can be reduced to the cases (32)(36) using (15)(28).
Observe that in d 6= 4 dimensions this algebra of covariants does not include 1 d , as
might be thought due to the presence of 2 , which is a 5 -like object anticommuting with
all gamma matrices. In fact tr 2 1 d = 0 due to (35).
In the case = 0, vanishes and 1 is an independent generator, since Eq. (29)
becomes trivial; in that case, add the defining relations
{1 , } = 0,
(43)
1 = 12 ,
(44)
I 1 = 1 I = 1 ,
(45)
which we obtained from (15)(19) in the case 6= 0.

In the case = 0, after identifying the formal sums (e.g., p ) as usual Einstein
convention sums, there is a homomorphism between the subalgebra generated by =
and , with the trace tr, and the usual four-dimensional Dirac algebra generated by 0 ,
with unit matrix I 0 and with the usual trace tr0 : this homomorphism is uniquely determined
by mapping the generators,
1
0 0 0 0
4!
and by the relations defining the homomorphism
0
( ) = 0 ,
() = 5
(46)
(I ) = I 0 ,
(X1 X2 ) = (X1 )(X2 ),
0
tr X = tr (X)
(aX1 + bX2 ) = a(X1 ) + b(X2 ),

(47)
for X, X1 and X2 belonging to this subalgebra; a and b are complex numbers. p , and
are mapped into the corresponding usual tensors in four dimensions.
To verify that this is a homomorphism with the properties (47) it is sufficient to verify
them on the relations (4), (15)(17), (20) defining the subalgebra for = 0.
The kernel of this homomorphism is obtained projecting this subalgebra with
(I 5 )/2, which commutes with all the elements of the subalgebra.
740
(I + 5 )/2 generates the orthogonal subalgebra, establishing an isomorphism

between this subalgebra and the usual Dirac algebra.
This homomorphism cannot be extended to 1 , since
0
0
4 = tr( 5 )1 6= tr0 5 5 (1 ) = 0
for any value of (1 ).
Furthermore it cannot be extended to the case 6= 0; for instance
0
0
8 = tr( 5 ) 6= tr0 5 5 ( ) = 0
(48)
for any value of ( ).

In order to check the consistency of the defining relations (15)(21), let us give an
explicit representation of this extension of the Dirac algebra for d = 2n even dimensions.
0
Denote by 0 the usual 2d/2 -dimensional gamma matrices; define further 5 10 20 30 40
and 0 i (d1)d/210 d0 (the 50 in the right-hand side of the latter expression is for
0
= 5 and should not be confused with the pseudoscalar object called 5 defined in the
former expression).
The extended algebra elements are given by the following tensor products:
I I 0 diag(1, 1, 1, 1),
0 diag(1, 1, 1, 1),
0
5 5 diag(1, 1, 1, 1),
(49)
50
1 diag(1, 1, 0, 0),
2 0 diag(0, 0, 1, 1),

tr X = 21(d/2)Sp I 0 diag(1, 1, 1, 1) X ,
where Sp is the ordinary trace on 2d/2+2 2d/2+2 matrices, while tr is not a matrix trace,
but it is a cyclic multi-linear operator on the space generated by the gamma matrices,
and 1 .
In this representation all the defining relations for the extended Dirac algebra hold. All
the properties which we derived above hold in this representation, including (35), which
however does not follow from (39) in the case r = 2n.
Let us finally give a somewhat intuitive interpretation of this tensor product representation. Consider the following action
N
X
r ( + iA )r
r=1
N
X

1r 0 ( + iA 50 )1r + 2r 0 ( iA 50 )2r
r=1

+ 3r 0 ( + iA 0 )3r + 4r 0 ( iA 0 )4r ,
where 1r , 3r and 4r are fermions, while 2r is a commuting fermion. Since 1r and
2r have opposite statistics and opposite parity, in a fermionic loop they contribute only
741
for an odd number of axial vector legs. Since 3r and 4r have the same statistics and
opposite parity, they contribute only for an even number of axial vector legs. The total
fermionic loop contribution of this multiplet is 2N times the fermionic loop contribution
( + iA ) using the extended Dirac algebra previously
of a fermion with action
described. Taking formally N = 1/2, the two contributions are the same. The trace defined
in (49) is interpreted as the supertrace for this supermultiplet; the factor 21(d/2) is N =
1/2 times the usual factor 22(d/2) introduced to normalize as in four dimensions the Dirac
trace in d dimensions; the minus sign in Sp[(I 0 diag(1, 1, 1, 1))X] is due to the bosonic
statistics of 2r .
2.3. Seminaive-dimensional regularization
In the seminaive-dimensional regularization (SNDR) the Feynman amplitudes are
expressed in terms of meromorphic functions of the complex parameter d = 4 ; in
the case of fermionic amplitudes, there is a meromorphic function for each element of the
generalized gamma matrix algebra described above, written in normal form (NF).
A basis for the gamma matrix normal form quantities belonging to the generalized Dirac
algebra consists in
{A , A , 1 A , 1 A },
(50)
{A } = {I, , [1 2 ] , . . . , [1 r ] , . . .},
(51)
where [ ] indicates antisymmetrization.

It can be decomposed in the following way:
c
NF = NF 0 + NF 1 + NF,

NF 0 = A , A ,

NF 1 = 1 A , 1 A ,

A = I, , [ ] , 5 , 5 ,
(52)
(53)
(54)
(55)
c consists of the remaining NF terms, which have at least a .

NF
Denote by A0 and A1 the subalgebras generated by NF 0 and NF1 , respectively, on the
complex numbers (constant in ).
c multiplied by an entire function in , and
Denote by A the subalgebra generated by NF,
by any NF term multiplied by an entire function with a zero in = 0. The objects in A are
evanescent.
One has
A0 A1 = A1 A0 = A1 A1 = A1 ,
A0 A0 = A0 ,
1 = A1 A = A.
0 = A0 A = AA
AA
(56)
Denote by T0 the algebra on constants in , whose generators are the normal form terms
in NF 0 , the momenta p , , and the normal forms obtained from these taking
tensor products of these (e.g., , I, p I I, I I , are normal forms in T0 ;
742
these tensor products are introduced to cover the case of Feynman graphs with more than
one open fermionic line).
Denote by T1 the algebra on constants in , whose generators are the normal form terms
in NF 1 and the normal forms obtained taking the tensor product of at least one NF1 term
and T0 normal forms.
Finally denote by Tb the space formed by normal forms with an hatted quantity, on
constants in , and by any normal form multiplied by an entire function with a zero in = 0,
and by the elements obtained making a tensor product of at least one of these elements and
those previously defined in T0 and T1 . This is the algebra of evanescent quantities. One has
T0 T1 = T1 T0 = T1 T1 = T1 ,
T 0 T0 = T0 ,
b
b
b=T
b.
b
T T 0 = T0 T = T T1 = T1 T
(57)
The homomorphism defined in the previous subsection extends canonically to T0 , but

b.
cannot be extended to T1 or to T
After the subtraction procedure, a renormalized but still regularized 1PI Feynman graph
b
Gren
n , written in normal form, belongs to the abstract algebra T0 + T1 + T . The subscript
n indicates the Lorentz and internal indices of the Feynman graph.
The algebraic consistence of the extended BMHV algebra guarantees that it is equivalent
directly in terms of convergent integrals, or to compute separately the
to compute Gren
n
and its counterdiagram graphs, and then sum them; the latter
divergent integrals for Gunren
n
procedure is much easier.
must be evaluated as a four-dimensional quantity Gren
Gren
n
n , expressed in terms of the
usual Dirac algebra. This is done in two steps:
the resulting quantity,
(i) set and the hatted objects to zero; denote by G=0ren
n
belonging to T0 + T1 ;
(ii) map the abstract objects in the generalized BMHV algebra to the usual Dirac algebra
quantities:

