Nucl - Phys.B v.790

Nuclear Physics B 790 (2008) 127
Higgs boson production via gluon fusion:

Squark loops at NLO QCD
Margarete Mhlleitner a,b , Michael Spira c,
a CERN, Theory Division, CH-1211 Genve 23, Switzerland
b Laboratoire dAnnecy-Le-Vieux de Physique Thorique, LAPTH, Annecy-Le-Vieux, France
c Paul Scherrer Institut, CH-5232 Villigen PSI, Switzerland
Received 28 March 2007; accepted 24 August 2007

Available online 1 September 2007
Abstract
The loop-induced processes gg h, H, A provide the dominant Higgs boson production mechanisms
at the Tevatron and LHC in a large range of the minimal supersymmetric extension of the Standard Model.
For squark masses below 400 GeV squark loop contributions become important in addition to the top
and bottom quark loops. The next-to-leading order QCD corrections to the squark contributions of these
processes are determined including the full squark and Higgs mass dependences. They turn out to be large
and thus important for the Tevatron and LHC experiments. Squark mass effects of the K factors can be of
O(20%). In addition we derive the QCD corrections to the squark contributions of the rare photonic Higgs
decays h, H , which play a role for the Higgs searches at the LHC.
2007 Elsevier B.V. All rights reserved.
1. Introduction
Higgs boson [1] searches belong to the major endeavors at present and future colliders within
the Standard Model (SM) and its minimal supersymmetric extension (MSSM). In the MSSM two
isospin Higgs doublets are introduced in order to generate masses of up- and down-type fermions [2]. After electroweak symmetry breaking three of the eight degrees of freedom are absorbed
by the Z and W gauge bosons, leaving five states as elementary Higgs particles. These consist
of two CP-even neutral (scalar) particles h, H , one CP-odd neutral (pseudoscalar) particle A and
Supported in part by the Swiss Bundesamt fr Bildung und Wissenschaft.
* Corresponding author.
E-mail address: michael.spira@psi.ch (M. Spira).

0550-3213/$ see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2007.08.011
M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127
Table 1
Higgs couplings in the MSSM to fermions and gauge bosons [V = W, Z] relative to the SM couplings
SM
MSSM
H
h
H
A
gu
gd
gV
1
cos / sin
sin / sin
1/ tg
1
sin / cos
cos / cos
tg
1
sin( )
cos( )
0
two charged bosons H . At leading order the MSSM Higgs sector is fixed by two independent
input parameters which are usually chosen as the pseudoscalar Higgs mass MA and tg = v2 /v1 ,
the ratio of the two vacuum expectation values. Including the one-loop and dominant two-loop
corrections the upper bound of the light scalar Higgs mass is Mh 140 GeV [3]. The couplings
of the various neutral Higgs bosons to fermions and gauge bosons, normalized to the SM Higgs
couplings, are listed in Table 1, where the angle denotes the mixing angle of the scalar Higgs
bosons h, H . An important property of the bottom Yukawa couplings is their enhancement for
large values of tg , while the top Yukawa couplings are suppressed for large tg [4].
For this work we need the Higgs couplings to stop and sbottom squarks in addition. The
scalar superpartners fL,R of the left- and right-handed fermion components mix with each other.
The mass eigenstates f1,2 of the sfermions are related to the current eigenstates fL,R by mixing
angles f ,
f1 = fL cos f + fR sin f ,
f2 = fL sin f + fR cos f .
(1)
The mass matrix of the sfermions in the leftright-basis is given by [5]1

M 2 + m2f
mf (Af rf )
fL
Mf =
,
mf (Af rf )
M 2 + m2f
(2)
fR
with the parameters rd = 1/ru = tg for down- and up-type fermions. The parameters Af denote
the trilinear scalar couplings of the soft supersymmetry breaking part of the Lagrangian, the
higgsino mass parameter and mf the fermion mass. The mixing angles acquire the form
sin 2f =
2mf (Af rf )
M 2
f1
M 2
f2
cos 2f =
M 2 M 2
fL
M 2
f1
fR
M 2
f2
and the masses of the squark mass eigenstates are given by

2
1
Mf2 = m2f + Mf2 + Mf2 M 2 M 2 + 4m2f (Af rf )2 .
fL
fR
L
R
1,2
2
(3)
(4)
Since the mixing angles are proportional to the masses of the ordinary fermions, mixing effects
are only important for the third-generation sfermions. The neutral Higgs couplings to sfermions
read [6]

g = m2f g1 + MZ2 I3f ef sin2 W g2 ,

fL fL
1 For simplicity, the D-terms have been absorbed in the sfermion-mass parameters M 2
.
fL/R
Table 2
Coefficients of the neutral MSSM Higgs couplings to up- and down-type sfermion pairs
f
g1
g2
g3
g4
h
H
A
h
H
A
cos / sin
sin / sin
0
sin / cos
cos / cos
0
sin( + )
cos( + )
0
sin( + )
cos( + )
0
sin / sin
cos / sin
1
cos / cos
sin / cos
1
cos / sin
sin / sin
1/ tg
sin / cos
cos / cos
tg
Fig. 1. MSSM Higgs boson production via gluon fusion mediated by top- and bottom quark as well as stop and sbottom
loops at leading order.
fR fR
fL fR
= m2f g1 + MZ2 ef sin2 W g2 ,

mf

=
g3 Af g4 ,
2
(5)
with the couplings gi listed in Table 2. I3f denotes the third component of the electroweak
isospin, ef the electric charge of the fermion f , W the Weinberg angle and MZ the Z-boson
mass. All these couplings have to be rotated to the mass eigenstates by the mixing angle f .
At hadron colliders as the Tevatron and LHC neutral Higgs bosons are copiously produced by
the gluon fusion processes gg h/H /A, which are mediated by top and bottom quark loops as
well as stop and sbottom loops in the MSSM (see Fig. 1). Due to the large size of the top Yukawa
couplings gluon fusion comprises the dominant Higgs boson production mechanism for small
and moderate tg . For large tg the leading role is taken over by Higgs radiation off bottom
quarks due to the strongly enhanced bottom Yukawa couplings [7].
The QCD corrections to the top and bottom quark loops are known since a long time including the full Higgs and quark mass dependences [8]. They increase the cross sections by up to
about 100%. The limit of very heavy top quarks provides an approximation within 2030%
for tg 5 [9]. In this limit the next-to-leading order (NLO) QCD corrections have been calculated before [10] and later the next-to-next-to-leading order (NNLO) QCD corrections [11]. The
NNLO corrections lead to a further moderate increase of the cross section by 2030%, so that
the dominant part are the NLO contributions. Very recently an estimate of the next-to-next-tonext-to-leading order (NNNLO) effects has been obtained [12] indicating improved perturbative
convergence. A full massive NNLO calculation is not available so far, so that the NNLO results
can only be used for small and moderate tg . On the other hand the NLO corrections to the
squark loops are only known in the limit of heavy squarks [13]. Moreover, the full SUSY-QCD
corrections have been obtained recently for heavy SUSY particle masses [14]. NLO computations for the last two contributions including the full mass dependences are missing so far. This
work presents the pure QCD corrections to the squark loops including the full squark and Higgs
mass dependences as a first step towards a full NLO SUSY-QCD calculation.
The reverse processes, gluonic Higgs decays, play a role for the decay profiles of the Higgs
particles. Their partial widths can be measured at a future linear e+ e collider with energy up
to about 1 TeV [15]. The gluonic branching ratios can reach a level of O(10%) in some regions
of the MSSM parameter space [4,7,16]. Analogously to the gluon fusion processes the QCD
corrections to the top and bottom quark loops have been calculated including the full quark and
Higgs mass dependences [8]. The NLO corrections in the limit of very heavy top quarks can
be found in [10,17]. This limit provides a reasonable approximation for small and moderate
values of tg . The NNLO QCD corrections are known for very heavy top quarks, too [18]. Very
recently, even the NNNLO QCD corrections have been obtained in the heavy top mass limit
[19]. While the NLO corrections enhance the partial decay width by up to about 70%, the NNLO
corrections turn out to be of moderate size O(20%) supplemented by the NNNLO corrections in
the per cent range. The quark mass effects at NNLO and beyond are unknown. The SUSY-QCD
corrections to the gluonic decay mode are identical to the gluon fusion process. The full mass
dependences of the NLO SUSY-QCD corrections are unknown. This work provides a first step
towards this aim by determining the pure QCD corrections to the squark loops.
A subsample of the diagrams describing the Higgs couplings to gluons determines the Higgs
couplings to a pair of photons. Since at NLO there is no gluon radiation, the LO and NLO
results for the photonic Higgs couplings are valid for the Higgs decays into photons, which are
important Higgs decay modes for the Higgs search at the LHC. Moreover, they determine Higgs
boson production at a future linear photon collider, a satellite mode of a linear e+ e collider built
up by Compton back-scattered laser light off the e beams [20,21]. The NLO QCD corrections
to the quark loops are known in the limit of a very heavy top quark [22,23] as well as in the fully
massive case [22], while the NNLO [24] and NLO SUSY-QCD [25] corrections are only known
for very heavy top and SUSY-particle masses. The results presented in this paper comprise a first
step towards a full massive SUSY-QCD calculation at NLO by means of the QCD corrections to
the squark loops.
This paper is organized as follows. In Section 2 we describe the NLO QCD calculation of the
squark loop contributions to the photonic and gluonic MSSM Higgs couplings and present the
results. Section 3 summarizes and concludes.
2. NLO QCD corrections
In order to compute the pure QCD corrections to the squark loops we need a modification of
the MSSM interaction Lagrangian allowing to separate gluon and gluino exchange contributions
in a renormalizable way. This is achieved by starting from the basic Lagrangian2
1
1
1
2
L = Ga Ga F F + ( H)2 MH
H2
4
4
2

m
H Q
+
/ mQ )Q gQ
QQH
Q(iD
v
Q
2 m2 |Q|
2 gH
|D Q|
Q
Q
m2
2H
|Q|
2 Quartic selfinteractions among the squarks are not taken into account in this work.
(6)
with the covariant derivative D = + igs Ga T a + ieA Q. Here Ga denotes the gluon field
strength tensor and Ga the gluon field accompanied by the color SU(3) generators T a (a =
1, . . . , 8), while F is the photon field strength tensor and A the photon field associated by
the electric charge operator Q. The Higgs field H represents generically either the light scalar h
H and g H are
or the heavy scalar H Higgs boson of the MSSM.3 In this work the couplings gQ
not renormalized, thus leading to a renormalizable model with strongly interacting scalars Q.
Only if gluino exchange contributions were taken into account, the full MSSM renormalization
would have to be performed. It should be noted that the gluino contributions can be separated
from the QCD corrections induced by light particles due to the gluino mass. Gluino corrections
are expected to be small [14].
The numerical results in this work will be presented for the gluophobic Higgs scenario [26],
which develops strong interference effects between quarks and squarks. It is defined by the following choices of the MSSM parameters [mt = 174.3 GeV],
MSUSY = 350 GeV,
= M2 = 300 GeV,
Xt = 770 GeV,
Ab = At ,
mg = 500 GeV,
(7)
where Xt = At / tg . We have used the program HDECAY [16] for the numerical determination of the SUSY particle masses and couplings. In this scenario the squark masses amount
to
tg = 3:
mt1 = 156 GeV,

mt2 = 517 GeV,
mb1 = 346 GeV,
mb2 = 358 GeV,
tg = 30:
mt1 = 155 GeV,

mt2 = 516 GeV,
mb1 = 314 GeV,
mb2 = 388 GeV.
(8)
The gluophobic scenario maximizes the destructive interference effects between top and stop
loop contributions to the light scalar Higgs coupling to gluons. The results of this work look
similar, whenever the squark masses turn out to be of the order of the top mass, or the Higgs
mass reaches values beyond the corresponding squark-antisquark threshold.
2.1. Scalar Higgs boson couplings to photons
At leading order (LO) the photonic MSSM Higgs couplings are mediated by top and bottom
quark loops as well as W loops for the scalar Higgs particles h, H . If the stop and sbottom masses
3 Since there are no squark loop contributions to the pseudoscalar Higgs boson couplings to photons and gluons at
leading order (LO), in this paper we will only deal with the scalar Higgs bosons h, H .
Fig. 2. MSSM Higgs boson couplings to photon pairs mediated by top- and bottom quark, stop and sbottom as well as
W boson loops at leading order.
range below 400 GeV, there are significant contributions from squark loops, too (see Fig. 2).
The LO photonic decay widths are given by [7,8,27]
LO (h/H )

3

GF 2 Mh/H
h/H h/H
h/H h/H
g
Ncf ef2 gf Af (f )
=
W AW (W ) +

3
36 2
2

2 h/H h/H
Ncf ef g A (f ) ,
+
f
(9)
h/H
AW ( ) = 2 + 3 + 3 (2 )f ( ) ,

h/H
Af ( ) = 2 1 + (1 )f ( ) ,

h/H
A ( ) = 1 f ( ) ,
f
1,
arcsin2 1
f ( ) =

1 log 1+1 i 2 < 1,

4
1 1
where we neglected contributions from charginos, charged Higgs bosons and the charged sleptons. The full expressions can be found e.g. in [7] and have been used in our work. Ncf (Ncf )
denote the color factors and ef (ef ) the electric charges of the (s)fermions in units of the proton
2 . For large loop particle masses
charge, while the scaling variables are defined as i = 4m2i /Mh/H
the form factors approach constant values,
4
2
( ) , for Mh/H
4m2f ,
3
1
h/H
2
A ( ) , for Mh/H
4m2f ,
f
3
h/H
2
2
4MW
.
AW ( ) 7, for Mh/H
h/H
Af
The reverse processes, h/H , play an important role for the MSSM Higgs search at
a photon collider at high energies [21,28]. Such a photon collider can be realized by Compton
Fig. 3. Generic diagrams for the NLO QCD corrections to the squark contributions to the photonic Higgs couplings.
back-scattering of laser light from high-energy electron beams in a linear e+ e collider [20]. In
this collider c.m. energies up to about 80% of the corresponding e+ e c.m. energy can be
reached. The s-channel Higgs boson production cross sections are then given by

dL
8 2
,
( h/H ) = 3 (h/H )
dh/H
Mh/H
(10)
2 /s
where dL /dh/H denotes the differential luminosity for h/H = Mh/H
with s being
the squared c.m. energy of the photon collider. This relation between the photon fusion cross
section and the photonic Higgs decay width holds in NLO QCD, too, since single gluon radiation
vanishes due to the conservation of color charges as well as due to the Furry theorem.
The NLO QCD corrections to the photonic Higgs decay modes and the photon fusion cross
section require the calculation of the two-loop diagrams depicted in Fig. 3. We have reduced the
5-dimensional Feynman parameter integrals to one-dimensional ones which have been integrated
Fig. 4. Real and imaginary parts of the QCD correction factor to the squark amplitudes of the two-photon couplings for
the scalar MSSM Higgs bosons. The renormalization scale of the running squark mass is taken to be = MH /2.
numerically.4 In a second calculation the QCD corrections have been obtained purely numerically. Both results agree within integration errors. The QCD corrections can be parametrized by
shifts of the quark and squark amplitudes according to

s
h/H
h/H
h/H
AQ (Q ) AQ (Q ) 1 + CQ (Q )
,

s
h/H
h/H
h/H
A (Q ) A (Q ) 1 + C (Q )
(11)
.
Q
Q
Q
The QCD corrections to the quark loops can be found in Ref. [22]. The corresponding coeffih/H
cient C (Q ) of the QCD corrections to the squark loops for finite Higgs and squark masses
Q
is presented in Fig. 4 as a function of Q . In order to improve the perturbative behavior of the
squark loop contributions they should be expressed preferably in terms of the running squark
masses mQ (MH /2), which are related to the pole masses MQ via

mQ () = MQ
s ()
s (MQ )
6
0
(12)
where 0 = 33 2NF with NF = 5 light flavors. Their scale is identified with = MH /2 within
thresholds
the photonic decay mode. These definitions imply a proper definition of the Q Q
MH = 2MQ , without artificial displacements due to finite shifts between the pole and running
squark masses, as is the case for the running MS masses. We have taken into account the LO
scale dependence of the squark masses generated by light particle contributions.
4 Analytical results can be derived, too, as in the case of the quark loops [29].
h/H
threshold and complex above. For = 1 it
Q
The coefficient C (Q ) is real below the Q
Q
Q
diverges, so that within a margin of a few GeV around the threshold the perturbative analysis
cannot be applied. This singular behavior originates from a Coulomb singularity at the threshold,
pairs can form 0++ states. This implies that the imaginary part m C h/H develops a
Q
since Q
Q
step due to Coulombic gluon exchange, thus resulting in a singular behavior of the real part at
threshold. The singular behavior can be quantified. The lowest order form factor of the squark
loops is given by
h/H,LO
( ) = 1 f ( )
Q
2
1 + i
4
for 1
(13)
where = 1 denotes the squark velocity above threshold. The QCD corrections to the
imaginary part at threshold can be derived from the Sommerfeld rescattering correction [30],
CCoul =

Z
Z
= 1 + + O s2
Z
1e
2
for Z =
4s
.
3
(14)
The QCD corrected imaginary part of the h/H couplings at threshold is now given as
2 2
s ,
(15)
3
so that the real part can be derived from a once-subtracted dispersion relation yielding the following expression in the threshold region
h/H
Q
m A
= Ccoul = +

2s
1 + i + const ,
log 1
(16)
Q
3
where the smooth non-singular constant is not relevant for the divergent behavior. Finally the
singular behavior of the QCD correction factor at threshold can be cast into the form
h/H
Q
h/H,LO
Q
h/H
Q
e C
h/H
Q
m C

8 2
log 1
1 + const,
2
Q
3( 4)
8 3
14.09.
3( 2 4)
(17)
The total size of the imaginary part and the singular behavior of the real part agree with the
numerical results depicted in Fig. 4. However, the singular behaviour at threshold is unphysical.
A proper inclusion of the finite decay widths of the virtual squarks as well as a resummation
of the Coulomb singularities [31] will regularize the divergences and improve the perturbative
results at threshold.
h/H
In the limit of heavy squark masses the coefficient C (Q ) approaches a finite and constant
Q
value5
h/H
(Q ) 3.
Q
(18)
This asymptotic value can also be derived from low-energy theorems [7,8,13,32], which are
based on the relation between the matrix elements with and without a light external Higgs boson.
5 This value differs from the heavy squark limit obtained in Ref. [25]. The difference can be traced back to a wrong
expression for the anomalous squark mass dimension used in [25].
10
This allows the derivation of an effective Lagrangian for the Higgs couplings to photons mediated
by squark loops in the heavy squark mass limit6
Leff =
e2
Q ()/
4 1 + mQ (s )
F F
H
,
v
(19)
contribution to the
where Q ()/ = (/2)[1 + 4s / + ] denotes the heavy squark Q
QED function and mQ (s ) = s / + the anomalous squark mass dimension. The NLO
expansion of the effective Lagrangian reads as
2
Leff = eQ

2

H
s
F F
1 + 3 + O s ,
8
v
h/H
-value
Q
which agrees with the C
(20)
of Eq. (18) in the heavy squark limit.
The numerical results are presented in Figs. 57 for the gluophobic Higgs scenario.7 Fig. 5
shows the partial photonic widths at NLO of the scalar Higgs bosons h, H for two values of
tg = 3, 30. The dotted lines exhibit the photonic widths without the contributions of SUSY
particles, the dashed lines include all SUSY particle contributions except those of the squarks,
while the full curves correspond to the full partial decay widths. For small as well as for large
values of tg the photonic widths range between about 0.1 and 10 keV. The importance of the
SUSY particles and in particular the squark contributions is clearly visible from the comparison
of the three different curves. The kinks, bumps and spikes correspond to the W W, t1 t1 , t t, b1 b1 ,
1 1 , 2 2 and b2 b2 thresholds in consecutive order.8 Due to the Coulomb singularities the stop
and sbottom thresholds develop spikes corresponding to the logarithmic singularities according
to Eq. (17).
The relative QCD corrections to the photonic Higgs decay widths are presented in Fig. 6 for
the two cases, in which SUSY particles have been taken into account or not. The QCD corrections
reach a size of 1020% for moderate and large Higgs masses apart from the threshold regions,
where the perturbative results are unreliable due to the Coulomb singularities. Since at a
collider the photon fusion cross section can be measured with an accuracy of a few per cent,
these corrections have to be taken into account properly. The size of the QCD corrections with
and without SUSY particle loops is of the same order of magnitude, but they can be of opposite
sign.
In order to quantify the size of squark mass effects beyond the heavy squark mass limit in the
relative QCD corrections, the ratio between the fully massive photonic decay width (h/H
) at NLO of Eqs. (9), (11) and the approximate width , where the heavy squark limit of
h/H
Eq. (18) has been inserted for the coefficient C instead of the fully massive result, is shown
Q
in Fig. 7. (The full squark mass dependence of the LO form factors has been taken into account
6 The anomalous dimension of the kinetic photon operator F F

does not contribute at NLO [9,17].
7 The massive QCD corrections to the quark and squark loops have been implemented in the program HDECAY [16].
8 In the gluophobic scenario the masses amount to m = 348 GeV, m = 356 GeV for tg = 3 and m =
1
2
1
327 GeV, m2 = 377 GeV for tg = 30.
11
Fig. 5. QCD corrected partial decay widths of the scalar MSSM Higgs bosons to two photons as functions of the corresponding Higgs masses for tg = 3 and 30. The full curves include all loop contributions, in the dashed lines the squark
contributions are omitted and in the dotted lines all SUSY particle loops are neglected. The kinks, bumps and spikes
correspond to the W W, t1 t1 , t t, b1 b1 , 1 1 , 2 2 and b2 b2 thresholds in consecutive order with rising Higgs mass. The
renormalization scale of the running quark and squark masses is chosen to be Mh/H /2, while the scale of s is taken to
be the corresponding Higgs mass.
in both expressions.) The squark mass effects reach a size of up to 30%, thus underlining the
relevance of including the full mass dependences.
2.2. Gluonic scalar Higgs decays
The gluonic Higgs boson decays h/H gg are mediated by quark and squark triangle loops
(see Fig. 1). The lowest order decay widths of the scalar MSSM Higgs boson decays are given
12
Fig. 6. Relative QCD corrections to the partial decay widths of the scalar MSSM Higgs bosons to two photons as functions
of the corresponding Higgs masses for tg = 3 and 30. The full curves include all loop contributions while in the dashed
lines the squark contributions are omitted. The kinks and spikes correspond to the W W, t1 t1 , t t, b1 b1 , 1 1 , 2 2 and
b2 b2 thresholds in consecutive order with rising Higgs mass. The renormalization scale of the running quark and squark
masses is chosen to be Mh/H /2, while the scale of s is taken to be the corresponding Higgs mass.
by [7,8]
LO (h/H gg) =
2

3
h/H h/H
GF s2 (R )Mh/H

h/H h/H
,

g
A
(
)
+
g
A
(
)
Q
Q
Q

Q
Q
36 2 3
Q
3
h/H
AQ ( ) = 1 + (1 )f ( ) ,
2
3
h/H
A ( ) = 1 f ( ) ,
Q
4
(21)
13
Fig. 7. Ratio of the QCD corrected partial decay widths of the scalar MSSM Higgs bosons to two photons including
the full squark mass dependence and those obtained by taking the relative QCD corrections to the squark loops in the
heavy mass limit as functions of the corresponding Higgs masses for tg = 3 and 30. The kinks and spikes correspond
to the W W, t1 t1 , t t, b1 b1 , 1 1 , 2 2 and b2 b2 thresholds in consecutive order with rising Higgs mass. The renormalization scale of the running quark and squark masses is chosen to be Mh/H /2, while the scale of s is taken to be the
corresponding Higgs mass.
using the same notation as for the photonic Higgs boson decays. For large loop particle masses
the form factors approach constant values,
h/H
AQ ( ) 1,
h/H
( )
Q
1
,
4
2
for Mh/H
4m2Q ,
2
4m2Q .
for Mh/H
14
The squark contributions become significant for squark masses below about 400 GeV.
The NLO QCD corrections to the gluonic decay widths are known to be large [8,1719].
They can be decomposed into the two-loop virtual corrections and the one-loop real corrections
determined by the real radiation processes
h/H ggg
and gq q.
The generic Feynman diagrams for the squark loop contributions are depicted in Figs. 8 and 9.
As in the photonic case the Feynman integrals of the virtual corrections have been reduced to
one-dimensional integrals, which have been evaluated numerically. In a second calculation the
results have been obtained purely numerically. Both results are in mutual agreement.
The strong coupling constant s is renormalized in the MS scheme, with the top quark and
squark contributions decoupled from the scale dependence, and the quark and squark masses are
renormalized on-shell. The result can be cast into the form

s
(h/H gg) = LO (h/H gg) 1 + E h/H
,
h/H h/H

gQ
)
Q
AQ (Q
95
7
7
h/H
E
=
NF + e h/H h/H
h/H h/H
4
6
2
g A (Q
)
Q gQ AQ (Q ) +
Q
Q
2
33 2NF
log 2R + E h/H ,
+
6
Mh/H
(22)
where the correction E h/H denotes the Higgs, quark and squark mass dependent part, while the
rest represents the total result in the limit of heavy loop particle masses. The latter can be derived
from low-energy theorems analogously to the photonic case. The squark contributions to the
gluonic Higgs couplings in the limit of large squark masses arise from the effective Lagrangian
[H = h, H ]
H
Leff = gQ
1 Q (s )/s a a H
G G ,
4 1 + mQ (s )
v
(23)
where Q (s )/s = (s /12)[1 + 11s /2 + ] denotes the heavy squark contribution to the
QCD function and mQ (s ) = s / + the anomalous squark mass dimension. The NLO
expansion of the effective Lagrangian reads as9

2
9 s
H s
a a H
Leff = gQ
G
G
1
+
+
O
s .
48
v
2
(24)
Including the quark contributions in the heavy quark limit, the total effective Lagrangian is given
by

gH

s a a H H
9 s
11 s
Q
Leff =
gQ 1 +
G G
+
1+
+ O s2 .
12
v
4
4
2
Q
(25)
9 The value for the QCD corrections in the heavy squark limit differs from the result obtained in Ref. [13]. The
difference can be traced back to a wrong expression for the anomalous squark mass dimension used in [13].
15
Fig. 8. Generic diagrams for the virtual NLO QCD corrections to the squark contributions to the gluonic Higgs couplings.
16
Fig. 9. Typical diagrams for the real NLO QCD corrections to the squark contributions to the gluonic Higgs decays.
The calculation based on this effective Lagrangian yields the following results for the finite parts
of the individual contributions to the coefficients E h/H ,
2
33 2NF
log 2R ,
6
Mh/H

h/H h/H

)
Q gQ
AQ
(Q
11 7
2
= +
+ e h/H h/H
,
h/H h/H
2
2
g A (Q
)
Q gQ AQ (Q ) +
Q
h/H
h/H
h/H
E h/H = Evirt + Eggg + Egq q +

h/H
Evirt
h/H
Eggg
h/H
Egq q
73
= 2 + ,
4
7
= NF ,
6
(26)
which agrees with the explicit results for the coefficients E h/H of Eq. (22) in the heavy quark
and squark limits, where E h/H vanishes.
thresholds 0++ states can form, the NLO QCD corrections exhibit Coulomb
Since at the Q Q
singularities as for the photonic Higgs couplings. The singular behavior can be derived from
the Sommerfeld rescattering corrections and leads to the following expressions at each specific
threshold,
0Q
Q
0
g h/H Ah/H ( ) 16 2 log( 1 1) + i + const

0 3( 2 4)
Q
0
0
0
Q
Q
Q
h/H
E
(27)
e
,
h/H h/H
h/H h/H
g A (Q )
Q gQ AQ (Q ) +
Q
Q
which agrees quantitatively with the numerical results.

The partial decay widths into gluons are presented in Fig. 10 with and without the squark
contributions for two values of tg = 3, 30 in the gluophobic Higgs scenario. The renormalization scale has been identified with the corresponding Higgs mass, R = Mh/H . The partial
widths range between 102 and 10 MeV. The comparison of the two curves underlines the large
size of the squark contributions, which are of the same order as the quark loops in this scenario.
17
Fig. 10. QCD corrected partial decay widths of the scalar MSSM Higgs bosons to two gluons as functions of the corresponding Higgs masses for tg = 3 and 30. The full curves include all loop contributions, while in the dashed lines
the squark contributions are omitted. The kinks, bumps and spikes correspond to the t1 t1 , t t, b1 b1 and b2 b2 thresholds in consecutive order with rising Higgs mass. The renormalization scale of the strong coupling s is chosen as the
corresponding Higgs mass.
The spikes of the full results originate from the Coulomb singularities of the t1 t1 , b1 b1 and b2 b2
thresholds in consecutive order.
Fig. 11 displays the relative QCD corrections to the gluonic Higgs decays with and without
squark contributions. Apart from the threshold singularities they are of similar size and increase
the decay widths by O(50%).
In order to quantify the squark mass effects on the relative QCD corrections, Fig. 12 depicts
the ratio between the fully massive gluonic decay widths (h/H gg) and the widths obtained
18
Fig. 11. Relative QCD corrections to the partial decay widths of the scalar MSSM Higgs bosons to two gluons as
functions of the corresponding Higgs masses for tg = 3 and 30. The full curves include all loop contributions, while
in the dashed lines the squark contributions are omitted. The kinks, bumps and spikes correspond to the t1 t1 , t t, b1 b1
and b2 b2 thresholds in consecutive order with rising Higgs mass. The renormalization scale of the strong coupling s is
chosen as the corresponding Higgs mass.
by taking the squark contributions to the coefficients E h/H in the infinite squark mass limits,
while the LO decay widths are used with the full squark mass dependence. It is clearly visible that
squark mass effects on the relative NLO corrections can reach 2030% in addition to the squark
mass dependence at LO. Thus the fully massive results are significant for a proper quantitative
prediction of the gluonic decay widths at NLO. The heavy squark mass limit turns out to be less
reliable than the heavy top mass limit for the top quark loops.
19
Fig. 12. Ratio of the QCD corrected partial decay widths of the scalar MSSM Higgs bosons to two gluons including
the full squark mass dependence and those obtained by taking the relative QCD corrections to the squark loops in the
heavy mass limit as functions of the corresponding Higgs masses for tg = 3 and 30. The kinks and spikes correspond to
the t1 t1 , b1 b1 and b2 b2 thresholds in consecutive order with rising Higgs mass. The renormalization scale of the strong
coupling s is chosen as the corresponding Higgs mass.
2.3. Gluon fusion

The gluon fusion processes gg h/H are mediated by heavy quark and squark triangle
loops with the latter contributing significantly for squark masses below 400 GeV. The lowest
order cross sections in the narrow-width approximation can be obtained from the gluonic decay
widths of the scalar Higgs bosons [7,8,33],
h/H
LO (pp h/H ) = 0
h/H
dLgg
,
dh/H
(28)
20
Fig. 13. Typical diagrams for the real NLO QCD corrections to the squark contributions to the gluon fusion processes.
h/H
h/H
0
2

h/H h/H

GF s2 (R ) h/H h/H
=
gQ AQ (Q ) +
g A (Q ) ,

Q
Q
288 2 Q
3
8Mh/H
LO (h/H gg),
(29)
2 /s with s specifying the squared hadronic c.m. energy. The LO form factors
where h/H = Mh/H
h/H
are identical to the corresponding form factors A of the gluonic decay modes, Eq. (21). The
Q/Q
gluon luminosity at the factorization scale F is defined as
dLgg
=
d
1

dx
g x, 2F g /x, 2F ,
x
g(x, 2F )
where
denotes the gluon parton density of the proton.
The NLO QCD corrections consist of the virtual two-loop corrections, corresponding to
the diagrams of Fig. 8, as well as the real corrections due to the radiation processes gg
gh/H, gq qh/H and q q gh/H . The diagrams of the real corrections are shown in Fig. 13.
While the Higgs bosons do not acquire any transverse momentum at LO, they appear at finite
transverse momenta in these radiation processes corresponding to Higgs + jet production. The
relevance of quark mass effects on the shapes of the transverse momentum distributions has
been demonstrated in Ref. [34], so that similar effects may be expected for the squark mass
dependence. The final results for the total hadronic cross sections can be split into five parts
accordingly,
(pp h/H + X)

s
dLgg
h/H
h/H
h/H
h/H
= 0
+ gg + gq + q q .
h/H
1 + C h/H
dh/H
(30)
The strong coupling constant is renormalized in the MS scheme, with the top quark and squark
contributions decoupled from the scale dependence. The quark and squark masses are renormal-
21
ized on-shell. The parton densities are defined in the MS scheme with five active flavors, i.e. the
top quark and the squarks are not included in the factorization scale dependence. The virtual coefficients split into the infrared 2 term, a logarithmic term including the renormalization scale
R and finite (s)quark mass dependent pieces ch/H (Q , Q ),
C h/H (Q , Q ) = 2 + ch/H (Q , Q ) +
2
33 2NF
log 2R .
6
Mh/H
(31)
The finite hard contributions from gluon radiation as well as gq and q q scattering depend after
mass factorization on the renormalization and factorization scales R , F and are given by
h/H
gg
1
=
h/H

2
dLgg s h/H
h/H
d
0
zPgg (z) log F + dgg (z, Q , Q )
d
s

+ 12
h/H
gq
1
=
d
1
=
h/H
z 2 z(1 z) log(1 z) ,

2F
s h/H
z
h/H
+ dgq (z, Q , Q ) ,
0
Pgq (z) log
d
2
s (1 z)2
dLgq
q,q
h/H
h/H
q q
log(1 z)
1z
dLq q
q
s h/H h/H
dq q (z, Q , Q ),
(32)
2 /
with z = h/H / = Mh/H
s , where s denotes the squared partonic c.m. energy; Pgg and Pgq are
the standard AltarelliParisi splitting functions [35]:

1
1
33 2NF
+ 2 + z(1 z) +
Pgg (z) = 6
(1 z),
1z + z
6
4 1 + (1 z)2
.
(33)
3
z
The natural scale choices turn out to be R = F = Mh/H . The quark and squark mass deh/H
pendences are contained in the kernels ch/H (Q , Q ) and dij (z, Q , Q ) in addition to the LO
Pgq (z) =
h/H
. In the heavy loop particle mass limit they reduce to simple expressions,
h/H h/H

gQ
)
Q
11
7
AQ (Q
h/H
c
(Q , Q )
+ e h/H h/H
,
h/H h/H
2
2
g
A
(Q ) +
A
( )
g
coefficients 0
Q Q
Q Q
11
h/H
dgg (z, Q , Q ) (1 z)3 ,
2
2 2
h/H
dgq (z, Q , Q ) z (1 z)2 ,
3
32
h/H
dq q (z, Q , Q ) (1 z)3 .
(34)
27
These can also be derived from the effective Lagrangian Eq. (25), so that the full calculation
agrees with the derivation from the low-energy theorems in the heavy loop particle mass limits.
22
thresholds the NLO QCD corrections develop

Q
Due to the formation of 0++ states at the Q
Coulomb singularities analogous to the photonic Higgs decay modes. The singular behavior can
be derived from the Sommerfeld rescattering corrections in the same way as in the photonic case,
threshold,
0Q
leading to the following expressions at each specific Q
0
ch/H
g h/H Ah/H ( ) 16 2 log( 1 1) + i + const

Q 0 3( 2 4)
0
0
0
Q
Q
Q
e
,
h/H h/H
h/H h/H
)
Q gQ AQ (Q ) +
Q g A (Q
Q
(35)
which agrees quantitatively with the numerical results.