(58)
= Gren
G=0ren
n
n ,
where Gren
n is the renormalized 1PI graph after the regulator has been removed. The map
must be the trace-preserving homomorphism previously defined.
It is necessary that this map be a trace-preserving homomorphism to preserve the
algebraic properties on convergent diagrams, like making a skeleton expansion of a
convergent 1PI Green function; these expansions involve making products of relevant 1PI
vertices and dressed propagators, sums over Lorentz indices and Dirac traces, which are
preserved by a homomorphism preserving the traces. It is clear that this homomorphism
is unique. Since it cannot be defined on T1 , it follows that a consistent renormalization
at all
scheme in the SNDR scheme must be such that the T1 terms are absent from G=0ren
n
,
which
therefore
orders in loops (and hence from the 0 term of the Laurent series for Gren
n
b, while G=0ren
must belong to T0 ).
must belong to T0 + T
n
This is a non-trivial requirement, since the T1 -dependent counterterms can only be local,
while non-local terms are produced in perturbation theory; for instance in a unsubtracted
two-loop fermionic two-point function (without its one-loop counterdiagrams) there are
743
generally both non-local poles ( this is true also in the BMHV case) and non-local finite A1
terms; the point is that these non-local terms are canceled by the one-loop counterdiagrams.
In the next subsection we will show that this non-trivial property holds at all orders in
perturbation theory.
The second step is trivial in the BMHV scheme, since the BMHV Dirac algebra for
= 0 is isomorphic with the usual Dirac algebra in four dimensions. This is not so in the
SNDR Dirac algebra for = 0.
We will call a quantity, belonging to the extended Dirac algebra, to be weakly evanescent
if it belongs to the kernel of the homomorphism , modulo evanescent terms. The space of
weakly evanescent quantities is called W .
The kernel of on the domain A0 is 12 (1 5 )A0 . The subspace 12 (1 + 5 )A0 is
isomorphic to the usual Dirac algebra. Clearly
W A0 = W W = W.
(59)
In the general tensor product case

560W T0 = W W = W.
(60)
Summarizing: the regularized quantities (bare action, Gren

n ) live in a space with four
lives in a space with twice
times as many Dirac algebra elements as in BMHV; G=0ren
n
ren
by evanescent
as many Dirac algebra elements as in BMHV; Gn differs from G=0ren
n
ren modulo weakly evanescent quantities; projecting
is
isomorphic
to
G
quantities; Gren
n
n
to 12 (1 + 5 )A0 one gets an object isomorphic to Gren
each A0 component of G=0ren
n
n .
2.4. Minimal subtraction in SNDR
The renormalization by minimal subtraction (MS) consists in a minimal subtraction
procedure and in the evaluation in four dimensions of the formal objects defined in
dimensional regularization.
We define it recursively starting from l = 1 loops.
(l)
Denote by Gn (k) a l-loop 1PI Feynman graph with momenta k, minimally renormalized according to our MS rules up to l 0 < l loops but yet unrenormalized at l loops. Denote
(l)
(l)
by n (k) the corresponding Green function, which is the sum of various Gn (k).
We define the minimal subtraction procedure for the SNDR in the following way.
The minimal subtraction on G(l)
n (k) consists in subtracting all the local terms,
polynomial in momenta and masses, which are:
(i) the poles of the Laurent expansion of the meromorphic functions defining such a
graph, both for hatted and non-hatted tensor structures in normal form;
(ii) the T1 terms.
(l)ren
is evaluated
The resulting renormalized but yet regularized 1PI Feynman graph Gn
in four dimensions following the procedure described in the previous subsection.
The l-loop bare action counterterms consist in the local terms subtracted from the 1PI
l-loop Green functions n(l) (k) according to these rules.
744
In a renormalizable theory without insertions of composite operators of dimension larger

than five, the only T1 counterterms are A1 terms (there are no relevant four-fermion terms).
(l)ren
thus defined has no non-local poles;
As in MSBMHV, we have to show that Gn
furthermore we have to show that it has no non-local T1 term.
0 (l)
Consider G0 (l)
n (k) and n (k) to be the corresponding quantity in the BMHV scheme.
Suppose by induction hypothesis that in the BMHV scheme finite local counterterms
can be chosen for the renormalized (but still regularized) relevant 1PI Green functions
ren(l 0 )
ren(l 0 )
ren(l 0 )
up to l 0 = l 1 loops such that n0
0 n0
is weakly evanescent, so
0 n0
that this difference vanishes after evaluation in four dimensions (in general the MSSNDR
scheme is equivalent to a non-minimal BMHV scheme).
Let us show that the same can be done at l loops.
r (l)
Let dn be the superficial degree of divergence of G(l)
n (k). For r > dn , k Gn (k) and
kr G0 (l)
n (k) admit a skeleton expansion.
kr denotes r derivatives with respect to the external momenta k, with indices
not saturated neither among themselves nor with the Lorentz indices included in the
subscript n.
The difference between the renormalized and still regularized Green functions in the
two schemes at l 0 < l loops is weakly evanescent by induction hypothesis, so that using the
skeleton expansion property and (57), (60) it follows that for r > dn also
r 0
kr G(l)
n (k) k G n (k).
(l)
(l)
This equation implies that Gn (k) G0 (l)
n (k) is a polynomial of degree dn
(61)
in the momenta
plus a weakly evanescent quantity. The polynomial part depends on the poles in (of
degree less or equal to l) and on the normal forms, including the T1 terms. An important
point in this kind of arguments [30] is that kr commutes with the operations of extracting
the poles or the 0 terms of the Laurent series on which the meromorphic functions are
expanded, and that it commutes also with the operation of projecting the T0 elements to
the kernel of and with the extraction of the T1 terms.
Making the minimal subtraction on G(l)
n (k) using our rules and making an appropriate
non-minimal subtraction in the BMHV sense on G0 (l)
n (k), the polynomial part of
(l)
(l)
0
Gn (k) G n (k) can be subtracted and the difference between the renormalized 1PI
Feynman graphs, yet regularized, is weakly evanescent.
We have therefore proven by induction that the MSSNDR scheme is equivalent to a
(generally non-minimal) BMHV scheme at all orders in perturbation theory.
The proof of polynomiality in masses of the minimal subtraction counterterms proceeds
in the same way [31] and holds also for the T1 counterterms.
Let us show from another point of view that step (ii) is necessary to define our minimal
subtraction scheme; consider a different definition of minimal subtraction for SNDR, in
which only poles are subtracted, and in which the evaluation procedure in four dimensions
includes the rule of setting 1 to zero and identifying with 5 (this map is not a tracepreserving homomorphism, as we saw in Section 2.2). Let us repeat the above argument,
(l)
in which now Gn refers to this modified scheme, while G0 (l)
n has the same interpretation
ren(l 0 )
of the skeleton expansion of
as above. In this case the renormalization parts n0
745
kr G(l)
n differ from the corresponding renormalization parts in the BMHV scheme by T1
terms and by weakly evanescent terms. Performing a trace involving a weakly evanescent
renormalization part and a T1 contribution in another renormalization part of the skeleton
expansion, one gets in general a finite non-local T0 contribution, which is absent in
kr G0 (l)
n (k). In such a case it is not possible to choose l-loop local counterterms to get
the same renormalized Green functions in the two schemes, up to weakly evanescent terms
or local T1 terms. Therefore one cannot go from the new scheme to the BMHV scheme
with a renormalization group transformation. This is not possible, so that the proposed
scheme is wrong.
The necessity of the subtraction of the T1 terms to have a consistent renormalization
scheme makes our subtraction scheme minimal; the polynomiality in the masses of these
T1 terms supports the notion that MSSNDR is really an MS scheme.
A characteristics of the usual MS scheme is that to compute the l-loop counterterms one
has to compute the O(1/) part of l-loop Feynman integrals, the O(1) part of (l 1)-loop
Feynman integrals, the O() part of (l 2)-loop Feynman integrals, and so on. In MS
(l)
SNDR the same happens. In fact the T1 l-loop terms in n (k) arise from a counterterm
graph with a counterterm containing a NF 1 factor (T1 terms are absent at tree level), from
a graph with vertices containing only NF 0 factors or from a graph containing at least a
c factor.
vertex with a NF
In the first case the Feynman integrals are clearly of (l 1)-loop order; in the second case
a NF 1 term is produced, in the reduction of the graph to normal form, by anticommuting
with , generating a 1 factor; to get a T1 term this must be contracted with another
, producing a , so that only the pole part of the l-loop Feynman integral is involved in
producing T1 ; the same is true in the third case.
(l)
Actually the subtraction of the l-loop T1 terms in n (k) can be done algebraically,
without doing any analytic computation; in fact one can subtract algebraically all the NF 1
(l)
terms (both poles and finite parts) from n (k), subtracting subsequently the remaining
poles. (In practice this algebraic subtraction can be done simply by treating in the open
fermionic lines as anticommuting with all the gamma matrices and by setting 1 to zero on
the same lines.) The analytic computation of the T1 l-loop counterterms matters only for
(l+1)ren
(k).
computing n
On the contrary, the subtraction of the l-loop poles from n(l)(k) requires some analytic
computation, so that from the analytic point of view the T1 counterterms require little effort
with respect to the pole counterterms.
Let us discuss briefly finite renormalization, which are needed to recover the chiral
symmetry or more in general to renormalize the Green functions in a non-minimal way,
e.g., at a momentum subtraction point.
b terms to have counterterms respecting
They belong to T0 , with the possible addition of T
d-dimensional Lorentz symmetry in the non-chiral cases; the discussion is analogous to the
one made in the first subsection for the non-minimal BMHV scheme.
If the tree-level action is reflection symmetric, the bare action in MSSNDR is also
reflection symmetric.
746
In a non-minimal SNDR scheme, reflection symmetry is preserved by choosing