The total gluon fusion cross section at NLO is displayed in Fig. 14 with and without squark
loop contributions. The renormalization and factorization scales have been identified with the
corresponding Higgs mass, R = F = Mh/H . In the gluophobic Higgs scenario the squark
loops alter the size of the cross sections by up to factors of about three. The spikes of the full
curves originate from the Coulomb singularities at the t1 t1 , b1 b1 and b2 b2 thresholds in consecutive order analogously to the photonic and gluonic Higgs couplings. It can be inferred from
Fig. 14 that squark effects are always important, if the Higgs mass exceeds the corresponding
squarkantisquark threshold. This is confirmed in particular for large values of tg , where the
top quark contribution is less important, but sizeable squark effects are visible close to and beyond the different thresholds. This feature also holds in other MSSM scenarios.
The LO and NLO cross sections are shown in Fig. 15. The QCD corrections increase the gluon
fusion cross sections by 10100%, but can be significantly larger in regions of large destructive
interferences between quark and squark loops, as is the case for very large Higgs masses for
tg = 30. The corrections are of very similar size for the quark and squark loops individually
in agreement with the results of Ref. [13]. In spite of the large corrections the residual scale
dependence is reduced from about 50% at LO to 20% at NLO and indicates a significant stabilization of the theoretical predictions. This agrees with the former results for the quark loop
contributions [8,10]. Based on the approximate NNLO and NNNLO results in the limit of heavy
top quarks a further moderate increase by less than 2030% can be expected beyond NLO for
small and moderate values of tg . For large values of tg , however, the size of the NLO corrections is moderate in regions without strong destructive interference effects between the quark
and squark loops and the scale dependence is small. This signalizes a much more reliable result
after including the NLO corrections. The residual theoretical uncertainties of our NLO results
can be estimated to less than about 20%.
The squark mass effects on the K factors are exemplified in Fig. 16, where the ratios of the
NLO cross sections are displayed, including the full mass dependence, and of the NLO cross
h/H
sections, where the coefficients ch/H (Q , Q ) and dij (z, Q , Q ) are used in the heavy squark
limits (Q ). In addition to the squark mass dependence of the LO cross sections, the K
factors develop a squark mass dependence of up to about 20%, thus supporting the relevance
of our results compared to the previous results of Ref. [13]. The squark mass effects on the K
factors turn out to be larger than the corresponding quark mass effects [9]. In addition they are
larger than the residual theoretical uncertainties and cannot be neglected in realistic analyses.
Since the gluino contributions are expected to be much smaller, the squark mass dependence
obtained in this work will be the dominant part of the differences between the heavy mass limits
and a full MSSM calculation at NLO.
23
Fig. 14. QCD corrected production cross sections of the scalar MSSM Higgs bosons via gluon fusion as functions of the
corresponding Higgs masses for tg = 3 and 30. The full curves include all loop contributions, while in the dashed lines
the squark contributions are omitted. The kinks, bumps and spikes correspond to the t1 t1 , t t, b1 b1 and b2 b2 thresholds in
consecutive order with rising Higgs mass. The renormalization and factorization scales are chosen as the corresponding
Higgs mass.
3. Conclusions
We have presented a NLO QCD calculation for the squark loop contributions to the production of neutral scalar MSSM Higgs bosons at the LHC and their decay modes into gluons and
photons. The photonic couplings of these Higgs particles play a significant role at a future photon collider, which may be built by Compton backscattering of laser light from electron beams
of a linear e+ e collider. The corrections stabilize the theoretical predictions compared to the
LO predictions. The QCD corrections turn out to be sizeable for the photonic Higgs couplings
24
Fig. 15. Production cross sections of the scalar MSSM Higgs bosons via gluon fusion as functions of the corresponding
Higgs masses for tg = 3 and 30. The full curves include the QCD corrections, while the dashed lines correspond to the
leading-order predictions. The kinks and spikes correspond to the t1 t1 , b1 b1 and b2 b2 thresholds in consecutive order
with rising Higgs mass. The renormalization and factorization scales are chosen as the corresponding Higgs mass.
and large for the gluonic Higgs decays as well as the gluon fusion processes. They increase the
latter cross sections and gluonic decay widths significantly so that the results have to be taken
into account for reliable analyses based on these processes. Despite of the large size of the NLO
corrections the approximate results in the heavy top mass limits for the gluonic decay widths and
gluon-fusion cross sections indicate sufficient perturbative convergence. The corrections beyond
NLO are expected to be of moderate size for the central scale choices. Squark mass effects on
the relative QCD corrections are sizeable and larger than the corresponding quark mass effects
for the quark loop contributions. It should be noted that squark mass effects are always relevant,
25
Fig. 16. Ratio of the QCD corrected production cross sections of the scalar MSSM Higgs bosons via gluon fusion
including the full squark mass dependence and those obtained by taking the relative QCD corrections to the squark
loops in the heavy mass limit as functions of the corresponding Higgs masses for tg = 3 and 30. The kinks and spikes
correspond to the t1 t1 , b1 b1 and b2 b2 thresholds in consecutive order with rising Higgs mass. The renormalization and
factorization scales are chosen as the corresponding Higgs mass.
if the squark masses are of the order of the top mass, or the Higgs mass is larger than the corresponding virtual squarkantisquark threshold. Our results have been implemented in the program
HIGLU [36] and can thus be applied to other MSSM scenarios, too.
Note added
During the final write-up of our work two independent papers appeared [37,38], where the
virtual corrections to the quark and squark loops of the gluon fusion processes gg h/H have
26
been derived analytically. However, a full numerical analysis of the gluon fusion processes at
NLO is not contained in these papers. We have compared our virtual corrections with the authors
of Ref. [38] and found full numerical agreement for the virtual corrections to the photonic and
gluonic Higgs couplings.
Acknowledgements
We would like to thank P.M. Zerwas for valuable comments on the manuscript. M.S. would
like to thank LAPTH for their very kind hospitality during his stay, where major parts of this
work have been performed. We are grateful to R. Bonciani for the detailed comparison of the
partial results of Ref. [38] with ours.
References
[1] P.W. Higgs, Phys. Lett. 12 (1964) 132;
F. Englert, R. Brout, Phys. Rev. Lett. 13 (1964) 321;
G.S. Guralnik, C.R. Hagen, T.W. Kibble, Phys. Rev. Lett. 13 (1964) 585.
[2] P. Fayet, Nucl. Phys. B 90 (1975) 104;
P. Fayet, Phys. Lett. B 64 (1976) 159;
P. Fayet, Phys. Lett. B 69 (1977) 489;
S. Dimopoulos, H. Georgi, Nucl. Phys. B 193 (1981) 150;
N. Sakai, Z. Phys. C 11 (1981) 153;
K. Inoue, A. Kakuto, H. Komatsu, S. Takeshita, Prog. Theor. Phys. 67 (1982) 1889;
E. Witten, Nucl. Phys. B 231 (1984) 419.
[3] See e.g. G. Degrassi, S. Heinemeyer, W. Hollik, P. Slavich, G. Weiglein, Eur. Phys. J. C 28 (2003) 133, hepph/0212020.
[4] M. Spira, P.M. Zerwas, hep-ph/9803257;
M. Gomez-Bock, M. Mondragon, M. Mhlleitner, R. Noriega-Papaqui, I. Pedraza, M. Spira, P.M. Zerwas, J. Phys.
Conf. Ser. 18 (2005) 74, hep-ph/0509077.
[5] H. Haber, G. Kane, Phys. Rep. 117 (1985) 75.
[6] A. Djouadi, J. Kalinowski, P. Ohmann, P.M. Zerwas, Z. Phys. C 74 (1997) 93, hep-ph/9605339.
[7] M. Spira, Fortschr. Phys. 46 (1998) 203, hep-ph/9705337;
A. Djouadi, hep-ph/0503172;
A. Djouadi, hep-ph/0503173.
[8] M. Spira, A. Djouadi, D. Graudenz, P.M. Zerwas, Phys. Lett. B 318 (1993) 347;
M. Spira, A. Djouadi, D. Graudenz, P.M. Zerwas, Nucl. Phys. B 453 (1995) 17.
[9] M. Krmer, E. Laenen, M. Spira, Nucl. Phys. B 511 (1998) 523;
M. Spira, hep-ph/9703355.
[10] A. Djouadi, M. Spira, P.M. Zerwas, Phys. Lett. B 264 (1991) 440;
S. Dawson, Nucl. Phys. B 359 (1991) 283;
D. Graudenz, M. Spira, P.M. Zerwas, Phys. Rev. Lett. 70 (1993) 1372;
R.P. Kauffman, W. Schaffer, Phys. Rev. D 49 (1994) 551;
S. Dawson, R.P. Kauffman, Phys. Rev. D 49 (1994) 2298.
[11] R.V. Harlander, W.B. Kilgore, Phys. Rev. Lett. 88 (2002) 201801;
R.V. Harlander, W.B. Kilgore, JHEP 0210 (2002) 017;
C. Anastasiou, K. Melnikov, Nucl. Phys. B 646 (2002) 220;
V. Ravindran, J. Smith, W.L. van Neerven, Nucl. Phys. B 665 (2003) 325.
[12] S. Moch, A. Vogt, Phys. Lett. B 631 (2005) 48, hep-ph/0508265;
V. Ravindran, Nucl. Phys. B 746 (2006) 58, hep-ph/0512249;
V. Ravindran, Nucl. Phys. B 752 (2006) 173, hep-ph/0603041.
[13] S. Dawson, A. Djouadi, M. Spira, Phys. Rev. Lett. 77 (1996) 16.
27
[14] R.V. Harlander, M. Steinhauser, Phys. Lett. B 574 (2003) 258, hep-ph/0307346;
R.V. Harlander, M. Steinhauser, Phys. Rev. D 68 (2003) 111701, hep-ph/0308210;
R.V. Harlander, M. Steinhauser, JHEP 0409 (2004) 066, hep-ph/0409010;
R.V. Harlander, F. Hofmann, JHEP 0603 (2006) 050, hep-ph/0507041.
[15] J.A. Aguilar-Saavedra, et al., ECFA/DESY LC Physics Working Group, hep-ph/0106315.
[16] A. Djouadi, J. Kalinowski, M. Spira, Comput. Phys. Commun. 108 (1998) 56, hep-ph/9704448.
[17] T. Inami, T. Kubota, Y. Okada, Z. Phys. C 18 (1983) 69.
[18] K.G. Chetyrkin, B.A. Kniehl, M. Steinhauser, Phys. Rev. Lett. 79 (1997) 353.
[19] P.A. Baikov, K.G. Chetyrkin, Phys. Rev. Lett. 97 (2006) 061803, hep-ph/0604194.
[20] I.F. Ginzburg, G.L. Kotkin, S.L. Panfil, V.G. Serbo, V.I. Telnov, Nucl. Instrum. Methods 219 (1984) 5.
[21] B. Badelek, et al., ECFA/DESY Photon Collider Working Group, Int. J. Mod. Phys. A 19 (2004) 5097, hepex/0108012;
M.M. Mhlleitner, P.M. Zerwas, Acta Phys. Pol. B 37 (2006) 1021, hep-ph/0511339.
[22] A. Djouadi, M. Spira, J. van der Bij, P.M. Zerwas, Phys. Lett. B 257 (1991) 187;
A. Djouadi, M. Spira, P.M. Zerwas, Phys. Lett. B 311 (1993) 255;
K. Melnikov, O. Yakovlev, Phys. Lett. B 312 (1993) 179;
M. Inoue, R. Najima, T. Oka, J. Saito, Mod. Phys. Lett. A 9 (1994) 1189;
M. Spira, hep-ph/9504339.
[23] H. Zheng, D. Wu, Phys. Rev. D 42 (1990) 3760;
S. Dawson, R.P. Kauffman, Phys. Rev. D 47 (1993) 1264.
[24] M. Steinhauser, Report MPI-PHT-96-130, in: Proceedings Ringberg Workshop on The Higgs PuzzleWhat can
we learn from LEP2, LHC, NLC and FMC?, Ringberg, Germany, 1996, hep-ph/9612395.
[25] A. Djouadi, V. Driesen, W. Hollik, J.I. Illana, Eur. Phys. J. C 1 (1998) 149, hep-ph/9612362.
[26] M. Carena, S. Heinemeyer, C.E.M. Wagner, G. Weiglein, Eur. Phys. J. C 26 (2003) 601, hep-ph/0202167.
[27] J.R. Ellis, M.K. Gaillard, D.V. Nanopoulos, Nucl. Phys. B 106 (1976) 292;
M.A. Shifman, A.I. Vainshtein, M.B. Voloshin, V.I. Zakharov, Sov. J. Nucl. Phys. 30 (1979) 711, Yad. Fiz. 30 (1979)
1368.
[28] M.M. Mhlleitner, M. Krmer, M. Spira, P.M. Zerwas, Phys. Lett. B 508 (2001) 311, hep-ph/0101083;
M.M. Mhlleitner, hep-ph/0008127;
D.M. Asner, J.B. Gronberg, J.F. Gunion, Phys. Rev. D 67 (2003) 035009, hep-ph/0110320;
P. Niezurawski, hep-ph/0503295;
P. Niezurawski, A.F. Zarnecki, M. Krawczyk, hep-ph/0307180;
P. Niezurawski, A.F. Zarnecki, M. Krawczyk, in: the Proceedings of 2005 International Linear Collider Workshop
(LCWS 2005), Stanford, CA, 1822 March 2005, pp. 0112, hep-ph/0507006;
M. Spira, P. Niezurawski, M. Krawczyk, A.F. Zarnecki, hep-ph/0612369.
[29] J. Fleischer, A.V. Kotikov, O.L. Veretin, Nucl. Phys. B 547 (1999) 343, hep-ph/9808242;
R. Harlander, P. Kant, JHEP 0512 (2005) 015, hep-ph/0509189.
[30] A. Sommerfeld, Atombau und Spektrallinien, Vieweg, Braunschweig, 1939.
[31] M.B. Voloshin, Sov. J. Nucl. Phys. 36 (1982) 143, Yad. Fiz. 36 (1982) 247;
V.S. Fadin, V.A. Khoze, JETP Lett. 46 (1987) 525, Pisma Zh. Eksp. Teor. Fiz. 46 (1987) 417;
V.S. Fadin, V.A. Khoze, Sov. J. Nucl. Phys. 48 (1988) 309, Yad. Fiz. 48 (1988) 487;
K. Melnikov, M. Spira, O.I. Yakovlev, Z. Phys. C 64 (1994) 401, hep-ph/9405301.
[32] B.A. Kniehl, M. Spira, Z. Phys. C 69 (1995) 77, hep-ph/9505225.
[33] H.M. Georgi, S.L. Glashow, M.E. Machacek, D.V. Nanopoulos, Phys. Rev. Lett. 40 (1978) 692.
[34] R.K. Ellis, I. Hinchliffe, M. Soldate, J.J. van der Bij, Nucl. Phys. B 297 (1988) 221;
U. Baur, E.W.N. Glover, Nucl. Phys. B 339 (1990) 38;
O. Brein, W. Hollik, Phys. Rev. D 68 (2003) 095006, hep-ph/0305321;
U. Langenegger, M. Spira, A. Starodumov, P. Trb, JHEP 0606 (2006) 035, hep-ph/0604156.
[35] G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298.
[36] M. Spira, hep-ph/9510347.
[37] C. Anastasiou, S. Beerli, S. Bucherer, A. Daleo, Z. Kunszt, JHEP 0701 (2007) 082, hep-ph/0611236.
[38] U. Aglietti, R. Bonciani, G. Degrassi, A. Vicini, JHEP 0701 (2007) 021, hep-ph/0611266.
Random walks of Wilson loops in the screening regime

P.V. Buividovich a, , M.I. Polikarpov b
a JIPNR, National Academy of Science, Acad. Krasin str. 99, 220109 Minsk, Belarus
b ITEP, B. Cheremushkinskaya str. 25, 117218 Moscow, Russia
Received 22 August 2007; accepted 31 August 2007

Abstract
Dynamics of Wilson loops in pure YangMills theories is analyzed in terms of random walks of the
holonomies of the gauge field on the gauge group manifold. It is shown that such random walks should
necessarily be free. The distribution of steps of these random walks is related to the spectrum of string
tensions of the theory and to certain cumulants of YangMills curvature tensor. It turns out that when
colour charges are completely screened, the holonomies of the gauge field can change only by the elements
of the group center, which indicates that in the screening regime confinement persists due to thin center
vortices. Thick center vortices are also considered and the emergence of such stepwise changes in the limits
of infinitely thin vortices and infinitely large loops is demonstrated.
PACS: 12.38.Aw; 05.40.Fb
1. Introduction
Wilson loops are the most popular order parameters for YangMills theories without dynamical quarks. Wilson loop WR [C] is defined as the expectation
value of the trace of the holonomy
of the gauge field over the loop C: WR [C] = TrR P exp(i C dx A ) [1]. Interaction potential
VR (r) between two static colour charges which transform under some irreducible representation R of the gauge group can be measured using the Wilson loop WR [Crt ], where Crt is
a rectangular loop of size r t with t . As the Wilson loop is in fact the amplitude of
propagation of static colour charges along the loop C [1], it is related to the potential VR (r) as
E-mail addresses: buividovich@tut.by (P.V. Buividovich), polykarp@itep.ru (M.I. Polikarpov).

doi:10.1016/j.nuclphysb.2007.08.013
P.V. Buividovich, M.I. Polikarpov / Nuclear Physics B 790 (2008) 2841
29
WR [Crt ] = exp(VR (r)t). It is known that in the confining phase of the theory static colour
charges are connected by a thick chromoelectric string with constant tension R , which gives
rise to linear interaction potential VR (r) = R r at sufficiently large distances. Correspondingly,
for sufficiently large loops Wilson loops WR [C] decay exponentially with the minimal area S[C]
of the surface spanned on the loop C, i.e. WR [C] = exp(R S[C]). Contribution of charge selfenergies shows up in the dependence of WR [C] on the perimeter of the loop C.
Thus confining properties of the theory are characterized by the whole spectrum of string
tensions R . The dependence of R on the representation of the gauge group gives some additional information on the properties of YangMills theory. For instance, lattice simulations show
that at intermediate distances (0.21 fm for SU(3) gauge group) R is proportional to the eigenvalue C2R of the second-order Casimir operator in the representation R (Casimir scaling) [2,3].
Casimir scaling is naturally explained in the framework of stochastic vacuum models [4]. At
very large distances Casimir scaling should be violated because of screening of colour charges
by gluons [5]. The reason is that when colour charges are separated by sufficiently large distance, it costs less energy to create a bound state of a colour charge and some number of gluons
than to create confining string between bare colour charges. For instance, when SU(N ) colour
charges are completely screened, string tension depends only on the N -ality of the representation
[57] and is proportional to the lowest eigenvalue of the second-order Casimir operator among
all irreducible representations with the same N -ality.
An interesting description of the spectrum of string tensions of YangMills theories was
proposed recently in [810], where all information about Wilson loops WR [C] in irreducible
representations of the gauge group was encoded in a single function on the gauge group manifold, namely, the probability distributionp[g; C] of the holonomies of the gauge field over the
loop C. The holonomy g[C] = P exp(i C dx A ) is defined only modulo gauge transformations g[C] hg[C]h1 , which implies that p[g; C] should be invariant under such transformations, i.e. p[g; C] = p[hgh1 ; C]. Thus p[g; C] should be defined as [810]:

p[g; C] = c g, g[C] ,
(1)

where c (g, g ) = R R (g)R (g ) is the delta-function on the group classes and R (g) are
the group characters. Using the character expansion of c (g, g ), one can immediately express
p[g; C] in terms of the Wilson loops WR [C]:

p[g; C] =
(2)
R (g)WR [C].
R
Conversely, all Wilson loops can be calculated if the probability distribution p[g; C] is known:
WR [C] = dg R (g)p[g; C]. It can be shown that p[g; C] is the Wilson loop in the regular
representation of the gauge group [10,11].
An advantage of such description is that the evolution of p[g; C] can be interpreted in terms
of a random walk of the holonomy g[C] on the gauge group manifold, the properties of this
random walk being directly related to the spatial distribution of non-Abelian magnetic flux. In a
particular configuration of gauge fields, the holonomy g[C] moves along some path on the group
manifold as the area of the loop is gradually increased. When one calculates expectation values
of Wilson loops in quantum theory and sums over all field configurations, weighted sum over
all such paths which start in the group identity and end in the element g yields the probability
distribution p[g; C]. In the case of pure YangMills theory in Euclidean spacetime all field
configurations and, consequently, all such paths are summed with non-negative weights, therefore such weighted sums over paths on the group manifold can be described as random walks
30
Fig. 1. Small increment of the area of the minimal surface spanned on the loop C.
on Lie groups [12,13]. The position of random walker corresponds then to the holonomy g[C]
modulo gauge transformations g[C] hg[C]h1 . If the loop C is slightly deformed in such a
way that the area of the minimal surface increases from S[C] to S[C ] = S[C] + S (see Fig. 1),
the holonomy g[C] changes as:

g[C ] = g[C] 1 + ih(x0 , x)F (x)h(x, x0 )S
(3)
y
where h(x, y) = P exp(i x dx A ) is the multiplicative integral of the gauge field over the
segment of the loop C bounded by the points x and
point x0 is the initial point used
x0 y. The
A ). The combination F (x) =
to calculate the holonomy g[C]: g[C] = P exp(i C;x
dx
0
h(x0 , x)F (x)h1 (x, x0 ) is usually called the shifted curvature tensor [4]. Thus each step of
the random walk of g[C] corresponds to a small change of non-Abelian magnetic flux through
the loop C.
One of the basic observations of [810] is that when Casimir scaling holds, the probability
distribution p[g; C] satisfies the free diffusion equation:
d
p[g; C] = 0 p[g; C]
dS[C]
(4)
where is the Laplace operator on the group manifold and 0 is some constant. Random walk
which is described by (4) is simply the Brownian motion on the group manifold, which consists of
a large number of small statistically independent random steps. Correspondingly, the distribution
of non-Abelian flux through the loops of intermediate sizes is almost random, which fits nicely
in the model of stochastic vacuum [4].
Free diffusion equation (4) and the corresponding random walk should be somehow modified
in order to reproduce screening effects. In [8,9] random walk in ZN -symmetric potential was
considered and it was shown that diffusion in such potential reproduces transition from Casimir
scaling to complete screening due to mixing of different representations with equal N -alities. An
advantage of such approach is that one can solve the diffusion equation in terms of path integral
and obtain the effective action for Wilson loops. However, this approach is to a large extent
phenomenologicalin particular, the choice of the potential which breaks SU(N ) symmetry of
the equation down to ZN is to a large extent arbitrary and affects only the transition from Casimir
scaling to screening.
In [11] the most general form of the diffusion equation for p[g; C] was obtained using the
classical loop equations [14] and the cumulant expansion theorem [15]. It turned out that this
diffusion equation should have the form of the so-called KramersMoyall cumulant expansion,
which describes a free random walk with an arbitrary distribution of steps. Free random walk
is understood here as a random walk with position-independent step distribution. Thus exact
equation for p[g; C] cannot include potential or drift terms.
31
The aim of this paper is to consider the most general free random walk on the group manifold
and to relate its properties to the spectrum of string tensions of YangMills theory. Although
in general it is not possible to find the effective action which describes a free random walk, the
distribution of steps of the random walk of g[C] has a more natural physical interpretation than
an external potential in the diffusion equation. In particular, it will be shown that the distribution
of steps of the random walk changes dramatically when Casimir scaling changes to screening.
While the free diffusion equation (4) reflects stochasticity of YangMills vacuum at small and
intermediate distances, random walk which describes diffusion of Wilson loops in the screening
regime can only consist of discrete jumps by the elements of the group center, which can be
naturally explained in terms of thin center vortices [16,17,21]. It will be also demonstrated that
in the case of center vortices with finite thickness the holonomy g[C] can change by any group
element, but the distribution of steps is still peaked near the elements of the group center, the
widths of the peaks being proportional to vortex thickness and inversely proportional to the area
S[C] of the minimal surface spanned on the loop.
The structure of the paper is the following: in Section 2 a general case of a free random walk
of g[C] on the group manifold is considered. The distribution of steps of such random walk is
expressed in terms of the spectrum of string tensions of the theory and in terms of the cumulants
of the shifted curvature tensor F . In Section 3 the random walk of g[C] in the screening
regime is investigated and it is shown that in the case of complete screening the holonomy g[C]
is only allowed to change by the elements of the group center. In Section 4 such discrete jumps
are interpreted in terms of thin center vortices. Vortices of finite thickness are considered in
Section 5, where it is shown how such jumps by the elements of the group center arise in the
limits of infinitely thin vortices or infinitely large loops.
2. Free random walks on the group manifold
In [11] it was shown that the most general form of the diffusion equation for the probability
distribution of the holonomies g[C] on the gauge group manifold is:

d
a1 ...ak [C]a1 ak p[g; C]
p[g; C] =
dS[C]
(5)
k=2
where a1 ...ak [C] are some coefficients which can in general depend on the loop C and a are
the generators of left shifts on the group manifold. The action of a is defined by the identity
(1 + a a )f (g) = f ((1 + i a Ta )g), where a are arbitrary infinitely small parameters, Ta are the
generators of the gauge group and f (g) is an arbitrary function on the group manifold. The term
with first-order derivative is prohibited in (5) by gauge invariance. It was shown in [11] that the
a ,
coefficients a1 ...ak [C] are directly related to the cumulants of the shifted curvature tensor F
which are the basic objects in the method of field correlators [4]. This relation can be formally
written as [11]:

(1)k d
dS 1 1 (x1 ) dS k k (xk )
a1 ...ak [C] =
k! dS[C]
S[C]
S[C]

Fa11 1 (x1 ) Fakk k (xk )
(6)
where the integration is performed over the surface of the minimal area S[C] spanned on the
a (x) transforms unloop C. It should be noted that for any point x the shifted curvature tensor F
der gauge transformations as the curvature tensor in some fixed point x0 , therefore the cumulants
32
a (x) are well-defined when all x , . . . , x are different. Gauge invariance implies only that
of F
1
k
a (x) and the coefficients a1 ...ak [C] should be colour singlets. In particular,
the cumulants of F
all a1 ...ak [C] with odd k should vanish.
Eq. (5) is a general equation which describes an arbitrary free random walk on the group
manifold, i.e. a random walk with position-independent distribution of steps. If P [g ; C]S is
the probability that g[C] changes to g[C ] = g g[C] as the area of the surface spanned on the
loop C increases from S[C] to S[C ] = S[C] + S, the evolution of the probability distribution
p[g; C] is described by the following equation:

d
p[g; C] = dg P g g 1 ; C p[g ; C] p[g; C] dg P [g ; C]
dS[C]

(7)
= dg P g g 1 ; C (g , g) dg P [g ; C] p[g ; C]
where (g , g) is the delta-function on the group manifold. Although the distribution of steps in
(7) can in general depend on the loop C, this dependence can be neglected if the Wilson area
law holds exactly and if the spectrum of string tensions R is known. Using the definition (1) the
following integral equation can be obtained for p[g; C]:

d
R (g)R WR [C] =
R (g) (R) dg p[g ; C]R (g )
p[g; C] =
dS[C]
R
R

= dh dg
dR R ghg 1 h1 (R)p[g ; C]

=

dR R gg 1 (R) p[g ; C]
dg
(8)

where the identity dhR (ghf h1 ) = dR1 R (g)R (f ) and the invariance of p[g; C] w.r.t.
gauge transformations g[C] hg[C]h1 were used. Comparing Eqs. (8) and (7), one can finally express the step distribution P [g; C] in terms of string tensions R :

P [g; C] =
(9)
dR R R (g).
R
If string tensions depend on the distance between colour charges, for instance, because of screening, P [g; C] should also depend on the loop C. This dependence will be investigated in the
Section 5 using the model of thick center vortices. It is also interesting to note that the positivity of P [g; C] for g = 1 is some constraint on possible spectrum of string tensions of the theory,
which follows from the positivity of path integral weight for pure YangMills theory in Euclidean
spacetime.
Eq. (5) is simply the gradient expansion of the general equation (7). Indeed, p(g ) in (7)
can be replaced by p[g g 1 g; C] = exp( a (gg 1 )a )p[g; C], where the function a (g) is
defined by the following identity:

exp i a (g)Ta = g.
(10)
Expanding the exponential in powers of a (gg 1 ), one can express the coefficients a1 ...ak [C]
in terms of the step distribution P [g; C]:

(1)k
dg P [g; C] a1 (g) ak (g).
a1 ...ak [C] =
(11)
k!
33
As all a1 ...ak [C] are colour singlets, the distribution of steps should satisfy P [hgh1 ; C] =
P [g; C].
For example, if Casimir scaling holds, R = 0 C2R , P [g; C] = 0 (g, 1) and only the coefficient ab [C] = 0 ab is not equal to zero. Correspondingly, only the second-order cumulant
of the shifted curvature tensor is nonzero, which corresponds to the limit of Gaussian-dominated
stochastic vacuum [4].
3. Complete screening and discrete jumps on the group manifold
In the case of complete screening string tensions R depend only on the N -ality of the
representation R: R = (R ). N -ality R of the representation R characterizes its transfor2i
mation properties w.r.t. the elements of the group center ZN = {e N k }, k = 0, 1, . . . , N 1:
R (gz) = R (g)zR , z ZN . A simple calculation shows that if R = (R ), P [g; C] is given
by a finite sum over the elements of the group center:

P [g; C] =
(12)
(z)(g, z)
zZN
where (z) is minus the Fourier transform of the spectrum of string tensions w.r.t. the group
center:
(z) = N 1
N1

()z .
(13)
=0
Thus when colour charges are completely screened, the evolution of the probability distribution
p[g; C] is described by the following equation:

d
(z)p[zg; C].
p[g; C] =
dS[C]
(14)
zZN
This equation implies that the random walk of g[C] consists of jumps by the elements of
the group center only. (z) with z = 1 is the probability of a jump by z per unit area. As the
string tension
between charges with zero N -ality should vanish due to complete screening by
gluons [5], zZN (z) = 0, which is simply the conservation of probability flow in the language
of random walks.
Thus the distribution of steps of the random walk of g[C] is completely different in Casimir
scaling and in screening regimes. When Casimir scaling holds, random walk of g[C] consists of a
large number of statistically independent small steps, which indicates that non-Abelian magnetic
flux which penetrates the loop C fluctuates randomly. On the other hand, Eq. (14) implies that
the cumulants in (6) are saturated by some singular field configurations.
4. Thin center vortices
It is straightforward to guess what is the structure of singular field configurations which correspond to discrete jumps described by (14). By definition the holonomy g[C] changes by the
element of the group center when the loop C is crossed by center vortex [16]. As follows from
Eq. (14), g[C] changes stepwise, which can only be explained if the vortices are mathematically
thin surfaces which are distributed in space with some finite density. Correspondingly, (z)S is
34
the probability to cross thin center vortex which carries magnetic flux z as the area of the loop is
increased from S[C] to S[C] + S.
The statement about the singular field configurations can be formulated more precisely using
Eqs. (6) and (11), which relate the step distribution P [g; C], the coefficients a1 ...ak [C] and the
cumulants of the shifted curvature tensor of YangMills fields. The coefficients a1 ...ak [C] can
be calculated if the field configurations which contribute to the cumulants in (6) are known. On
the other hand, these coefficients can be found from the step distribution P [g; C]. In this section
it will be shown that the coefficients a1 ...ak [C] obtained from the distribution (12) agree with
the result obtained from (6) if center vortices are infinitely thin non-interacting random surfaces.
This analysis will be extended in the next section, where center vortices of finite thickness will be
considered and the emergence of the step distribution (12) in the limit of infinitely thin vortices
will be demonstrated.
In order to calculate the coefficients
a1 ...ak [C] for the step distribution (12), one should cal
culate the integrals of the form dg (g, z) a1 (g) ak (g). Such integrals, however, cannot be
calculated by direct substitution g z, as for the elements of the group center the choice of
a (g) is not unique. In fact, the points g = z with z = 1 are the only points on the group manifold where the function a (g) is not uniquely defined. Indeed, as exp(i a (z)Ta ) = z, for any
group element h one has h exp(i a (z)Ta )h1 = exp(i a (z)hTa h1 ) = exp(i b (z)Oba (h)Ta ) =
a
hzh1 =
are the matrices of the adjoint representation of the gauge group). The in z (here Ob (h)
tegrals dg (g, z) a1 (g) ak (g) can nevertheless be calculated if the delta-functions
in (12)

are defined as the limits
of
some
smooth
functions,
which
immediately
yields
dg
(g,
z)

a1 (g) ak (g) = dh Oba11 (h) b1 (z) Obakk (h) bk (z). It is convenient to introduce the nota
a
tion a1 (z) ak (z)h = dh Oba11 (h) b1 (z) Obkk (h) bk (z), so that the coefficients a1 ...ak [C]
which correspond to the step distribution (12) can be written as:
a1 ...ak [C] =

(1)k
(z) a1 (z) ak (z) h .
k!
(15)
zZN
Eqs. (6) and (15) imply that in this case the integrals of the cumulants of the shifted curvature
tensor should have the following form:

dS 1 1 (x1 ) dS k k (xk ) Fa11 1 (x1 ) Fakk k (xk )

S[C]
S[C]

(z)S[C] a1 (z) ak (z) h .
(16)
zZN
On the other hand, the integrals in (16) can be calculated using the exact expression for the
field strength of thin center vortex [18]:

a
a
F (x) = (z) d (y) 4 (x y),
(17)
where is the vortex surface. The choice of a (z) in (17) is also not unique, and it is usually assumed that physical observables should be averaged over all possible a (z) [19,20]. It is
convenient now to introduce the linking number of the vortex surface and the loop C:

dS (x) d (y) 4 (x y).

L[C; ] =
(18)
S[C]
35
The linking number L[C; ] counts how many times the surface winds around the loop C.
Using the expressions (17), (18) and averaging over all possible vortex configurations and over
all a (z), one can rewrite the integrals in (16) as:


S[C]
S[C]

Lk [C; z ] a1 (z) ak (z) ,
(19)
zZN
where z denotes the union of the surfaces of all vortices which carry magnetic flux z. As
only the cumulants of even order are different from zero, all indices a1 , . . . , ak can be pairwise
contracted in both (19) and (16). As for all possible a (z) the squares a (z) a (z) are equal,
this gives equal constant factors in front of both expressions. After that direct comparison of
(19) and (16) shows that all even-order cumulants of the linking number L[C; z ] are equal:
L2k [C; z ] = (z)S[C], L2k+1 [C; z ] = 0. This property is enough to recover the probability distribution of the linking number L[C; z ]it is distributed as the difference of two
Poisson-distributed positive integer numbers k+ and k with mean values (z)S[C]/2 each. The
numbers k+ and k can be readily interpreted as the numbers of vortices which wind around C
leftwards and rightwards. Explicit expression for the probability distribution of L[C; z ] is:

p L[C; z ] = exp (z)S[C]

k+ k =L[C;z ]
((z)S[C]/2)k+ +k
.
k+ !k !
(20)
The fact that k+ and k are distributed by Poisson implies that thin vortices do not interact,
since all intersections of vortex surfaces z and the loop C are statistically independent events.
Recent results of lattice simulations indicate that center vortices are indeed mathematically thin
random surfaces which are distributed in space with finite density vort 24 fm2 [21]. A typical
vortex configuration on the lattice consists of a single percolating vortex which stretches through
the whole physical space plus a large number of small vortices, whose sizes do not exceed several
lattice spacings [21,22]. As such small vortices can only lead to perimeter-dependent effects, the
existence of the percolating vortex is crucial for the area dependence of the expectation values
k+ = k = (z)S[C]/2. The results obtained in [23] also give some preliminary indications
that the effective interaction between thin vortices is rather small. Thus the picture of thin noninteracting center vortices may be a good approximation for the low-energy dynamics of pure
YangMills theories.
5. Thick center vortices
The analysis presented in the previous section is based on the assumption that thin center
vortices are the only field configurations which contribute to the cumulants of the shifted curvature tensor in (6), which is justified if Eq. (14) is exact for loops of finite size. However,
since complete screening is only an asymptotic law which holds in the limit of very large loops
with S[C] , (14) could also be the limiting case of some more general diffusion equation
which takes into account the contribution of other field configurations. In order to derive such
more general equation, one should make some model-dependent assumptions on the form of the
cumulants of the shifted curvature tensor, so that the coefficients a1 ...ak [C] can be calculated
explicitly. In this section the diffusion equation (7) will be analyzed under the assumption that
36
the dominating field configurations in the vacuum of YangMills theory are thick center vortices,
which roughly corresponds to the picture of spaghetti vacuum [19,20]. Although the assumptions and approximations made below by no means capture all properties of the theory, they are
enough to demonstrate how Eq. (14) emerges in the limit of infinitely thin vortices or in the limit
of very large loops. Vortices of finite thickness can be considered as an effective description of
the interference between thin center vortices which were observed in lattice simulations [21] and
other field configurations, as far as such picture of the vacuum of YangMills theory correctly
reproduces the spectrum of string tensions.
Field strength of a thick center vortex can be obtained by smoothing the delta-function in
(17) [18]:

a
F
(21)
(x) = a (z) d (y)f (x y)
where f (x y) is some smooth function which decays at distances of order of the vortex thickness lvort .
Proceeding as in the previous section, one can calculate the integrals of the cumulants of the
shifted curvature tensor for the field configurations (21), assuming that vortex surfaces z are
random. However, for thick center vortices the linking number L[C; ] cannot be rigorously
defined. Its role is now played by the integral of the following form:

L[C; ] =
(22)
dS (x) d (y)f (x y).
S[C]
Just as in (19), the cumulants of the shifted curvature tensor can be expressed in terms of
L[C;
]:


S[C]
S[C]

k [C; z ] a1 (z) ak (z) .
L
(23)
zZN
On the other hand, the integrals in (23) can be related to the coefficients a1 ...ak [C] in (5) using
Eq. (6). These coefficients can be also obtained from P [g; C] using Eq. (11). Comparing (6), (23)
and (11), one can obtain an equation which relates the cumulants of L[C;
z ] and the averages
a
of (g):

k [C; z ] a1 (z) ak (z) .
S[C] dg P [g; C] a1 (g) ak (g) =
(24)
L
zZN
This relation can be used to find the step distribution P [g; C] if the distribution of L[C;
z ]
is known. L[C; z ] is not in general integer-valued, but it takes integer values when all thick
vortices which contribute to the integral (22) are completely inside the loop C. Deviations from
integer values occur if some vortices are only partially within the loop. If the loop is pierced
by k+ right-winding vortices and k left-winding vortices, the average number of such vortices
can be estimated as L[C;

z ] (k+ + k )lvort P [C]/S[C], where P [C] is the perimeter of the
loop C. For sufficiently large loops and sufficiently small vortex densities one can assume that
L[C;
z ] and k+ , k are also distributed by Poisson, i.e. that the interaction between vortices
37
Fig. 2. The probability distribution of L[C;

z ] for different values of (z)S[C] and [C]: (z)S[C] = 3, [C] = 0.05
for plot 1, (z)S[C] = 1, [C] = 0.05 for plot 2, (z)S[C] = 3, [C] = 0.1 for plot 3.
can be neglected. As a nearest extension of the results obtained in the previous section, one can
try to smear the original distribution of linking numbers (20) in such a way that the probability
z ]) is still peaked around integer values, the widths of the peaks being
distribution p(L[C;
proportional to L[C; z ]. Without loss of generality the shape of the peaks can be approximated
by a Gaussian distribution. The resulting probability distribution is:

p L[C;
z ] = exp (z)S[C]
((z)S[C]/2)k+ +k

G L[C;
z ] (k+ k ), (k+ + k )[C]
k+ !k !
k+ ,k
(25)
where [C] = lvort P [C]/S[C] is assumed to be small and G(x, 2 ) = (2 2 )1/2 exp(x 2 /
(2 2 )) is the Gaussian distribution with dispersion . Such probability distribution for some
values of (z)S[C] and [C] is plotted on Fig. 2. It turns out that for the distribution (25) the
generating function for the cumulants can be found explicitly:
W (q, z) =
+

(iq)k
k=1
k!

k [C; z ] = (z)S[C] cos q exp [C]q 2 1 .
L
(26)
For comparison, for the distribution (20) W (q, z) = (z)S[C](cos q 1).