reflection symmetric T0 finite renormalization terms.
Obviously non-minimal T1 counterterms are forbidden.
3. Applications
After having shown that the MSSNDR is a consistent renormalization scheme, let us
show that it is convenient from a computational point of view.
In this scheme the renormalized Green functions are the same as in MSNDR as long as
fermionic loops with an odd number of 5 does not occur. In these cases the computations
are exactly the same in the two schemes and the chiral Ward identities are satisfied.
In the remaining cases, in MSSNDR the traces with odd number of or 1 enter
in the game; in these cases one must check explicitly the chiral Ward identities; if they
are broken, they can be recovered as usual by adding local finite counterterms to the MS
counterterms, order by order in perturbation theory. These are the cases in which NDR
becomes ambiguous or inconsistent.
In the case of external vector gauge fields, the vectorial Ward identities are preserved in
the MSSNDR as well as in the MSBMHV scheme; in fact, since the vectorial gauge
transformations do not involve neither the dimension, nor or 1 , the operations of
extracting the poles and the T1 terms commute with gauge transformations, so that the
bare action is vector gauge invariant. The extension of this argument to gauge theories will
be discussed later.
3.1. Yukawa models
In the Yukawa models it is possible to use NDR with anticommuting 5 without finding
inconsistencies; MSNDR preserves the rigid chiral symmetries. However coupling these
models to external gauge fields and considering the corresponding Ward identities, NDR
is not anymore a consistent regularization scheme, due to the presence of the chiral
anomalies.
In the MSBMHV scheme the chiral Ward identities are broken by a large number
of spurious terms, which must be subtracted by introducing non-invariant local finite
counterterms in the bare action. This has been done at one-loop and partly at two loops
in [18], choosing renormalization conditions which, in the cases in which fermionic loops
with an odd number of 5 do not occur, give the same renormalized Green functions as in
MSNDR.
In SNDR the tree-level bare action is
Z
1
1
1

(D i )2 +
S (0) =
D (D )
2
2
2
i i + 1 hij kl i j k l .
(62)
+ i y
4!
We have set the dimensional regularization scale to one.
747
We use the same notation as in [18], with the difference that in the tree-level action
and in the gauge transformations on the fermions we replace 5 with ; the difference
between the two cases being weakly evanescent, this is an equivalent starting point for
making perturbation theory. We restrict our attention to the massless Yukawa theories.
Furthermore we do not require that A a = 0, which was chosen in [18] to reduce the number
of counterterms in the BMHV bare action.
The fermions have internal indices which are not indicated (I ). The matrices Si , Pi are
hermitian.
We will consider a group G which is not necessarily semi-simple.
The fermions belong to a (reducible) representation t a ; under an infinitesimal transformation
= i a t a .
(63)
We will consider chiral representations

t a = tRa PR + tLa PL = tsa I + tpa ,
ta = tLa PR + tRa PL = tsa I tpa ,
tLa = tsa tpa ,

tRa = tsa + tpa ,
I +
I
,
PL =
,
PR =
2
2
where tRa and tLa belong in general to different representations.
The Dirac conjugate fermion transforms as
= ia ta .
(64)
(65)
The scalars belong to a real (reducible) representation ( a )ij ; under an infinitesimal

transformation
i = ia ija j .
(66)
The covariant derivatives are

D = I + iAa t a ,
D = iAa ta ,

D i = ij + iAa ija j .
(67)
i i , with
The Yukawa coupling y
yi = Si I + iPi = Y i PR + Y i PL ,
Yi = Si + iPi
(68)
is invariant under these transformations provided

y j jai + y i t a ta y i = 0
(69)
or equivalently
Y j jai + Y i tRa tLa Y i = 0,
Y j jai + Y i tLa tRa Y i = 0.
The tree-level action is reflection symmetric.
748
We use a modified subtraction scheme [33], in which the bare constants are chosen of
the form
X
(l)
h l Ndl cA (),
cA =
l>1
Nd = (4)
/22

,
1+
2
(l)
cA
() =
(l)
cA,r
r .
(70)
r>0
(l)
For all counterterms which do not involve NF 1 , in minimal subtraction cA () has no 0

term.
The considerations made in the first section are easily generalized to the modified
subtraction schemes.
Let us discuss the MSSNDR renormalization at one and two loops, comparing it with
the MSNDR computations in the MSNDR scheme [32] and in the non-minimal BMHV
scheme [18].
Consider first the scalar Green functions; since it is not possible to produce renormalization parts involving , these Green functions have only fermionic loops with an even
number of s, which behave in these traces as the anticommuting objects 2 . Therefore in
this case our MS scheme is the same as the MSNDR scheme; only poles are subtracted,
respecting the chiral Ward identities. All the spurious terms present in BMHV [18] are
absent.
The same is true for the -independent parts of the Green functions involving
external gauge fields with or without external scalar fields. There is none of the spurious
terms present in BMHV, with or without A a (the former can be avoided in the case of
external gauge fields setting A a = 0, as in [18], but they must be included when the
gauge fields are promoted to quantum fields, leading to many other spurious terms in
BMHV, besides those computed in [18]). Let us now consider the relevant fermionic Green
functions. At one loop, multiplying them by 22 , they become identical to those in NDR
(after the replacement 5 2 ), and are minimal subtracted as in that case; multiplying
them by 12 , they become identical to those in BMHV (after the replacement 5 1 ).
Expressing the sum of the two sectors in the base of NF terms involving and 1 ,
MS consists in subtracting, besides the poles, the finite NF 1 terms, that is the A1 terms.
Projecting the MS subtracted graphs with 22 , one has MSNDR; projecting the MS
subtracted graphs with 12 , one has BMHV graphs renormalized in a non-minimal way (in
the BMHV sense), with a finite subtraction such that in four dimensions the renormalized
graphs are the same as in MSNDR. This is precisely the renormalization choice made
in [18].
To renormalize the fermionic Green functions at two loops in our scheme it is sufficient
(2)
to multiply them by 22 and perform MSNDR. The terms SNF1 are useful only to compute
renormalized Green functions at three or more loops, so we will not compute them here;
(2)
let us only remark that all contributions to SNF 1 apart from those with a fermionic loop
subdiagram can be read out from the corresponding spurious terms in [18], replacing 5
with 1 and projecting them with 12 . The difference in the remaining diagrams is that
749
in [18] the fermionic loop contains 5 , not 2 , leading to some difference in the coefficients
of those counterterms, but to the same renormalized Green functions.
The last renormalization parts to be considered are those involving , which are
those with fermionic loops involving an odd number of s, which behave as the non-naive
5 -objects 1 . At one loop the relevant graphs are the -dependent parts of the threevector and four-vector graphs, which are the same as in BMHV, since 1 acts as 5 in these
cases. In our MS scheme, there are no spurious terms in the bare action, corresponding to
the fact that the anomaly appears in the Bardeen form [28], preserving vectorial gauge
invariance (for a discussion, see [18]).
At two loops, the -dependent part of the three-vector graphs involves again 1 ;
the fermionic counterterms are therefore those obtained projecting the one-loop fermionic
counterterms with 12 , that is are the same as in BMHV in [18]. Therefore the two-loop
anomaly terms cancel as in [18], in agreement with the AdlerBardeen theorem [13].
The complete bare action in MSSNDR scheme is
Z
1
1
(PR c + PL c )D
cij D i D j +
S=
2
2
1
i i
(D )(P
L c + PR c ) + i c
2
1
1
a
b
+ cij kl i j k l + cab F
F
+ SNF 1 + 1S[Aa ],
(71)
4!
4
where

c c Yi , Yi c Yi , Yi ,

(72)
c c Yi , Yi
are hermitian due to reflection symmetry and 1S[Aa ] is the non-naive part of the pure
gauge part of the bare action. It is due to Feynman graphs containing one or more fermionic
traces with odd number of or 1 , giving each a Levi-Civita tensor.
At one loop one has
4
(1)
cij = Yij ,
1
(1)
c = Yi Yi ,
2
(1)
ci = yj yi yj ,
48
3
(1)
hrs ,
cij kl = Y(ij kl) + hrs
(ij kl)
8
1
(1)
cab = S2 (F )ab S2 (S)ab
3
3
and
(1)
SNF1
Z
=
i
i
1 Pi yi D + (D )
1 yi Pi

1 yi tpa y Aa ,
1 yj Pi yj i i
+
i
(73)
750
1S (1) [Aa ] = 0,
(74)
where Yi1 i2 i2n1 i2n , S2 (F ) and S2 (S) are defined as in [18], as well as the group covariant
K2ab in the next formula.
We checked explicitly that SNF 1 is invariant under vectorial gauge transformations, i.e.,
under transformations satisfying the constraint a tpa = 0 (see (64)), as expected by the
argument given at the beginning of this section.
At two loops one has [18,32]

1
8
2
4
3
hikmn hj kmn + 2 +
Yikj k + 2 +
Yij kk ,
cij(2) =
12

1
1
2
2
3
(2)
Yj Yi Yi Yj 2 Yj Yi Yj Yi + 2 +
Yij Yj Yi ,
c = 2 +
8
2
2

4
2
2
1
(2)
Yj k yk yi yj + 2
yk yj yi yj yk
ci = 2

1
1
2
yk yi yj yj + yj yj yi yk + 2 yk yi yj yk + yj yk yi yj
+ 2
2
1
1
+ yk yj yi yk yj hij kl yk yj yl ,