For illustration, the step distribution P [g; C] which solves (24) will be found for SU(2) gauge
group. For SU(2) group z = 1, 2 (1) = 0 and 2 (1) = 2 if the generators Ta are the Pauli
matrices. For the cumulants of even order the indices a1 , . . . , a2k can be again pairwise contracted. As P [g; C] is the function on the group classes, it depends only on the absolute value of
a (g). Taking into account that the Haar measure on SU(2) group classes is dg = 2/ sin2 d ,
one can rewrite the relation (24) as:

2/S[C]
0

d sin2 P [ ; C](/)2k = L 2k [C; 1 ] .
(27)
38
It is also convenient to multiply (27) by (1)k q 2k /(2k)! and to sum over all k, which yields:

d sin2 P [ ; C] cos(q/) = W (q, 1).
2/S[C]
(28)
Using the expression (26) and performing the inverse Fourier transform w.r.t. the variable q =
k, the step distribution P [g; C] can be finally found:
sin2 P [ ; C] =
+

cos(k ) exp 2 k 2 [C] cos(k) 1 ,
(29)
k=
where = (1).
For sufficiently small (C) the function sin2 P [ ; C] (which is actually the probability distribution of steps over the coordinate ) looks like a Gaussian peak near = . There is also
a -function singularity
at = 0, which is irrelevant for diffusion. The width of the peak near
= is proportional to lvort P [C]/S[C]. Thus assuming finite thickness of vortices smears the
-functions in (14) and allows steps by any elements of the gauge group. For a fixed loop C the
width of the step distribution is proportional to vortex thickness, but if the vortex thickness lvort
is kept fixed, for large loops deviations from Eq. (14) tend to zero as P [C]/S[C] S[C]1/2 ,
thus asymptotically screening is recovered even for vortices of finite size, as it should be [17,24].
The spectrum of string tensions which corresponds to the step distribution (29) can be found
using Eq. (9). As follows from (9),string tensions R are proportional to the expectation values
of the characters R (g): R = dR1 dg P [g; C]R (g). Direct calculation gives to the first order
in [C]:
k = 2 2/3k(k + 1)[C],
k = 2/3k(k + 1)[C],
k = 1/2, 3/2, . . . ,
k = 0, 1, 2, . . . .
(30)
Remembering that [C] = lvort P [C]/S[C], one can conclude that the difference between thin and
thick vortices shows up only in perimeter-dependent effects, which can be taken into account if
the step distribution P [g; C] is allowed to depend on the loop C. Since perimeter dependence
of Wilson loops corresponds to the contribution of charge self-energies, it follows from (30) that
the energy required to screen a colour charge in SU(2) representation with spin k is proportional
to k(k + 1), i.e. to the second-order Casimir operator.
It is interesting to note that Casimir scaling of string tensions can also be formally described
using the expression (28), although the validity of such description should be discussed separately. In [24] it was shown that Casimir scaling at intermediate distances and asymptotic
screening can be obtained for YangMills vacuum dominated by center vortices if the field distribution a (z)f (x) inside each center vortex is random at some scale lc lvort . In this case the
distribution of L[C;
z ] should be completely different from the distributions (20) and (25) if
the size of the loop is comparable with the vortex thickness lvort . By virtue of the Central Limit
Theorem it should look like a Gaussian distribution with the dispersion f 2 lc2 S[C], where f 2
is the dispersion of f (x). For such distribution W (q) = f 2 lc2 S[C]q 2 , and Eq. (28) yields the
following result:
sin2 P [ ; C] = 2 f 2 lc2
+

k=
k 2 cos(k ) = f 2 lc2
d2
( ).
d 2
(31)
39
The distribution described by (31) is singular, but the string tensions k can nevertheless be
calculated using Eq. (9):
2
k =
dk
=
0
2
2f lc2
d sin2 P [ ; C]k ( )
d 2 sin(2k + 1)
8f 2 lc2
=
k(k + 1).
2k + 1 0 d 2
sin
3
lim
(32)
Thus such rather general assumptions concerning the distribution of the modified linking number
L[C;
z ] lead to Casimir scaling at intermediate distances and to screening at asymptotically
large distances, in full accordance with the results obtained in [24].
Finally, it should be noted that as the step distribution (29) is nonzero on the whole group
manifold, it still describes discontinuous changes of the holonomy g[C]. Step distribution which
describes continuous changes of the holonomy g[C] in nonsingular gauge fields can only contain
singularities of the form a1 ak (g, 1) with finite k, as in (31). In order to obtain such step
distribution for the model of thick center vortices, some more advanced assumptions on the
distribution of linking numbers and on the distribution of fields inside vortices should be made. In
particular, one could take into account that interactions between vortices should lead to deviations
from Poisson-like distribution of the linking number. Although it could be extremely interesting
to investigate such effects, this is not the aim of this paper. The derivation presented above is only
intended to illustrate how Eq. (14) can arise as the limiting case of some more general diffusion
equation and how the spectrum of string tensions can change in this case.
6. Conclusions
In this paper a phenomenological analysis of [810] was extended to the case of a general
free random walk on the group manifold basing on the exact results obtained in [11]. It was
shown that the distribution of steps of a random walk of the holonomy g[C] is directly related
to the spectrum of string tensions of the theory and also to certain vacuum expectation values
of the curvature tensor of YangMills fields. In some sense the spectrum of string tensions of
the theory is a dual image of the spatial distribution of non-Abelian magnetic flux, just as the
amplitude of particle scattering in an external potential is proportional to the Fourier transform
of this potential. The description of the dynamics of Wilson loops in terms of random walks on
the gauge group manifold makes this correspondence more apparent.
The relation between the step distribution of the random walk of g[C] and the spectrum of
string tensions was used to analyze the random walk of Wilson loops in the case of complete
screening, when string tension R depends only on the N -ality of the representation R. It turned
out that when colour charges are completely screened, the holonomy g[C] can only change by the
elements of the group center, which implies that the only field configurations which contribute to
the cumulants of the shifted curvature tensor are thin center vortices [16]. Complete screening,
however, is only an asymptotic property, and Eq. (14) can as well be the S[C] limit of some
more general diffusion equation. This possibility was investigated for the case of thick center vortices in YangMills theory with SU(2) gauge group. It was shown that indeed for loops of finite
size Eq. (14) can only be valid if center vortices are infinitely thin, but in the limit S[C]
it is recovered even for vortices of finite thickness. For sufficiently small vortex sizes or for sufficiently large loops the distribution of steps P [g; C] is a Gaussian-like peak near g = 1 with
40
Fig. 3. Time dependence of 1/2 Tr g for a particular implementation of a random walk on SU(2) group manifold with
external potential V (g) = Tr g 2 .
the width [C] = lvort P [C]/S[C]. Of course these conclusions are model-dependent, but at least
qualitatively the emergence of jumps by the elements of the group center can be understood.
It could be also interesting to study how Eq. (14) is modified due to contribution of other field
configurations such as instantons or monopoles.
It seems that the most important property of random walks on Lie groups which lead to asymptotical screening-like effects is the possibility of fast jump-like transitions between points
on the group manifold which differ by the elements of the group center. For instance, if the
free diffusion equation (4) is modified by introducing external ZN -symmetric potential V (g),
as proposed in [8,9], asymptotically stable probability distribution p(g) exp(V (g)) is also
ZN -symmetric and has N equal maxima at some points {gz}, z ZN . Random walker spends
most time in the vicinity of such points and can only travel from one such point to another in
a relatively short time. This effect is somewhat similar to the restoration of classically broken
symmetries due to tunneling. In order to illustrate this example, random walk on SU(2) group
with the external potential V (g) = Tr g 2 was simulated. Time dependence of 1/2 Tr g for a
particular implementation of such random walk is plotted on Fig. 3. It can be seen that indeed
the random walker mostly wanders near the points g = 1, while the transitions between the
regions close to these points are very fast and rare. It is reasonable to conjecture that such transitions simulate the discrete jumps described by Eq. (14) and lead to asymptotic screening [8,9].
It could be very interesting to investigate whether such jumps could emerge in some explicitly
SU(N )-symmetric dynamical system. It could be also useful to consider some generalizations of
a free random walk described by Eq. (7). For instance, one could consider random walks with
memory, which can be more appropriate for the description of domain-like structure of Yang
Mills vacuum [4,19,20,24].
Acknowledgements
P.V. Buividovich is grateful to all members of the ITEP lattice group (ITEP, Moscow) for support and stimulating discussions, and especially to E. Luschevskaya for her kind hospitality. Illuminating discussions with V.I. Zakharov are also acknowledged. M.I. Polikarpov was partially
supported by grants RFBR-05-02-16306a, RFBR-0402-16079a, RFBR-0602-04010-NNIOa and
41
EU Integrated Infrastructure Initiative Hadron Physics (I3HP) under contract RII3-CT-2004506078. P.V. Buividovich was partially supported by the grant of ICTP office of external activities
for his participation in the 7th international conference Symmetry in non-linear mathematical
physics (Kyiv, 2430 June 2007), where a part of this work was reported.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
K.G. Wilson, Phys. Rev. D 10 (1974) 2445, http://prola.aps.org/abstract/PRD/v10/i8/p2445_1.

G.S. Bali, Phys. Rev. D 62 (2000) 114503, hep-lat/0006022.
S. Deldar, Phys. Rev. D 62 (2000) 034509, hep-lat/9911008.
A. Di Giacomo, H.G. Dosch, V.I. Shevchenko, Y.A. Simonov, Phys. Rep. 372 (2002) 319, hep-ph/0007223.
G. Mack, Phys. Lett. B 78 (1978) 263.
J. Greensite, M.B. Halpern, Phys. Rev. D 27 (1983) 2545.
M.R. Douglas, S.H. Shenker, Nucl. Phys. B 447 (1995) 271, hep-th/9503163.
A.M. Brzoska, F. Lenz, J.W. Negele, M. Thies, Phys. Rev. D 71 (2005) 034008, hep-th/0412003.
G. Arcioni, S. de Haro, P. Gao, Phys. Rev. D 73 (2006) 074508, hep-th/0511213.
P.V. Buividovich, V.I. Kuvshinov, Phys. Lett. B 634 (2006) 262, hep-th/0602154.
P.V. Buividovich, V.I. Kuvshinov, Phys. Rev. D 73 (2006) 094015, hep-th/0605207.
N. Varopoulos, Canad. J. Math. 46 (1994) 438.
Y. Guivarch, Development of Mathematics 19502000, Birkhuser, Basel, 2000, pp. 577608.
A.M. Polyakov, Gauge Fields and Strings, Harwood Academic Publishers, 1987.
N.G. van Kampen, Stochastic Processes in Physics and Chemistry, North-Holland, Amsterdam, 1981.
G. t Hooft, Nucl. Phys. B 138 (1978) 1.
L. Del Debbio, M. Faber, J. Greensite, S. Olejnik, Phys. Rev. D 55 (1997) 2298, hep-lat/9708023.
M. Engelhardt, H. Reinhardt, Nucl. Phys. B 567 (2000) 249, hep-th/9907139.
J. Ambjrn, P. Olesen, Nucl. Phys. B 170 (1980) 60.
H.B. Nielsen, P. Olesen, Nucl. Phys. B 160 (1979) 380.
F.V. Gubarev, A.V. Kovalenko, M.I. Polikarpov, S.N. Syritsyn, V.I. Zakharov, Phys. Lett. B 574 (2003) 136, heplat/0212003.
[22] V.G. Bornyakov, P.Y. Boyko, M.I. Polikarpov, V.I. Zakharov, Nucl. Phys. B 672 (2003) 222, hep-lat/0305021.
[23] P.V. Buividovich, M.I. Polikarpov, Center vortices as rigid strings, Report ITEP-LAT/2007-08 (2007), arXiv:
0705.3745.
[24] J. Greensite, K. Langfeld, S. Olejnik, H. Reinhardt, T. Tok, Phys. Rev. D 75 (2007) 034501, hep-lat/0609050.
Master field treatment of metric perturbations sourced

by the trailing string
Steven S. Gubser, Silviu S. Pufu
Joseph Henry Laboratories, Princeton University, Princeton, NJ 08544, USA
Received 2 July 2007; accepted 27 August 2007
Abstract
We present decoupled, separable forms of the linearized Einstein equations sourced by a string trailing
behind an external quark moving through a thermal state of N = 4 super-YangMills theory. We solve these
equations in the approximations of large and small wave-numbers.
1. Introduction
In [1], using the AdS/CFT correspondence [24] (for a review see [5]), the expectation
value Tmn of the stress tensor of N = 4 super-YangMills theory at finite temperature was
calculated in the presence of an external quark moving a constant speed v. The calculation is
based on the trailing string configuration of [6,7] (see also [8] for a closely related development,
[9] for earlier work in a somewhat similar vein, and, for example, [10] for a brief survey of other
recent literature on related topics). Conceptually, all that one requires in order to calculate Tmn
is a solution of the linearized Einstein equations in the AdS5 -Schwarzschild background, sourced
by the trailing string, with infalling boundary conditions imposed at the horizon and additional
boundary conditions, appropriate to the absence of deformations of the Lagrangian of N = 4
gauge theory, imposed at the boundary of AdS5 -Schwarzschild. In practice, the calculation is
somewhat tedious because there are fifteen linearized Einstein equations. In [1] these equations
were somewhat streamlined from their original form in axial gauge, where all metric perturbations with an index in the direction orthogonal to the boundary are set to zero. However, the most
E-mail address: spufu@princeton.edu (S.S. Pufu).

doi:10.1016/j.nuclphysb.2007.08.015
S.S. Gubser, S.S. Pufu / Nuclear Physics B 790 (2008) 4271
43
interesting set of equations in [1], namely the one that determines T00 , still involves four coupled second order differential equations for four unknown functions, together with three first
order constraints. The complicated form of these equations makes them hard to work with, challenging to check independently, and difficult to extend to more general settingsfor example,
backgrounds where conformal invariance is broken or the motion of the quark is not uniform.
There exists in the GR literature a highly developed master field formalism, dating back to
such works as [1113], whose aim is to obtain fully decoupled and separable forms of the linearized Einstein equations, as well as equations for other fields with spin, in curved backgrounds
with some symmetry. As a toy version of the calculation, consider a massive scalar, with action

1
1
S = d 5 x g ()2 m2 2 + J ,
(1)
2
2
in the curved background

ds 2 = a(r)2 h(r) dt 2 + d x2 +
dr 2
.
h(r)a(r)2
(2)
The first step is to consider a specific Fourier mode of and J in the x directions, with wave The equation of motion for this mode is
number k.

1
1
k2
t2 + 3 r a 3 f r 2 m2 = J,
(3)
f
a
a
and we have defined f (r) = h(r)a(r)2 . The corresponding master field and master
where k = |k|
equation are an equivalent way of writing (3):

1 2
(2 Vm )m = t + r f r Vm m = Jm = a 3/2 J,
f
m = a 3/2 ,
Vm (r) = m2 +
k2 3 a 3 a 2
3 a
+
+
f
+
f
f,
4 a2
2 a
a2 2 a
(4)
where primes denote d/dr and

1
2 = t2 + r f r
f
(5)
is the Laplacian on the orbit spacetime

ds22 = f dt 2 +
1 2
dr .
f
(6)
The first aim of this paper is to reduce the linearized Einstein equations, sourced by the trailing
string, to five master equations, each one similar to (4). The homogeneous parts of these equations are well known, having been worked out in full in [14] following earlier work including
[15,16]. The source terms were partly worked out in [17], whose methods we follow.1 With an
eye toward future applications to non-conformal backgrounds, we derive the master equations
in the general warped background (2) dual to a finite-temperature field theory on R3,1 , without
making assumptions about the stress tensor that supports the five-dimensional curvature.
1 Our methods are also reminiscent of those used in [18] to study sound waves in N = 2 gauge theory via a gaugeinvariant formalism.
44
With the master equations in hand, we specialize to the case of the trailing string in AdS5 Schwarzschild and reproduce some of the results of [1]. Also we consider an analytic approximation for the large k region, following methods developed in [19]. We explain in Appendix A
how the parameterization of metric perturbations used in this paper relates to the axial gauge
parameterization used in [1].
The current work has some overlap with the independent study [20].
2. Decoupling the linearized Einstein equations
The first step in formulating the master equations is to expand the metric and stress tensor perturbations into Fourier harmonics involving scalar-, vector-, and tensor-valued functions on R3 .
As we describe in Section 2.1, the scalar case involves nothing but plane waves, and the other
two cases involve plane waves times appropriate polarization tensors. The coefficients of these
harmonics are functions of t and r which in general have some tensor structure in the orbit
space (6). We refer to terms in the Fourier expansions as scalar, vector, and tensor perturbations
according to the type of harmonic involved rather than their tensor structure in the orbit space:
indeed, one finds that the scalar perturbations are expressed in terms of a symmetric tensor in
the orbit space; the vector perturbations can be expressed in terms of a vector in the orbit space;
and the tensor perturbations can be expressed in terms of an orbit-space scalar. In the case of
tensor perturbations, the orbit-space scalar is (up to an overall factor) the master field T , as
described in Section 2.3. In the case of vector perturbations, the orbit-space vector can be expressed in terms of an orbit-space scalar V by use of one of the linearized Einstein equations,
as described in Section 2.4. Likewise, in the case of scalar perturbations, the orbit-space tensor
can be expressed in terms of an orbit-space scalar S , as described in Section 2.5. Each master
field satisfies a second order master equation, which implies the linearized Einstein equations not
already used in constructing the master field.2 There is an intrinsic parity for tensor and vector
perturbations which affects the polarization tensors but not the form of the master equation. Thus
the five master fields are Teven , Todd ,Veven , Vodd , and S . The corresponding master equations
(without reference to parity) are given in (35), (42), and (55).
2.1. Scalar, vector, and tensor harmonics
The defining equations for scalar, vector, and tensor harmonics on R3 are
i

i + k 2 S = 0,

i
i + k 2 Vj = 0,

i
i + k 2 Tj h = 0,
(7)
j Vj = 0,
(8)
Tj h = 0 = T j ,
j
(9)
where indices are raised and lowered using the standard metric on R3 .
The scalar harmonics can be chosen to be pure exponentials:

x) = ei kx .
S(k,
(10)
2 We do not know of a general theorem that master fields satisfying second-order master equations can always be
found.
They satisfy the normalization condition

S(k ) = (2)3 3 (k k ),
S(k),
45
d 3 x S1 (
x )S2 (
x ).
S1 , S2
(11)
R3

Vector harmonics may be assumed to be proportional to ei kx , but for each choice of wavenumber k there are two distinct solutions, which are conventionally specified by their transforma which preserves both the defintion property under the parity transformation x
x , k k,

and k = k 2 + k 2 ,
ing equation (8) and the exponential part of the harmonic. Denoting k = |k|
2
one may express the solutions as

1 i k
2
e x k
k1 k 2
kk
x
) = 1 ei k
Vodd
( 0 k3 k2 ).
i (k, x
k
x) =
(k,
Veven
i

k1 k 3 ,
(12)
These vector harmonics (12) satisfy the normalization conditions
even
Veven (k ) = (2)3 3 (k k ) = Vodd (k),
Vodd (k ) ,
V (k),
even
Vodd (k ) = 0
V (k),
(13)
under the inner product

j
x )V2 (
x ).
V1 , V2 = d 3 x V1,j (
(14)
R3
It is convenient for us to make the x 1 direction privileged, because our eventual aim is to describe
the emission from an external quark moving in the x 1 direction through a thermal plasma of
x ) on R3 admits the Fourier expansion
N = 4 gauge theory. A general vector field Xi (

d 3 k even even
odd (k,
x) + XS (k)S
i (k,
x) ,
Xi (
(15)
XV (k)Vi (k, x) + XVodd (k)V
x) =
i
3
(2)
X odd (k),
and XS (k),
where
for some set of Fourier coefficients XVeven (k),
V
x) = 1 i S(k,
x).
Si (k,
(16)
k
The expansion (15) is orthonormal with respect to the inner product (14).

Tensor harmonics may also be assumed to be proportional to ei kx , and there are again two
distinct polarization tensors with definite parity:
2
k1 k2
k1 k3
k
k 2 k 2 +k 2 k 2
2k 2 +k 2
x
k1 k2 1 k22 3 k2 k3 1k 2 ,
x) = 1 ei k
(
k,
Teven
(17)
ij
k2
2
k12 k32 +k 2 k22
2k12 +k
k1 k3 k2 k3 2
2
k
k
0
k
k
x
k3
) = 1 ei k
Todd
ij (k, x
k
k2
3
2k1 k2 k3
2
k
k1 (k32 k22 )
2
k
2
k1 (k32 k22 )
2
k
2k1 k22 k3
k
(18)
46
With the inner product

1
jh
T1 , T2 =
d 3 x T1,j h (
x )T2 (
x)
2
(19)
R3
we have the normalization condition

even
Teven (k ) = (2)3 3 (k k ) = Todd (k),

Todd (k ) ,
T (k),
even
Todd (k ) = 0.
T (k),
(20)
R3
A general symmetric tensor field Xij (

x ) on
admits the Fourier decomposition

d 3k S
x) + X S (k)S
ij (k,
x)
Xij (
X (k)ij S(k,
x) =
T
(2)3 L
even (k,
x) + X odd (k)V
odd (k,
x)
+ XVeven (k)V
ij
V
ij

even (k,
x) + X odd (k)T
odd (k,
x) ,
+ XTeven (k)T
ij
T
ij
(21)
where
) = 1 (i Veven ,
) = 1 (i Vodd ,
Vodd
Veven
ij (k, x
ij (k, x
j)
j)
k
k
x) = 1 i j S + 1 ij S,
Sij (k,
3
k2
(22)
and we use the notation (ij ) = 12 (ij + j i). The decomposition (21) is orthogonal with respect to
the inner product (20), but not orthonormal because of some k-independent factors arising from
the inner products of the derived harmonics in (22).
2.2. Fourier decomposition of the perturbations
The background (2) is, by assumption, a solution of the five-dimensional Einstein equations,
1
R RG + G = T .
(23)
2
The explicit cosmological term is redundant because it could have been soaked into T ; however, retaining it explicitly makes it easier to apply the formalism to an anti-de Sitter space
example. T includes any contributions from bulk scalar fields or other matter. Let the back(0)
ground metric and stress tensor be denoted G(0)
and T . Consider a perturbation of both the
metric and the stress tensor:
G = G(0)
+ h ,
(0)
T = T
+ ,
(24)
where is a formal expansion parameter which we eventually set to 1.

Let indices b and c run over t and r, while indices i and j run over the three x directions. The
metric perturbations can be decomposed into three functions hbc which are scalars on R3 , two
vector-valued functions hbi , and one tensor-valued function hij . These in turn may be expanded
in Fourier modes as indicated in (15) and (21):

d 3k S
x),
hbc (t, r, x) =
(25)
f (k, t, r)S(k,
(2)3 bc
47
d 3k S
x)
f (k, t, r)Si (k,
(2)3 b

t, r)Veven (k,
x) + f V ,odd (k,
t, r)Vodd (k,
x) ,
+ fbV ,even (k,
i
i
b

d 3k S
2
x) + H S (k,
t, r)Sij (k,
x)
H (k, t, r)ij S(k,
hij (t, r, x) = 2a(r)
T
(2)3 L
t, r)Veven (k,
x) + H V ,odd (k,
t, r)Vodd (k,
x)
+ HTV ,even (k,
ij
ij
T

t, r)Teven (k,
x) + H T ,odd (k,
t, r)Todd (k,
x) ,
+ HTT ,even (k,
ij
ij
T
hbi (t, r, x) = a(r)
(26)
(27)
where the factors of a(r) are chosen for convenience. Likewise one may expand

d 3k S
bc (t, r, x) =
(k)S(k, x),
(2)3 bc

d 3k S
x)
bi (t, r, x) = a(r)
(k, t, r)Si (k,
(2)3 b

t, r)Veven (k,
x) + V ,odd (k,
t, r)Vodd (k,
x) ,
+ bV ,even (k,
i
i
b

d 3k S
x) + S (k,
t, r)Sij (k,
x)
p (k, t, r)ij S(k,
ij (t, r, x) = 2a(r)2
(2)3
t, r)Veven (k,
x) + V ,odd (k,
t, r)Vodd (k,
x)
+ V ,even (k,
ij
ij

t, r)Teven (k,
x) + T ,odd (k)T
k,odd
(
x) .
+ T ,even (k,
ij
ij
(30)
It is also useful to express a general vector v as

d 3k S
x),
v (k, t, r)S(k,
vb =
(2)3 b

d 3 k V ,even
2
x) + v V ,odd (k,
t, r)Vodd (k,
x)
vi = a(r)
v
(k, t, r)Veven
(k,
i
i
(2)3

t, r)Si (k,
x) .
+ vVS (k,
(31)
(28)
(29)
The diffeomorphism symmetry of the Einstein equations may be expressed as

v G = Lv G = ( v + v ),

v T = Lv T = v T + T v + T v ,
(32)
where v is an arbitrary vector, and for convenience we have quantified the smallness of the
coordinate deformation in terms of the same formal expansion parameter that we used in the
metric expansion (24). Thus, at linearized level, (32) becomes
v h = (0) v + (0) v ,
(0)
(0)
(0)
v = v T
+ T
v + T
v .
(33)
For the purpose of formulating master equations for metric perturbations, we do not need to
know details about how the zeroth order stress tensor arises from matter fields. Instead, we may
(0)
extract T in terms of the unperturbed metric using the zeroth order Einstein equations:
(0) (0) (0) (0) (0)
1
(0)
(0)
(0)
= R
R (0) G(0)
T
+ G = diag Ttt , Trr , Txx , Txx , Txx ,
2
a 2
3 a
a
(0)
Ttt = f 3 2 f 2
ff 3 f 2 ,
2a
a
a
48
Trr(0) =
a 2 3 a f
+3 2 +
,
f
2a f
a
1
(0)
= a 2 + f a 2 + 2aa f + 2aa f + a 2 f .
Txx
2
(34)
2.3. Master equation for the tensor modes

Decoupled equations of motion for the tensor modes may be found directly by plugging the
expansions (27) and (30) into the linearized Einstein equations. The result is the same for even
and odd modes, so we will simply omit to specify parity in the following. The master equation is
(2 VT )T = JT ,
(35)
where
T = a 3/2 HTT ,
VT = 2 +
JT = 2a 3/2 T ,
k 2 11 a 11 a 2
11 a
+
f+
f +
f + f
2
2
2 a
4 a
2 a
a
(36)
and 2 = f1 t2 + r f r , as in (5).
2.4. Master equation for the vector modes
The results for vector modes are the same for even and odd parity, so we will not refer to parity
explicitly in the rest of this section. As explained in [14,17], the first step is to consider quantities
that are invariant under diffeomorphisms. To this end, consider the transformation properties of
vector components of the metric and tensor perturbations, as can be derived by substituting the
expansions (25)(31) into (33):
fbV fbV + ab vV ,
HTV HTV kvV ,
(0)
(0)
Txx
kTxx
V V 2 vV .
b vV ,
a
a
One may form diffeomorphism-invariant combinations as follows:
bV bV +
(0)
(37)
(0)
a
Txx
Txx
bV = bV +
V = V 2 HTV .
b HTV ,
fbV = fbV + b HTV ,
k
ka
a
If the fV are expressed in terms of the master field as follows3 :
(38)

2a
1
t fVt = V 2 r a 3/2 V ,
k
a
V
fVr = ,
a
(39)
then the bc and ij components of the linearized Einstein equations are satisfied automatically.
The tj components lead to an equation for V involving up to third order derivatives in r, while
the rj components lead to a second order equation for V . In deriving these equations, we found
3 Two degrees of freedom, namely fV and fV , can be reduced to one, namely , because of a first order constraint
V
r
t
that would emerge if we substituted general fbV into the linearized Einstein equations. This is explained, for example, in
Section IV.B.2 of [17].
49
it convenient to pass to a gauge where HTV = 0, and to assume harmonic time dependence, eit ,
for fbV , bV , V , and V (which are now indistinguishable from the corresponding unhatted
quantities). A crucial point is that the third order equation for V follows from the second order
equation plus a relation that follows from the conservation of the stress tensor. More explicitly,
from the O() term in the equation

(0)
(0) + (1) T
(40)
+ = 0
one may derive the relation in question:
(0)
(0)
1 b 4 V
k 2 + 2Txx V r Txx
(41)
a
=
+
V
b
ka
a4
a 5/2
where Db is the covariant derivative with respect to the orbit spacetime (6).
The master equation for the vector modes is just the second order equation following from
the rj Einstein equations:
(2 VV )V = JV ,
(42)
where
k 2 5 a 23 a 2
5 a
+
+
f
+
f
f + f ,
2
2
2
a
4
2
a
a
a

a V
V
r f
JV = 2 af r +
kf
VV = 2 +
(43)
and 2 is again the Laplacian in the two-dimensional orbit spacetime (6).

2.5. Master equation for the scalar modes
For scalar modes, there is no longer a notion of parity. The diffeomorphism transformations
of the metric perturbations are
S
S
S
fbc
fbc
+ 2D(b vc)
,
HTS HTS kvVS ,
k
fbS fbS vbS + ab vVS ,
a
k
a
HLS HLS + vVS + f vrS .
3
a
(44)
Because of the appearance of vbS without derivatives in the transformation of fbS and of vVS
without derivatives in the transformation of HTS , these quantities may be gauged away or used to
S and H S :
construct diffeomorphism-invariant variants of fbc
L
S
S
= fbc
+ 2D(b Xc) ,
fbc
1
a
H LS = HLS + HTS + f Xr
3
a
(45)
where

a
a
S
S
Xb =
f + b HT
k b
k
(46)
is invariant under diffeomorphisms with only vVS non-zero. Also,

(0)
(0)
S
S
bc
+ v S,d d Tbc + 2Td(b c) v S,d ,
bc
50
k (0)
1 (0) S
bS bS Tbd v S,d + Txx
b vV ,
a
a
k (0) S
vV ,
S S 2 Txx
a
k (0) S
f
(0)
p S p S + 2 Txx
vV + 2 vrS r Txx
3a
2a
and the gauge-invariant quantities are
(0)
(47)
(0)
S
S
= bc
+ X d d Tbc + 2Td(b c) X d ,
bc
k (0)
1 (0)
b HTS ,
bS = bS Tbd X d + Txx
a
ka
1 (0) S
HT ,
S = S 2 Txx
a
1 (0) S
f
(0)
p S = p S + 2 Txx
HT + 2 XrS r Txx
.
3a
2a
(48)
In manipulating the linearized Einstein equations, we employed a gauge where fbS = 0 and
HTS = 0. One may however pass immediately to diffeomorphism-invariant formulas by replacing
S , H S , S , S , and S by their diffeomorphism-invariant relatives as defined in (45) and (46),
fbc
L
b
bc
and we will quote all formulas in diffeomorphism-invariant form. One may eliminate H LS by
noting that the ij Einstein equations with i = j that is, the off-diagonal equations in the lowerright 3 3 blocklead directly to
1
2a 2 S
S
H LS = Gbc
(0) fbc 2 .
2
k
(49)
S , and using
The tr, rr, ti, and ri Einstein equations involve only first order derivatives of fbc
stress tensor conservation equations one may show that the other Einstein equations are implied
S as
by these four. Approximately following [14] we express the symmetric tensor fbc
f
fttS = (2X Y ),
3a
i
ftrS =
Z,
af
1
frrS =
(2Y + X).
3af
(50)
Then the first order equations just mentioned reduce to

(0)
a
a
f
1
2 k 2 2Ttt

Y+
+ 2
Z = X ,
X X+ 2
a
a
f
f
f
f
a
f
f
2
Z + X = Z ,
X+
Y + 2 Z = Y ,
2f
2f
f

f 2 2
f 2 k2
4
a
3 a
a 2
+
f+
ff X + 2 +
2f
f 3 2 f 2 3 ff Y
2 +
4
3
2a
4
3
a
a
a

2

a
k f
(0) f
+ 32
(51)
+ Ttt
Z = C ,
2
a
2 a
f
Y
where

2 S 2a 2 S 2a 2
f

+ 2 2ar 3a + 10a S ,
X = a tr
i
k r
f
k
51
2a 2 S 8a 2
+ 2 (ar + 2a ) S ,
k r
k
2
2a S 8a 3 S
Z =
+ 2 ,
ik t
k

aff S
6aa f 2 S a 3
8k 2
6f a f S
S
C =
tr + 2af 2 rr
r + 2 122 + 3f 2 2 f
.
i
k
a
k
a
(52)
Y =
Because the system of Eqs. (51) consists of three first order differential equations and one algebraic constraint, it can be reduced to a single second order differential equation satisfied by any
desired linear combination of X, Y , and Z. In order to put this equation in a form that resembles
the other master equations, we define the scalar master field as

a
S = A X + Y + 3 Z ,
(53)
a
where
A
a 1/2
,
H 1/2 (H + 3af a )1/2
9
H k 2 + 2a 2 + 3f a 2 + a f .
2
(54)
The master equation then reads:

(2 + WS r VS )S = JS ,
(55)
with 2 = f1 t2 + r f r , as in (5), and

WS =

3f
4aa + 9f a 2 + 4f a a + 5af a + 3aa f .
2H
(56)
The expressions for VS and JS are too long to be reproduced here. In the general warped
background, VS includes a term 3aa t2 /H in addition to non-derivative terms. This term,
as well as WS , vanish when we specialize to the case where the unperturbed metric (2) is
4 /r 4 , where L is related to the cosmoAdS5 -Schwarzschild, i.e. when a = r/L and h = 1 rH
2
logical constant by = 6/L . In this case, the master equation (55) takes the standard form
(2 VS )S = JS with
4

4
4
4
12k 2 L4 r 2 rH
9r 13rH
+ k 4 L8 r 4 r 4 9rH
4k 6 L12 r 6
4

8
4
5r + 3rH
,
108rH

2
4
4
2r 4 + 4rH
24rH
f
3
a
2

+
C +
JS = 3/2
(X + Y ) + a X + Y + Z
r
a H f2
H L6 r
f L4 r

4
4)
72rH
3(r 4 + 5rH
H
+
+
Z ,
6
2
4
2
f
HL r
fL r
VS =
4L10 r 8 H 2
H = k2 +
4
6rH
.
L4 r 2
(Note that H as quoted in (57) is the same one defined in (54).)
(57)
52
3. The trailing string in AdS5 -Schwarzschild

Now, as promised, we turn to an application of the master field formalism to computations
of Tmn as sourced by the trailing string configuration in AdS5 -Schwarzschild. In Section 3.1
we work out the stress tensor in five dimensions created by the trailing string, in the Fourierexpanded forms (28)(30). The resulting expressions may be inserted into the master equations
(35), (42), and (55) to obtain explicit differential equations in r for the radial dependence of the
master fields, which we give explicitly in (69)(73). In Sections 3.33.5 we obtain asymptotic
forms near the horizon and near the boundary of AdS5 -Schwarzschild for the solutions of the
master equations. As explained in Section 3.2, the asymptotic forms near the boundary translate
into the desired quantities Tmn in the gauge theory.
3.1. The stress tensor of the trailing string
In Poincar coordinates, the AdS5 -Schwarzschild metric can be realized as a particular case
of (2) with
r4
r
(58)
,
h(r) = 1 H4 .
L
r
Here, rH denotes the location of the black hole horizon, and the AdS radius L is related to the
cosmological constant term via
a(r) =
6
.
L2
(59)
It is useful to note that the horizon temperature is T = rH /L2 . The equations of motion follow
from the action

G(R + 12/L2 )
1
2
5
d
x
det
g
X
(
)
,
S = d 5x
(60)
2
252
where g is the worldsheet metric. In static gauge, the worldsheet metric that minimizes the
action is, to leading order, given by [6,7]

X (t, r) t r X 1 (t, r) 0 0 ,
g G X X ,

L2 v
r + rH
r irH
+ i log
X 1 (t, r) = vt + (r), (r) =
(61)
log
.
4irH
r + irH
r rH
We can use the above expressions to compute the source term T that appears in the Einstein
equations (23). Because AdS5 -Schwarzschild is the solution to the Einstein equations with negative cosmological constant but vanishing stress-energy tensor of the matter fields, T comes
only from the trailing string: that is,
T =
h+v 2 r 4 /r 4
H
h2
2
rH v 2
L2 h

52
1
L5 3 i
i
=
x X
v

5
2
2 1 v r
0
0
r 2 v2
LH2 h
r4
(v 2 h)
L4
r2 v
LH2
0
0
v
h
r2 v
LH2
v2
0
0
0
0
0
0
0
, (62)
0
0
0
53
where we have set the formal expansion parameter = 1. The notation 3 (x i X i ) is short for
(x 1 vt (r))(x 2 )(x 3 ). The expression (62) (with indices lowered) can then be decomposed
as in (28)(30), with the coefficients of the decomposition given by
aV ,odd = V ,odd = T ,odd = 0,
(63)
3k12 )
,
4k 2 L2
r 2 v 2 (k 2
r 2v2
,
S = T
6L2
2 r2
v 2 k
iv 2 k1 k r 2
T ,even
V ,even = T
,
=
T
,
k 2 L2
4k 2 L2
S

V ,even

t
ivk1 L hr 4 /L4
vk L hr 4 /L4
t
= T
,
= T
,
2 /L2
2 /L2
rH
rH
S
krh
krh
rV ,even

rS
2 /L2
1 h2 (v 2 + h v 2 h)r 4 /L4 hv 2 rH
tt trS
=T 2
,
2
2
2
2
S
S
hv rH /L
v h
rt rr
h
pS = T
(64)
(65)
(66)
where the common prefactor T is

T = v
L5 ik1 (vt+(r))
e
,
r5
v
52
1
.