192 96
48
96 48
(2)
Yninj kl + Ynij nkl + 2 +
Ynij kln
cij kl = 2 +

48 24
12 6
96
Ymknl +
Ymp hpnkl
+ hmnij 2 Ymkln + 2 +

3
6
3
+ 2 hmnpq hpqkl + 2
hnpqk hmpql ,
2
(2)
= K2ab ,
(75)
cab
(2)
where in the expression for cij kl the symmetrization in the indices i, j, k, l is understood.
We didnt compute 1S (2) [Aa ], which might be different from zero; in that case, it must
be an term, and it is the same as in BMHV scheme used in [18]. Observe that
1S (2) [Aa ] does not have to do with the anomaly, since it does not involve the totally
symmetric tensor d abc .
(2)
which, as explained in the previous section, is not necessary to
We didnt compute SNF
1
compute the two-loop renormalized Green functions.
In absence of the external gauge fields MSSNDR preserves the rigid chiral Ward
identities at all loops. In fact it is not possible to produce the relevant terms with ,
which are at the origin of the breaking of the symmetries.
In presence of the external gauge fields the chiral Ward identities might be broken by
the contributions, so that a finite counterterm 1S (2) [Aa ]non min might be necessary
to recover the chiral Ward identities; at more than two loops 1S (l) [Aa ] can contain also
parity even terms, coming from an even number of Levi-Civita tensors.
Let us finally discuss what would have happened if, as discussed at the end of the
previous section, we had not subtracted the A1 terms.
751
Fig. 1.
In the computation of the anomaly, the difference from the correctly MS-subtracted case
and the one now discussed consists of the counterterm graph in Fig. 1, in which there
is the insertion of the A1 counterterm in (74), leading to a non-local finite contribution,
which conspires to the absence of two-loop corrections to the anomaly; apart from non
satisfying the AdlerBardeen theorem, the suggested theory would have the following
inconsistency: in the one-loop approximation it has the same renormalized Green functions
as in a consistent renormalization scheme, the non-minimal BMHV scheme used in [18];
from general renormalization theory arguments it follows that the renormalized Green
functions in two different renormalization schemes can be connected by a renormalization
group transformation; being the one-loop relevant terms equal, it follows that the marginal
Green functions can differ at two loops only by a local term; however we just found a case
in which this difference is non-local; this is not possible, so that the proposed scheme is
inconsistent, in agreement with the arguments in the previous subsection.
3.2. Gauge theories
In the Yukawa theories only the pure gauge Green functions do not satisfy the chiral
Ward identities in the MSSNDR scheme; the remaining Green functions satisfy them at
all orders in perturbation theory. The vector Ward identities are respected in this scheme.
In chiral gauge theories no Green function satisfies the chiral Ward identity in this
scheme (at order high enough in loops); however SNDR breaks the chiral symmetries in a
gentler way than BMHV.
As long as graphs involving fermionic loops with an odd number of 5 -like objects are
absent, the chiral Ward identities will be satisfied. In general, we can expect that the level at
which MS is not sufficient in SNDR is the same at which NDR reveals its inconsistencies.
Apart from the graphs mentioned in the previous section, it is reasonable to expect that
other superficially divergent graphs containing the subgraph in Fig. 2(a) (giving nonvanishing contributions to the two-loop anomaly [18]), like the one in Fig. 2(b) need to
be subtracted in a non-minimal way.
In presence of composite operators the need of introducing a finite renormalization
appears already at low loop order; for instance coupling an external gauge current to the
theory, the minimally subtracted triangle graph involving it and two chiral gauge fields
752
Fig. 2.
gives the anomaly in the Bardeen form which does not respect the chiral Ward identities
so that a finite counterterm must be added to restore them.
Apart from this issue, the MSSNDR scheme for a chiral gauge theory differs
qualitatively from that in Yukawa theories for the appearance of 5 itself (not or 1 ) in the
bare action. It is only at this point that the bare action starts living in the extended algebra
four times bigger than the BMHV algebra, as can be seen observing that, for instance, in
absence of counterterms with 5 in a chiral gauge theory the relevant bare vectorial part of
the vertices contains and 1 , but not 5 and 1 5 .
For example consider the graph in Fig. 2(c), contributing to the vertex for an external
gauge field (it is chosen to be external to avoid anomaly cancellation); in MS it generates
counterterms involving 3!1 / = 5 /, while at lower loop level only
vertices involving and 1 appeared in the bare action. Similar counterterms can be
expected in the graph in Fig. 2(b); this counterterm cannot be replaced by /, since
the difference is not weakly evanescent (see the example in Eq. (48)).
As far as vector gauge invariance is concerned, let us argue that MSSNDR preserves
it in the case of a non-abelian vector gauge theory with chiral Yukawa couplings (i.e., the
gauged version of the Yukawa model in the previous subsection, with tpa = 0).
The bare action is BRST-invariant at tree level. Suppose that it is BRST-invariant at
(l 1) loops; then the l-loop functional generator of the 1PI Green functions renormalized
up to (l 1) loops, called (l) , satisfies the SlavnovTaylor identities, which in the Zinn
Justin form [34] read
S (0) (l) + (l) S (0) =
l1
X
(m) (lm) .
(76)
m=1
0
In MSSNDR (l ) , for l 0 < l, is finite and belongs to T0 for = 0; due to (57), for = 0
the right-hand side of (76) belongs also to T0 ; it follows that the T1 local term of (l) is
BRST invariant, as well as its poles; therefore the bare action is BRST invariant also at l
loops.
3.3. Comparison with other NDR schemes
The NDR prescriptions proposed in the past have, as regularization schemes, some
inconsistency, so that they require some care in using them, usually on a diagram-by-
753
diagram basis, while in a true regularization scheme one should be able to get unambiguous
results by using the bare action and the Feynman rules, whatever computational routine is
followed.
Furthermore no proof has been given, in any of these schemes, that the renormalized
Green functions are equal, modulo finite local renormalization terms, to those in some
consistent scheme, e.g., the BMHV scheme, at all orders in perturbation theory.
The issue of preserving the Ward identities comes only after settling these points, and
it is solved by adding local finite terms to the bare action. In absence of a consistent
regularization system it is not true that checking the Ward identities on the relevant vertices
one is guaranteed that the theory is renormalized consistently. In fact consider a set of
renormalized Green functions which differs from any consistent set of renormalized
Green functions by non-local quantities. 1 Then check the Ward identities on the relevant
terms defined at a subtraction point. If there are no anomalies, one can add local finite
counterterms to restore the Ward identities on the relevant terms. Clearly the new set of
Green functions is still inconsistent, since non-local l-loop terms cannot be canceled by
local l-loop counterterms.
The dangers of using algebraically inconsistent regularization systems have been
reviewed in [26], where it is emphasized the fact that the renormalization conditions are
compatible with the Ward identities does not guarantee the validity of the quantum action
principle, in absence of a true regularization scheme.
Let us consider briefly some of the NDR schemes discussed in the literature.
To avoid confusion with previously defined objects, call 5 the algebraically undefined
5 -like object; 5 is defined in (9).
In [20] 5 is chosen to be anticommuting in open fermionic lines and in loops with an
even number of 5 , while in the loops with an odd number of 5 the trace rules are given
modulo some evanescent term, to be fixed by the Ward identities. The authors emphasize
that this is not a true regularization scheme, but that it is a convenient prescription for
one-loop computations.
In [22] 5 is chosen to be anticommuting in open fermionic lines and in loops with an
even number of 5 , while in the loops with an odd number of 5 one reduces the traces
with an odd number of fermions, before evaluating them, to the case of one 5 only, by
using anticommuting 5 ; this rule does not preserve the Bose symmetry.
In [2325] a non-cyclic definition of the trace is used, together with a totally
anticommuting 5 . Bose symmetry is broken in this scheme; to avoid this problem one
can add the rule of choosing an appropriate reading point in the graphs.
In [35] it is suggested the use of 5 together with a trace rule which, in the case of
even number of 5 , acts as if the 5 were naive, while in the case of odd number of
1 I.e., such that there is no renormalization group transformation transforming one set of Green functions in the
other; one can construct easily similar pathological cases by choosing, within a consistent regularization system,
two different one-loop sets of renormalization conditions, say A and B, and by computing a subset of the twoloop Green functions in system A, another subset in system B; the resulting mixing of Green functions differs by
non-local terms from the set of Green functions computed within a single scheme. Working without a consistent
regularization system without great care it is likely that a similarly inconsistent system is produced at some point.
754
Fig. 3.
5 it acts as the BMHV trace. In this case there are ambiguities in the treatment of
objects like tr({ 5 , }{ 5 , }) , which is zero or not depending on the fact that
({ 5 , }{ 5 , }) is given or not its value 4. For instance using this trace rule the
renormalization of the two-loop graph in Fig. 3 requires some care due to this ambiguity.
Trueman [36] has used the trick of renormalizing the axial vectorfermionfermion
vertex as the corresponding vectorfermionfermion vertex multiplied by 5 , as a way of
restoring the Ward identities and for treating 5 as if it were anticommuting, while keeping
the BMHV rules. Unlike the other prescriptions reviewed in this subsection, this one is
consistent; however, apart from the fact that this trick has been defined only for certain
cases, it has the drawback of using the BMHV scheme, with all its spurious anomalies
(i.e., finite counterterms which must be added to restore the Ward identities); for instance
in the case of the Yukawa theory in [18] a trick similar to the one proposed by Trueman has
been used, with the result that the number of spurious anomalies is conspicuous. Truemans
trick works well as long as one is able to make computations without resorting to the full
form of the bare action, implicitly defined; for a three-loop application of this trick see [12].
Let us finally describe the MSSNDR as a NDR prescription.
5 is chosen to be anticommuting in open fermionic lines and in loops with an even
number of 5 , while in the loops with an odd number of 5 it is considered to be equal
to 5 .
Apart from subtracting the poles, the MSSNDR requires a finite subtraction in the
fermionic subdiagrams which occur in the fermionic traces with an odd number of 5 ;
this finite subtraction is chosen such that the corresponding fermionic renormalization part
has the same value, apart from evanescent terms, as the fermionic renormalization part
occurring in an open fermionic line, or in a closed fermionic line with an even number of
5 (minimal subtraction of the T1 terms).
In MSSNDR the BMHV 5 does not appear in the tree-level action, but can be
produced in divergent graphs, by application of the Levi-Civita tensor on the gamma
matrices. The resulting counterterms must not be confused with those with 5 ; in particular
the counterterms ( 5 5 )/ cannot be neglected (( 5 )/ is not weakly evanescent).
Acknowledgement
I thank M. Raciti and F. Riva for help and discussions.
755
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
G. t Hooft, M. Veltman, Nucl. Phys. B 44 (1972) 189.