2 1 v 2
(67)
The odd components are zero because of the choice of the even and odd vector and tensor harmonics that we made in Section 2.1. As a result, the odd-parity master equations may be solved
trivially by setting Todd = Vodd = 0. We will henceforth be concerned only with the even parity
cases.
The time-dependence of the source is eivk1 t , and we will make the steady-state assumption
that the master fields T , V , and S have a similar time dependence. It is convenient to make
the following definitions:
52 L5 3/2
a T (r)eivk1 t ,
2
2i52 vk1 L3 3/2
V =
a V (r)eivk1 t ,
k2
6 2 L
S = 54 a 3/2 S (r)eivk1 t .
k
The equations satisfied by I with I = T , V , and S can be written as

v 2 k12 k 2 4 I2
1
3
r a f r +
+ VI I = JI ,
2+
f
a3
a
L2
T =
(68)
(69)
where T = 2, V = 1, and S = 0 (compare with Eq. (3)). The quantities VI and JI are given by
4
4
4
9r 4
12rH
12rH
2k 2 rH
k4
2k 2
VT = 0,
(70)
VV = 2H4 ,
VS =
+
+
,
L r
H 2 L6 r 2 L2 r 4 L10 r 4
L6 r 6
2 eik1
v v 2 k
,
JT = 2
5 k 2 L2 r 3

4
2
1
5vk1 rH
v k eik1
vk1
2
k v 2 k12 5 +
2 2 + irH
,
JV = 2
L r
r
L2 r 6
5 kk1
(71)
(72)
54

v eik1
k1 k 2 2 + v 2 3v 2 k12
JS = 2
6
2
5 6k1 L H

2 8 2
2 41
2 2
4 1
3k L + k L 2k 3v k1 90rH 3
r
r

1
2
4
2k 4 + v 2 k12 k 2 3k12 4 + 3k1 L4 rH
18v 4 k14 + k 4 2 5v 2
3ivk 2 L6 rH
r

1

1
6 2
+ 3v 2 k 2 k12 1 2v 2 5 + 6ivrH
L 2k 4 + 9v 2 k12 k 2 3k12 6
r
r

2

8
2
2 2 1
+ 18k1 rH k 2 + 3v 9v k1 7 ,
r
(73)
4 /L4 r 2 as in (57). Note that the V decay at the boundary as 1/r 2 or faster.
where H = k 2 + 6rH
I
3.2. The holographic stress tensor

To understand the holographic relation of the master fields to the gauge theory stress tensor Tmn , a useful preliminary is to count the independent components of Tmn . Because the stress
tensor is symmetric, we start with 10 components. Conservation eliminates four, and conformal
symmetry eliminates one more, so there are five left. These five independent components are
related holographically to the five master fields. One can be more explicit by expanding

d 3k
t)S(k,
x),
T00 conserved =
(74)
qS (k,
(2)3
T0i conserved

d 3k 1
even
even
odd
odd

=
q
(
k,
t)S
+
q
(
k,
t)V
(
k,
x

)
+
q
(
k,
t)V
(
k,
x

)
,
(75)
t S
i
V
i
V
i
(2)3 k

t)

d 3 k qS (k,
x) + 1 3t2 + k 2 qS (k,
t)Sij (k,
x)
ij S(k,
Tij conserved = 2
3
2
6
(2)
4k
1
t)Veven (k,
x) + 1 t q odd (k,
t)Vodd (k,
x)
+ t qVeven (k,
ij
ij
k
k V

t)Teven (k,
x) + q odd (k,
t)Todd (k,
x) .
+ qTeven (k,
ij
T
ij
(76)
The first term on the right-hand side of (75) must take the form indicated in order to have
m Tm0 conserved = 0. The first term on the right-hand side of (76) is dictated by the conformal condition T m m conserved = 0, and the next three terms are consequences of the conservation
equations m Tmi conserved = 0.
The boundary asymptotics of each master field encodes the value of the corresponding q coefficient in (74)(76), as we explain more precisely in Sections 3.33.5. We have already remarked
that the odd perturbations vanish, and so in the following we omit parity labels. A subtlety in the
expansions (74)(76) is that they actually do not capture the full stress tensor, which is
Tmn = Tmn bath + Tmn conserved + Tmn drag .
2
(77)
Here Tmn bath = 8 N 2 T 4 diag{3, 1, 1, 1} is the stress-energy dual to the black hole in AdS5 Schwarzschild. The last term is not conserved and is due to an external force required to balance
55
the drag force of [6,7]. The presence of this term was noted in [1], and we rederive it in Section 3.6.
In order to compute Tmn using AdS/CFT, the most straightforward approach is to express
the perturbation (24) in axial gauge, hr = 0. As will be explained in detail below, the non-zero
components of the metric perturbations have the following large r behavior4 :
3
1
1
,
hmn = Rmn r 2 + Pmn + Qtot
mn 2 + O r
r
r
(78)
m
) but not on r. Up to a subtlety regardwhere Rmn , Pmn , and Qtot
mn may depend on x = (t, x
5
ing residual gauge freedom, the boundary condition Rmn = 0 corresponds to requiring that the
metric on the boundary should remain fixed, or in other words not adding components of Tmn to
the gauge theory Lagrangian. Another boundary condition comes from requiring purely infalling
modes at the black hole horizon.
Because hmn is a perturbation around AdS5 -Schwarzschild rather than pure AdS5 , it relates
only to the second and third terms in (77). More precisely, as argued in [1], one has
Tmn tot Tmn conserved + Tmn drag =
2
52 L3
Qtot
mn .
(79)
The notation Tmn tot is meant to indicate the total VEV arising from the trailing string: it excludes Tmn bath . Actually, it excludes one other thing: the Pmn /r term in (78) signals the need
for a counter-term subtraction to make Tmn well-defined. As discussed in [1], this subtraction
is a delta-function supported at the location of the quark.
The main purpose of Sections 3.33.5 is to relate the boundary asymptotics of solutions to the
radial master equations (69) to the quantities Qtot
mn entering (78). In so doing we will discover a
natural split of Qtot
into
a
non-conserved
piece
which can be computed analytically and a conmn
served piece which must in general be obtained by solving the master equations with appropriate
boundary conditions. We also explain what these boundary conditions are, thereby completing,
in the master field formalism, a specification of the boundary value problem that determines the
holographic stress tensor.
3.3. The asymptotics of the tensor master equation
Near the boundary of AdS5 -Schwarzschild, the solution to the tensor master equation behaves
as

5
1
1
2
T (r) = RT 1 2
,
log
r
+ PT 3 + Qtot
T 4 +O r
4
4r
16r
r
r
(80)
where PT and are given by

PT =
2
v v 2 k
,
52 3k 2

L4 k 2 v 2 k12
(81)
4 The quantities P , Qtot , and R

mn
mn differ from the corresponding quantities in [1] by different normalization factors.
mn
5 The subtlety is that the boundary condition R
mn = 0 is actually a little more restrictive than necessary. It should be
allowed to have Rmn gmn , corresponding to adding T m m = 0 to the Lagrangian, or Rmn v(m kn) , corresponding
to adding a multiple of m T mn to the Lagrangian. Such deformations correspond to metric perturbations in AdS5 -
Schwarzschild which are pure gauge.
56
T ,even
and Qtot
= 12 52 L5 T (r)eivk1 t , the
T and RT are the two integration constants. Since HT
requirement Rmn = 0 translates into RT = 0. Assuming RT = 0, we have
HTT ,even = v
2
6

2
v 2 k
L5
ivk1 t L
eivk1 t 3 + 5 Qtot
+ O r 5 ,
T e
2
4
2L
6k
r
r
(82)
which can be used to find the tensor components of the Qtot

mn coefficients via (25)(27). Feeding
the result through (79) and comparing to the expansion (76), one finds that
t) = Qtot (k)e
ivk1 t .
qT (k,
T
(83)
(Recall that because odd parity perturbations vanish for the trailing string, we are always referring
to even parity; however, for a more complicated gravitational source, (83) could be used equally
in reference to odd parity modes.) The normalization factor between T and T in (68) was
chosen to make the relation (83) simple.
Near the horizon, the leading behavior of T is
T (r) = UT (r rH )ivk1 L
2 /4r
2 /4r
+ VT (r rH )ivk1 L
+ O(r rH ),
(84)
and the infalling boundary condition is VT = 0.

3.4. The asymptotics of the vector master equation
In the vector case, the master field has the following asymptotic behavior at large r:

4
1
1
1
V (r) = RV
,
1 2 log r + PV 2 + Qtot
V 3 +O r
r
2r
r
r
(85)
where
PV =
v vk
,
52 k

L4 k 2 v 2 k12
(86)
and Qtot
V together with RV are integration constants. Relating this large r series expansion to (78)
is not hard: one can use the definition (68) and plug (85) into (39) to obtain large r asymptotic
expressions for the gauge-invariant quantities fb , and then use the expression for fb in (38) as
well as the axial gauge condition frV = 0 to find large r series expansions for both ftV and HTV .
When the latter quantities are plugged into (25)(27), one can easily check that all non-zero
components of Rmn turn out to be proportional to RV , so setting Rmn = 0 means setting RV = 0.
When RV = 0, the large r behavior of the functions that enter in the metric perturbations is
given in axial gauge by

2k
5

52
2vk ivk1 t L4
v irH
V ,even
tot
ivk1 t L
ft
= v
+
+ O r 4 ,
e
Q
e
V
2
2
2
3
3k
2L
r
r
5 kk1 L

52 ivk1 tot ivk1 t L6
2iv 2 k1 k ivk1 t L5
e
+ O r 5 .
QV e
2
3
4
2L k
3k
r
r
Feeding (87) through (79) and comparing to (75), one finds that
HTV ,even = v
(87)
t) = Qtot (k)e
ivk1 t ,
qV (k,
V
(88)
57
provided one identifies

v ir 2
T0i drag = 2 H2
5 L
d 3 k k ivk1 t even
e
Vi (k, x).
(2)3 kk1
(89)
At the horizon, the situation is similar to the one encountered in the tensor case. The leading
behavior of V is
V (r) = UV (r rH )ivk1 L
2 /4r
2 /4r
+ VV (r rH )ivk1 L
+ O(r rH ),
(90)
and again the infalling boundary condition is VV = 0.

3.5. The asymptotics of the scalar master equation
Near the boundary, the asymptotic behavior of S (r) obeying (69) is
3
1
1
1
,
+ RS 2 log r + Qtot
S 2 +O r
r
r
r
with PS explicitly given by
S (r) = PS
PS =
v k 2 (2 + v 2 ) 3v 2 k12
,
2k 2
52
(91)
(92)
and Qtot
S and RS being regarded as the two integration constants of the second order equation (69).
The asymptotic behavior of S (r) given above can be related to the asymptotic behavior of
the metric perturbations (78) through a straightforward but tedious calculation: one just has to
use (68) to go back to the master field S and then trace back the steps outlined in Section 2.5,
in the special case where the background is AdS5 -Schwarzschild. It is important to note that
in doing so we only need to consider large r series solutions to the differential equations we
encountera much easier task than solving these equations for all r. We will now provide an
outline of the computation.
S and H S in terms of (r) that are valid for
First, one can easily obtain expressions for fbc
S
L
all r from combining the definitions (49) and (50) with Eqs. (51) and with the definition (53) of
S and H S by replacing
the master field. Next, one can find large r asymptotic expansions for fbc
L
S (r) with a large r series solution to the master equation. While the first few terms of this
solution are given in (91), an improved series expansion (up to order O(r 8 )) is needed to arrive
at the results listed below. The last step of the computation is to assume axial gauge, namely
S and H S to find series solutions
frS = ftrS = frrS = 0, and to use the asymptotic expansions for fbc
L
for the quantities fttS , ftS , HTS , and HLS that obey the system of differential equations (45). The
asymptotics of the non-zero metric perturbations hmn can then be found from (25)(27), and
one can check they have the form (78). Moreover, all scalar components of Rmn end up being
proportional to RS , so again Rmn = 0 means RS = 0.
Assuming RS = 0, the expressions one ends up with for the axial gauge metric perturbations
are

2
4

52
2(2 + v 2 ) ivk1 t L3
v ivrH
S
tot
ivk1 t L
+ O r 3 ,
e
+
QS + 2
e
ftt = v
2
2
9
r
2L
k
L
r
5 1
ftS = v
4
52 ivk1 tot ivk1 t L5
2ivk1 ivk1 t L4
,
e
+
O
r
e
Q
S
3k
2L k
r2
r3
58

2

2 1 tot v ivrH
1
L5
L6
HLS = v eivk1 t 3 + 5
QS + 2
eivk1 t 4 + O r 5 ,
2
9
2L 6
r
r
5 6k1 L

52 k 2 3v 2 k12 tot
v 2 (k 2 3k12 ) ivk1 t L5
HTS = v
e
+
QS
2L
6k 2
r3
4k 2

v ir 2 v(k 2 + 3k12 ) ivk1 t L6
+ 2 H 2
+ O r 5 .
e
2
4
4k k1 L
r
5
(93)
As in the tensor and vector cases, feeding (93) through (79) and comparing to (74), one finds that
t) = Qtot (k)e
ivk1 t ,
qS (k,
S
(94)
provided one identifies

v ivr 2
T00 drag = 2 2H
5 L
d 3 k 1 ivk1 t
e
S(k, x),
(2)3 k1

2
k 2 + 3k12
d 3 k ivk1 t 1
v irH

e
ij S(k, x) +
Sij (k, x) .
Tij drag = 2 2 2
6k1
(2)3
4k 2 k1
5 L
(95)
(96)
From (93) one may also find explicit expressions for the quantities Pmn appearing in (78).
Using (A.3) in Appendix A one can check that the expressions so derived match with the corresponding Pmn coefficients calculated in [1].
At the horizon, the series expansion for S is
S (r) = US (r rH )ivk1 L
2 /4r
2 /4r
+ VS (r rH )ivk1 L
+ O(r rH ).
(97)
The infalling boundary condition is VS = 0.

3.6. The force on the moving quark
In Sections 3.4 and 3.5, we saw the non-conserved part of the stress tensor, Tmn drag , emerging from mismatches between the Fourier expansion (74)(76) of a general conserved, traceless
stress tensor and the boundary asymptotics of the metric perturbation hmn , as computed by tracing backward through the master field definitions in Sections 2.4 and 2.5. In this section we
recheck an observation of [1], namely that the non-conservation of Tmn drag is the result of an
external force on the quark that precisely cancels the drag force so as to keep the quark moving at
constant velocity. Equivalently, one can regard this non-conservation of Tmn drag as describing
the rate at which energy and momentum are supplied to the bath by the external quark.
The external force on the quark is

F i = d 3 x m T mi drag ,
(98)
V
where the integration volume V is any three-dimensional volume enclosing the quark. The quantity m T mn can be easily calculated from (89), (95), and (96):

v vr 2 1

m T mn = 2 H
x vt x 2 x 3 (v
2
5 L
1 0 0).
(99)
59
Evaluating the integral and using v = 52 /2 1 v 2 , we obtain

F1 =
2
vrH
2 L2 1 v 2

2 N
gYM
2
v
,
T2
1 v2
(100)
which is indeed minus the drag

force [6,7]. In obtaining the last equality we have used T =
2 N.
2
2

rH /L as well as L / = gYM
4. Small k approximation
If k T , the equations for the functions I , where I = T , V , or S, simplify and can be
solved analytically. In fact, we can use the regular perturbation theory technique to find small k
series solutions for each of these equations. The series solutions will in turn enable us to extract
tot
tot
small k series expansions for Qtot
T , QV , and QS , which are needed to find the small k behavior
of Tmn .
4.1. Tensor set
The tensor master equation (35) is the same as the dilaton equation in [21], or the A-equation
in [1]. While the derivation of the small k solution of this equation can be also found in Section 3.5 of [1], we will now outline the main steps of the calculation.
We start with the equation for T in (69) and develop a series expansion in k:
T (r) = 0 (r) + k1 (r) + k 2 2 (r) + .
(101)
The equations satisfied by each i can be found from equating like powers of k in (69). In
doing so, we consider k1 and k to be of the same order of smallness as k, so for example a
term containing k1 k k 2 will be considered to be O(k 4 ). These equations can then be solved
by requiring that the boundary and horizon asymptotics of the solutions be nothing but small k
expansions of (80) and (84) with RT = 0 and VT = 0. For example, the equation for 0 is
r2

5 h
v v 2 k 2
+
+
r 0 = 2 2 ,
r
h
5 k rh
(102)
and its solution is

4
2
r 4 rH
v v 2 k
r + rH
1 r
0 (r) = 2
+ log
+ log
2 tan
.
3
rH
r rH
r4
5 4k 2 rH
(103)
Expanding this solution in series at large r, we obtain

0 (r) =
2
2r

1
v v 2 k
v v 2 k
H 1
2
+ O r 5 ,
2
2
3
2
4
r
5 3k r
5 4k
from which we can identify the coefficient of 1/r 4 with the leading contribution to Qtot
T .
Performing a similar analysis for 1 and 2 , we find
(104)
60

2
2 L2 log 2
v rH v 2 k
v iv 3 k1 k
+ 2
2
2
8k 2
5 4k
5
2 L4 (6 12 log 2)
2
2

v v 2 k
2
2 2
2 2 12(log 2)
+ 2
v
k
+
v
k
+ O k3 .
k
1
1
2
6 12 log 2
192k rH
5
Qtot
T =
(105)
4.2. Vector set

The small k analysis in the vector case goes mostly the same way as in the tensor case: one
starts by expanding V in a low-k series
V (r) = 0 (r) + k1 (r) + k 2 2 (r) + ,
(106)
and then the equations for i follow from equating like powers of k in the expansion of the V
equation in (69). The difference from the tensor case is that the matching of the i to the asymptotics (85) and (90) with RV = VV = 0, appropriately expanded in k, results in one undetermined
integration constant at each order i. Fortunately, this integration constant can be determined simply by looking at the horizon asymptotics of i+1 . For example, 0 is given exactly by
0 (r) =
3
rH
v vk (r rH )
+
U
,
0
kr 3
r3
52
(107)
where U0 is an integration constant of order O(k 0 ). It is not hard to see that, to order O(k 0 ), this
expression satisfies the required boundary conditions: 0 looks like an infalling wave at r = rH
and there is no 1/r piece at large r regardless of the value of U0 . The only way to determine U0
is to go to order O(k).
We obtain the following small k expansion for Qtot
V :
Qtot
V =
v rH k (4v 2 k12 k 2 ) v ik L2 [k 4 (8 log 2)v 4 k14 + (2 log 2)v 2 k 2 k12 ]

+ 2
52
4vkk12
5
16v 2 kk13

+ O k2 .
(108)
4.3. Scalar set

Obtaining a small k solution for the scalar equation in (69) is a bit tricky, because the quantity S does not have a small k expansion itself. Instead, we can find a small k approximation for
the rescaled field

6r 4
S (r) = k 2 + 4H2 S (r).
(109)
L r
To do so, we can follow the same steps as the ones outlined in the tensor case in Section 4.1 and
replace T by S . It is not hard to check that the zeroth order solution 0 is zero in this case, so
the first non-trivial term should be first order in k:
S (r) = k1 (r) + k 2 2 (r) + k 3 3 (r) + .
(110)
One has to be careful though when matching the solutions for i to the asymptotics (91) and (97):
the rescaling factor in (109) reduces to k 2 in the large r limit, so the order of the S expansion
61
changes by two when we look at the boundary asymptotics. With this minor technical difference,
the matching proceeds in the same way as in the tensor case.
By reading off the coefficient of the 1/r 2 term in the large r expansion of S , we find Qtot
S to
be
2 (k 2 + 3k 2 )
v ivrH
v 3v 2 rH k12 [3v 2 k12 + k 2 (2 + v 2 )]
1
+ O(k).
+
(111)
52 L2 k1 (k 2 3v 2 k12 ) 52
(k 2 3v 2 k12 )2
This formula diverges at the Mach angle, namely where k1 /k = 1/v 3, signaling a sonic boom
in the thermal medium. While the divergence appears to get worse as one goes to higher orders
in the expansion, the singular behavior disappears if the series expansion (111) is resummed into
a form that resembles a Lorentzian lineshape. For a more detailed discussion, see Section 3.5
of [1].
Qtot
S =
5. Large k approximation
In [19], a method of Greens functions was employed to determine the large k asymptotic
form of the VEV of the operator O dual to the dilaton, as explored numerically in [21]. This
method can be understood as an expansion in powers of rH , which makes sense because rH T
and all k-dependence of dimensionless quantities arises in series expansions in k/T . Suppose
one starts with an ordinary differential equation of the form

L(r) = r2 + t (r)r + u(r) (r) = j (r)
(112)
where there is an expansion
2
L2 + ,
L = L0 + rH L1 + rH
2
= 0 + rH 1 + rH
2 + ,
2
j = j0 + rH j1 + rH
j2 + .
(113)
The radial master equations (69) fit this form after both sides are divided by f = (1
4 /r 4 )r 2 /L2 . At least formally, one may solve (112) order-by-order in r . At the lowest orrH
H
der, L0 is a differential operator in pure AdS5 , which is the rH 0 limit of AdS5 -Schwarzschild.
Construct two solutions of the homogeneous equation,
L0 H = 0 = L0 VEV ,
(114)
where H satisfies appropriate boundary conditions at the degenerate horizon r = 0, while VEV
has appropriate falloff at the boundary of AdS5 to describe a VEV rather than a deformation. Then
a Greens function inverse to L0 , satisfying
L0 G0 (r; r0 ) = (r r0 )
(115)
as well as the boundary conditions at the horizon and infinity satisfied by H and VEV , respectively, may be constructed as

H (r0 )VEV (r) for r > r0 ,
1
G0 (r; r0 ) =
(116)
W [H , VEV ](r0 ) VEV (r0 )H (r) for r < r0 ,
where the Wronskian is

W [H , VEV ](r0 ) = H (r0 )VEV
(r0 ) VEV (r0 )H
(r0 ).
(117)
62
The leading order solution 0 may be determined as follows:

0 (r) = (G0 j0 )(r)
dr0 G0 (r; r0 )j0 (r0 )

0
= A0 (r)H (r) + B0 (r)VEV (r)
(118)
where

A0 (r) =
r
VEV (r0 )
dr0
j0 (r0 ),
W (r0 )
r
B0 (r) =
dr0
0
H (r0 )
j0 (r0 ).
W (r0 )
(119)
In optimal circumstances, B0 () = limr B0 (r) is finite and A0 (r) 0 fast enough as r 0

so that the boundary asymptotics of 0 (r) may be expressed as
0 (r) = B0 ()VEV (r) + (sub-leading).
(120)
0)
O(rH
conThe form (120) is convenient because, up to a normalizing factor, B0 () is just the

tot , or Qtot . Unfortunately, there is a complication in the cases we will study:
tribution to Qtot
,
Q
T
V
S
B0 (r) does not have a finite limit. The solution is to isolate a simple analytic form 0CT (r), capturing that part of 0 (r) whose boundary asymptotics are more singular than indicated in (120),
and split both 0 and j0 so that
0 = 0CT + 0R ,
j0 = j0CT + j0R ,
L0 0CT = j0CT ,
L0 0R = j0R .
(121)
One may then follow through the Greens function analysis of the renormalized equation
L0 0R = j0R as indicated in (118)(119) and find that the renormalized analog of (120) holds
with finite B0R (). The splitting (121) is in the spirit of holographic renormalization as reviewed
for example in [22], although it also follows an older and more elementary line of thinking where
one looks for leading non-analytic contributions to Greens functions in momentum space: see
for example [3].
Higher-order i may be determined in a similar fashion:
1 (r) = (G0 j1 )(r),
2 (r) = (G0 j2 )(r),
....
j1 = j1 + L1 0 ,
j2 = j2 + L2 0 + L1 1 ,
(122)
A renormalization process similar to (121) could be employed, if needed, at higher orders; but
need for it would be unexpected given that the counter-term contributions usually correspond
to contact terms in the gauge theory which are temperature-independent.
A significant obstacle to the treatment just described is that the expansion (113) is not uniform
in r: near the horizon it breaks down. It is not clear whether the boundary conditions appropriate
to the zero-temperature Greens function G0 lead, through the expansion (113), to a solution
satisfying standard boundary conditions at the horizon. This issue was confronted in [19] with
a WKB treatment. We will take the less rigorous approach of computing the leading non-trivial
tot
tot
k-dependence of Qtot
T , QV , and QS and comparing to numerical evaluations from [1]. The
eventual aim is to give an all-scales description of Tmn , both in k-space and position space,
using a combination of numerical methods for intermediate scales and analytical approximations
for the IR and UV limits.
63
5.1. Tensor set

This is essentially the case considered in [19], because the tensor master equation is the same
as the dilaton equation [21]. The leading non-zero terms in the expansions (113) are
5
L4
L0 = r2 + r 4 k 2 ,
r
r
3
ivk1 L2
j2 =
PT ,
j0 = 5 PT ,
r
r8
where PT is as defined in (81) and

2
.
k 2 k 2 v 2 k12 = 1 v 2 k12 + k
(123)
(124)
The zero temperature solutions and Wronskian are

VEV (r) =
1 2
I2 kL /r ,
r2
W (r) = W [H , VEV ](r) =
H (r) =
2
1
/r ,
K2 kL
2
r
1
,
r5
(125)
where I and K denote Bessel functions. It is significant that k 2 > 0 always because it results in
a Greens function that vanishes faster than any power of r as r 0, i.e. in the depths of AdS5 .6
Using 0CT (r) = PT /r 3 , one may straightforwardly show that

2
5
5
3 kL
1
ivk1
,
2 (r) = PT
.
+O r
+O r
0 (r) = PT 3
(126)
r
16r 4
k 2 L2 r 4
A counter-term is not needed for the analysis of 2 (r), as expected on general grounds. Comparing (126) with (80) leads immediately to

2
4
ivk1 2
3 kL
tot
QT = PT
(127)
rH + O rH .
+
16
k 2 L2
5.2. Vector set
The leading non-zero terms in the expansion (113) are
5
L4
3
L0 = r2 + r 4 k 2 + 2 ,
r
r
r

1
L2 2
3k v 2 k12 PV ,
j2 =
j0 = 4 PV ,
7
3ivk1 r
r
(128)
where PV is as defined in (86). The zero temperature solutions and Wronskian are
VEV (r) =
1 2
I1 kL /r ,
r2
H (r) =
2
1
/r ,
K1 kL
2
r
2 /rH
6 Probably this means that putting the horizon back at finite r
kL
) effects for some
H corresponds to O(e
c-number b0 . If so, it means that the series expansion (113) for in powers of rH is only an asymptotic series. Be 2
cause rH /L2 = T , one sees that ekL /rH has essentially the form of a MaxwellBoltzmann factor.
64
W (r) = W [H , VEV ](r) =
1
.
r5
Using 0CT (r) = PV /r 2 one obtains

2
4
kL
1
,
+O r
0 (r) = PV 2
r
4r 3
(129)

2 (r) = PV
Comparing to (85), one sees that

2 3k 2 v 2 k12 2
4
kL
tot
rH + O rH .
+
QV = PV
4
3ivk1 k 2 L2

4
3k 2 v 2 k12
.
+O r
3ivk1 k 2 L2 r 3
(130)
(131)
5.3. Scalar set

The leading non-zero terms in the expansion (113) are
5
L4
4
L0 = r2 + r 4 k 2 + 2 ,
r
r
r

(2k 2 v 2 k12 )L4
1
j0 = 3 +
PS ,
r
3r 5

2iL2 v 3k 4 5v 2 k14 (1 v 2 ) + k12 k 2 (1 + 8v 2 ) 1
ivk1 6 2 2k 2 v 2 k12
+
r
L
j2 =
PS ,
H
3k1
9
r6
r8
k 2 (2 + v 2 ) v 2 k12 (1 v 2 )
(132)
where PS is as defined in (92). The zero temperature solutions and Wronskian are
VEV (r) =
1 2
I0 kL /r ,
r2
W (r) = W [H , VEV ](r) =
H (r) =
1
.
r5
2
1
/r ,
K0 kL
2
r
(133)
Using 0CT (r) = PS /r one obtains

1 L2 (k 2 + v 2 k12 )
+ O r 3 ,
0 (r) = PS
2
r
6kr

2
3
2iv(k + v 2 k12 ) 9k 4 2v 2 (1 v 2 )k14 + k12 k 2 (5 + 11v 2 )
+O r
.
2 (r) = PS
9k1 k 4 L2
(2 + v 2 )k 2 v 2 k12 (1 v 2 )
(134)
Comparing to (91), one sees that

L2 [(2 + v 2 )k 2 v 2 (1 v 2 )k12 ]
v
Qtot
=
S
52
12k

4
2
2
4
r v 9k 2v (1 v 2 )k14 + k12 k 2 (5 + 11v 2 )
+ H2
(135)
.
+ O rH
9L
k1 k 4
or, for v not too close to 1, in
The result (135) is more properly an expansion in rH /(L2 k),
2
inverse powers of K kL /rH , which is the dimensionless wave-number used in [1].
65
Fig. 1. Comparisons between the large K analytic approximations found in (135) (plots (a), (b)), (131) (plots (c), (d)),
and (127) (plots (e), (f)) and corresponding numerical solutions of the linearized Einstein equations in the form studied
in [1]. The unbroken colored lines come from numerical solutions at the values of K = k/T indicated, while the dashed
lines come from evaluations of (135), suitably transformed into evaluations of QE , QD , and QA as described in the text.
The values of K used in plots (c), (e) are the same as in (a), and those in (d), (f) are the same as in (b). For values of K
larger than the largest one shown in each plot, the numerical evaluations are almost indistinguishable from the K
approximation. The Mach angle is indicated, but there is no reason to expect structure near this angle because large K
is the opposite limit of where hydrodynamics is expected to apply. (For interpretation of the references to colour in this
figure legend, the reader is referred to the web version of this article.)
To get an idea of when the next term in the expansion (135) is significant, we have plotted
in Fig. 1(a), (b) a comparison of (135) with numerical evaluations using the approach of [1].
The most immediate conclusion is that the value of K where (135) becomes reasonably accurate
depends significantly on v. To understand the comparison with numerics, first note that we are
tot
plotting not Qtot
S but QE as defined in [1]: this quantity is proportional to QS minus the leading
large k behavior in (135) (corresponding to the pure Coulombic near-field of the quark) plus the
non-conserved contribution from (95). The constant of proportionality is the same as the one in
(A.4). Also, because of the comparisons with away-side jet-splitting that motivated the study [1],
66
we choose for an angular variable not the angle = arctan k /k1 between the wave-vector and
the direction of the quarks motion, but instead = . Thus = corresponds to wavevectors exactly along the direction of the quarks motion. Finally, all dimensionful quantities are

rendered dimensionless using factors of T = rH /L2 : for example, K = k/T
.
In Fig. 1(c)(f) we have made corresponding comparisons of the analytic forms (127)
and (131) to numerical evaluations following [1]. As in the comparison of Qtot
S to QE , there
is a non-trivial affine relation between Qtot
and
the
quantity
Q
appearing
in
the
plots, and beD
V
and
Q
.
These
relations
can
be
worked
out
using
(A.4),
keeping
in
mind that the
tween Qtot
A
T
quantities plotted exclude the purely Coulombic near-field.
5.4. The position-space near field
The large k results from the previous sections can be Fourier transformed to position space
tot
tot
without much difficulty. It is important to remember that Qtot
T , QV , and QS give only the contribution Tmn conserved to the total stress tensor in the plasma, and that the stress tensor of the
thermal bath Tmn bath together with a non-conserved piece Tmn drag that is responsible for the
drag force need to be added in order to recover the full stress tensor (see Eq. (77)). From now on
we will focus on the quantity Tmn tot = Tmn conserved + Tmn drag , as defined in (79).
Combining (135) with the non-conserved contribution (95) we obtain for the energy density

d 3 k ivk1 t+i k
x
e
(k),
(t, x) T00 (t, x) tot =
(2)3

= (0) (k)
+ (2) (k)T
2+O T4 ,
(k)

v 2 k12 (2 + v 2 )k 2
2 N,
=
gYM
(0) (k)
24k 1 v 2

2v 2 k12 + (5 11v 2 )k 2
2 N,
= iv k1
gYM
(2) (k)
(136)
18k 4 (1 v 2 )
2 as in (124). The integrals in (136) can
where we have set k12 = (1 v 2 )k12 and used k 2 = k12 + k
be easily computed by differentiating appropriately modified versions of the standard formulas

1
|
x|
d 3 k ei kx
d 3 k ei kx
=
,
= .
(137)

4
8
(2)3 |k|
2 2 |
x |2
(2)3 |k|
The energy density in position space can thus be written

(t, x) = (0) (t, x) + (2) (t, x)T 2 + O T 4 ,
2
(1 v 2 )x12 + (1 + v 2 )x
2 N,
gYM
(0) (t, x) =
2 )3
12 2 (1 v 2 )(x12 + x
2
(5 11v 2 )x12 + (5 8v 2 )x
2 N,
gYM
(2) (t, x) = v x1
(138)
2 )5/2
72(1 v 2 )3/2 (x12 + x
the x 1 -axis in the co-moving

where x1 = (x1 vt)/(1 v 2 )1/2 measures the distance along

2 N , (0) and (2) are functions
frame, and x = (x22 + x32 )1/2 . At t = 0 and for fixed v and gYM

2 . In
only of = arctan x /x1 , up to overall power-law dependences on x = x12 + x
Fig. 2 we illustrate in two ways the effect of the O(T 2 ) term on the near-field energy density.
67
Fig. 2. (a) A contour plot of the near-field energy density as computed from (138) at time t = 0. The red regions in the
left side of the plot represent negative T00 more precisely, an energy deficit compared to the average energy density
of the thermal bath. The deficit arises because the O(T 2 ) dominates over the O(1) term, which leads us to question the
accuracy of the asymptotics in these regions. (b) The O(T 2 ) correction (2)as a function of = arctan x /x1
2 was chosen for each velocity in

for several different velocities v. (2) 1/x 2 , and a different value of x = x12 + x
order to make the qualitative effects easy to discern. (For interpretation of the references to colour in this figure legend,
the reader is referred to the web version of this article.)
The main qualitative effects are Lorentz flattening of the region of large energy density, visible
in Fig. 2(a), and a surprising pattern of energy build-up in front of the quark with a corresponding
deficit of energy behind the quark, visible both in Fig. 2(a) and in the v = 0.9 curve in Fig. 2(b).
To understand this latter effect, recall that the subleading term (2) encodes the leading behavior
of dissipative effects at large k. For v 0.67, this shift in the energy density is positive behind
the quark (i.e. for < /2). This coincides with our intuition that the trailing string represents
color flux spreading out behind the quark as it moves forward through the plasma. But for v
0.79, the shift in the energy density reverses sign as a function of .7 A possible interpretation is
that there is a bull-dozer effect where the quarks near field creates a pile-up of thermal gluons
and other matter just ahead of it, like dirt piling up in front of the blade of a bull-dozer. It would
be interesting to study this effect further and see how far it survives beyond the leading correction
term (2) .
Other components of the stress tensor may also be straightforwardly computed given Qtot
S ,
tot
QV , Qtot
,
and
the
non-conserved
terms
(89),
(95),
and
(96).
In
particular,
the
Poynting
vector
is
T
as follows:

(0)
(2)
S i = S i + Si T 2 + O T 4 ,

d 3 k ivk1 t+i k
x

Si (t, x) = T0i (t, x) tot =
e
Si (k),
(2)3
S = (S1 , S2 , S3 ),
7 In the intermediate range of velocities, 0.67 v 0.79, the regions where (2) is positive comprise a forwardpointing lobe and a backward-leaning, thickened cone. This is exemplified in the v = 0.7 curve in Fig. 2(b).
68
Fig. 3. The near-field Poynting vector to the approximation shown in (140) in the plane T x3 = 0.02. The arrows show
where red means
the direction of the projection of S into the plane, and the color corresponds to the magnitude of S,
large and blue means small. (For interpretation of the references to colour in this figure legend, the reader is referred to
the web version of this article.)

2

2k12 + 3k
8i k1 2
2 N 1+ v
= v
S1 (k)
gYM
T +O T4 ,
24k
1 v2
1 v 2 3k 3

2)
2
4
4(7k + 9k
k1 k2
v
2
2

,
gYM N 1 +
T +O T
S2 (k) = v
24k
3i k1 k 3
1 v2

2)
4
4(7k 2 + 9k
k1 k3
v
2
2

S3 (k) = v
.
gYM N 1 +
T +O T
24k
3i k1 k 3
1 v2
(139)
The Fourier integrals are again derivatives of the standard forms (137), and the position-space
forms are8

2
2
2
4
2 x1 x 2
v x gYM N
v
2
2x1 + x T + O T
,
S1 (t, x) =
1+
2
1 v 2 x 6 6 2
1 v 2 4 x

2
2
4
x1 x2 gYM N
2 x
v
v
2
2
S2 (t, x) =
11
x
T
,
+
8x
+
O
T
1
1
6 2
1 v 2 x 6
1 v 2 12 x1

2
2
4
x1 x3 gYM N
2 x
v
v
2
2
S3 (t, x) =
11
x
T
.
+
8x
+
O
T
1
1
6 2
1 v 2 x 6
1 v 2 12 x1
(140)
In Fig. 3 we show a cross-section of the approximation (140) to the Poynting vector field for
v = 0.5 and v = 0.95. For the purposes of plotting, a convenient way to regularize the divergence
at x = 0 is to take a cross-section at fixed but small x3 .
8 We obtained the explicit position-space result (140) for the near-field Poynting vector after seeing a comparable result
in a summary draft of [20]. We thank A. Yarom for correspondence about this result.
69
6. Conclusions
We have derived fully decoupled forms of the Einstein equations linearized around a general
five-dimensional warped background (2) appropriate to describing holographic duals of finite
temperature field theories on R3,1 . The final equations, (35), (42), and (55), are expressed in
terms of diffeomorphism-invariant master fields, from which one may extract all components
of the metric perturbations. Indeed, the entire derivation of the master equations has been cast
in an explicitly diffeomorphism-invariant formalism. Although our results for the scalar master
equation are incomplete in that we did not give the full expression for the source term except in
the AdS5 -Schwarzschild background, this general source term may be readily obtained starting
from (51).
We have applied the master field formalism to reproduce aspects of the results of [1], which
dealt with the stress tensor produced by an external quark moving through a thermal bath of
N = 4 gauge theory. We have also extended these results by calculating the leading correction at
large wave-number to the Coulombic near-field of the moving quark. This calculation was done
using a method recently developed in [19]. The result for the energy density has an unexpected
feature in position space: for v 0.79, the leading correction indicates a pile-up of energy just
in front of the quark.
Following [1] we attempted to Fourier transform the numerical results to position space. We
found this difficult because even after the subtraction of the Coulombic near-field, the Fourier

integrals have poor convergence properties: without the oscillating factor ei kx , they are quadratically divergent in the ultraviolet. With the leading correction subtracted off as well, this divergence is reduced to logarithmic, leading to a significantly more tractable numerical problem.
Ideally one should calculate to one more order of accuracy in the large k asymptotic forms, so
as to cure the divergences altogether and work with absolutely convergent Fourier integrals. Obtaining the next correction is significantly harder than the calculations we did in Section 5.15.3,
but it may be tractable.
Acknowledgements
This work was supported in part by the Department of Energy under Grant No. DE-FG0291ER40671, and by the Sloan Foundation. We gratefully acknowledge collaborative contributions by F. Rocha in the early stages of this project, and we also thank J. Friess and G. Michalogiorgakis for useful discussions and A. Yarom for correspondence.
Appendix A. Relation to the variables used in [1]
In [1], the metric perturbations h were written as

L6
d 3 k k i k

h e x ,
h = 3
3
(2)
rH
where in axial gauge
hk
H00
2
r 0
= v 3 H01
L H
02
H03
0
0
0
0
0
H01
0
H11
H12
H13
H02
0
H12
H22
H23
H03
0
H13 .
H23
H33
(A.1)
(A.2)
70
The equations of motion were decoupled into five sets of differential equations in quantities A,
B1 , B2 , C, D1 , D2 , E1 , E2 , E3 , and E4 , which were defined as particular linear combinations of
the Hmn s (see [1]). Setting k3 = 0 and k2 = k , and decomposing the above expression for h
as in (25)(27), we get
HTT ,even = v
2
L5 v 2 k
A,
3 k2
2rH
HTV ,even = v
2iL5 v 2 k1 k
D2 ,
3
k2
rH
ftV ,even = v
HLS = v
ftS = v
HTT ,odd = v
2L4 rv k
D1 ,
3
k
rH
L5
E ,
3 3
3rH
2iL4 rv k
3
rH
E2 ,
HTV ,odd = v
ftV ,odd = v
HTS = v
L5
3
2rH
L7
kC,
4
2rH
L8 rv
5
rH
iL9 v 2
5
rH
k1 kB2 ,
k 2 B1 ,
E4 ,
fttS = v
2L3 r 2 h(r)
3
3rH
(E3 E1 ),
(A.3)
and of course frV ,even = frV ,odd = frrS = ftrS = 0 in axial gauge. We can use these formulas to
tot
tot
tot
tot
tot
relate Qtot
T , QV , and QS to the quantities QA , QD , and QE , respectively, which were defined
in [1]:

2
v rH v 2
v 4k rH v
irH
tot
tot
tot
tot
QT = 2
QA ,
QV = 2
QD +
,
k
k2
4k1 L2 v
5
5

v
irH v
tot
=
2r
+
Q
.
Qtot
(A.4)
H
S
E
2k1 L2
52
References
[1] J.J. Friess, S.S. Gubser, G. Michalogiorgakis, S.S. Pufu, The stress tensor of a quark moving through N = 4 thermal
plasma, Phys. Rev. D 75 (2007) 106003, hep-th/0607022.
[2] J.M. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231252, hep-th/9711200.
[3] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string theory, Phys. Lett.
B 428 (1998) 105114, hep-th/9802109.
[4] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253291, hep-th/9802150.
[5] O. Aharony, S.S. Gubser, J.M. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity, Phys.
Rep. 323 (2000) 183386, hep-th/9905111.
[6] C.P. Herzog, A. Karch, P. Kovtun, C. Kozcaz, L.G. Yaffe, Energy loss of a heavy quark moving through N = 4
supersymmetric YangMills plasma, hep-th/0605158.
[7] S.S. Gubser, Drag force in AdS/CFT, hep-th/0605182.
[8] J. Casalderrey-Solana, D. Teaney, Heavy quark diffusion in strongly coupled N = 4 YangMills, hep-ph/0605199.
[9] S.-J. Sin, I. Zahed, Holography of radiation and jet quenching, Phys. Lett. B 608 (2005) 265273, hep-th/0407215.
[10] S.S. Gubser, Jet-quenching and momentum correlators from the gauge-string duality, hep-th/0612143.
[11] T. Regge, J.A. Wheeler, Phys. Rev. 108 (1957) 1063.
[12] F.J. Zerilli, Effective potential for even parity ReggeWheeler gravitational perturbation equations, Phys. Rev.
Lett. 24 (1970) 737738.
[13] S.A. Teukolsky, Perturbations of a rotating black hole. 1. Fundamental equations for gravitational electromagnetic,
and neutrino field perturbations, Astrophys. J. 185 (1973) 635647.
71
[14] H. Kodama, A. Ishibashi, A master equation for gravitational perturbations of maximally symmetric black holes in
higher dimensions, Prog. Theor. Phys. 110 (2003) 701722, hep-th/0305147.
[15] J.M. Bardeen, Gauge invariant cosmological perturbations, Phys. Rev. D 22 (1980) 18821905.
[16] H. Kodama, M. Sasaki, Cosmological perturbation theory, Prog. Theor. Phys. Suppl. 78 (1984) 1166.
[17] H. Kodama, A. Ishibashi, O. Seto, Brane world cosmology: Gauge-invariant formalism for perturbation, Phys. Rev.
D 62 (2000) 064022, hep-th/0004160.
[18] P. Benincasa, A. Buchel, A.O. Starinets, Sound waves in strongly coupled non-conformal gauge theory plasma,
Nucl. Phys. B 733 (2006) 160187, hep-th/0507026.
[19] A. Yarom, The high momentum behavior of a quark wake, hep-th/0702164.
[20] A. Yarom, On the energy deposited by a quark moving in an N = 4 SYM plasma, hep-th/0703095.
[21] J.J. Friess, S.S. Gubser, G. Michalogiorgakis, Dissipation from a heavy quark moving through N = 4 super-Yang
Mills plasma, JHEP 0609 (2006) 072, hep-th/0605292.
[22] K. Skenderis, Lecture notes on holographic renormalization, Class. Quantum Grav. 19 (2002) 58495876, hepth/0209067.
Constructing the AdS/CFT dressing factor

Nikolay Gromov a,b, , Pedro Vieira a,c
a Laboratoire de Physique Thorique de lEcole Normale Suprieure et lUniversit Paris-VI, Paris 75231, France
b St. Petersburg INP, Gatchina, 188 300 St. Petersburg, Russia
c Departamento de Fsica e Centro de Fsica do Porto, Faculdade de Cincias da Universidade do Porto,
Rua do Campo Alegre, 687, 4169-007 Porto, Portugal

Received 4 June 2007; accepted 29 August 2007
Abstract
We prove the universality of the HernandezLopez phase by deriving it from first principles. We find a
very simple integral representation for the phase and discuss its possible origin from a nested Bethe ansatz
structure. Hopefully, the same kind of derivation could be used to constrain higher orders of the full quantum
dressing factor.
PACS: 11.25.Tq
Keywords: Duality; Integrability
1. Introduction
Bethe ansatz equations, first written by Bethe in the study of one-dimensional metals [1], seem
to be a key ingredient in the AdS/CFT duality [2] between N = 4 SYM and type IIB superstring
theory on AdS5 S 5 .
The N = 4 SYM dilatation operator in the planar limit can be perturbatively computed in
powers of the t Hooft coupling . In the seminal work of Minahan and Zarembo [3] it was
shown that the 1-loop dilatation operator acts on the six real scalars of the theory exactly like an
* Corresponding author at: Laboratoire de Physique Thorique de lEcole Normale Suprieure et lUniversit Paris-VI,
Paris 75231, France.
E-mail addresses: nikgromov@gmail.com (N. Gromov), pedrogvieira@gmail.com (P. Vieira).
doi:10.1016/j.nuclphysb.2007.08.019
N. Gromov, P. Vieira / Nuclear Physics B 790 (2008) 7288
73
integrable SO(6) spin chain Hamiltonian. Restricting ourselves to two complex scalars we obtain
the same Hamiltonian considered by Bethe for the Heisemberg spin chain. The eigenstates are
K magnon states with momentum and energy parameterized by the roots ui which satisfy Bethe
equations
ui + i/2
ui i/2
L
=
K

ui uj + i
.
ui uj i
j =i
The full N = 4 1-loop dilatation operator [4] is also given by an integrable Hamiltonian whose
spectrum is dictated by a system of seven Bethe equations [5], corresponding to the seven nodes
of the psu(2, 2|4) Dynkin diagram. In [6] the all loop version of the Bethe equation for the SU(2)
sector was conjectured to be
yi+
L
=
yi
K

ui uj + i
,
ui uj i
(1)
j =i
where yj (uj ) and yj (uj ) are given by

1
4
y + = u,
y

1
4
i
.
y + = u
y
2
On the other hand, for the same sector but from the string side of the correspondence, a map
between classical string solutions and Riemann surfaces was proposed [7]. The resemblance
between the cuts connecting the different sheets of these Riemann surfaces and the distribution
of roots of the Bethe equations in the scaling limit seemed to indicate that the former could be
the continuous limit of some quantum string Bethe ansatz. In [8] these equations were proposed
to be
yi+
yi
L
=
K

ui uj + i 2
(ui , uj ),
ui uj i AFS
(2)
j =i
where
AFS (ui , uj ) =

1 1/(yj+ yi ) yj yi 1 yj+ yi+ 1 i(uj ui )
1 1/(yj yi+ ) yj yi+ 1 yj+ yi 1
(3)
The striking similarity between (1) and (2) naturally led to the proposal that both sides of the
correspondence would be described by the same equation which a scalar factor 2 interpolating
2
from AFS
for large t Hooft coupling to 1 for small .
In [9] Beisert and Staudacher (BS) conjectured the all-loop Bethe equations for the full
PSU(2, 2|4) group and in [10] these equations were brought to firmer ground. As before, one
of the main tools used to guess the form of these equations was the existence of the classical algebraic curve for the full superstring on AdS5 S 5 [21]. The seven equations for the seven types
74
of roots ua,j are entangled and the middle equation, the most complicated one, can be written
as1
+
y4,k
L
y4,k
= ei V (y4,k )
K4

uk,4 uj,4 + i 2
(u4,k , u4,j )
uk,4 uj,4 i AFS
j =k
K1

j =1
K3
K5
K7
1 1/y4,k
y1,j
y4,k y3,j
y4,k y5,j
1 1/y4,k
y7,j
+
1 1/y4,k
y1,j
y+
j =1 4,k
y3,j
y+
j =1 4,k
y5,j
j =1
+
1 1/y4,k
y7,j
. (4)
Finally, the energy of the string states (or anomalous dimension of the SYM operators) is carried
by the middle root only

K4

i
i
=
.
+
2
y4,j
y4,j
i=1
(5)
The potential phase V should be responsible for the interpolation between the YM and the string
equations for small and large t Hooft coupling .
Based on the 1-loop shift analysis of some
classical circular strings [11] Hernandez and Lopez
proposed a universal form for the first 1/ correction in V [12] which should be able to reproduce, together with the finite size corrections to the scaling limit [1315], the 1-loop shift around
any classical solution. This was quite a bold proposal since it relied solely on the analysis of rank
one rigid circular strings. In [16] the proposed phase passed a very nontrivial testit was shown
to reproduce the non-analytic contribution to the 1-loop shift around a simple SU(3) solution.
However, at the present stage, only for the sl(2) one cut solution the 1-loop shift from the Bethe
ansatz equations is completely understood [12,13].
Recently, using the crossing constraint [17] and transcendentality principles [18], the full form
of the scalar factor was conjectured from the string [19] and gauge [20] theory points of view.
It is therefore fair to say that the advance in the last four years was spectacular. On the other
hand it is also true that there is a great deal of conjectures involved one should both check and,
hopefully, proof (or disproof). In this article we establish rigorously the HernandezLopez phase
by fully proving its universality.
Finally we discuss a possible origin for this phase from a nested Bethe ansatz approach.
2. Brief derivation of the HernandezLopez coefficients
One can compute the fluctuation energies, i.e. the energy level spacing in the spectrum of
the string, around any classical string solution in AdS5 S 5 . To do so one should first map the
classical configuration to an 8-sheet Riemann surface [21] described by a set of 8 quasi-momenta
pi (x).2
1 We choose to write it in this very general formthis might at first seem strange in the sense that, for a generic
potential V depending on the position of all the roots, this equation does not seem to describe a factorized scattering
process. Obviously, eventually the phase must be such that this property is satisfied.
2 Throughout all this paper we shall use regular Latin indices i, j, k to label the quasi-momenta. According to Fig. 3
2,
3,
4,
1,
2,
3,
4.
they can take values 1,
75
In particular the classical energy Ecl of the solution is encoded in the large x asymptotics of
ij
ij
the pi (x). Then, to compute the fluctuation energies En one adds Nn poles with residue
4 x 2
(x) = 2
x 1
to the quasi-momenta pi and pj (this procedure was first suggested in [14] in connection with
the study of the 1-loop super YangMills Bethe ansatz). The choice of the pairs of sheets ij
corresponds to the several string polarizations and the mode number n, the Fourier mode of the
excitation, fixes the position xn of the new poles by [22]
pi (xn ) pj (xn ) = 2n.
(6)
Then the excitation energies can be read from the large x asymptotic of the perturbed quasimomenta [22]
E = Ecl +

ij ij
Nn En .
n= ij
An object of prime interest obtained from these fluctuation energies is the 1-loop shift [23]
1
ij
E0 =
(1)F En
2 n=
(7)
ij
where (1)F = 1 for bosonic and fermionic excitations respectively. To sum over all fluctuation energies it is extremely useful to employ the technique used by Schafer-Nameki in [24] and
integrate them in the n plane using the poles of the cotangent to pick the integer values of the
mode numbers n,

1
ij
cot(n)
(1)F En dn
E0 =
(8)
4i
C
ij
where C encircles all the poles of cot (and only them). The structure of singularities of the fluctuation energies in the complex n plane is intricate and we shall discuss it in more detail elsewhere
[25]. For a simple su(2) circular string solution
this analysis was carried out in [24]. Let us consider the sum of the BMN frequencies [26] n2 + J 2 from n = N to n = N with N large.
In the n plane, for each frequency, the integral (8) over n can be deformed to run along the
cuts in the imaginary n axis with branchpoints n = iJ as depicted in Fig. 1(a). Moreover, for
large J , we can replace the cot in (8) by i for the lower/upper half plane. Through (6) we can
map the integration contour to the x plane. For this solution the quasi-momenta are very simple
and
4J x
pi pj = 2
x 1
so that the branchcuts in the n plane are mapped to the unit circle with the branchpoints n = iJ
sent to x = isee Fig. 1(b). The integration contour for large N is mapped to the solid (blue)
contour in Fig. 1(b) and tends to the unit circle as N .
For general classical
solutions the picture is similar. For large n the fluctuation frequencies should behave like n2 so that we will always have two branchpoints like as Fig. 1(a).
These branchpoints are always large if the classical solution has some large global charge [25].
76
(a)
(b)

Fig. 1. (a) Analytical structure of the BMN frequencies n2 + J 2 and integration contour for (8). (b) Same picture in
the x plane obtained through the map 2x = 2nJ . The branchpoints are mapped to x = i. As N the integration
x 1
path is mapped to the unit circle.
(a)
(b)
Fig. 2. (a) Typical analytical structure of the excitation energies as a function of the mode number n. The branchpoints
associated to the cuts going to infinity are large if some charge of the classical solution is large. There could also be extra
cuts in the n plane. The integral (8) can be then split into two contributions Iphase and Ianomaly as depicted in the figure.
(b) The contour Iphase going along the large cuts in the n plane is mapped into some ellipsoidal form in the x plane. The
contours around the extra cuts in the n plane are mapped to the cycles around the cuts of the classical solution around
which we are quantizing.
As before, when computing the integral along these branchcuts we can replace cot(n)
i sign(Im(n)) with exponential precision. Let us call this contribution by Iphase the solid
(blue) contour in Fig. 2(a). Generically, contrary to what happened in the previous example,
this is not the end of the story. There could be additional cuts in the n plane whose contribution
we have to subtract in order to pick the poles in cot (and only these poles) in (8). We denote the
contribution from these integrals over the new cuts in the n plane by Ianomaly the dashed (blue)
contours in Fig. 2(a). For the simple solutions considered in the literature these two terms usually
bear the names of non-analytic and analytic contributions.
77
The large branchcuts in the n plane will then be mapped to some ellipsoidal curve in the
x plane passing through3 x = 1 whereas the integrals around the extra cuts in the n plane
are mapped precisely to the integrals over the cuts of the original classical solution [25]see
Fig. 2(b). Finally, to pass from an integral over the n plane to an integral in the x plane we just
need to use the map (6),
dn =
pi pj
2
(9)
dxn ,
so that we see the typical p and cot(p) which always appear in the finite size correction analysis
[1315]! We will study the general finite size corrections in a forthcoming publication [25] and
elucidate its relation to Ianomaly . In this paper we prove the universality of the HernandezLopez
(HL) scalar factor by analyzing the Iphase contribution around any classical configuration.
As we will see in the following section, the addition of the dressing factor ei V in the middle
equation (4) amounts to adding the potential to each quasi-momenta pi pi V/2. Having this
in mind, let us give a sketch of the proof. As we mentioned in the beginning of this section, by
adding4
(x)
ij
x xn
ij
to the quasi-momenta pi and pj with xn fixed by (6) we are considering a quantum fluctuation
with mode number n and polarization ij. Then, if we want to get the contribution Iphase we should
integrate this pole over n using (9) and sum over the different polarizations with the appropriate
3),
(1,
4)
bosonic excitation coming with a
signs (8). For example for p 1 we have (i, j) = (1,
3),
(1,
4)
for the fermionic excitations summed with a minus signsee
plus sign and (i, j) = (1,
Fig. 3. Then we find that the contribution we must add for each pi is the same and reads
V(x) =
+1
p 3
+ p 4
p 3
p 4

(x)
(1/x) dy
.
x y 1/x y 2
(10)
It is interesting to see that this potential is explicitly x 1/x odd,

V(x) = V(1/x).
(11)
Finally, as we will see in the next section, the combination of quasi-momenta appearing in (10)
is precisely the one from which one reads the global charges Qn ,

p 3 + p 4 p 3 p 4 = y G4 (y) G4 (1/y)
(12)
where
G4 (y) =
Qn+1 y n ,
(13)
n=0
3 For large n the position of the pole should always be close to x = 1 where all quasi-momenta have simple poles
[27].
4 And also image of this pole (1/x) according to x 1/x symmetry (A.1).
ij
1/xxn
78
Fig. 3. The 16 elementary physical excitations are the stacks (bound states) containing the middle node root. From the
left to the right we have four S 5 fluctuations, four AdS5 modes and eight fermionic excitations. The bosonic (fermionic)
stacks contain an even (odd) number of fermionic roots signaled by a cross in the Dynkin diagram of psu(2, 2|4) in the
left.
so that we can expand the denominators in (10) for large x and obtain

1 (r 1)(s 1)
Qs
Qr
V(x) = (x)
(s r)(r + s 2) x s
xr
(14)
r,s=2
r+sOdd
where we recognize precisely the HernandezLopez coefficients! To obtain the values of the
potential for |x| < 1 we can simply use the exact symmetry (11) which is not manifest in the
form (14). In the next section we shall explain the tight relation between this potential V and the
BS equations and provide a detailed derivation of the scalar factor.
3. Constraining the scalar factor
In this section we shall fill the gaps in the sketchy derivation above. First we will start by
explaining the relation between the quasi-momenta and the BeisertStaudacher (BS) equations
in the large limit. We will see that relation (12) follows immediately from the definition of
the quasi-momenta and that the phase V(x) appearing in (4) is simply translated into a potential
for the quasi-momenta pi pi V/2. Then, we clarify the steps leading to (10), preform the
large N limit more carefully.
3.1. Classical limit
One of the main ingredients in the construction of the BS equations was the requirement that
these equations reproduce the classical algebraical curve of [21]. For the sake of completeness,
let us we make the passage to the classical limit explicit and at the same time introduce some
important notations.
The seven BS equations, one for each node of a super Dynkin diagram of the psu(2, 2|4)
algebra, give us the position of the roots ua,j where a = 1, . . . , 7 denotes the Dynkin node and
j = 1, . . . , Ka . The equation associated to the middle node of the Dynkin diagram takes the
form (4). In fact, for a given number of roots, the Bethe equations have several solutions. To
79
classify them one takes the log of the Bethe equation associated with each root. The different
choices of the branch of the log correspond to the different solutions. In other words, to each
root ua,j one should associate a mode number na,j . Thus, a choice of mode numbers amounts to
fixing the quantum state.
To study the large scaling limit,
uj L Ki ,
(15)
of the BS equations it is useful to introduce 8 functions {p 1 , p 2 , p 3 , p 4 , p 1 , p 2 , p 3 , p 4 }. First we
define the resolvents Ga and Ha for each type of roots
Ka

(x)
.
x ya,j
j =1
j =1
Then, denoting H a (x) = Ha (1/x) and J = L/ , we have
Ga (x) =
Ka

(ya,j )
,
x ya,j
Ha (x) =
(16)
2J x G4 (0)x
1
H1 + H2 + H 2 H 3 + V,
2
x2 1
2J x + G4 (0)
1
H1 H 3 + H 4 + V,
p 1 = +
2
x2 1
2J x G4 (0)x
1
H2 + H3 + H 1 H 2 + V,
p 2 = +
2
2
x 1
2J x + G4 (0)
1
+ H3 H4 + H 1 + V,
p 2 = +
2
x2 1
2J x G4 (0)x
1
H5 + H6 + H 6 H 7 V,
p 3 =
2
2
x 1
2J x + G4 (0)
1
H5 + H4 H 7 V,
p 3 =
2
x2 1
2J x G4 (0)x
1
H6 + H7 + H 5 H 6 V,
p 4 =
2
x2 1
2J x + G4 (0)
1
+ H7 + H 5 H 4 V.
p 4 =
(17)
2
x2 1
In the continuous limit, with a large number of roots for each mode number, roots with the
same na,j will condense into square root cuts. Moreover roots belonging to consecutive nodes
of the Dynkin diagram can form bound states and in this way give rise to a cut Cij connecting
non-consecutive Dynkin nodes. As mentioned in the introduction only the middle roots u4 carry
energy (5). Then the 16 elementary physical excitations are the bound states represented in Fig. 1,
each bound state corresponding to a different string polarization. These bound states, named
stacks, were first found in [28]we refer to this article for more details. We denote the values of
a function p(x) above and below some of these cuts Cij by p . Then, in the large limit, we can
recast the seven BS equations as5
p 1 = +
pi+ pj = 2nij ,
x Cij .
(18)
5 In the notation of [9] we use the grading = = 1 corresponding to the Dynkin diagram of Fig. 3. Moreover we
1
2
are considering the 1/ corrections coming from V O(1/ ) but we drop in this paper the finite size corrections
usually called by anomaly terms. We shall discuss them separately [25].
80
From the definition of the quasi-momenta we can read the large and small x asymptotics, the
behavior at the x = 1 poles and understand how the several quasi-momenta are related by the
x to 1/x inversion symmetry. To analyze
the classical limit, moreover, we can drop the potential
V whose contribution is of order 1/ . We conclude that the analytical properties of p, together

with the equations just mentioned are exactly the same as those of the quasi-momenta defining
the 8-sheet Riemann surface of Beisert, Kazakov, Sakai and Zarembo [21]. Thus, the classical
limit coincides with this continuous limit.
3.2. Deriving the HernandezLopez scalar factor
Until the end of this section we drop the phase V in the BS equations (and thus also in (17)).
If we add a stack connecting sheets i and j to some configuration of Bethe roots with all roots
condensed into some cuts as described above, the position of the new stack will be given by (6)
and all the other roots will be slightly shifted uj u j . Then the energy of the new configuration is given by the energy of the original configuration plus the fluctuation energy with mode
number n associated to the corresponding string polarization [22]
ij
= + En .
(19)
Let us now perform a simple rewriting exercise and treat each of the roots of this new stack
separately in pk . That is, if the stack contains a root associated with the Dynkin node a we write
Ga (x) Ga (x) +
(xn )
,
x xn
Ha (x) Ha (x) +
(x)
x xn
where Ga and Ha are now defined with the sum over roots going only over a = 1, . . . , Ka where
Ka is the original number of roots of type ua,j . Then, with this new stack, each quasi-momentum
pk can be written as before but using the new resolvents Ga and Ha containing only the Ka
ij
original roots plus an extra term Vk (indices i, j indicate the stack we added) which we call
potential and read6
+
+
V1 (x) ij
+1
1i
2i
ij
(x)
)
x
(x
V2 (x)
+1
+2i
+1i
(1/x)
n

+

,
(20)
=
2
ij
ij
ij
3j
4j
1 x 1 (x )2
V3 (x)
x xn
1/x xn
n
4j
3j
1
V4 (x)
and
V1 (x) ij
1
+1i
+2i
ij
V2 (x)
1 1 (xn ) +2i
+ (1/x)
(x)

+ 1i
.
ij
ij
V3 (x)
+1 x 2 1 x ij
3j
4j
x xn
1/x xn
n
4j
3j
+1
V4 (x)
(21)
In (17) we saw that the inclusion of the phase V in the middle node of the BS equations amounts
to adding the same potential V to all quasi-momenta. The main difference to what we have there
is that now the xn roots hidden in the potential also contribute to the charges (because every stack
6 For example, consider a fermionic stack i, j = 2,
3 connecting p 2 and p 3 . As we see from Fig. 3 this stack is made
of two almost coincident u4 and u5 roots. The first term in the potentials comes from the resolvent of the middle node
though the G4 (0) and G4 (0) terms present in all quasi-momenta (17). The new terms in p 1 , p 2 , p 3 , p 4 come from the
resolvents H4 and H5 which, for the other quasi-momenta, are either not present or appear with opposite signs.
(a)
81
(b)
Fig. 4. (a) The non-analytic contribution Iphase is given by the integral (24) whose integration path goes along the
large cuts discussed in Section 2. The difference in orientations with respect to Fig. 2(a) is due to the absence of cot in
expression (24) compared to (8). (b) In the x plane the integral can safely be deformed to go over the upper and lower

+1
to denote 12 C + 12 C . The relation between the
halves of the unit circle. In the main text we use the shorthand 1
1
2
large N regularization in the n plane and the
regularization in the x plane is discussed in Appendix A.
contains a u4 root)

dx G4 (x) (xn )
+ m .
Qm =
2i x m
xn
(22)
ij
Moreover the potentials Vk are different for different quasi-momenta.

Suppose that instead of (19) we want
= + Iphase
(23)
with7
1
Iphase
lim
2 N
N
ij
(1)F En dn
(24)
ij
where the contour in the n plane is as depicted in Fig. 4(a). Then, by linearity, we need only to
ij
replace Vk by

1
lim
Vk (x) =
2 N
N
ij
ij
(1)F Vk x, xn dn.
(25)
ij N
Let us now show that all the potentials Vk are the same up to a sign and are equal to V (10) from
the previous section. Indeed
(1) For each summand in (25) we can pass to the x plane through (9). As we explain in
Appendix A we can assume that for all fluctuation energies the corresponding integral in the
7 As it is discussed in [22] in order to define this quantity unambiguously a precise prescription for the labeling of the
fluctuation energies is needed. This point is discussed in Appendix A.
82
x plane is the same and goes over the upper and lower halves of the circle of radius 1 +
centered at the origin as plotted in Fig. 4(b).

(2) The first terms in (20) and (21) do not contribute to Vk . Indeed, if we integrate some
ij
function of xn summed over the 16 possible excitations listed in Fig. 3 with a (1)F weight
N

ij
F
(1)
f xn dn
ij
we obtain, using (9),8

+1+
f (y)
i=1,2,j=3,4

dy
p i p j + p i p j p i p j p i p j
= 0.
2

(26)
(3) Finally, consider for example V1 . We have

1
V1 (x) =
2
1+

(x) dy
p 1 p 3 + p 1 p 4 p 1 p 3 p 1 p 4
x y 2
1+

(1/x) dy
p 2 p 3 + p 2 p 4 p 2 p 3 p 2 p 4
.
1/x y 2
We see that p 1 and p 2 drop out so that the expression simplifies considerably. The same happens
for the other Vi and moreover, due to the super-tracelessness of the monodromy matrix, p 1 +
p 2 + p 3 + p 4 = p 1 + p 2 + p 3 + p 4 , and all the potentials are equal. Using (12) we have
1
V1,
2,
1,
2 (x) = V3,
4,
3,
4 (x) V(x) =
1

(x)
(1/x) dy
y G4 (y) G4 (1/y)
.
x y 1/x y 2
(27)
Notice also that due to (3) the extra terms in the charges (22) give no contribution! Thus, since
now all potentials Vk are equal to V, we proved that for arbitrary configuration of Bethe roots,
the addition of the HernandezLopez phase will lead to an energy shift given by Iphase .
As we saw in the previous section, to obtain the HernandezLopez phase as usually written in
terms of charges it suffices to use (13). If we want, on the other hand, to write
ei V (y4,k ) =
K4
ei(y4,k ,y4,j )
j =k
8 We can as well use the quasi-momenta with the resolvents G and H summed only over the original roots because
a
a
the inclusion of the potentials in (9) is an higher order effect.
83
where the factorized scattering property is manifest we just need to use the definition (16) and
integrate over y to get9

1
1
x +1y 1
+
log
x 1y +1
(x y)2 (xy 1)2

2
.
+
(x y)(xy 1)
(x)(y)
(x, y) =

(28)
The real scattering phase, the phase that describes the scattering between two magnons in the
Bethe ansatz equation, must inherit the explicit x to 1/x oddness (11) of the potential. To obtain
the values of the phase for |x| < 1 we use (1/x, y) = (x, y). Alternatively, we recall that
the contour in Fig. 4(b) tells us that to be completely rigorous we should replace the log in
(28) by 12 (log+ (. . .) + log (. . .)) where log has a branchcut in the upper/lower half of the unit
circlesee Fig. 4(b). Then the expression for (x, y) becomes explicitly x to 1/x odd and is
discontinuous on the unit circle. If, on the other hand, we analytically continue the expression
(28) from some point x outside the unit circle up to some point 1/x inside the unit circle we get
2i from one of the log so that we trivially find
i(x, y) + i (1/x, y) = (x)(y)

1
1
+
,
(x y)2 (xy 1)2
which is precisely Janiks crossing relation [17] for the dressing factor at 1/ order [29].
4. Speculations on nesting
There are several indications pointing towards the existence of an extra hidden level in
the Bethe ansatz equations. In [31,32] the classical algebraic curve for the string moving in
S 5 AdS5 S 5 was obtained as the classical limit of the quantum nested Bethe ansatz equations
coming from the Zamolodchikovs bootstrap procedure [33] where in addition to the magnons
described by the roots uj we have the rapidities of the relativistic particles with O(N) isotopic degree of freedom. In [34] it was observed that the BDS equations [6] mentioned in the
introduction (1) could be obtained from the Hubbard model where the electron has a spin which
can create spin waves described by the roots uj , but also has momentum p . In both cases the
introduction of the extra level simplifies the structure of the Bethe equations considerably. Moreover, in the same setup as in [32] it was also found [35] that the elimination of the rapidities
from the simple Bethe equations would lead to the more complicated AFS equations (2). More
recently it was argued [36] that the conjectured all loop dressing factor should also bear its origin from an extra level in the Bethe ansatz equations.10 Thus we would like to understand if
(27) could have imprinted signs of a nested Bethe ansatz structure. Since the generalization of
the Bootstrap method for the supercoset PSU(2, 2|4) is not known we will proceed at a rather
speculative and schematic level. Let us recall that in the scaling limit the S 5 nested Bethe ansatz
9 By resuming the HernandezLopez coefficients the phase (x, y) was written down in [29], see also Appendix B
in [30].
10 In [37] it was argued that the dressing factor could instead come from a dressing of the bare vacuum in the original
[9] equations with no scalar factor at all.
84
equations read (see e.g. Eqs. (6.3) or (3.15) in [32])

G
/ + Ka Ga = 2m,
z [2, 2],
G
/ a + Kab Gb + Ka G = 2na ,
z Ca ,
(29)
where the Ks are numeric coefficients and where the cut from [2, 2] is the image of the unit
circle under the Zhukovsky map. The discontinuity of G is related to the density of the extra
level particles and is given on the x plane by, roughly speaking,
=
p2 (x) p2 (1/x)
.
2i
(30)
Suppose we want to take into account 1/ corrections coming from the extra level. We denote
the extra term in the first equation in (29) by 1 V (, {u}). Then, if we want to eliminate s
from this equation and plug them into the second one to obtain some effective equation for the
magnons as in [35], we must solve G
/ = 1 V ( ) and plug the solution of this Hilbert problem
into the second equation. In the second equation this will then appear as an extra phase. The
HernandezLopez phase seems to fit into this construction. Indeed, if we go to the Zhukovsky
plane in (27) through z = x + 1/x and w = y + 1/y we will have
4
V(x) =
2
2
4 w 2 dw
V (w)
,
z2 4 w z
V (w) =
w [G4 (y) G4 (1/y)]

,
2i
which precisely indicates that it came from the solution of a Hilbert problem with V (w) resembling the derivative of the density in (30). Indeed

p 2 (y) p 2 (1/y) p 2 (y) p 2 (1/y)
V (w) = w
+
.
2i
2i
As explained in Section 2 we are considering classical solutions with some large conserved
charge. Then the quasi-momenta p(x) scales like that charge close to the unit circle [25] and so
will the density (30). Thus if we assume that the anomaly for these roots is of the usual form
coth we see that we can drop the cotangent with exponential precision and we are left
precisely with the derivative of the density of particles which, as we argued above, strongly
resembles V (w)! Moreover, the integral over the unit circle in Fig. 2(b) will be mapped to the
cycle around the from z = 2 to z = 2 thus leading to the very democratic Fig. 5.
Fig. 5. On z plane the contributions Iphase and Ianomaly can be treated in absolutely equal footing. This hints the existence
of an extra level in the Bethe ansatz equations whose finite size corrections would give the contribution Iphase .
85
5. Conclusions
The fluctuation energies around any classical solution can be computed from the classical
algebraic
curve [22] which, a priori, contains no information about the HLphase suppressed
as 1/ . On the other hand,if we expand the energy of some state in 1/ we will obtain
the classical energy of order plus a finite correction which is known [23] to be the sum of
these fluctuation frequencies. In other words, the sub-leading order is constrained by the leading
order!
This interconnection between the leading and sub-leading terms means that, if one believes
in the existence of a Bethe ansatz description of the quantum string, then the first correction to
the dressing factor is completely constrained. Thus we might wonder if this procedure can be
iterated to fix order by order, the full scalar factor.
In this paper we found that the first correction to the dressing factor must have the form
1
V(x) =
1

(x)
(1/x) dy
y G4 (y) G4 (1/y)
,
x y 1/x y 2
in order to accommodate for the 1/ non-analytical part of the 1-loop shift for any classical
configuration.
One can see that the 1/ corrections in the BeisertStaudacher equations with the
HernandezLopez phase to any conserved charge Qn is given by one half of the graded sum
of the corresponding charges of the fluctuationsas it was the case for energy.
Moreover, this representation of the phase has several interesting features. First of all, it is
extremely simplethe HernandezLopez coefficients follow after a trivial expansion of the resolvent in conserved charges and to find the scattering phase (x, y) between two magnons one
merely needs to plug the definition of the resolvent into the integral without the need to perform
any re-summation. By construction, the cut structure is also very clear and thus the crossing
relation becomes transparent.
Finally this phase has imprinted signs pointing towards the existence of an extra level of
roots (which are in fact rapidities of physical particles of the theory) in the Bethe ansatz equations. These extra roots live in the unit circle and in the quasi-classical limit condense into some
smooth distribution whose density resembles the G4 (y) G4 (1/y) [32] appearing in V(x). It is
very likely that the anomaly associated with the roots of this new level naturally reproduce the
HernandezLopez phase.
Acknowledgements
We would like to thank K. Zarembo and V. Kazakov for many useful discussions. The work
of N.G. was partially supported by French Government PhD fellowship, by RSGSS-1124.2003.2
and by RFFI project grant 06-02-16786. P.V. is funded by the Fundao para a Cincia e Tecnologia fellowship SFRH/BD/17959/2004/0WA9.
Appendix A. Large N limit
In the x plane the contour in Fig. 4(a) is mapped to that in Fig. 4(b). For large N the contour
ij
ij
starts at 1
(N ) and ends at +1 +
+ (N ). In this appendix we perform a careful analysis of
the large N limit.
86
A.1. Asymptotics of quasi-momenta and expansion of xn