C.G. Bollini, J.J. Giambiagi, Phys. Lett. 40B (1972) 566.
J.F. Ashmore, Nuovo Cimento Lett. 4 (1972) 289.
G.M. Cicuta, E. Montaldi, Nuovo Cimento Lett. 4 (1972) 329.
D.A. Akyeampong, R. Delburgo, Nuovo Cimento 17A (1973) 578.
G. t Hooft, Nucl. Phys. B 61 (1973) 455.
P. Breitenlohner, D. Maison, Comm. Math. Phys. 52 (1977) 11.
G. Bonneau, Nucl. Phys. B 171 (1980) 477; Nucl. Phys. B 177 (1981) 523.
J.G. Korner, N. Nasrallah, K. Schilcher, Phys. Rev. D 41 (1990) 888.
R. Ferrari, A. Le Yaouanc, L. Oliver, J.C. Raynal, Phys. Rev. D 52 (1995) 3036.
A.J. Buras, P.H. Weisz, Nucl. Phys. B 333 (1990) 66.
S.A. Larin, Phys. Lett. B 303 (1993) 113.
S. Adler, W. Bardeen, Phys. Rev. 182 (1969) 1517.
D.R.T. Jones, J.P. Leveille, Nucl. Phys. B 206 (1982) 473.
M. Bos, Nucl. Phys. B 404 (1993) 215.
C.P. Martin, D. Sanchez-Ruiz, Action principles, restoration of BRS symmetry and the renormalization group equation for chiral non-Abelian gauge theories in dimensional renormalization
with a non-anticommuting 5 , preprint hep-th/9905076.
C. Schubert, Nucl. Phys. B 323 (1989) 478.
M. Pernici, M. Raciti, F. Riva, Dimensional renormalization of Yukawa theories via Wilsonian
methods, preprint hep-th/9912248.
W.A. Bardeen, R. Gastmans, B. Lautrup, Nucl. Phys. B 46 (1972) 319.
M. Chanowitz, M. Furman, I. Hinchliffe, Nucl. Phys. B 159 (1979) 225.
S. Gottlieb, J.T. Donohue, Phys. Rev. D 20 (1979) 3378.
M.I. Adelhafiz, M. Zralek, Acta Phys. Pol. B 18 (1987) 21.
H. Nicolai, P.K. Townsend, Phys. Lett. B 93 (1980) 111.
D. Kreimer, Phys. Lett. B 237 (1990) 59.
J.G. Korner, D. Kreimer, K. Schilcher, Z. Phys. C 54 (1992) 503.
G. Bonneau, Int. J. Mod. Phys. A 5 (1990) 3831; Phys. Lett. 96B (1980) 147.
P.A. Baikov, V.A. Ilin, Theor. Math. Phys. 88 (1991) 789.
W.A. Bardeen, Phys. Rev. 184 (1969) 1848.
J. Collins, Renormalization, Cambridge Monographs on Mathematical Physics, Cambridge
Univ. Press, 1984.
W.E. Caswell, A.D. Kennedy, Phys. Rev. D 25 (1982) 392.
J.C. Collins, Nucl. Phys. B 92 (1975) 477.
M.E. Machacek, M.T. Vaughn, Nucl. Phys. B 222 (1983) 83; Nucl. Phys. B 236 (1984) 221;
Nucl. Phys. B 249 (1985) 70.
E. Braaten, J.P. Leveille, Phys. Rev. D 24 (1981) 1369.
J. Zinn-Justin, Quantum Field Theory and Critical Phenomena, Clarendon Press, Oxford, 1989.
M. Veltman, Nucl. Phys. B 319 (1989) 253.
T.L. Trueman, Phys. Lett. 88B (1979) 331; Z. Phys. C 69 (1996) 525.

Erratum
Erratum to Signals of a superlight gravitino

at hadron colliders when the other superparticles
are heavy [Nucl. Phys. B 526 (1998) 136]
Andrea Brignole, Ferruccio Feruglio, Michelangelo L. Mangano,
Fabio Zwirner
Due to an inconsistency between the sign convention on the covariant derivatives and
other sign conventions, inherited from Ref. [5], the sign of some interference terms was
incorrectly taken. The correction of this sign error results in partonic cross-sections that
now fall off morerapidly for increasing energy of the final photon or parton. As a result,
the sensitivity to F (m3/2 ) is slightly weakened, as summarized by the following table:
Machine/signal
F (GeV)
m3/2 (105 eV)
TeV I, j
310
2.3
TeV II, j
410
4.0
LHC, j
1600
62
D0, 1a,
210
1.1
TeV I,
265
1.7
TeV II,
370
3.3
LHC,
1350
44
All the numbers given in the paper should be changed accordingly. All the figures should
be modified as well: their revised version can be found on hep-ph/9801329, revised v2.
We list here the most important changes to formulas and comments.
Eq. (6) should read:
(D ) = ( + ieQq A ) + igS (A ) .
The right-hand side of Eq. (10) should change sign.
PII of original article: S0550-3213(98)00254-5.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 1 9 - 4
760
Erratum / Nuclear Physics B 582 (2000) 759761

A(p1 , p2 , q1 , q2 , k)
= (kp1 )(kp2 )2 (kq1 )(p1 p2 )(p1 q2 ) (kp1 )(kp2 )(kq1 )2 (p1 p2 )(p1 q2 )
(kp1 )(kp2 )2 (kq1 )(p1 q1 )(p1 q2 ) + (kp1 )(kp2 )(kq1 )2 (p1 q1 )(p1 q2 )
2(kp1 )(kp2 )(kq1 )(p1 p2 )(p1 q2 )2 + (kp1 )(kq1 )2 (p1 p2 )(p1 q2 )2
+ (kp1 )(kp2 )(kq1 )(p1 q1 )(p1 q2 )2 + (kp1 )2 (kp2 )(kq1 )(p1 q2 )(p2 q1 )
(kp2 )(kq2 )(p1 p2 )(p1 q1 )(p1 q2 )(p2 q1 ) + (kp2 )(kq2 )(p1 q1 )2 (p1 q2 )(p2 q1 )
(kp1 )2 (kq1 )(p1 q2 )2 (p2 q1 ) + (kp2 )(kq1 )(p1 p2 )(p1 q2 )2 (p2 q1 )
(kp2 )(kq1 )(p1 q1 )(p1 q2 )2 (p2 q1 ) (kp1 )(kq2 )(p1 q1 )(p1 q2 )(p2 q1 )2
+ (kp2 )(kq2 )(p1 p2 )(p2 q1 )3 (kq2 )(p1 p2 )(p1 q2 )(p2 q1 )3
+ (kp1 )(kq1 )(p1 q1 )(p1 q2 )(p2 q1 )(p2 q2 ) + (kq1 )(p1 p2 )(p1 q2 )(p2 q1 )2 (p2 q2 )
+ (kp1 )(kp2 )(p1 p2 )(p1 q2 )(p2 q1 )(q1 q2 ) (kp1 )(kq1 )(p1 p2 )(p1 q2 )(p2 q1 )(q1 q2 )
(kp1 )(kp2 )(p1 q1 )(p1 q2 )(p2 q1 )(q1 q2 ) + (kp1 )2 (p1 q2 )(p2 q1 )2 (q1 q2 )
(kp2 )(p1 p2 )(p1 q2 )(p2 q1 )2 (q1 q2 ) + (p1 p2 )(p1 q2 )2 (p2 q1 )2 (q1 q2 ).
f (x , cos ) = (1 x )2

g(xq , cos q ) = (1 xq )
(1 x )(2 2x + x2 )
x sin2

x3 sin2
.
32
1 + xq2
16 sin2 (q /2)
+
x (6 + 6x + x2 )
16
(1 xq )(1 2xq + 2xq2 )

4 cos2 (q /2)
xq (12 27xq + 12xq2)

32
xq2 (3 4xq ) cos q

32

.
In the 4th paragraph of Section 4, the sentences First of all, . . . the background is
smallest. should be replaced by the following ones:
First of all, notice that the shape of
the background is very similar to that of the signal.
This suggests that the best limit on F will be extracted from a comparison of signal
and background in the widest possible region of ET compatible with other experimental
requirements.
The considerations at the end of Section 4, These limits are obtained for . . . ET distribution as well., should be replaced by the following ones:
These limits are obtained for ETmin = 100 GeV in the case of TeV I and TeV II,
and for ETmin = 200 GeV in the case of the LHC. Lower values of ETmin could lead
to more stringent limits, but in this case other backgrounds could become important,
Erratum / Nuclear Physics B 582 (2000) 759761
761
and a detailed experimental analysis would be required. We should also mention

that our procedure relies on the detailed prediction of the background rates. Notice
nevertheless that in the regions of ET where a large background is expected, its absolute
normalization can be determined with high accuracy by measuring the (Z `+ ` )+jet
rates. In these regions of large rates, furthermore, one could use not just the total rate
above a given threshold, but the ET -distribution as well.

Erratum
Erratum to Non-perturbative quark mass

renormalization in quenched lattice QCD
[Nucl. Phys. B 544 (1999) 669]
Stefano Capitani, Martin Lscher, Rainer Sommer, Hartmut Wittig
The penultimate line in Table A.1 on p. 687 of our paper contains a wrong value for .
The correct entry in that line should read:
7.1190
L/a
g 2 (L)
g 2 (2L)
6(u, a/L)
2.4484(58)
3.496(42)
3.496(44)
CERN, Theory Division, CH-1211, Geneva 23, Switzerland. E-mail: hartmut.wittig@cern.ch
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 1 6 3 - 2

Erratum
Erratum to The dynamics of AffleckDine

condensate collapse [Nucl. Phys. B 570
(2000) 407]
Kari Enqvist, John McDonald
On p. 416, at the bottom:
Compared with the natural time scale t = 1/m, the time taken to reach the Q-axiton state
is long, of the order of 1600/m for the case K = 0.05 and Q = Qmax , which is about
five expansion times (Hi1 300/m).
should read:
Compared with the natural time scale t = 1/m, the time taken to reach the Q-axiton state
is long, of the order of 16000/m for the case K = 0.05 and Q = Qmax , which is about
fifty expansion times (Hi1 300/m).

E-mail addresses: enqvist@pcu.helsinki.fi (K. Enqvist), mcdonald@physics.gla.ac.uk (J. McDonald).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 0 4 - 7

CUMULATIVE AUTHOR INDEX B581B582
Alvarez, O.
Arhrib, A.
B582 (2000) 139

B581 (2000) 34
Br, O.
Baulieu, L.
Blanger, G.
Bertola, M.
Bertolini, M.
Bialas, P.
Bilal, A.
Blumenhagen, R.
Blmlein, J.
Bohm, A.R.
Boudjema, F.
Brignole, A.
Bros, J.
B581 (2000) 499

B581 (2000) 604
B581 (2000) 3
B581 (2000) 575
B582 (2000) 393
B581 (2000) 477
B582 (2000) 65
B582 (2000) 44
B581 (2000) 449
B581 (2000) 91
B581 (2000) 3
B582 (2000) 759
B581 (2000) 575
Caffo, M.
Campbell, B.A.
Campos, I.
Capdequi Peyranre, M.
Capitani, S.
Carter, G.W.
Casas, J.A.
Castro Alvaredo, O.A.
Chakrabarti, S.K.
Chan, Ch.S.
Chu, C.-S.
Coste, A.
Cski, C.
Czyz, H.
B581 (2000) 274

B581 (2000) 240
B581 (2000) 499
B581 (2000) 34
B582 (2000) 762
B582 (2000) 571
B581 (2000) 61
B581 (2000) 643
B582 (2000) 627
B581 (2000) 156
B582 (2000) 65
B581 (2000) 679
B581 (2000) 309
B581 (2000) 274
de Azcrraga, J.A.
Denef, F.
Derendinger, J.-P.
Diakonov, D.
Di Clemente, V.
Donato, F.
Dubin, A.Yu.
B581 (2000) 743

B581 (2000) 135
B582 (2000) 231
B582 (2000) 571
B581 (2000) 61
B581 (2000) 3
B582 (2000) 677
Eden, B.
Einhorn, M.B.
Enqvist, K.
B581 (2000) 523

B582 (2000) 216
B582 (2000) 763
Erickson, J.K.
Erlich, J.
Evans, N.
B582 (2000) 155

B581 (2000) 309
B581 (2000) 391
Ferreira, C.N.
Feruglio, F.
Fosco, C.D.
Fries, R.J.
B581 (2000) 165

B582 (2000) 759
B582 (2000) 716
B582 (2000) 537
Gannon, T.
Geyer, B.
Godbole, R.
Gonzlez-Sprinberg, G.A.
Gorini, V.
Grlich, L.
Guimares, M.E.X.
B581 (2000) 679

B581 (2000) 341
B581 (2000) 3
B582 (2000) 3
B581 (2000) 575
B582 (2000) 44
B581 (2000) 165
Harshman, N.L.
Helayl-Neto, J.A.
Herdeiro, C.A.R.
Hikami, K.
Hollik, W.
Hollowood, T.J.
Hong, D.K.
Hormuzdiar, J.
Howe, P.S.
Hsu, S.D.H.
B581 (2000) 91
B581 (2000) 165
B582 (2000) 363
B581 (2000) 761
B581 (2000) 34
B581 (2000) 309
B582 (2000) 451
B581 (2000) 391
B581 (2000) 523
B581 (2000) 391
Inoue, R.
Intriligator, K.
B581 (2000) 761

B581 (2000) 257
Kaminsky, K.
Kaste, P.
Ketov, S.V.
Ketov, S.V.
Kim, J.E.
Kimura, Y.
Kiselev, V.V.
Kitazawa, Y.
Kniehl, B.A.
Kogut, J.B.
Krs, B.
Kotikov, A.V.
B581 (2000) 240