Large ns are mapped to the vicinity of 1 where

p 2 +
a n (x 1)n ,
p 2 +
a n (x 1)n ,
+
+
x1
x 1
n=0
n=0

p 3
p 3
+
+
bn (x 1) ,
bn (x 1)n .
x1
x 1
n=0
n=0
The remaining quasi-momenta are fixed by the x 1/x symmetry

p 1,2 (x) = 2m p 2,1 (1/x),
p 3,4 (x) = +2m p 4,3 (1/x),
p 1,2,3,4 (x) = p 2,1,4,3 (1/x).
(A.1)
ij
From this expansion we can read the large n behavior of xn defined by (6). Let us, however,
use a more general definition
ij
ij
pi xn pj xn = 2(n mi + mj ).
(A.2)
ij
For n all xn are close to 1 and we find

+ O 1/n2
(A.3)
n
where we notice that the first 1/n coefficient is universal and fixed uniquely by the residues of
the quasi-momenta.
ij
xn = 1 +
A.2. Large N versus

regularization
The main goal of this appendix is to justify the integration path used in the main text where
for all ij the integral in the x plane starts from 1
and ends at 1 +
as depicted on the
Fig. 5(b). However by definition (25) we have to start from the large N regularization. These two
ij
regularization in principal are not equivalent, since xN s are not exactly equal for all ij and thus
we should calculate the difference between both regularizations. For example
reg
Vk (x) =
1
lim
2 N,
0
xNij

ij
ij
xN
1+

ij
(1)F Vk (x, y)
pi (y) pj (y)

2
dy.
(A.4)
ij
The difference of the integrals above can be rewritten as a sum of two integrals, one from xN to
ij
1
and another from 1 +
to xN , and thus we can use expansion of quasi-momenta around
reg
1 to evaluate this Vk (x). One can see that only first terms in (20), (21) could be responsible
for a non-zero difference given by

2x
1
1
reg
V (x) =
+
(m + m1 + m2 m1 m2 )
1,2
(x 2 1) +
(m + m3 + m4 m3 m4 ),

reg
1,2
87

1
1
(m + m1 + m2 m1 m2 )
(x 2 1) +
(m + m3 + m4 m3 m4 ).
V (x) =
2x
If the potentials Vk are to be originated from a phase in the middle note of the BS equations then
they should all be equal. This means that in order to be consistent with the phase origin, these
two terms should be zero. Fortunately it is possible to choose mi in such a way that it is so. For
example
m1 = m,
m4 = m
(A.5)
and all the others mi are zero. This amounts to a different prescription for the mode numbers
comparatively to [22]. For obvious reasons let us denote it by Bethe ansatz friendly prescription.
Contrary to what we had in [22] we have no obvious argument, except the obvious Bethe ansatz
friendliness, in favor of this new prescription. For the sl(2) and su(2) one cut solutions this
prescription gives the same result (with exponential precision in large J ) as in [11,13,24].
By the same means we can see that in (22) the last term does not contribute in the Bethe ansatz
friendly prescription.
References
[1] H. Bethe, On the theory of metals. 1. Eigenvalues and eigenfunctions for the linear atomic chain, Z. Phys. 71 (1931)
205.
(1998) 231, Int. J. Theor. Phys. 38 (1999) 1113, hep-th/9711200;
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string theory, Phys. Lett.
B 428 (1998) 105, hep-th/9802109;
E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[3] J.A. Minahan, K. Zarembo, The Bethe-ansatz for N = 4 super YangMills, JHEP 0303 (2003) 013, hep-th/0212208.
[4] N. Beisert, The complete one-loop dilatation operator of N = 4 super YangMills theory, Nucl. Phys. B 676 (2004)
3, hep-th/0307015.
[5] N. Beisert, M. Staudacher, The N = 4 SYM integrable super spin chain, Nucl. Phys. B 670 (2003) 439, hep-th/
0307042.
[6] N. Beisert, V. Dippel, M. Staudacher, A novel long range spin chain and planar N = 4 super YangMills,
JHEP 0407 (2004) 075, hep-th/0405001.
[7] V.A. Kazakov, A. Marshakov, J.A. Minahan, K. Zarembo, Classical/quantum integrability in AdS/CFT, JHEP 0405
(2004) 024, hep-th/0402207.
[8] G. Arutyunov, S. Frolov, M. Staudacher, Bethe ansatz for quantum strings, Phys. Rev. D 66 (2002) 010001,
hep-th/0406256.
[9] N. Beisert, M. Staudacher, Long-range PSU(2, 2|4) Bethe ansaetze for gauge theory and strings, Nucl. Phys. B 727
(2005) 1, hep-th/0504190.
[10] N. Beisert, The su(2|2) dynamic S-matrix, hep-th/0511082.
[11] S. Frolov, A.A. Tseytlin, Multi-spin string solutions in AdS5 S 5 , Nucl. Phys. B 668 (2003) 77, hep-th/0304255;
S. Frolov, A.A. Tseytlin, Quantizing three-spin string solution in AdS5 S 5 , JHEP 0307 (2003) 016, hep-th/
0306130;
G. Arutyunov, J. Russo, A.A. Tseytlin, Spinning strings in AdS5 S 5 : New integrable system relations, Phys. Rev.
D 69 (2004) 086009, hep-th/0311004;
I.Y. Park, A. Tirziu, A.A. Tseytlin, Spinning strings in AdS5 S 5 : One-loop correction to energy in SL(2) sector,
JHEP 0503 (2005) 013, hep-th/0501203;
N. Beisert, A.A. Tseytlin, On quantum corrections to spinning strings and Bethe equations, Phys. Lett. B 629
(2005) 102, hep-th/0509084.
[12] R. Hernandez, E. Lopez, Quantum corrections to the string Bethe ansatz, JHEP 0607 (2006) 004, hep-th/0603204.
[13] S. Schafer-Nameki, M. Zamaklar, K. Zarembo, Quantum corrections to spinning strings in AdS5 S 5 and Bethe
ansatz: A comparative study, JHEP 0509 (2005) 051, hep-th/0507189.
88
[14] N. Beisert, L. Freyhult, Fluctuations and energy shifts in the Bethe ansatz, Phys. Lett. B 622 (2005) 343, hep-th/
0506243.
[15] N. Beisert, A.A. Tseytlin, K. Zarembo, Matching quantum strings to quantum spins: One-loop vs. finite-size
corrections, Nucl. Phys. B 715 (2005) 190, hep-th/0502173;
R. Hernandez, E. Lopez, A. Perianez, G. Sierra, Finite size effects in ferromagnetic spin chains and quantum
corrections to classical strings, JHEP 0506 (2005) 011, hep-th/0502188;
N. Gromov, V. Kazakov, Double scaling and finite size corrections in sl(2) spin chain, Nucl. Phys. B 736 (2006)
199, hep-th/0510194.
[16] L. Freyhult, C. Kristjansen, A universality test of the quantum string Bethe ansatz, Phys. Lett. B 638 (2006) 258,
hep-th/0604069.
[17] R.A. Janik, The AdS5 S 5 superstring worldsheet S-matrix and crossing symmetry, Phys. Rev. D 73 (2006)
086006, hep-th/0603038.
[18] A.V. Kotikov, L.N. Lipatov, DGLAP and BFKL equations in the N = 4 supersymmetric gauge theory, Nucl. Phys.
B 661 (2003) 19, hep-ph/0208220;
A.V. Kotikov, L.N. Lipatov, DGLAP and BFKL equations in the N = 4 supersymmetric gauge theory, Nucl. Phys.
B 685 (2004) 405, Erratum.
[19] N. Beisert, R. Hernandez, E. Lopez, A crossing-symmetric phase for AdS5 S 5 strings, JHEP 0611 (2006) 070,
hep-th/0609044;
N. Beisert, On the scattering phase for AdS5 S 5 strings, Mod. Phys. Lett. A 22 (2007) 415, hep-th/0606214.
[20] N. Beisert, B. Eden, M. Staudacher, Transcendentality and crossing, J. Stat. Mech. 0701 (2007) P021, hep-th/
0610251;
B. Eden, M. Staudacher, Integrability and transcendentality, J. Stat. Mech. 0611 (2006) P014, hep-th/0603157.
[21] N. Beisert, V.A. Kazakov, K. Sakai, K. Zarembo, The algebraic curve of classical superstrings on AdS5 S 5 ,
Commun. Math. Phys. 263 (2006) 659, hep-th/0502226.
[22] N. Gromov, P. Vieira, The AdS5 S 5 superstring quantum spectrum from the algebraic curve, hep-th/0703191.
[23] S. Frolov, A.A. Tseytlin, Semiclassical quantization of rotating superstring in AdS5 S 5 , JHEP 0206 (2002) 007,
hep-th/0204226.
[24] S. Schafer-Nameki, Exact expressions for quantum corrections to spinning strings, Phys. Lett. B 639 (2006) 571,
hep-th/0602214.
[25] N. Gromov, P. Vieira, Complete 1-loop test of AdS/CFT, arXiv: 0709.3487 [hep-th].
[26] D. Berenstein, J.M. Maldacena, H. Nastase, Strings in flat space and pp waves from N = 4 super YangMills, AIP
Conf. Proc. 646 (2003) 3.
[27] G. Arutyunov, S. Frolov, Integrable Hamiltonian for classical strings on AdS5 S 5 , JHEP 0502 (2005) 059,
hep-th/0411089.
[28] N. Beisert, V.A. Kazakov, K. Sakai, K. Zarembo, Complete spectrum of long operators in N = 4 SYM at one loop,
JHEP 0507 (2005) 030, hep-th/0503200.
[29] G. Arutyunov, S. Frolov, On AdS5 S 5 string S-matrix, Phys. Lett. B 639 (2006) 378, hep-th/0604043.
[30] S. Schafer-Nameki, M. Zamaklar, K. Zarembo, How accurate is the quantum string Bethe ansatz?, JHEP 0612
(2006) 020, hep-th/0610250.
[31] N. Mann, J. Polchinski, Bethe ansatz for a quantum supercoset sigma model, Phys. Rev. D 72 (2005) 086002,
hep-th/0508232.
[32] N. Gromov, V. Kazakov, K. Sakai, P. Vieira, Strings as multi-particle states of quantum sigma-models, Nucl. Phys.
B 764 (2007) 15, hep-th/0603043.
[33] A.B. Zamolodchikov, A.B. Zamolodchikov, Relativistic factorized S-matrix in two-dimensions having O(N )
isotopic symmetry, Nucl. Phys. B 133 (1978) 525, JETP Lett. 26 (1977) 457.
[34] A. Rej, D. Serban, M. Staudacher, Planar N = 4 gauge theory and the Hubbard model, JHEP 0603 (2006) 018,
hep-th/0512077.
[35] N. Gromov, V. Kazakov, Asymptotic Bethe ansatz from string sigma model on S 3 R, hep-th/0605026.
[36] A. Rej, M. Staudacher, S. Zieme, Nesting and dressing, hep-th/0702151.
[37] K. Sakai, Y. Satoh, Origin of dressing phase in N = 4 super YangMills, hep-th/0703177.
KaluzaKlein induced supersymmetry breaking

for braneworlds in type IIB supergravity
Jean-Luc Lehners a,,1 , Paul Smyth b,2 , K.S. Stelle c,3
a DAMTP, CMS, Wilberforce Road, CB3 0WA, Cambridge, UK
b Institute for Theoretical Physics, K.U. Leuven, Celestijnenlaan 200D, B-3001 Leuven, Belgium
c The Blackett Laboratory, Imperial College London, Prince Consort Road, London SW7 2AZ, UK
Received 16 May 2007; accepted 31 August 2007

Abstract
We consider Z2 -symmetric braneworlds arising from 5-sphere compactifications with 5-form flux in
type IIB supergravity. This KaluzaKlein reduction produces a D = 5 theory which supports 12 -supersymmetric Z2 -symmetric domain-wall solutions. However, upon lifting such solutions back to D = 10, one
finds that supersymmetry is broken by 5-sphere KaluzaKlein effects. This happens owing to the action on
the Killing spinor of the Z2 SO(1, 9) symmetry, which requires an orientation-reversing transformation in
the 5-sphere directions together with the flip of the orbifold coordinate. We study the consequences of this
supersymmetry breaking for the masses of fermion fluctuation modes about the brane background and find
a natural two-scale hierarchy: some bulk modes have characteristic masses of order L1 but other modes
5
more closely associated to the branes have an additional factor exp( L ), where L5 is the AdS5 length
5
parameter and is the orbifold size.
PACS: 04.65.+e; 04.50.-h; 11.27.+d
Keywords: Supersymmetry breaking; Domain walls; Compactifications
E-mail addresses: j.lehners@damtp.cam.ac.uk (J.-L. Lehners), paul.smyth@fys.kuleuven.be (P. Smyth),

k.stelle@imperial.ac.uk (K.S. Stelle).
1 Research supported by PPARC.
2 Research supported in part by the Belgian Federal Office for Scientific, Technical and Cultural Affairs through the
Interuniversity Attraction Poles Programme Belgian Science Policy P5/27 and by the EU under contract MRTN-CT2004-005104.
3 Research supported in part by the EU under MRTN contract MRTN-CT-2004-005104 and by PPARC under rolling
grant PP/D0744X/1.
doi:10.1016/j.nuclphysb.2007.08.016
90
J.-L. Lehners et al. / Nuclear Physics B 790 (2008) 89110
1. Introduction
Perhaps the most dramatic way in which string theory has modified our picture of spacetime
so far is by the inclusion of extra dimensions and branes. Branes have the remarkable property
that they can localise YangMills gauge theories on themthis then leads to the braneworld
picture, in which our universe is a brane embedded in a higher-dimensional bulk spacetime. The
RandallSundrum models in particular, in which the bulk is 5-dimensional anti-de Sitter space,
have been studied extensively due to their simplicity and because they provide a possible solution
to the hierarchy problem [1], while also being able to localise gravitons on a brane [2]. However,
in order to take these models seriously, one would like to see them emerge as a solution of the
supergravity approximations to string theory. Such a solution was presented in [3,4] in the context
of 5-sphere compactifications of type IIB supergravity [5] (it is clear from the fact that type IIB
supergravity admits an AdS5 S 5 vacuum that this is a natural place to start). In the present paper
we will study this embedding of the RandallSundrum geometry, and the more general family of
Z2 -symmetric braneworlds of which it is a limiting case, in more detail from a 10-dimensional
point of view.
One of the main outstanding questions is how supersymmetry may be broken in braneworld
models, and this is the question that we address here. We find that supersymmetry is in fact
automatically broken at the location of the Z2 symmetric branes in all the present family of solutions, owing to the geometry of the internal 5-sphere. The reason is the following: due to its
chirality (in particular the self-duality of the 5-form), type IIB supergravity actually admits an
SO(1, 9) rather than an O(1, 9) symmetry. This means that when one is forming a Z2 -symmetric
braneworld, one needs to mod out by a Z2 that is an element of SO(1, 9). Thus, if we flip the
orbifold coordinate y y, we must accompany this transformation with a reversal of the orientation of the internal 5-sphere. However, because of the curvature of the 5-sphere, the Killing
spinor equation is sensitive to the spheres orientation, and this makes the Killing spinor discontinuous at the location of the branes. Consequently, supersymmetry is broken on the branes,
while being preserved in the bulka phenomenologically attractive setup.
As we will show, this breaking of supersymmetry is not manifest from the 5-dimensional point
of view, but can only be appreciated by including the internal manifold in the analysis. This is
why we call this type of supersymmetry breaking KaluzaKlein induced. It is also clear from the
general argument just presented that this mechanism will apply to all Z2 -symmetric braneworlds
in type IIB supergravity, as long as the internal manifold is curved.
In order to illustrate the effects of this supersymmetry breaking, we study a class of bosonic
zero modes as well as their fermionic superpartners. We perform our analysis in linearised perturbation theory about the braneworld background, taking into account the corresponding brane
actions. The modes that we focus on are those which are factorisable with regard to their worldvolume and orbifold dependencies, and which have a profile in the orbifold direction such that,
were supersymmetry not broken, they would appear as massless fields from the 4-dimensional
point of view. Here, however, the fermionic modes acquire a mass, while the bosons (which are
insensitive to the orientation of the 5-sphere) remain massless. The mass of the fermions depends
crucially on their y-dependence. In the most common case, the resulting mass is naturally of the
order of the compactification scale L5 , which may be taken to be near the GUT or Planck scale.
However, if the fermionic modes are such that they have a y-dependence that evolves contrary
to the bulk warping, then their mass is suppressed by an additional bulk warp factor. In this way
one obtains two scales of supersymmetry breaking, and thus both heavy and light fermions, by
the same mechanism.
91
2. Dimensional reduction of type IIB supergravity on a 5-sphere

Type IIB supergravity can be given a formulation in terms of a Lagrangian, supplemented by
the self-duality condition on the 5-form field. Keeping only the graviton, the gravitino and the
5-form, we have

1 2
LIIB = g R
(2.1)
F[5] M MNP D N P ,
4 5!
F[5] = F[5] .
(2.2)
Here D M denotes the supercovariant derivative which also appears in the supersymmetry transformation of the gravitino:

i
M = D M = M +
(2.3)
FN P QRS N P QRS M ,
16 5!
where is the (chiral) spinorial parameter of the transformation. We will dimensionally reduce
this theory on a 5-sphere S 5 . In this section we are only interested in the dimensional reduction
of the bulk. We will find a domain wall solution to this theory, and in the subsequent sections we
will be concerned with the modifications required by the presence of these lower-dimensional
hypersurfaces. The -matrices are decomposed according to
m = m 1 1 ,
a = 1 a 2 ,
(2.4)
(2.5)
where the i are the Pauli matrices. With this decomposition, the ten-dimensional chirality operator is given by 11 = 1 1 3 . The 10-dimensional fields and the supersymmetry parameter
are then dimensionally reduced according to [5,6]

2
ds10
(2.6)
= e2 ds52 + e2 ds 2 S 5 ,
5
8
F[5] = 4me [5] + 4m[5] S ,
(2.7)

1
1
m = e 2 (m + m )
(2.8)
,
0

3i 11
1
,
e 10 a
a =
(2.9)
0
5

1
1
= e 2
(2.10)
0
where is the breathing mode of the sphere, i.e. determines the volume of the sphere; denotes
a Killing spinor on the 5-sphere. Note that we have chosen and such that they are of positive
(10-dimensional) chirality. In order to obtain canonically normalised fields in 5 dimensions, one
also has to impose

3
1 5
,
= .
=
(2.11)
4 3
5
The resulting 5-dimensional bulk theory is the maximal (32-supercharge) SO(6) gauged supergravity [7]. The 32-supercharge structure is generated by four complex, independent 4component D = 5 spinors arising from a 5-sphere Killing spinor in the 4 of SU(4) SO(6).
92
The Z2 symmetry we are interested in acts in the same way on each of these spinors, and so
for our purposes it shall suffice to consider just one of them. Thus, from now on we shall focus
on a single gravitino and adopt a minimal D = 5 (8 real spinor component) notation (see [6,8]).
The graviton supermultiplet contains the gravitino m and a vector. The breathing mode scalar
belongs to a massive vector multiplet [6,7] which also contains a spinor . Since the two vectors
play no role in what follows, we will set them to zero henceforth. For the reduced set of fields
that we are considering, the 5-dimensional theory is described by the Lagrangian4

2

1
1 m
1 W
W
2
mnp
L5 = g R () V () m
Dn p m
2
2
4 2
16

1 W

1
n m
m n
m
m
+ ,m m + m +
(2.12)
m m ,
4
8
where V () has the double exponential form [5]
16
V = 8m2 e8 R5 e 5
(2.13)
and R5 represents the Ricci scalar of S 5 . The 5-dimensional supercovariant derivative Dn is

defined in terms of the superpotential W by
D n = n +
1
W n ,
24
(2.14)
where
R5 8
e5 .
20
As usual, the potential V can be written in terms of the superpotential W according to

2
1
2
W2 .
V = W,
8
3
W () = 8me
+ 20
The fermionic supersymmetry transformations are

1
m = Dm = m + m W ,
24

1 m
1
=
m W, .
2
4
(2.15)
(2.16)
(2.17)
(2.18)
This 5-dimensional theory admits a two-parameter domain wall solution given by [5]
ds52 = e2A dx dx + e2B dy 2 ,
e
e
7
15
4A
= H = k|y| + c,
= b1 H
2/7
+ b2 H
5/7
B = 4A,
k > 0,
(2.19)
where 3kb1 = 28m and 3kb2 = +28 R205 and c is a constant. The linear harmonic function
H (y) is taken to admit a second (trough-like) kink at y = ; we thus have a positive-tension
4 We are not considering higher-order terms in the fermions here. See also [9] and [10] for a discussion of the fermionic
equations of motion.
93
brane at y = 0 and a negative-tension one at y = in a HoravaWitten-like setup with the two

branes located at the endpoints of an S 1 /Z2 orbifold. In the limit where the scalar sits at the
minimum of its potential, the bulk becomes AdS5 and the double domain wall configuration then
represents the embedding of the RandallSundrum model [1] in type IIB supergravity [3,4].
To display the RandallSundrum AdS5 patch geometry explicitly [3], consider the positive 7
tension brane at y = 0 and take H (0) = c > H = e 15 + k, > 0, where is the constant
24
R5
scalar field in the AdS5 S 5 solution (satisfying e 5 = 20m
2 ). Then take the limit k 0+ and
change coordinates using |y| = e
metric
2|z|
L5
ds 2 = e
L4 |z|
5
to obtain the Poincar-coordinate form of the AdS5
dx dx + dz2 ,
(2.20)
where
1
L5 = m
20m2
R5
5
6
(2.21)
is the length parameter of the AdS5 space.

3. The supersymmetric theory in 5 dimensions
In order to fully account for the kinks in the double domain wall solution presentedabove,
we have to extend the 5-dimensional theory so as to allow the coupling constants m and R5 to
change sign when crossing a domain wall. This approach was developed by Bergshoeff, Kallosh
and Van Proeyen (BKvP) [11], and it allows for a complete characterisation of D = 5 supersymmetry, even at the singular brane hypersurfaces. The easiest way to implement this procedure is
to let

m m (y),
(3.1)
R5 R5 (y),
with

(y) =
+1 for 0 y < ,
1 for y < 0,
(3.2)
and we impose the upstairs-picture identification y y + 2 . Note that, consequently, the superpotential should be redefined as

R5 8
4
5
20
e
.
W (y, ) = (y) 8me
(3.3)
20
Then, the potential of the 5-dimensional theory can still be expressed by the relation (2.16) and
the 5-dimensional supersymmetry variations of the gravitino and dilatino remain unchanged,
except for the above new definition of the superpotential.5 The Killing spinor equations, i.e. the
vanishing of the above supersymmetry transformations, are then solved exactly by

1/8
= b1 H 2/7 + b2 H 5/7
(3.4)
+ ,
5 The supersymmetric bosonic theory, including brane actions, has been presented previously in [12].
94
subject to the projection condition y + = + , where + is a constant spinor. Thus, in the extended theory with couplings changing sign across domain wall hypersurfaces, the 5-dimensional
braneworld solutions preserve half of the supersymmetries.
For completeness, and in order to contrast with the 10-dimensional calculation that will follow
shortly, we write out the integrability condition for the Killing spinor equations. Since the Killing
spinors have already been found, the integrability condition is necessarily satisfied; however, it
can be instructive to see the details of how this goes:
0 = pmn [Dm , Dn ]

1
1
1
= n Gn p gn p W 2 + gn p m W,m W ,p .
24
4
4
(3.5)
For p = y, using y = , the last two terms cancel, which is in agreement with the fact that
there are no singular contributions to Gyy . For p = the last term vanishes, while the singular
terms arising from W,y , where the y derivative acts on the (y) inside W , are cancelled by the
singular contributions to the Einstein tensor

1
3kb1 5/7 15kb2 2/7
+
G = Regular g (y)
(3.6)
H
H
.
gyy
7
14
4. Oxidising back to 10 dimensionsBreaking supersymmetry
The 5-dimensional domain wall solution can be oxidised back to 10 dimensions, resulting in
the metric [5]
1/2
2 2

2
= b1 H 3/7 + b2
dx dx + b1 H 13/28 + b2 H 25/28
dy
ds10

3/14
2 5
+H
ds S .
(4.1)
Now we can check again the integrability condition resulting from the Killing spinor equation in
10 dimensions, remembering that we now have F[5] m (y) rather than just F[5] m. We get
0 = PMN [D M , D N ]

P N 1 FP2 N 2i QRS N FP QRSN .
= N G
96
4!
(4.2)
The bulk terms are easily seen to satisfy this equation. The singular terms in the Einstein tensor
in 10 dimensions are [13]:

1
3kb1 15/28 15kb2 3/28
+
H
H
,
G = Regular g (y)
(4.3)
gyy
7
14
yy = Regular + 0,
G
(4.4)
ab = Regular g ab (y) 1 6kb2 H 3/28 .

G
gyy 7
(4.5)
For P = y, the last term in the integrability condition (4.2) vanishes, in agreement with the ab yy . For P = , the last term in the integrability condition adequately
sence of singular terms in G
; however, there is nothing there to cancel
cancels the b1 contribution to the singular terms in G
the singular terms proportional to b2 (and likewise for the b2 terms when P = a). We are thus
led to conclude that the oxidised domain wall solution does not preserve any supersymmetry!
95
It is however supersymmetric away from the branes in the bulk spacetime. Note that this result
also means that the extension of the ordinary 5-dimensional supergravity theory along the lines
advocated by BKvP cannot be obtained by dimensionally reducing type IIB supergravity.
What, however, is the problem with supersymmetry more concretely? It is enlightening to
study the Killing spinor equations directly; they are given by the condition

i
N P QRS
0 = M = M +
(4.6)
FN P QRS
M .
16 5!
Let us write out this calculation in detail: the -matrices are dimensionally reduced to 4 + 1 + 5
dimensions according to
1/4

= b1 H 3/7 + b2
(4.7)
1 1 ,

1
y = b1 H 13/28 + b2 H 25/28 y 1 1 ,
(4.8)
a = H 3/28 a 2
(4.9)
where there is no y-dependence left in , y , a (thus are the 4-dimensional -matrices

with indices raised and lowered with , y is the 4-dimensional chirality matrix, and a are
the -matrices on the internal 5-sphere).
We are now in a position to analyse the Killing spinor equations, using F[5] =
4i5!m (y)H 15/28 [1 1 (1 + i2 )]:

1/4

3kb1
0 = = +
1 (1 + 3 ) ,
(y)H 15/28 b1 H 3/7 + b2
112

1
3kb
1
(y)H 15/28 b1 H 13/28 + b2 H 25/28
0 = y = y +
112

y 1 (1 + 3 ) ,

3kb1
(y)H 15/28 H 3/28 1 a i(3 + 1) ,
0 = a = a +
112
where the spin covariant derivatives are given by
(4.10)
(4.11)
(4.12)

1/4
3kb1
y 1 1,
(y)H 15/28 b1 H 3/7 + b2
=
(4.13)
56
y = y ,
(4.14)

3ik
(y) b1 H 3/7 + b2 y a 3 .
a = a
(4.15)
56
From the expression for the oxidised metric (4.1), we expect the Killing spinor to be of the
form

1/8
1
3/7
= b1 H
(4.16)
+ b2
+
,
0
with y + = + . This ansatz indeed solves the first two Killing spinor equations (4.10) and
(4.11), but (4.12) reduces to

i R5
a = (y)
(4.17)
a .
2 20
96
Using the explicitly known expressions for Killing spinors on spheres [14], we can write down
a solution to the above equation as
4

1 3kb2 2
3kb2
i
= (y, a ) = e 2 (y)5 28 5
(4.18)
e 2 j ( 28 ) j,j +1 0 ,
j =1
where 0 is a constant spinor and a with a, b, . . . = 1, 2, 3, 4, 5 are the angular coordinates on

the 5-sphere. Thus, we see that we are forced to introduce y-dependence into the spherical part
of the candidate Killing spinor in the form of a -function. This makes the angular dependence
in the spherical part discontinuous and in this way supersymmetry is necessarily broken, because
a Killing spinor must be continuous. Note that the change in sign in the angular dependence
corresponds to a reversal of orientation of the 5-sphere. Thus we obtain the geometrical picture
that the orientation of the 5-sphere changes as we cross a brane. This is consistent with the fact
that the type IIB theory admits an SO(1, 9) symmetry rather than an O(1, 9) symmetry (this is
because the self-duality of the 5-form must be preserved). Indeed, the Z2 symmetry by which we
are modding out at the location of the branes, must be contained within SO(1, 9), and therefore
the flip y y must be accompanied by a reversal of orientation of the five-sphere.6
We should note that there also exists a supersymmetric limit, namely b2 0. In this limit
the troublesome term proportional to (y) disappears in the Killing spinor equation (4.17) of
the sphere, and in fact that condition reduces simply to the condition of having a covariantly
constant spinor. However, this limit is really the decompactification limit in which the sphere
becomes larger and larger, as well as flatter and flatter, and one ends up with an ordinary 3-brane
in 10 dimensions.
Another aspect of the decompactification/supersymmetry-restoring limit b2 0 is the structure of the metric warp factor. In Poincar coordinates (where the transverse term is simply dz2 ),
the D = 5 metric in the b2 0 limit has a power-law warp factor:

1
5k 5
dx dx + dz2 .
ds52 = 1
7|z|
(4.19)
This should be compared with the structure of the metric in the RandallSundrum limit [3]
k 0, where the Poincar-coordinate metric (2.20) is composed of patches of anti-de Sitter
space with an exponential warp factor:
ds52 = e
2|z|
L5
dx dx + dz2 .
(4.20)
The exponential warp factor underlies many of the proposed physical applications of the
RandallSundrum schemes, be it the effective concentration of gravity near the D = 4 positive
tension brane in RSII, or possible applications to the hierarchy problem arising from exponential
differences in coupling constants on opposing RSI braneworlds. These features disappear with
the power-law warp factor (4.19) which arises as supersymmetry is restored in the b2 0 limit.
Let us take stock at this point of what we have learned: we have a double domain wall solution in 5 dimensions, which upon oxidation to 10 dimensions on a 5-sphere leads to another
6 Note that for odd-dimensional spheres such an orientation-reversing map admits at least two fixed points. From (4.18)
one sees that our choice of Z2 has a fixed-point set on the locus 5 = 0, i.e. at the equator of the 5-sphere. Note also that
the antipodal map is orientation-preserving for odd-dimensional spheres, so this cannot be used as the S 5 part of the Z2
action.
97
double domain wall solution. This solution has the particular property that it is supersymmetric
everywhere in the bulk spacetime, but breaks supersymmetry completely at the locations of the
domain walls. This immediately raises two questions:
1. Since supersymmetry is broken on the domain wall, what is the mass scale of this breaking,
as seen from the viewpoint of a 4-dimensional observer on the domain wall? We will treat this
question in Section 5.
2. Is this solution stable? Indeed one might speculate that since this non-supersymmetric
solution is surrounded by a supersymmetric spacetime, it might be kept stable by the surrounding
bulk (in the fully supersymmetric case solutions of this type are known to be stable despite the
presence of a negative-tension brane [12]). A detailed calculation of the stability properties of
this solution would be very interesting. We leave it for future work.
5. Fermionic modes and the scale of supersymmetry breaking
If supersymmetry were not broken, we would expect the theory on the 4-dimensional branes
to be an ungauged supergravity theory with half the number of supercharges as compared with
the bulk theory [8]. Then, there would be fermionic modes which, from the 4-dimensional point
of view, would be massless. In this section, we will present modes of this type, which we obtain as superpartners of linearised massless bosonic perturbations (see Appendix A for a detailed
derivation, following the linearised supersymmetry procedure of [15]). However, since supersymmetry is actually broken, we know that the fermionic excitations will pick up mass terms
(while the bosons remain massless at this level). The easiest way to derive these mass terms is by
dimensionally reducing the 10-dimensional RaritaSchwinger action for these modes in order to
find their 4-dimensional effective actions. The mass terms then arise when a y-derivative hits the
discontinuity in the spherical spinor part (y, a ) at the location of the branes.7
5.1. The gravitino
As shown in Appendix A, one of the would-be massless perturbation modes of the braneworld
geometry that we are considering is a worldvolume gravitino given by
1/8

= b1 H 2/7 + b2 H 5/7
(x),
(5.1)
y = 0,
(5.2)
= 0.
(5.3)
This mode can be lifted to 10 dimensions, where it reads

1/8

1
,
= b1 H 2/7 + b2 H 5/7
(y, a )
0
(5.4)
y = 0,
(5.5)
a = 0.
(5.6)
7 The setup described here thus provides a concrete example of the general framework for brane supersymmetry breaking of Bagger and Belyaev [16,17].
98
The discontinuous spherical part of the gravitino should really be seen as an approximation; one
would expect the gravitino to be continuous but interpolating between two different bulk profiles
on either side of a brane. For our purposes, though, this approximation is accurate enough. The
action for the gravitino can then be dimensionally reduced as follows8

gi
M MNP D N P
10d

=
4d

g
5/14
dy
i b1 H
sin4 (5 ) d5
d4
0
3/7
1/2
+ b2

[1

1
0]2 1
0

3/4 5/14 3kb2

i b1 H 2/7 + b2 H 5/7
H
i
5 (y) (y )
28

1
y 5 [1 0]2 1
.
0
Now, if we assume that 5 = (thus the SO(6) symmetry of the five-sphere also gets broken
at the location of the branes), and further use the fact that y = as well as the integrals

sin4 (5 ) d5 =
3
,
8
(5.7)

sin4 (5 )5 d5 =
3 2
,
16
(5.8)
then we get the 4-dimensional effective action for :

S =

g i
m(3/2) ,
(5.9)
4d
where
m(3/2) =
3kb2 [(b1 H 2/7 + b2 H 5/7 )3/4 ]0

.
2/7 + b H 5/7 )1/2
56
2
0 dy (b1 H
(5.10)
Thus, as expected, we find an ungauged supergravity in 4 dimensions, broken by the above mass
term.
8 We define
= A. The 10-dimensional intertwiner A1,9 is dimensionally reduced according to A1,9 = A1,4
A0,5 2 . Our conventions are as in Sohnius [18].
99
The expression for the mass term is a bit unwieldy, which is why it is instructive to write out
the RandallSundrum limit of the above formulae. The 10-dimensional metric is then given by
2
=e
ds10,RS
2|y|
L
5
dx dx + dy 2 + L25 d52 ,
(5.11)
where L5 is the length parameter of both AdS5 and S 5 as given in (2.21). The gravitino reduces
to
= e
|y|
2L
(x),
(5.12)
y = 0,
(5.13)
= 0,
(5.14)
which in 10 dimensions is expressed as

1
|y|
,
= e 2L5 (x) (y, a )
0
y = 0,
(5.15)
(5.16)
a = 0.
(5.17)
In the RandallSundrum limit the discontinuous spherical spinor is given by

4

1
i
(y)5 5
2 j j,j +1
2
(y, a ) = e
e
0 .
(5.18)
j =1
The calculation of the effective action proceeds along the same lines as above, and this time we
find

=
m(3/2,RS) ,
g i
S,RS
(5.19)
4d
where
3
(1 e L5 )
.
m(3/2,RS) =
2
L5
(1 e L5 )
We can see that m(3/2,RS) varies between
always close to the L1
5
(5.20)
3
2L5
(as
L5
0) and
L5
(as
L5
), and is therefore
scale of S 5 -compactification, which one may take to be close to the GUT
or Planck scales. Thus, from the 4-dimensional point of view, the gravitino is heavy.
5.2. Other modes
Let us now turn our attention to the other fermionic modes discussed in Appendix A. These
modes have a different y-profile, but nevertheless, in a supersymmetry-preserving context, they
would appear to be massless from a 4-dimensional perspective. First of all, we have the fermionic
partner of the Goldstone boson associated with the y-translation symmetry that is broken by the
brane. This mode is given by
=

3/8
k
+ ,
s, 2b1 H 2/7 + 5b2 H 5/7 b1 H 2/7 + b2 H 5/7
8H
(5.21)
100

13/8
k
s, +
2b1 H 2/7 + 5b2 H 5/7 b1 H 2/7 + b2 H 5/7
4H

k 2 (y)
2/7
5/7 11/8
b
24b12 H 4/7 + 60b1 b2 H
+
H
+
b
H
1
2
112H 2

+ 45b22 H 10/7 s(x)+ ,
3/8
15k
s, + ,
=
b1 H 2/7 + b2 H 5/7
2H
which in the RandallSundrum limit simplifies to
y =
(5.22)
(5.23)
1 3|y|
= e 2L5 s, + ,
4
5|y|
3|y|
1 2L
(y) 2L
e 5 s(x)+ ,
y = e 5 s, +
2
2L5
= 0.
(5.24)
(5.25)
(5.26)
If we now let
s, +
(5.27)
and take this as the definition of the mode (x), then we can derive the effective 4-dimensional
action for :

S,RS =
(5.28)
g i m(,RS)
,
4d
where the mass term is given by
6 (e L5 1)
.
m(,RS) =
5L5 L25
(e 1)
(5.29)
3
6
We can see that this time m(,RS) varies between 5L
(as L5 0) and e L5 5L
(as L5 ).
5
5
Thus we get an exponential mass suppression when L5 is large. It seems reasonable on phenomenological grounds to assume that might be an order of magnitude larger than L5 [1], and
therefore can be a light fermion from the 4-dimensional effective theory point of view. This
is because has a profile along the orbifold direction y which evolves in the opposite way as
compared to the bulk warp factor, and therefore is localised mainly near the negative-tension
brane at y = .
The last mode that we will consider is the fermionic partner to the third bosonic mode
presented in Appendix A. This bosonic mode has the particular property that in the Randall
Sundrum limit it reduces to a pure scalar field perturbation, the metric remaining unchanged. In
general, its fermionic partner is given by

17/8
k
s, b1 H 2/7 + b2 H 5/7
+ ,
10/7
8H
7/8
k
y =
b1 H 2/7 + b2 H 5/7
s, +
4H 10/7

3k 2 b1 (y)
2/7
5/7 9/8
b
+
H
+
b
H
s(x)+ ,
1
2
28H 15/7
(5.30)
(5.31)
101
1/8 2 1/7

15k
2b1 H + b1 b2 H 4/7 b22 H s, + ,
b1 H 2/7 + b2 H 5/7
8H
which in the RandallSundrum limit reduces to a pure dilatino perturbation:
=
(5.32)
= 0,
(5.33)
y = 0,
1 7|y|
= e 2L5 s, + .
2
(5.34)
(5.35)
It is again helpful to define by

1
s, + ,
2
in terms of which the effective action is given by

S ,RS =
g i m( ,RS) ,
(5.36)
(5.37)
4d
with the mass term

11
5 (1 e L5 )
.
m( ,RS) =
10
L5
(1 e L5 )
m( ,RS) thus varies between
of a heavy fermion.
11
2L5
(as
(5.38)
L5
0) and
5
L5
(as
L5
), and so is another example
6. Discussion
We have seen that Z2 -symmetric braneworlds in type IIB supergravity necessarily break supersymmetry owing to the chiral nature of the theory and the curvature of the internal manifold.
Supersymmetry is broken only at the location of the branes, and this can also be traced
back to
the presence of source terms that are proportional to the square root of the curvature R5 b2
of the internal 5-sphere.
We have shown how fermionic modes, which would have been massless in a supersymmetric
context, thus acquire masses. Moreover, depending on their profiles along the orbifold direction y, the effective 4-dimensional fermionic modes can appear either heavy or light. The heavy
modes are those whose profiles along the orbifold direction evolve similarly to the metric warp
factor, i.e. they are mainly associated to the bulk geometry, and they tend to have a mass comparable to the L1
5 compactification scale. The light fermions, by contrast, are those modes which
have profiles that evolve in the opposite way as compared to the metric warp factor and are more
specifically associated to a brane. In the RandallSundrum limit, for example, the light fermions
are those that have a y-dependence proportional to
c|y|
e L5
(6.1)
with c > 1.
This ensures that these modes are mostly localised near the negative-tension brane at y = . For
large values of L5 , their masses are suppressed by a factor of e L5 , which is certainly attractive
for phenomenological reasons. We recall the discussion of Section 4 on the exponential warp
102
factor in the general solutions of Bremer et al. [5], and particularly in the RandallSundrum
limit [8], as compared to the power-law warp factor occurring in the supersymmetry-restoring
b2 0 limit. This exponential warp factor is also seen to be at the root of the orbifold exponential
hierarchy of masses for fermionic fluctuations that we have found.
For simplicity, we have focused in this paper on the minimal D = 5 supersymmetric structure with 8 supercharges. The full S 5 -reduced theory, of course, has an extended 32-supercharge
supersymmetry organized into a 4 of SU(4) SO(6). In the bulk spacetime of the brane solutions we consider, each of these 4 D = 5 spinor supercharges splits up into two D = 4 spinors
of opposite y chirality, one of which becomes spontaneously broken, with a corresponding
massive gravitino in the usual fashion. The remaining y chirality gives the erstwhile unbroken
supersymmetry, which however is broken by the Z2 structure of the brane system as we have
shown. Choosing a specific Z2 action in the 5-sphere directions necessitates picking an equator
of the 5-sphere, which becomes the fixed-point surface for the chosen Z2 . This breaks the surviving automorphism symmetry down from SU(4) SO(6) to USp(4) SO(5). However, the
4 representation remains irreducible with respect to USp(4), so all 4 of the D = 5 theorys supersymmetries get broken by the Z2 action in the same way. Accordingly, the full story is just a
four-fold replication of the minimal D = 5 story that we have presented.
Let us conclude with a few remarks about the nature of the supersymmetry-breaking sources.
The RandallSundrum scenario has been associated to a combination of D3 branes and 7branes [19]. There is a possible association of the b2 term in our construction to 7-branes, as
noted already in [13]. Note that the singular terms in Gab are a factor of 45 smaller than the ones
in G , suggesting that the upper 4-dimensional parts of the worldvolumes of the 7-branes might
be averaged over the 5 spherical dimensions. This association is supported furthermore by the
fact that the y-dependence of the singular terms in (4.3)(4.5) would be consistent with the presence of two transverse directions instead of just one, e.g. the y-direction and one of the spherical
directions. For the b2 part of the solution, we explicitly have

(y)
,
G Gab b2 (y) b1 H 5/14 + b2 H 11/14 = b2
gtransverse
(6.2)
where (with no summation implied on a)

gtransverse = gyy gaa .
(6.3)
Note also that the Z2 symmetry chosen in our construction has a fixed-point set on the locus
y = 0, 5 = 0, i.e. an 8-dimensional surface which could be associated to a 7-sphere worldvolume.
Going against the 7-brane interpretation of the b2 part of the solution, however, is the fact that
the IIB axionic scalar that would support a standard 7-brane is zero in the background considered
here. Of course it could be that the precise smearing of 7-branes needed has to be such that the
axion charge averages to zero.9
A final question is that of stability. Even if the background solution we consider can be
associated to a smeared set of 7-branes taken together with the D3 brane, the breaking of supersymmetry that we have found raises the question of whether this construction has tachyonic
instabilities. But since the bulk spacetime remains perfectly supersymmetric away from the
9 Another puzzle with such an interpretation arises in the analogous case of 11-dimensional supergravity compactified
on a 7-sphere. In that case, the analogous source would have to be made out of 8-branes, but no 8-brane solutions are
known in D = 11 supergravity, although they do exist in massive type IIA supergravity [20].
103
branes, one is led to speculate that the bulk supersymmetry might be enough to stabilise the
boundary branes where supersymmetry is broken, perhaps in a manner similar to the fake supergravity framework of Ref. [21].
Acknowledgements
The authors would like to thank Jussi Kalkkinen for collaboration during the early stages of
this work. They would also like to acknowledge useful and stimulating discussions with Ben
Allanach, Kevin Costello, Mirjam Cvetic, Ruth Durrer, Joel Fine, Gary Gibbons, Ulf Gran, Neil
Lambert, Jim Liu and Antoine Van Proeyen. K.S.S. would like to thank the TH Unit at CERN
and the Galileo Galilei Institute for Theoretical Physics for hospitality and would like to thank
the INFN for partial support during the completion of this work.
Appendix A. Linearised domain wall perturbation modes
In this appendix, we explicitly derive the form of linearised bosonic perturbations about domain wall geometries, which, from the 4-dimensional point of view, are massless. We also show
how one can then determine the fermionic superpartners of these modes. It should be noted that
the method employed here is not the same as determining the moduli of a domain wall solution
and then promoting those moduli to spacetime-dependent fields (see for example [22] for an exposition of the latter method). Here, we allow the various modes to have different y-dependent
profiles along the orbifold direction, chosen such that the modes appear massless from the brane
worldvolume perspective (when supersymmetry is not broken). The existence of this type of zero
mode is really a particular feature of braneworld KaluzaKlein reductions. Consider theories of
the form

1
g R ()2 V ()
gW (),
S=
(A.1)
2
5d
4d,y=0
where, in this appendix, we are considering a single positive tension domain wall residing at
y = 0. In static gauge, the equations of motion are
1
1
1
1 (y)
g W (),
Gmn = ,m ,n gmn ,p ,p gmn V ()
2
4
2
2 gyy m n
(y) W
V
+
.
2 =
gyy
(A.2)
(A.3)
We write the fields as

(0)
+ hmn ,
gmn = gmn
(A.4)
(A.5)
where
(0)
(0)
(1)
quantities correspond to the unperturbed domain wall solutions. We then have
g mn = g mn(0) hmn ,
(A.6)
1 (0)pl
(1)p
mn = g
(A.7)
(hlm;n + hln;m hmn;l ).
2
When we perturb the geometry, we choose coordinates such that the domain wall always remains
at y = 0 [23]. The linearised equations of motion then are
104
1 (0) pl

1 p
h m;np + hp n;mp hmn;p p h;mn gmn
h ;lp h;p p
2
2
1
1 (0) pl (0)
(0)
hmn R + gmn h Rpl
2
2
1 (0) (1) 1 (0) (1) 1
1 (0) pl (0) (0)
(0)
= ,m ,n + ,n ,m hmn (0),p ,p
+ gmn
h ,p ,l
2
2
4
4
1 (0) (0),p (1) 1
1 V (1)
gmn
,p hmn V
2
2
2

1 (y)
1
(0) W (1)
(0)

m n h W + g
W
hy y g
2
2
(0)
gyy
(A.8)
and
1
(0)
(0)
(0)
2(0) (1) hmn ;mn hmn ;n ;m + h,m ,m
2

2 V (1)
(y) 2 W (1) 1 y W
h
.
=
y
2
2
2
(0)
gyy
(A.9)
For a background metric of the form

ds 2 = e2A(y) dx dx + e2B(y) dy 2 ,
(A.10)
the non-zero connections are

(0)y
= A,y e2A2B ,
(0)
y
= A,y ,
(0)y
yy
= B,y ,
(A.11)
and thus we have

(0)
R
= e2A2B A,yy 4A2,y + A,y B,y ,
(A.12)
(0)
Ryy
= 4A,yy 4A2,y + 4A,y B,y ,

2A2B
3A,yy + 6A2,y 3A,y B,y ,
G(0)
= e
(A.13)
2
G(0)
yy = 6A,y ,
(0)
2(0) (0) = e2B ,yy
(A.15)
(0)
(0)
+ 4,y
A,y ,y
B,y .
(A.14)
(A.16)
Taking into account that we are imposing a Z2 symmetry at the location of the domain wall, we
can see that the junction conditions (i.e. the matching conditions for the singular pieces in the
Einstein equations) at the location of the domain wall become

12A,y = eB W y=0 ,
(A.17)

W
(0)
= eB
.
2,y
(A.18)
y=0
Note that the yy Einstein equation is given by
1 (0)2 1 2B
6A2,y = ,y
(A.19)
e V.
4
2
This does not involve second derivatives in y, which is consistent with the fact that there are no
singular source terms in that direction. If we evaluate this equation at the location of the domain
105
wall, we can substitute in the junction conditions derived above, to find

1
V=
8

2
W2
3

(A.20)
y=0
This is of course the relation between the superpotential W and the potential V in supersymmetric theories. In a supersymmetric context, the junction conditions above are actually the
Bogomolnyi equations, and they are then valid throughout the bulk. Furthermore, this shows
that the domain wall couples to the bulk via the superpotential.
Looking at the terms containing two y derivatives in the linearised equations of motion, one
can write down the linearised junction conditions in this background (making use of the 0th order
junction conditions):

h h

,y
1

1
= eB W h h + e2AB W hyy
6
4
W (1)
1 2A+B
+ e
,
2
y=0
1 2 W (1) 1 B W
(1)
,y
= eB
+ e
hyy .
2
4
2
y=0
(A.21)
(A.22)
There is also a junction condition associated with the y linearised Einstein equation, and it
reads
hy = 0
(A.23)
y=0
This condition was already implied by the imposition of the Z2 symmetry at y = 0 under which
hy is odd.
A.1. Examples of bosonic modes
We will now give explicit expressions for these modes in the Bremer et al. case [5] as well as
the RandallSundrum limit [3,4].
In the Bremer et al. case [5], the background solution is given by
e
7
15
= H = k|y| + c,
e4A = b1 H 2/7 + b2 H 5/7 ,
B = 4A.
(A.24)
We then have the following expressions for the superpotential and the potential:

3k
2b1 H 5/7 + 5b2 H 2/7 (y),
7

9k 2 2 10/7
V=
5b22 H 4/7 .
2b1 H
196
W=
(A.25)
(A.26)
The = linearised junction conditions are solved for h (b1 H 2/7 + b2 H 5/7 )1/2 , unless
h . In the first case, this ansatz also solves the other junction conditions and the linearised
equations of motion (it should be noted that, although the linearised junction conditions would
also allow for hyy and (1) contributions, say proportional to a mode c(x), the = linearised
106
Einstein equations would then demand c, = 0 and thus we set hyy = 0 = (1) ). This mode
represents a 4-dimensional worldvolume graviton excitation:
1/2

h = b1 H 2/7 + b2 H 5/7
(A.27)
h (x),
hyy = 0,
(A.28)
= 0,
(A.29)
(1)
where h (x) obeys the 4-dimensional linearised Einstein equations.

In the second case, our ansatz for the metric perturbations is

h = a s x f (y),

hyy = s x j (y).
(A.30)
(A.31)
Then, looking at the linearised equations for = we find

1
a= ,
2

5/2
j (y) = b1 H 2/7 + b2 H 5/7
f (y).
(A.32)
(A.33)
(1)
Note that these conditions also automatically ensure that there are no s; terms present in G
for any , . Next we look at the y equations, from which we can infer that
1/2
21 (y)
(1) =
Hf,y .
s x b1 H 2/7 + b2 H 5/7
2 15k
(A.34)
At this point all m equations are identically satisfied for all f (y). The linearised yy and
equations demand

2(4d) s x = 0
(A.35)
and the additional constraint

0 = f,yy 98b12 H 4/7 + 98b22 H 10/7 + 196b1 b2 H

+ kf,y 154b12 H 3/7 + 91b22 H 3/7 + 245b1 b2

+ k 2 f 10b12 H 10/7 10b22 H 3/7 20b1 b2 H 1 .
(A.36)
This has the following two solutions:

f (y) = 2b1 H 5/7 + 5b2 H 2/7 b1 H 2/7 + b2 H 5/7 ,y ,
5/2

.
f (y) = b1 H 2/7 + b2 H 1/7
(A.37)
(A.38)
Thus, explicitly, we have the Goldstone mode

1
h = s x 2b1 H 5/7 + 5b2 H 2/7 k,
2
5/2

hyy = s x b1 H 2/7 + b2 H 5/7
2b1 H 5/7 + 5b2 H 2/7 k,
1/2 1

H k
(1) = s x 15 b1 H 2/7 + b2 H 5/7
and a third mode
(A.39)
(A.40)
(A.41)
107
5/2

1
h = s x b1 H 2/7 + b2 H 1/7
k,
(A.42)
2
10/7
k,
hyy = s x H
(A.43)

15
(1) =
(A.44)
s x b1 H 2/7 + b2 H 5/7 2b1 H 8/7 b2 H 5/7 k.
4
It is then straightforward to verify that the latter two modes also satisfy the linearised junction
conditions. The Goldstone mode takes its name from the fact that, in the bulk, its general form
can be obtained by a y-dependent diffeomorphism with parameter
= 0,
3/2

s,
y = 7 b1 H 2/7 + b2 H 5/7
(A.45)
(A.46)
where
hmn = m;n + n;m ,
(A.47)
= ,m .
(A.48)
(1)
If we then promote s to a function s(x), this is not a diffeomorphism anymore, and we obtain
the above non-trivial mode. In this sense, this mode is a Goldstone mode corresponding to the
translational symmetry that is broken by the domain wall (see [24] for a general treatment of
these types of modes).
For the RandallSundrum model [1,3], we have the following expressions for the superpotential and the potential:
W=
12
(y),
L5
V =
12
,
L25
W
= 0,
V
= 0,
8
2W
= (y),
L5
2
32
2V
= 2,
2
L5
(A.49)
(A.50)
where L5 is the AdS5 radius of curvature.10 The background metric is given by

ds52 = e
2|y|
L
5
dx dx + dy 2 .
(A.51)
The perturbation modes can simply be determined by taking the appropriate limit of the Bremer
et al. modes [3]. This gives the graviton excitation
h = e
2|y|
L
5
h (x),
(A.52)
hyy = 0,
(A.53)
(1) = 0,
(A.54)
where h (x) obeys the 4-dimensional linearised Einstein equations. The Goldstone mode is
now given by
h = s(x) ,
(A.55)
10 Incidentally, the second derivative of the potential indicates that the breathing mode has mass squared equal to
32/L25 , in agreement with [7].
108

2|y|
hyy = 2e L5 s(x),
(A.56)
= 0,
(A.57)
2(4d) s(x) = 0.
(A.58)
(1)
with
This is the radion mode of Ref. [23] (see also [25] for a heterotic M-theory equivalent), and
2|y|
it can be obtained by starting with a diffeomorphism with parameter y = L25 e L5 s. The third
mode reduces to a pure scalar field perturbation:
h = 0,
(A.59)
hyy = 0,
(A.60)
(1) = e
again with
4|y|
L5
(A.61)
s(x),
2(4d) s(x) = 0.
A.2. Fermionic partners

The fermionic superpartners of the bosonic modes that we have just derived can be obtained
by using the linearised form of the supersymmetry transformations (2.17), (2.18) [15]. In this
way it is guaranteed that the resulting fermions are also solutions of the linearised equations of
motion. In general, the fermions are given by
m = (D)(1)
m ,

1
1
1
1 2 W (1)
(1)
(1)
= hy y y ,y + ,
+ y ,y
,
4
2
2
4 2
(A.62)
(A.63)
with

1
1
+ h,y A,y h g A,y hy y y

(D)(1)
= h,
4
4
W
1 W (1)
+ h +
,
48
24
1
W
1 W (1)
y
(D)(1)
+ hy y y +
y .
y = hyy,
4
48
24
(A.64)
(A.65)
In the case of a graviton perturbation

h = e2A h (x),
hyy = 0 = (1) ,
(A.66)
the fermionic superpartner is the gravitino given by

A
A
A
1
1
= h, e 2 + = h , e 2 + e 2 (x),
4
4
y = 0,
= 0.
(A.67)
(A.68)
(A.69)
Consequently the chirality of the gravitino is given by

y = + .
(A.70)
109
Our remaining bosonic modes are of the form

1
h = s(x)f (y),
2
hyy = s(x)e2B2A f (y).
(A.71)
(A.72)
In this case the linearised junction conditions (A.21) and (A.22) simplify the resulting expressions for the fermionic partners, and we end up with
1 3
= e 2 A f s, + ,
8
1 5 A+B
W 3
1 W (1) A +B
f s, + + e 2 A+B f s(x)+ +
e 2 + ,
y = e 2
4
48
24
1
(1) A2
e + .
= ,
2
Thus we can see that we have the following chiralities
(A.73)
(A.74)
(A.75)
y = + ,
(A.76)
y = ,
(A.77)
whereas y contains terms of both chiralities.

As a consistency check, it is straightforward to verify that all the above fermionic modes
satisfy their equations of motion:

1 W m
1
g mn mn ,n = 0,
8
4
2

1 W
W
1 W m
1
m
m +
m n ,n m +
m = 0.
2
2
8
2
4
mnp Dn p
(A.78)
(A.79)
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-ph/9905221.

L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690, hep-th/9906064.
M.J. Duff, J.T. Liu, K.S. Stelle, J. Math. Phys. 42 (2001) 3027, hep-th/0007120.
M. Cvetic, H. Lu, C.N. Pope, Class. Quantum Grav. 17 (2000) 4867, hep-th/0001002.
M.S. Bremer, M.J. Duff, H. Lu, C.N. Pope, K.S. Stelle, Nucl. Phys. B 543 (1999) 321, hep-th/9807051.
J.T. Liu, H. Sati, Nucl. Phys. B 605 (2001) 116, hep-th/0009184.
H.J. Kim, L.J. Romans, P. van Nieuwenhuizen, Phys. Rev. D 32 (1985) 389.
M.J. Duff, J.T. Liu, W.A. Sabra, Nucl. Phys. B 605 (2001) 234, hep-th/0009212.
G.W. Gibbons, N.D. Lambert, Phys. Lett. B 488 (2000) 90, hep-th/0003197.
F. Brito, M. Cvetic, S. Yoon, Phys. Rev. D 64 (2001) 064021, hep-ph/0105010.
E. Bergshoeff, R. Kallosh, A. Van Proeyen, JHEP 0010 (2000) 033, hep-th/0007044.
J.L. Lehners, P. Smyth, K.S. Stelle, Class. Quantum Grav. 22 (2005) 2589, hep-th/0501212.
M. Cvetic, M.J. Duff, J.T. Liu, H. Lu, C.N. Pope, K.S. Stelle, Nucl. Phys. B 605 (2001) 141, hep-th/0011167.
H. Lu, C.N. Pope, J. Rahmfeld, J. Math. Phys. 40 (1999) 4518, hep-th/9805151.
M. Cvetic, N.D. Lambert, Phys. Lett. B 540 (2002) 301, hep-th/0205247.
J.A. Bagger, D.V. Belyaev, Phys. Rev. D 67 (2003) 025004, hep-th/0206024.
J.A. Bagger, D.V. Belyaev, Phys. Rev. D 72 (2005) 065007, hep-th/0406126.
M.F. Sohnius, Phys. Rep. 128 (1985) 39.
C.S. Chan, P.L. Paul, H.L. Verlinde, Nucl. Phys. B 581 (2000) 156, hep-th/0003236.
L.J. Romans, Phys. Lett. B 169 (1986) 374.
110
[21]
[22]
[23]
[24]
[25]
D.Z. Freedman, C. Nunez, M. Schnabl, K. Skenderis, Phys. Rev. D 69 (2004) 104027, hep-th/0312055.
J.-L. Lehners, P. McFadden, N. Turok, hep-th/0612026.
C. Charmousis, R. Gregory, V.A. Rubakov, Phys. Rev. D 62 (2000) 067505, hep-th/9912160.
T. Adawi, M. Cederwall, U. Gran, B.E.W. Nilsson, B. Razaznejad, JHEP 9902 (1999) 001, hep-th/9811145.
J.-L. Lehners, K.S. Stelle, Nucl. Phys. B 661 (2003) 273, hep-th/0210228.
Constraints on the electroweak chiral Lagrangian

from the precision data
Sukanta Dutta a,b , Kaoru Hagiwara b,c , Qi-Shu Yan d,e, , Kentaroh Yoshida f
a SGTB Khalsa College, University of Delhi, Delhi-110007, India
b KEK Theory Division, Tsukuba 305-0801, Japan
c The Graduate University for Advanced Studies (SOKENDAI), Tsukuba 305-0801, Japan
d Department of Physics, National Tsing Hua University, Hsinchu, Taiwan
e National Center of Theoretical Sciences (Theory Division), 101, Section 2 Kuang Fu road, Hsinchu, Taiwan
f Kavli Institute for Theoretical Physics, University of California, Santa Barbara, CA 93106-4030, USA
Received 18 May 2007; accepted 31 August 2007

Abstract
In the framework of the effective field theory method, we use the experimental data and the perturbative
unitarity bounds to determine the values and uncertainty of all the 11 chiral coefficients (i , i = 0, . . . , 10)
of the standard electroweak chiral Lagrangian. Up to linear terms in i , we provide the one-loop renormalization group equations of all the chiral coefficients, which are calculated in the Feynmant Hooft gauge
using the modified minimal subtraction scheme. With the improved renormalization group equations to
sum over the logarithmic corrections, we analyze the current experimental uncertainty of oblique correction
parameters, S() and T (). We find that, due to the large uncertainty in the triple gauge-boson coupling
measurements, the parameter space of positive S() for > 1 TeV is still allowed by the current experimental data. T () tends to increase with even in the presence of the operators that contribute to the triple
and quartic gauge-boson couplings.
PACS: 11.10.Gh; 11.10.Hi; 12.15.Ji; 12.15.Lk
Keywords: Electroweak chiral Lagrangian; Precision data; Triple gauge-boson couplings
* Corresponding author at: National Center of Theoretical Sciences (Theory Division), 101, Section 2 Kuang Fu road,
Hsinchu, Taiwan.
E-mail address: yanqs@phys.nthu.edu.tw (Q.-S. Yan).
doi:10.1016/j.nuclphysb.2007.08.017
112
S. Dutta et al. / Nuclear Physics B 790 (2008) 111137
1. Introduction
Electroweak symmetry breaking (EWSB) mechanism is the most important issue which will
be explored at the Large Hadronic Collider (LHC). TeV-scale supersymmetric theories suggest
that the electroweak symmetry is spontaneously broken by fundamental Higgs fields. On the contrary, QCD-like theories do not have fundamental scalar fields and suggest that the electroweak
symmetry is dynamically broken by fermion-pair condensates [1,2].
In order to obtain hints on the TeV scale physics that leads to the EWSB, it is useful to
examine the consequence of the electroweak precision data of the bosonic sector in an integrated
fashion. In this paper, we study the scale dependence of the coefficients of the electroweak chiral
Lagrangian, and examine carefully if the present precision data exclude models with S(
1 TeV) < 0 conclusively. We pay particular attention to the magnitude and sign of the coefficients
of the operators that contribute only to the triple and quartic gauge-boson couplings and hence
weakly constrained by the present data, because they enter in the functions of the running S
and T parameters. Part of our findings have been reported in Ref. [3].
We follow the standard analysis of the chiral Lagrangian method [46] and include 11 operators up to mass dimension four in the electroweak chiral Lagrangian (EWCL) [710]. This study
extends the work of Bagger et al. [11], who considered the effects of those operators that contribute to the weak boson two point functions only, i.e., 3 out of the 11 operators. We consider
all the 11 operators, among which 5 operators contribute to the triple gauge-boson couplings
(TGCs) and 9 contribute to the quartic gauge-boson couplings (QGCs). We consider constraints
from the TGC measurements at LEP2 [1215] and at the Tevatron [16] as well as those from the
perturbative unitarity to bound the QGCs.
In the framework of the effective field theory method [17], all the 11 couplings in the EWCL
are renormalization scale dependent quantities. Hence the electroweak precision data constrain
the chiral coefficients at the low scale, = mZ . In order to find the connection between S(mZ )
and S(), we extend the previous one-loop RGE [18] of the EWCL used by Bagger et al. [11].
Along with the previous studies, we assume the validity of the perturbation theory and the absence of any additional resonances between mZ and . The new functions take account of all
the 11 chiral coefficients. The improved RGE makes it possible to analyze the effects of those
operators which contribute only to the three-point and four-point gauge-boson couplings on the
uncertainty of S() and T () at the scale of new physics.
By utilizing the RGE to sum over the logarithmic corrections of quantum fluctuations and
by analyzing the current experimental uncertainty of TGC and QGC, we find that in the most
conservative perturbative calculation, the central value of S(1 TeV) and its corresponding 1
error reads as
S(1 TeV) = 0.02 0.20.
(1)
We observe that the current electroweak precision data, especially the data from the TGC measurements, have not reached the precision to fix the signs of S() above > 1 TeV. If we
remove the effects of the chiral coefficients that contribute only to TGC and QGC, the central
value and its 1 error reads as
S(1 TeV) = 0.14 0.09,
(2)
reproducing the result of Ref. [11].

In our analysis, we will use the formalism of nonlinear realization of EWSB, i.e., the gauged
nonlinear model. We consider the set of all the dimension 4 operators which are even under
113
the transformation of discrete symmetries, C and P . We omit all dimension 6 or higher order
operators. There are works, for instance Ref. [19], which attempt to constrain the electroweak
symmetry breaking models by including some of dimension 6 operators. By including more
operators one can only make the allowed area of the S()T () plane larger, and hence our
conclusion that S( = 1 TeV) > 0 is still allowed persists.
This paper is organized as follows. In Section 2, the bosonic sector of the EWCL in our analysis is introduced. In Section 3, we briefly review the constraints on the chiral coefficients from
the electroweak precision data on the gauge-boson two point functions (Section 3.1), those on the
TGC (Section 3.2), and those from the perturbative unitarity of the weak boson scattering amplitudes (Section 3.3). In Section 4, we list the improved renormalization group equations (RGEs).
In Section 5, we study the experimental uncertainty of S()T () for = 0.3 TeV, 1 TeV,
and 3 TeV by using the improved RGEs. We close the paper in Section 6 with discussions and
conclusions. Appendices A and B are added to introduce our method to calculate the functions
of the chiral coefficients (Appendix A) and the treatment of the ghost terms (Appendix B).
2. The operators in our analysis
In our analysis, we consider the following 14 bosonic operators which preserve discrete symmetries, C and P [710]:

1
1
i L i ,
H1 2 H 2 v 2 L W/Z + 0 v 2 L 0 +
2
g
g
10
LEW =
(3)
i=1

1
H 1 = tr W W ,
2
1
H 2 = B B ,
4

1
L W/Z = tr V V ,
4
(4)
(5)
(6)
where g and g are the gauge couplings of SU(2)L and U (1)Y gauge groups, respectively. The
NambuGoldstone bosons are parameterized in the nonlinear form as

2 a T a
.
U = exp i
(7)
v
The gauge covariant derivative, local gauge fields and their gauge covariant field strength are
given as
V = ( U )U + iW iU BY U ,
(8)
W = Wa T a ,
(9)
BY
= B T ,
3
(10)
W = W W + i[W , W ],
(11)
B = B B ,
(12)
where T a = a /2, and a are the Pauli matrices.
114
The couplings i form the 11-dimensional parameter space of the EWCL, where the corresponding operators Li are given as

1
L 0 = tr(T V ) tr T V ,
4

1
L 1 = B tr T W ,
2

1
L 2 = i B tr T V , V ,
2

L 3 = i tr W V , V ,

2
L 4 = tr(V V ) ,
2

L 5 = tr V V ,

L 6 = tr(V V ) tr T V tr T V ,

2

L 7 = tr V V tr T V ,
2
1
L 8 = tr(T W ) ,
4

1
L 9 = i tr(T W ) tr T V , V ,
2
2
1
L 10 = tr(T V ) tr(T V ) ,
2
(18)
T 2U T 3 U .
(24)
(13)
(14)
(15)
(16)
(17)
(19)
(20)
(21)
(22)
(23)
with
Each operator in the Lagrangian LEW is invariant under the following local SU(2)L U (1)Y
gauge transformation
U gL UgY ,
W gL W gL igL gL ,
BY BY igY gY ,
B B ,
W gL W gL ,
V gL V gL ,
where the gauge transformation factors gL and gY are defined as

gL (x) exp ig a (x)T a ,
gY (x) exp ig (x)T 3 .
T gL T gL ,
(25)
(26)
a (x)
Here
and (x) are the real parameters of the gauge transformation of SU(2)L and U (1)Y ,
respectively.
While operators L W/Z and L 0 contribute to the vector boson mass term, the operators L i , i =
1, . . . , 10, contribute to their kinetic terms, TGC and QGC, as tabulated in Table 1.
Accordingly we can classify the chiral coefficients into three groups: (1) 0 , 1 , and 8 contribute to the weak boson two-point functions, and are constrained by the electroweak precision
data [23]; (2) 2 , 3 , and 9 contribute to the three-point couplings but not to the two-point
functions, and are constrained by the TGC measurements [1215]; (3) 4 , 5 , 6 , 7 , and 10
contribute only to the weak boson four-point couplings (QGC).
The typical size of the allowed range of 11 the chiral coefficients are hence:
O(0 , 1 , 8 ) 103 ,
O(2 , 3 , 9 ) 10
(27)
(28)
115
Table 1
Operators and their contributions to 2, 3, and 4 point gauge-boson vertices
v 2 L
L 1
L 2
L 3
L 4
L 5
L 6
L 7
L 8
L 9
L 10
2 pt. vtx.
3 pt. vtx. (TGC)
4 pt. vtx. (QGC)
O(4 , 5 , 6 , 7 , 10 ) 1.
(29)
We will see in the following section, however, that the QGC couplings in Eq. (29) are constrained
severely by the perturbation unitarity conditions for > 1 TeV.
3. Constraints on the chiral coefficients
If we do not consider a particular class of the underlying theory that leads to the electroweak
symmetry breaking, all the chiral coefficients are arbitrary parameters. In this pure phenomenological viewpoint, we study constraints on their magnitude from the electroweak precision data
of the gauge-boson two-point functions, the TGC measurements, and from the perturbation unitarity conditions from the weak boson scattering amplitudes.
In our analysis, the bounds on the chiral coefficients listed in this section are determined at the
scale = mZ . Strictly speaking, the constraints on 2 , 3 , and 9 from the TGC measurements at
LEP2 are obtained at the scale 2mZ , and those on the 4 , 5 , 6 , 7 , and 10 are derived from
the perturbative unitarity by assuming its validity up to the scale . However, since we consider
the effects of these loosely constrained chiral coefficients to the running of the most precisely
measured coefficients, 0 , 1 , and 8 only, the scale dependence of all the other couplings give
negligibly small effects on our results. They can be considered as higher order corrections of our
leading order analysis.
Here we consider the most general case by retaining all the operators including those which violate the custodial SU(2)c symmetry, since the underlying dynamics can break it explicitly [20].
If we impose the custodial symmetry, the following chiral coefficients vanishes:
0 , 6 , 7 , 8 , 9 , 10 0.
(30)
We discuss the implication of the custodial symmetry in the latter sections of this paper.
3.1. Constraints from the two-point vertices
By using three accurately measured quantities in Table 2, 1/EM , GF , and mZ , we fix the
vacuum expectation value v, and the gauge couplings g and g . The parameter v is identified as

v(mZ ) = 1/
(31)
2GF = 246.26 GeV,
116
Table 2
The inputs to fix g(mZ ), g (mZ ), and v(mZ )
Inputs
Value
1/EM (mZ )
GF
mZ
127.87
1.166 105 GeV2
91.18 GeV
Table 3
Current values of the three best measured electroweak parameters [24] and mt [27]
Parameter
Current value
mW

eff
sin2 W
mt
80.403 0.029 GeV

83.984 0.086 MeV
0.23152 0.00014
171.4 2.2 GeV
and the two gauge couplings are identified as

g(mZ ) = 0.66,
g (mZ ) = 0.36,
(32)
by using the tree-level relations. We retain only two digits in Eq. (32), and do not consider their
errors, because their small variations do not affect our studies on the ST parameters in the
leading order.
To make a global fit with the oblique parameters S, T , and U [21], we follow the strategy of
Peskin and Wells [22], and consider only three most precisely measured quantities, the average
value of charged leptonic partial decay width of Z, the effective Weinberg mixing angle
eff , and m . The latest results from the LEP, SLC, and Tevatron [24] are shown in Table 3.
sin2 W
W
eff , m and S, T , U can be expressed
In the Standard Model, the relations among , sin2 W
W
as [25]
mW (GeV) = 80.377 0.288 S + 0.418 T + 0.337 U,
(33)
(GeV) = 0.08395 0.00018 S + 0.00075 T ,
(34)
eff
sin2 W
= 0.23148 + 0.00359 S 0.00241 T ,
(35)
ref
where the central values are obtained by using the Zfitter 6.41 [26] with mref
t = 175 GeV, mH =
100 GeV, s (mZ ) = 0.1176, and 5h = 0.0279 as inputs. The top and Higgs mass dependence
of the SM predictions are incorporated in the shifts S, T , and U [25].
In order to utilize the above formula designed for the theories with a Higgs boson to theories
like the EWCL without a Higgs boson, we follow the prescription of Bagger, Falk and Swartz,
[11]:
ND
S = SSM SHiggs
+ S,
ND
U = USM UHiggs
+ U.
ND
T = TSM THiggs
+ T,
(36)
Here S, T , and U , are the chiral couplings in the EWCL, which do not have dependence on mH .
ND , T ND , and U ND are the Higgs boson contributions in the heavy Higgs limit and can be
SHiggs
Higgs
Higgs
simply expressed as

ND
SHiggs

1 5
mH
=
ln
,
6 12
mZ
ND
THiggs
117

3 5
1
mH
=
ln
,
mZ
cos2 W 8 12
ND
UHiggs
= 0.
(37)
By using the parameterization of SSM , TSM , and USM in Ref. [25], we observe that the mH
dependence cancel accurately for mH > 300 GeV. For definiteness, we set mH = 500 GeV and
find
ND
SSM SHiggs
= 0.057 0.007xt ,
ND
TSM THiggs
= 0.004 + 0.125xt + 0.003xt2 ,
ND
USM UHiggs
= 0.003 + 0.022xt ,
(38)
where xt = (mt 175)/10 parameterize the remaining mt dependence.

It is now straightforward to find constraints of S(mZ ), T (mZ ), and U (mZ ) from the data of
Table 3, by using the parameterizations Eqs. (33)(38), we find
S(mZ ) = (0.01 0.10)
T (mZ ) = (+0.09 0.14)

U (mZ ) = (+0.06 0.13)

=

1
,
0.93
1
0.55 0.69 1
(39)
where the 1 errors and their correlations are given. In Fig. 1, we show the 1 (39% C.L.)
allowed region by a solid contour in the ST plane.
Fig. 1. The S(mZ )T (mZ ) contours with 1 error from the electroweak data of Table 3. The solid-line contour shows the
constraint without the custodial symmetry while the dotted-line contour shows the constraint with the custodial symmetry
(U (mZ ) = 0).
118
We can make one-to-one correspondences between the chiral coefficients 1 , 8 , 0 and the
oblique parameters S, T , U as in Ref. [10]
1
EM
1
S(),
0 ()
T (),
8 ()
U ().
(40)
16
2
16
Here the magnitudes of i (mZ ) are fixed by S, T , and U at = mZ , where EM = 1/137.36
is the fine structure constant to conform with the T parameter definition of by Peskin and
Takeuchi [21]. The dependence of the i () parameters on are determined by the RGE provided in Section 4. Eq. (40) can be interpreted as the definition of the running S, T , and U
parameters at = mZ . In terms of the chiral coefficients 1 (mZ ), 0 (mZ ), and 8 (mZ ), the
constraint Eq. (39) gives
1 ()
1 (mZ ) = (+0.02 0.20) 102 ,
0 (mZ ) = (+0.03 0.05) 102 ,
8 (mZ ) = (0.12 0.25) 102 ,
(41)
with the same correlation matrix.