B582 (2000) 203
B582 (2000) 95
B582 (2000) 119
B582 (2000) 296
B581 (2000) 295
B581 (2000) 432
B581 (2000) 295
B582 (2000) 514
B582 (2000) 477
B582 (2000) 44
B582 (2000) 19
766
Kramer, G.
Kyae, B.
B582 (2000) 514

B582 (2000) 296
Laine, M.
Lazar, M.
Lee, H.M.
Lerche, W.
Lesgourgues, J.
Li, T.
Lipatov, L.N.
Lpez, A.
Losada, M.
Lscher, M.
Lst, D.
Ltken, C.A.
B582 (2000) 277

B581 (2000) 341
B582 (2000) 296
B582 (2000) 203
B582 (2000) 593
B582 (2000) 176
B582 (2000) 19
B582 (2000) 716
B582 (2000) 277
B582 (2000) 762
B582 (2000) 44
B582 (2000) 203
Macfarlane, A.J.
Mangano, M.L.
McDonald, J.
Miramontes, J.L.
Montes, X.
Morel, A.
Moschella, U.
Moultaka, G.
Mukhopadhyay, B.
Mller, B.
B581 (2000) 743

B582 (2000) 759
B582 (2000) 763
B581 (2000) 643
B582 (2000) 259
B581 (2000) 477
B581 (2000) 575
B581 (2000) 34
B582 (2000) 627
B582 (2000) 537
Nastase, H.
B581 (2000) 179
Oeckl, R.
Ohsawa, A.
Onishchenko, A.I.
B581 (2000) 559

B581 (2000) 73
B581 (2000) 432
Paul, P.L.
Pernici, M.
Petersson, B.
Petrov, K.
Pickering, A.
Polyakov, A.
Ptter, B.
B581 (2000) 156

B582 (2000) 733
B581 (2000) 477
B581 (2000) 477
B581 (2000) 523
B581 (2000) 116
B582 (2000) 514
Quirs, M.
B581 (2000) 61
Rasmussen, J.
Reisz, T.
Remiddi, E.
Robaschik, D.
B582 (2000) 649

B581 (2000) 477
B581 (2000) 274
B581 (2000) 449
Rosier-Lees, S.
Ruelle, P.
Russo, R.
Rychkov, V.
B581 (2000) 3
B581 (2000) 679
B582 (2000) 65
B581 (2000) 116
Santamaria, A.
Sauser, R.
Schaeffer, R.
Schfer, A.
Schaposnik, F.A.
Scharnhorst, K.
Schwetz, M.
Semenoff, G.W.
Sevrin, A.
Shirman, Y.
Shore, G.M.
Sokatchev, E.
Sommer, R.
Stein, E.
Stephanov, M.A.
B582 (2000) 3
B582 (2000) 231
B581 (2000) 575
B582 (2000) 537
B582 (2000) 716
B581 (2000) 718
B581 (2000) 391
B582 (2000) 155
B581 (2000) 135
B581 (2000) 309
B581 (2000) 409
B581 (2000) 523
B582 (2000) 762
B582 (2000) 537
B582 (2000) 477
Tamada, M.
Tanaka, T.
Toublan, D.
Trigiante, M.
Troost, J.
B581 (2000) 73
B582 (2000) 259
B582 (2000) 477
B582 (2000) 393
B581 (2000) 135
Vaman, D.
van Nieuwenhuizen, P.
Verbaarschot, J.J.M.
Verlinde, H.
Vidal, J.
Volkov, M.S.
B581 (2000) 179

B581 (2000) 179
B582 (2000) 477
B581 (2000) 156
B582 (2000) 3
B582 (2000) 313
Walcher, J.
West, P.C.
White, B.E.
Wiedemann, U.A.
Wipf, A.
Wittig, H.
B582 (2000) 203

B581 (2000) 523
B581 (2000) 409
B582 (2000) 409
B582 (2000) 313
B582 (2000) 762
Zarembo, K.
Zayas, L.A.P.
Zhitnitsky, A.
Zwanziger, D.
Zwirner, F.
B582 (2000) 155

B582 (2000) 216
B582 (2000) 477
B581 (2000) 604
B582 (2000) 759

Nucl - Phys.B v.582

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Nucl - Phys.B v.582

Загружено:

Авторское право:

Доступные форматы

Nuclear Physics B 582 (2000) 318

Model independent bounds on the tau lepton

Igu 4225, 11400 Montevideo, Uruguay

E-46100 Burjassot, Valncia, Spain

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

2. The effective Lagrangian for tau magnetic moments

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

2 GF = 246 GeV, and the interactions (2.4) can be written

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

where, as usual, we define cW = cos W , sW = sin W , tan W = g 0 /g and e = gsW . Thus,

where F is the electromagnetic field strength tensor, Z = Z Z and W+ =

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

where the momenta p1 , p2 , p3 , p4 , correspond to e , e+ , , + respectively, q = p1 +

Here the couplings are:

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

3.1. Total rates and cross sections

where the coefficients i are given by:

1 + 2v ve Re{} + v2 ve2 + ae2 ||2

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

This equation leads to the following bounds on Z :

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

= 1 + F1 (s)  + F2 (s) 2 + F1Z (s) Z + F2Z (s) Z2 + F Z (s) Z  .

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

is defined in the following angular regions

(+) = cos > 0, cos h > 0 + cos < 0, cos h < 0 ,

() = cos > 0, cos h < 0 + cos < 0, cos h > 0 .

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

5. Combined limits on electromagnetic and weak magnetic moments of the tau

(Eq. (3.25)) measured at SLD and LEP1 (Eq. (3.30));

(Eq. (4.1)) measured at LEP2 and pp colliders (Eq. (4.4)).

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

6. Discussion and conclusions

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

0.06 < W < 0.07,

0.003 < W < 0.004.

0.068 < ha i < 0.065 [OPAL].

G.A. Gonzlez-Sprinberg et al. / Nuclear Physics B 582 (2000) 318

J. Vidal, J. Bernabu, G. Gonzlez-Sprinberg, Nucl. Phys. B (Proc. Suppl.) 76 (1999) 221.

Nuclear Physics B 582 (2000) 1943

NLO corrections to the BFKL equation in QCD

188300 St. Petersburg, Russia

A.V. Kotikov, L.N. Lipatov / Nuclear Physics B 582 (2000) 1943

gluons having their initial and final transverse momenta

the angle between the two-dimensional vectors

Hkl = hkl + hkl ,

A.V. Kotikov, L.N. Lipatov / Nuclear Physics B 582 (2000) 1943

is the Euler constant and kl = k l .

s are solutions of the Schrdinger equations:

with the properties:

Moreover, it coincides with the local Hamiltonian of an integrable Heisenberg spin

combined with the transposition of these operators.

the duality equation for m,e

A.V. Kotikov, L.N. Lipatov / Nuclear Physics B 582 (2000) 1943

2. NLO corrections to BFKL kernel in QCD

A.V. Kotikov, L.N. Lipatov / Nuclear Physics B 582 (2000) 1943

for the gluongluon scattering is

s0 = |q| |q 0|. (15)

and the spacetime dimension D = 4 + 2 for 0. The gluon Regge trajectory

A.V. Kotikov, L.N. Lipatov / Nuclear Physics B 582 (2000) 1943

Here we introduced also

This equation leads to the following bounds on Z :

= 1 + F1 (s) + F2 (s) 2 + F1Z (s) Z + F2Z (s) Z2 + F Z (s) Z .

0.06 < W < 0.07,