When we impose the custodial symmetry. The constraint makes the allowed area smaller,
and the optimal values of S(mZ ) and T (mZ ) moves up slightly, because of the positive value of
U (mZ ) in the 3-parameter fit and the negative correlations in the third row of Eq. (39). We obtain

S(mZ ) = (+0.02 0.09)
1
(42)
=
.
0.91 1
T (mZ ) = (+0.13 0.10)
The corresponding 1 contour is shown in Fig. 1 by a dotted-line contour. In terms of the chiral
coefficients, we impost 8 (mZ ) = 0 and find
1 (mZ ) = (0.04 0.17) 102 ,
0 (mZ ) = (+0.05 0.04) 102
(43)
with the same correlation, = 0.91. We will use the constraints Eq. (42) or equivalently Eq. (43)
in the analysis when the custodial SU(2)c symmetry is assumed.
These results roughly agree with those given in Ref. [11], where the small differences can be
attributed to the changes in the input electroweak data.
3.2. Constraints from the TGC
There are three chiral coefficients 2 , 3 , 9 in the EWCL which contribute to the triple
gauge-boson couplings (TGC) but not to the two-point functions. The relations between the
experimentally measured anomalous TGC [28] and the three-point chiral coefficients are given
in Ref. [10]:
e2
(1 8 + 2 + 3 + 9 ),
s2
1
e2
e2
e2
+
+
(
)
+
(8 + 3 + 9 ),
kz = 2
0
1
1
2
c s2
c2 (c2 s 2 )
c2
s2
1
e2
e2
gZ1 = 2
+
+
3 .
0
1
c s2
c2 (c2 s 2 )
s 2 c2
k =
(44)
Because the constraints of 1 , 0 , and 8 in Eq. (41) are much more stringent than those of 2 ,
3 , and 9 from the TGC measurements, we can simplify the above relations to
k = (2 + 3 + 9 )g 2 ,
kZ = 2 g 2 + (3 + 9 )g 2 ,
119
g1Z = 3 gZ2 ,
(45)
in our analysis.
The unitarity bounds for anomalous TGC, k , kz and g1Z derived [29] from processes
1
f f 2 V 1 V 2 are summarized as
Z 0.87
1.86
0.85
g <
(46)
,
|kZ | < 2 ,
.
1
2
2
Here (TeV) is the cutoff scale, up to which we request the validity of the perturbation theory.
For = 1 TeV, these unitarity bounds constrain the magnitudes of anomalous TGC to be of
order one. Hence, the unitarity constraints in Eq. (46) can be neglected when compared with the
constraints from the TGC measurement even up to 3 TeV.
Constraints from the D experiment at the Tevatron p p collider [16] on anomalous TGC are
not much stronger than the unitarity bounds given in Eq. (46) at 1 TeV. It is the TGC measurements at LEP2 from the process e+ e W + W [1215] that give the best constraints on these
parameters. We use the constraints on 2 , 3 , and 9 from the LEP2 TGC measurements [1215]
for the following three cases.
In case (1), we impose the custodial symmetry on the anomalous dimensionless EWCL
couplings, rendering 9 = 0, i.e. Eq. (30). Eq. (45) leads to a relation among the three TGC
observables,
|k | <
Z = tan2 w + g1Z .
(47)
In this scenario, we use the following results of the TGC measurements by the LEP working
group [30], each of which has been obtained from one-parameter fit:
k (mZ ) = 0.03 0.05,
g1Z (mZ ) = 0.02 0.02.
(48)
Although there may be correlations between these two parameters in the two parameter fit, it can
be small [30]. If we assume that the correlation between the errors is negligibly small, then from
Eq. (48) and 9 = 0, we find

2 (mZ ) = (0.04 0.12)
1
(49)
=
.
0.46 1
3 (mZ ) = (0.03 0.04)
In case (2), we adopt the two parameter-fitting result of the L3 Collaboration [14], which has
also been obtained under the custodial symmetry (47), and reads:

k (mZ ) = (+0.16 0.13)
1
.
=
(50)
0.71 1
g Z (mZ ) = (0.09 0.05)
1
In terms of the chiral coefficients, we find

2 (mZ ) = (+0.54 0.36)
1
=
.
0.82 1
3 (mZ ) = (0.16 0.10)
(51)
In the above two cases we set 9 = 0 by appealing to the custodial symmetry. In general,
the custodial SU(2)c symmetry that may explain the smallness of 0 does not necessarily imply
the suppression of 9 . This motivates us to examine the third case where the chiral coefficients
are analyzed in the absence of the constraint from Eq. (47). Since the experimental analysis for
generic TGC without the constraint cannot be found, we use the measurement of kZ (mZ ) from
120
the L3 Collaboration [14] as an input, even though the experimental analysis was carried out
by assuming the symmetry. Along with this analysis we choose the data on the other two TGC
observables from the LEP combined limits, Eq. (48), on k (mZ ) and g1Z (mZ ) [30], as the
case (3):
kZ (mZ ) = 0.08 0.06,
g1Z (mZ ) = 0.02 0.02.
k (mZ ) = 0.03 0.05,

(52)
Although the above limits should be much stronger than the true constraints when all the three
TGC are allowed to vary freely, we adopt the above bounds as inputs of our general analysis without the custodial symmetry constraints. Again by neglecting the correlations among the errors of
Eq. (52), we find

2 (mZ ) = (+0.09 0.14)
1
3 (mZ ) = (0.03 0.04)

= 0.00
.
(53)
1
0.65 0.32 1
(m ) = (0.12 0.12)
9
The bound is too stringent because the individual results of Eq. (52) are obtained under the
constraint of Eq. (47) and also because the possible correlations among those measurements are
neglected. However, it serves our purpose of showing the possible impacts of custodial SU(2)c
symmetry violation.
In all the above three cases, we observe that 3 is more stringently constrained by the LEP2
data than 2 and 9 . Fig. 2 shows the 1 error contour for 2 3 for these three cases. We note
that the central values of the L3 data deviate significantly from the prediction of the Standard
Model.
Fig. 2. The 2 3 contours with 1 error (39% confidence level). In case (1), the data is given in Eq. (48) and case (2),
the data is given in Eq. (50). In both cases (1) and (2), the custodial symmetry condition 9 = 0 is imposed. In case (3),
the data is given in Eq. (52) where 9 is taken as a free parameter in the fit.
121
3.3. Perturbative unitarity constraints

The most stringent theoretical constraint for the four-point chiral coefficients comes from the
partial wave unitarity conditions of the longitudinally polarized vector boson scattering amplitudes. These scattering amplitudes grow with energy if there are no new resonances up to the
cutoff scale [31].
Here we only consider the processes with J = 0 channels and find
|44 + 25 | < 3
v4
,
4
4

4 + 6 + 3(5 + 7 ) < 3 v ,
4

6 v 4
4 + 5 + 2(6 + 7 + 10 ) <
.
(54)
5 4
|34 + 45 | < 3
4

2(4 + 6 ) + 5 + 7 < 3 v ,
4
v4
,
4
The five constraints are obtained from WL+ WL+ WL+ WL+ , WL+ WL WL+ WL , WL+ WL
ZL ZL , WL+ ZL WL+ ZL , and ZL ZL ZL ZL , respectively. In our analysis, the effects of the
terms proportional to v 2 /2 are found to be negligibly small and hence are dropped.
Recently, Distler et al. [32] found that from the amplitudes of the scattering processes
ZL ZL ZL ZL and WL ZL WL ZL , by using dispersion relations and by assuming Lorentz
invariance, analyticity, unitarity, and custodinal symmetry, it is possible to derive the lower
bounds for the chiral parameters 4 and 5 , which read
5 + 24
1
1.08,
96 2
1
0.31.
96 2
To quote these numbers, we set = v and = 1/5.
We now summarize the current bounds on the chiral coefficients of the EWCL in Table 4.
Among the six terms that contribute to the two-point functions, the first three coefficients, g, g ,
and v, are fixed by EM (mZ ), GF , and mZ in Table 2, and the remaining three chiral coefficients,
0 , 1 , and 8 are constrained from the Z pole data, the precise W mass measurement, and the
top quark mass measurement, see Table 3. The three three-point chiral coefficients, 2 , 3 , and
4
Table 4
Summary on our knowledge of chiral coefficients, from Eqs. (32), (41), (53) (here we take
the most stringent bounds), and Eq. (54)
Chiral coefficients
Central values
Error bars ()
0 (mZ )
1 (mZ )
2 (mZ )
3 (mZ )
+0.0003
+0.0002
0.09
+0.03
0.0005
0.0020
0.14
0.04
4 (mZ )
4
v4
5 (mZ )
4
v4
6 (mZ )
4
v4
7 (mZ )
8 (mZ )
9 (mZ )
v4
0.0025
0.12
10 (mZ )
0.0012
+0.13
4
v4
122
9 are determined from the LEP2 W pair production measurements, and the five four-point chiral
coefficients, 4 , 5 , 6 , 7 , and 10 are constrained by the five perturbative unitarity conditions
of Eq. (54) if there is no new resonances up to the scale .
We observe from Table 4 that the two-point coefficients 0 , 1 , 8 are constrained to be
103 , the TGC coefficients 2 , 3 , and 9 can be 101 , while the remaining coefficients
should be smaller than (v/)4 , or 102 for 1 TeV.
It may be instructive to compare the above constraints with the corresponding ones of the QCD
chiral theory. Ref. [5] presents the fit of these dimensionless couplings in the chiral perturbation
theory describing low energy QCD. The constraints found in Ref. [6] are L9 (m ) = (7.4 0.7
103 ) and L10 (m ) = (6.0 0.7 103 ), which correspond to 2 (= 3 ) and 1 respectively,
in the electroweak theory.
The large and positive value of 1 at 1 TeV, favored by the electroweak data, has been
confronted against the large and negative value of L10 (m ), which led to the assertion [21] that
the electroweak symmetry breaking does not mimic QCD. Although strictly QCD-like theories
may also give 2 3 102 as L9 (m ), we examine possible implications of models with
2 , 3 , 9 101 , which are still allowed by the present TGC measurements.
4. The function of chiral coefficients at one-loop level
In order to find the connection between S(mZ )T (mZ ) and S()T (), by using the method
developed in [33], we extend the previous one-loop RGE [18] of the EWCL by including the
three-point and four-point chiral coefficients in the functions of S and T (equivalently, the
functions of the chiral coefficients 1 and 0 ). In the Wilsonian renormalization group concept,
at the one-loop level, all dimensionless chiral coefficients i should run in a logarithmic way
from mZ to , just like the gauge couplings, due to the screening effects of the active quantum
degree of freedoms (such as the Goldstone particles, the gauge-bosons, and the ghosts). This
logarithmic running is obtained by using the dimensional regularization in our calculation.
The RGE for the gauge couplings g and g , for the vev term v 2 , as well as for the chiral
coefficients i can be simply expressed as
d
1
g ,
g=
dt
8 2
d
1
g ,
g =
dt
8 2
d 2
1
2,
v =
dt
8 2 v
d
1
,
i =
dt
8 2 i
(55)
where t = ln(/mZ ). The g,g ,v 2 ,i are the beta functions for the running of g and g , v 2 , as
well as i . Below we list the functions of the chiral coefficients of the EWCL at one-loop.
Technical details of our calculation is described in Appendix A. The gauge fixing terms and the
treatment of the ghost terms are presented in Appendix B.
We organize and group these functions in the order of their contributions to the multi-point
functions. The functions are written for those operators which contribute to the two point
functions (0 , 1 , and 8 ), three point functions (2 , 3 , and 9 ), and four point functions (4 ,
5 , 6 , 7 , and 10 ). In the standard derivative power counting rule, these terms are counted as
O(p 6 ) order effects.
We first write g,g for the gauge couplings:

g3
52 g 2
29 0
139 g 2
28g 2 g 2
2
2
g =

1 g 48 g +
3
+
, (56)
2
4
6
6
3
2
6

g 3 1
53 g 2
0
21 g 2 2 g 2 +
.
g =
(57)
2 12
3
3
123
It is noted that the g and g both have constant terms which are expected by naively counting of
the active bosonic degrees of freedom. Compared to the SM functions, only the physical Higgs
boson contributions are absent. We have not included the contribution of active fermionic degree
of freedoms which can be fixed in a straightforward manner. Also we find that the three-point
chiral coefficients 2 , 3 , and 9 , contribute to the functions of the gauge couplings, and can
modify the asymptotic behavior of gauge couplings, which should match the gauge couplings of
the ultra-violet theory at the scale .
The function of v 2 is given as

3g 2 3g 2
3g 4
31 g 2 g 2 38 g 4
3
+
+ 0 g 2 3g 2
+
+ 2 3g 2 g 2
v 2 =
2
4
2
2
4
2

4
2

2

3g g
21 4 3gZ
3 2g 4 3g 2 g 2 9 g 4
4
g +
2
4
4

4
36 gZ
15 4
5
(58)
g + 3gZ4
37 gZ4 .
2
4
For v 2 we notice that it has no dependence on 10 . This can be attributed to the fact that the use
of dimensional regularization results in retaining only the logarithmic running of v 2 .
The functions of chiral couplings, 0 , 1 , and 8 , are given as

2
2 2
3g 2
9g 2
31 g 2 g 2 38 g 4
3g 4
9g
3g g
+ 0
+
+ 2
0 =
8
4
4
4
8
2
4

4

2

2
2

2
2

2

4
3g g
15g
g
15g g
33 g g
+ 9 +
+ 4
+
+
2
2
4
4
8

2 2
4
4

4

910 gZ4
33gZ
3g
3g g
3g
+ 5
+
+ 6
+
+ 7 3g 4 + 3gZ4 +
, (59)
2
4
4
8
2
1
52 g 2 53 g 2 9 g 2
+ 41 g 2 8 g 2
+
,
(60)
12
2
6
2
0
52 g 2 3 g 2 179 g 2
+ 1 g 2 + 128 g 2
+
.
8 =
(61)
2
6
2
6
We remark here that the three-point chiral coefficients 2 , 3 , and 9 , affect the running behavior
of the two-point chiral coefficients 1 , 8 , and 0 , parameters. We observe that the functions
for 1 and 8 do not contain the four-point chiral coefficients 4 , 5 , 6 , 7 , and 10 but 0
does. Interpreting this exception in term of Feynman diagrams, we observe that 1 and 8 only
receives the radiative corrections through the diagram of Fig. 3(a) while 0 receives the radiative
1 =
(a)
(b)
Fig. 3. Feynman diagrams for two-point functions. Wavy line loops should include vector, Goldstone, and ghost loops.
The solid circles show the chiral operators Li .
124
corrections from both Figs. 3(a) and 3(b). It is the diagram Fig. 3(b) which renders the entry of
the four-point chiral coefficients in the function of 0 .
The functions of the three-point chiral coefficients 2 , 3 and 9 , are given as

1
g 2
0 1 g 2
53 g 2 9 g 2
2
2 =
+
+ 2 2g +
+
24
6
2
12
12
12

2
2

2

2
g
g
g
6 g
+
5
g 2
7 g 2 ,
+ 4
(62)
4
2
2
2

1
0 1 g 2 2 g 2
59 g 2
59g 2 g 2
+
+ 3
+
3 =
24
6
4
6
12
4
3

2
2
2
2
3g 2
g 2
3g 2
g 2
5g
5g
5g
5g
+
5
+ 6
+
7
,
+ 4
4
8
2
4
4
8
2
4
(63)
0 1 g 2 38 g 2 2 g 2 53 g 2 1199 g 2
+
+
+
9 =
2
4
4
12
4
12

74 g 2 55 g 2
5g 2 5g 2
5g 2 7g 2
(64)
6
+
7
+
.
8
4
4
8
2
4
The functions of the four-point chiral coefficients, 4 , 5 , 6 , 7 , and 10 are given as

1
52 g 2 3 g 2 9 g 2
11g 2
2
4 = 0
+ 4
+ 6g
+ 5 2g 2 + 3g 2
12
6
2
6
2

2
7g 2
g
+ 27 g 2 ,
+ 6 +
(65)
2
2

1
15g 2
3g 2
0 2 g 2 3 g 2 9 g 2
4 g 2 +
+ 5 g 2
5 = +
24
2
3
2
3
4
2

2

2
2

2
3g
3g
3g
g
6
(66)
+
+ 7
+
,
4
4
2
2

30
254 g 2 75 g 2
25g 2 5g 2
2 g 2 +
+
6
+ 27 g 2
6 =
2
4
2
4
4

+ 10 g 2 + 5g 2 ,
(67)

2

2

2

2
2

2
30 32 g
9 g
414 g
115 g
33g
5g
7 =
+
+
+
+ 6
+
4
4
4
8
4
8
8

2

2

7g
13g
+ 7
(68)
+
+ 10 2g 2 + 4g 2 ,
4
4

176 g 2 77 g 2
+
+ 910 g 2 g 2 .
10 =
(69)
4
2
In calculating the above functions, we have performed the following consistency checks:
(1) We ensure that when all the anomalous couplings are set to vanish, our calculation reduces
not only to the standard functions for gauge couplings but also to the constant terms in the
functions for the chiral coefficients [18]. (2) The electromagnetic symmetry UEM (1) is checked
and found to hold in our calculation at each step. (3) Whenever g is set to vanish, Z can be
125
expressed in terms of W 3 , and similarly W can be expressed in terms of W 1 and W 2 . Thus

we find that the global SU(2)c custodial symmetry explicitly validated in our calculation at each
step. (4) When we only keep 0 in the functions, our results agree with those obtained by
M. Tanabashi [34] and in Ref. [35]. (5) Using the alternative parameterization for Goldstone
boson as

1 1

2 T + 2 2 T 2
23 T 3
exp i
,
U = exp i
(70)
v
v
our computational method yields the same answer.
5. Current experimental uncertainty in S()T ()
In this section we analyze the current experimental uncertainty of S()T () obtained from
the constraints at the mZ scale listed in Table 4. First, we analyze the experimental uncertainty
of S()T () induced by the uncertainty of the three-point chiral coefficients. Second, we will
analyze the experimental uncertainty of T () induced by both the uncertainty of the three-point
and the four-point chiral coefficients.
Numerically, we find that S(), T (), and U () (or 1 (), 0 (), and 8 () via Eq. (40))
are not sensitive to the running of the other chiral coefficients and that of the gauge couplings.
Therefore we can use linear approximation of the RGE solutions:
2
,
1 ln
mZ
2
U () = U (mZ ) 8 ln
.
mZ
S() = S(mZ )
T () = T (mZ ) +
4 2 EM 0
ln
,
mZ
(71)
Here we explain the basic difference between the experimental uncertainty analysis of ST
given in [23] and ours. The ST uncertainty contour figures given in [23] are obtained in the
minimal Standard Model with a Higgs boson as a regulator, along with the introduction of three
extra two point operators, L 0 , L 1 , and L 8 , to describe the new physics effects. In this analysis,
although the central values of S, T , and U vary with the Higgs boson mass, their error bars are
insensitive to mH and are determined by the experimental errors. To have an analogy with the
prediction of the QCD-like theories, the experimental values of ST are determined by choosing
mH = 1 TeV (1 TeV is assumed to be the compositeness scale).
In our uncertainty analysis, we do not use Higgs as regulator but adopt the dimensional regularization and the MS scheme. The uncertainties of S(mZ ), T (mZ ), U (mZ ) and their correlations
are then determined by the electroweak data listed in Table 4. Therefore, the errors at the scale
= mZ are essentially the same as those of Ref. [23]. In our analysis, however, the error bars of
S()T () become larger and larger when we extrapolate the low energy constraints to high energy scale, since more and more uncertainty from the other operators creeps in the computation.
5.1. The experimental uncertainty of S()T () from the anomalous TGC measurement
The parameters, S(), T () and U () are the values of parameters S, T and U at the matching scale , where the EWCL matches with the underlying fundamental theories such as the
QCD-like models, extra dimension models, Higgsless models, etc. In the perturbation method,
S(), T (), U () and S(mZ ), T (mZ ), U (mZ ) are connected by the improved renormalization
group equations (60), (59), and (61), respectively.
126
Fig. 4. The S()T () contours for various cutoff scales . The solid contours are obtained when the TGC data is taken
from the case (1), Eq. (48), and U (mZ ) = +0.00. The dashed contours are obtained when all the three and four point
chiral coefficients are set to be zero.
Fig. 5. The S()T () contours for various cutoff scales . The solid contours are obtained when the TGC data is taken
from the case (2), Eq. (50), and U (mZ ) = +0.00. The dashed contours are obtained when all the three and four point
chiral coefficients are set to be zero.
With the S, T , U values at = mZ in Section 3.1 and the three-point chiral coefficients at
= mZ for all the three cases as given in Section 3.2, we evolute the values of S, T , U up
to the cutoff scale . The results are shown in Fig. 4 for the case (1) with the constraints in
Eq. (49), Fig. 5 for the case (2) with the constraints in Eq. (51), and Fig. 6 for the case (3) with
127
Fig. 6. The S()T () contours for various cutoff scales when the custodial symmetry is not imposed on the chiral
coefficients. The solid contours are obtained when the TGC data is taken from the case (1), Eq. (52) and S, T , and U
at = mZ are constrained by Eq. (39). The dashed contours are obtained when all the three and four point chiral
coefficients are set to be zero.
the constraints in Eq. (53). In Figs. 4 and 5 for the cases (1) and (2), we set all the coefficients that
violate the custodial symmetry to zero at the scale mZ except 0 , and therefore we use Eq. (42)
for U (mZ ) = 0 as the input. Because the hypercharge gauge interactions violate the custodial
symmetry, we cannot impose the conditions at all the scales. According to Eq. (61), however,
U () remains negligibly small at = 1 TeV, U (1 TeV) = 0.05 0.20 for the case (1) and
U (1 TeV) = 0.08 0.55 for the case (2). In Fig. 6 for the case (3), we allow all the coefficients
to vary, and use Eq. (39) as the input. Results for at 300 GeV, 1 TeV, and 3 TeV are shown in
these figures.
The dashed line contours correspond to the analysis without the contributions of the threepoint chiral coefficients, i.e., they include only the 1 error of the two-point chiral coefficients.
Therefore the error contours do not change their shapes and sizes while the central values of
the contours vary with the cutoff scale . The solid-line contours show the analysis which includes the contributions of the three-point chiral coefficients in addition to the two point chiral
coefficients.
Although the contribution of the three-point chiral coefficients to 1 ()0 () (or S()
T ()) in Eqs. (60) and (59) are loop factor suppressed, the experimental uncertainty of these
chiral coefficients is almost two orders of magnitude larger than those of the two-point chiral
coefficients, as shown in Table 4. Therefore, if we do not make any theoretical assumptions on the
magnitude of the chiral coefficients, the uncertainties of S()T () can be dominated by those
of three-point chiral coefficients at large . The central values of S()T () move according
to the central values of the present TGC measurements, and the size of the solid contours grows
with increasing .
We would like to mention some salient features with respect to Figs. 46: (1) In the absence of
the TGC contribution (dashed line contours), S() decreases as increases from the reference
128
LEP1 constraint at = mZ . This is in agreement with the observation of Refs. [11,22]. The
dashed contours for = 1 TeV essentially agree with the ST contour of Ref. [23] at mH =
1 TeV.
(2) When the operators that contribute to the TGC are taken into account, the dependences
of the center of the ST contour are governed by the central values of the TGC measurements
in Eqs. (48), (50), and (52) for the case (1) (Fig. 4), the case (2) (Fig. 5), and the case (3)
(Fig. 6), respectively. For instance, central value of the TGC contribution to 1 , (152
53 + 39 )g 2 /6, is 0.03 for the case (1), Eq. (48), which adds up to the leading contribution
of 1/12, and hence the central value of S() gets even more negative than that of the dashed
contours for 2 = 3 = 9 = 0. The central value of the same combination is 0.65 for the case (2),
Eq. (50), from the L3 measurement [14], which makes the central value of S() positive at
large in Fig. 5. Large error of this single experiment constraint explains the rapid growth of
the 1 allowed region with increasing . In the case (3), where the custodial symmetry violating
coefficient 9 is allowed to take an arbitrary value, the central values of Eq. (52) make the same
combination about 0.08 which approximately cancels the constant term of 1/12. This results
in the almost -independence of the central value of S() as observed in Fig. 6.
In summary, those models which satisfy the uncertainty
(152 53 + 39 )g 2 /6 1.092 0.363 + 0.229 > 1/12
(72)
give S() S(mZ ), and hence QCD-like models can still be consistent with the electroweak
data provided that they give rise to the TGC anomaly which satisfy the above condition. In
case of the -dependence of the T parameter, the relevant contribution of the TGC operators
in Eq. (59) gives 0.072 + 0.083 0.059 , which tends to be smaller than the leading term of
3g 2 /8. Accordingly, in all our examples, T () grows with increasing as can be read out from
Figs. 46 for the cases (1), (2), and (3), respectively.
The central values of S()T () and the errors induced by TGC measurements can easily
be read off from Figs. 46. We find
S(1 TeV) = 0.15 0.22,
T (1 TeV) = +0.47 0.12,
(73)
for the case (1), Eq. (48), with one-parameter fit data and the custodial symmetry constraint.
S(1 TeV) = +0.84 0.60,
T (1 TeV) = +0.68 0.18,
(74)
for the case (2), Eq. (50), with two-parameter fit data from L3 measurement [14] and the custodial
symmetry constraint, and
S(1 TeV) = 0.02 0.20,
T (1 TeV) = +0.52 0.12,
(75)
for the case (3), Eq. (52), without the custodial symmetry constraint.
The central values of the above results can easily be obtained from the TGC contribution to
the functions of 1 and 0
S(1 TeV) S(mZ ) = 0.13 + 1.662 0.553 + 0.339 ,
T (1 TeV) T (mZ ) = 0.40 + 0.602 + 0.703 0.449 .
(76)
These numbers and figures demonstrate an apparent fact that the uncertainty of the TGC
measurement can significantly affect the value of S()T (). We point out the fact that the
current precision of the TGC measurement is not good enough to fix the sign of S().
129
Table 5
Values of T () and its uncertainty. Individual 1 errors from TZ , T TGC and T QGC are also shown
T () 1
TZ
T TGC
T QGC
0.3 TeV
0.5 TeV
1 TeV
3 TeV
0.37 8.91
0.41 1.16
0.52 0.22
0.73 0.25
0.14
0.14
0.14
0.14
0.06
0.08
0.12
0.17
8.91
1.15
0.10
0.04
5.2. The uncertainty of T () from the three-point and four-point chiral coefficients
In Figs. 46, the TGC contributions can at most modify the central value of T (1 TeV) by
| T (1 TeV)| 0.20. Thus the contribution of TGC is not large enough to cancel the large leading contribution from 3g 2 /8 in the function of the 0 (or equally T ) parameter, which makes
T () positive.
In this section, we analyze the effect of the four-point chiral coefficients to the T () parameter. Numerical results are given in Table 5. The columns TZ and T TGC list the 1 uncertainty
from the measurement of T and the TGC at = mZ , respectively. The column T QGC lists
the uncertainty from the QGC constrained by the theoretical unitary bounds. For the TGC constraints, we adopt the case (3), or Eq. (53) as an input.
We notice that with the increasing the TGC uncertainty T TGC increases logarithmically
while the QGC uncertainty T QGC decreases rapidly. This is because of the power dependence
of the unitary bounds given in Eq. (54). Consequently we find that T QGC > T TGC for <
950 GeV but T QGC < T TGC for larger .
From Table 5, we can conclude that in the constrained EWCL parameter space with 1 error in
TGC and with theoretical unitary bounds on QGC, it is unlikely to have a scenario with vanishing
T ( = 1 TeV) while keeping T (mZ ) = 0.09.
6. Conclusions and discussions
In this work, we study the impacts of all the 11 dimensionless chiral coefficients (i , i =
0, . . . , 10) of the standard electroweak chiral Lagrangian, under the constraints from the experimental data and the perturbative unitarity bounds. We provide the improved RGEs by including
the linear terms of all chiral coefficients. By using the improved RGEs, we examine the renormalization scale dependence of the oblique parameters and update the experimental uncertainty
analysis on S() and T ().
We observe that the electroweak precision measurements have constrained the oblique parameters, S, T , and U , and the anomalous TGC. According to the above experimental uncertainty
analysis, we find that current precision data for the chiral coefficients of the EWCL (as given in
Table 4) still allow positive S() parameter space, as shown in Figs. 46, due to the large uncertainty in the TGC chiral coefficients. Therefore, it is premature to claim that the sign of S()
conclusively rules out the QCD-like EWSB mechanism. However, the upcoming colliders, with
higher sensitivity to the TGC and QGC, can further reduce the allowed parameter space and help
to pinpoint the correct model of the EWSB.
Before closing, let us briefly discuss the limitations of our analysis. Most importantly, we
have included neither the two-loop effects of O(p 2 ) operators nor the tree-level contributions
of O(p 6 ) operators. In the cases when the derivative power counting rule holds, the effective
couplings corresponding to the O(p 6 ) operators must be suppressed by 1/(16 2 )2 , in contrast
130
to the 1/(16 2 ) factor for the O(p 4 ) operators. Meanwhile, up to O(p 6 ) order, the running
of i s are not affected by the effective couplings corresponding to the O(p 6 ) operators. Two
loop effects of O(p 2 ) operators must be suppressed by the two loop factor, and they must be
tiny. Therefore, we expect that the uncertainties induced by TGC and QGC are dominant when
= 1 TeV.
We have restricted our functions of chiral coefficients, to retain linear terms and neglect the
terms proportional to (i g 2 )2 and (i g 2 )2 . These terms can be classified as p 8 order effects in
the derivative power counting rule as depicted in Feynman diagram Fig. 3(a). Since the numerical
values of i are small, inclusion of these terms can only render negligible correction to our ST
contours.
The derivative power counting rule in the system has achieved great phenomenological
success and its validity in the EWCL domain is yet to be tested by the experiments. From the
pure theoretical viewpoint, there exist possible theories (e.g., large NT C theories and models with
large extra dimensions) which can have large deviations in the triple and quartic gauge couplings
from the prediction of the Standard Model. For example, in the large NT C theories, we can have
the following power counting rules for the chiral coefficients [5]:

i O(NT C ).
0 O NT0 C ,
(77)
In a situation where the two-point chiral coefficients 0 , 1 , and 8 are already tightly constrained, one may look for symmetry breaking models which can have large triple and quartic
gauge couplings.
Acknowledgements
We would like to thank Ulrich Parzefall for communication on the TGC measurements at
LEP2, Masaharu Tanabashi and Masayasu Harada for stimulating discussions, and Benjamin
Grinstein for communication on theoretical bounds to chiral coefficients. This work is supported
in part by Grant-in-Aid for Scientific Research (#18340060) from Ministry of Education, Culture,
Science and Technology of Japan, and the JSPS core university programs. Q.-S.Y. thanks the
JSPS fellowship program (#P03194) and NCTS (Hsinchu, Taiwan) for financial support. The
work of K.Y. is supported in part by JSPS Postdoctoral Fellowships for Research Abroad and the
National Science Foundation under Grant No. NSF PHY05-51164.
Appendix A. The calculation of the functions of two-point chiral coefficients
We describe the method of calculating the functions of chiral coefficients given in Eqs. (56)
(69) through the following steps [33].
(1) In the background field method [36], we decompose fields in the Lagrangian of Eq. (3) as
follows:
W W + W ,
B B + B,
(A.2)
U U U ,
(A.3)
(A.1)
and U are the background fields of vector boson and background Goldstone
where W , B,
and U are the quantum fluctuations.
bosons in the nonlinear realization, and W , B,
(2) We parameterize U as
a a
2i T
U = exp
,
v
131
(A.4)
and expand the Lagrangian given in Eq. (3) up to the quadratic terms
LEW a LEW LEW a
B +
W +
a
W a
B
and a + .
+ bilinear terms of W a , B,
U ) +
LEW = LEW (W , B,
(A.5)
The structure of the divergences upto dimension 4 operators induced by the quantum fluctuations are organized as

2
2
U ) = 2 g 1 H 1 + 2 g 1 H 2 + v v 2 L W/Z + 0 + 0 v v 2 L 0
LEW (W , B,
g g2
g g 2
v2
v2
+
10
i L i ,
(A.6)
i=1
due to the local gauge symmetries of the classical fields, divergences of higher order operators are not relevant to our purpose and are simply thrown away.
(3) We cast the quadratic terms in quantum fields (the vector bosons V a , the Goldstone bosons
i , and the ghosts v a and ub ) in a compact form with the appropriate gauge fixing terms as
described in Appendix B at one loop level

1 2
V
X
b
V
V
Lquad = V ,
(A.7)
v a 2ab

vu u .
2
2
X
Here we have made the partial integrals and organized the quadratic terms in the mass eigenstate basis. Calculation in the weak interaction eigenstate basis yields the same results, which
serves as one extra checking point for our computation method.
(4) We express the one-loop divergences induced by the quantum fluctuations by using the
Gaussian integral over the d-dimensional spacetime

1
1
.
L1-loop = Tr ln 2V + Tr ln 2 2 Tr ln 2vu + Tr ln 1 X 21
V ; X 2
2
x
(A.8)
(5) Finally by using the heat kernel technique [37], we extract the divergent terms from Eq. (A.8)
which are linear in i . The divergences emanating from the first three terms in Eq. (A.8) are
listed in Tables 6, 7, and 8, respectively. Tables 9 and 10 lists divergences contributed from
the fourth term of Eq. (A.8). The operator v 2 L0 gives two terms, as shown in Eq. (A.6).
The divergences listed in these tables are those of the 0 v 2 L0 , while those corresponding
to 0 v 2 L0 are neglected.
Combination of all the divergences determines the counter terms given in Eq. (A.6) and the
RGE given in Eqs. (56)(58). We have used the following equation of motion to remove the
ambiguity of the parameterization of the Goldstone fields
Z =
O4Z (i )
,
v2
(A.9)
132
Table 6
Divergences contributed from Tr ln 2V /2 up to linear terms of i . The unit is 1/8 2
0 1
1 H
g2 1
1 H
g 2 2
L 1
40g 2
3
20
3
4g 2
4g 2
8g 2
3
10
8g3
4g 2
L 2
L 3
20g
3
2g 2
4g 2
2g 2
4g 2
L 4
L 5
L 6
L 7
L 8
2
12g 2 16
3 g
L 9
2g 2
20g
3
4g 2
L 10
v 2 L W/Z
4
2g 2 g 2 2g 2 g 2 2g 2 g 2 7g 4 + gZ
v 2 L
g2 g 2
g 2 g 2
4
4
4
10g 4 + 4gZ
gZ
4gZ
5
11
2
2
4
4)

2
2

2
2
4
2 g (g + gZ ) g (g + gZ ) g 2 gZ 4(g 4 + gZ
g 2 g 2
4
6gZ
Table 7
Divergences contributed from Tr ln 2 /2 up to linear terms of i . The unit is 1/8 2
0
1 H
g2 1
1 H
g 2 2
1
12
1
6
1
12
1
3
L 1
1
12
L 2
1
24
16
L 3
1
24
1
6
L 4
1
12
1
12
L 5
L 6
L 7
L 8
L 9
g 2 +g 2
2 Z
10
g 2
g 2
g2
g2
2
12
32
3
4
12
12
g2 g2
g2
2
g 2
2
g 2
2
g2
4
2g 2 +g 2
8
4g 2 +g 2
2
2 2
6g +g
4

2
g
4
2
3g8
g2
2
g8
g2
2
2 2
2g +g
4
g 2
2
2
g4
2g 2 +g 2
8
2 2
g g
2
2 2
g +g
4
11g 2 +3g 2
4
2
2
9g +7g
8
g 2
4
2 2
2g +g
8
2g 2 +g 2
4
2 2
2g +g
4
g2
2 2
4g +g
2
g2
2 2
3g +g
2
g2 + g 2
g 2 + g 2
2
2
g 3g
4
2g 2 g 2
3g 2 g 2
2g 2
L 10
5g 2
2)
(g 2 +gZ
v 2 L W/Z
4
g 2
v 2 L 0
8
g2 g 2
3 g2 g2
Z 2
2
3 g2 3 g2 g2 g 2
2
4 Z
4
d W = i
4
4 5 g 4 g 4 gZ
4
74 g 4 14 gZ
gZ
Z
2
4
2 ) g 2 (g 2 +g 2 )
5g 2 (g 2 +gZ
Z 2 g 4 + 11 g 4 g 4 + g 4
Z
8
4
8
8 Z
O W (i )
gZ2 g 2
Z W + 4 2 ,
gZ
v
1 g4
4
g4
8
4
3gZ
2
(A.10)
133
Table 8
Divergences contributed from Tr ln 2vu up to linear terms of i . The unit is 1/8 2
0
1 H
g2 1
2g 2
3
2
3
1 H
g 2 2
10
g2
3
L 1
g3
L 2
2
L 3
g3
L 4
L 5
L 6
L 7
2
L 8
g3
L 9
g3
L 10
v 2 L W/Z
v 2 L 0
2g 2 g 2
g2 g 2
g2 g 2
g 2 g 2
1 2 2
2g g
1 2 2
2g g
Table 9

1
2
Divergences (part A) contributed from Tr ln(1 X 21
V ; X 2 )/2 up to linear terms of i . The unit is 1/8
1 H
g2 1
g 2
2
5g6
2
2
11g
3 +g
5g 2
3
8g 2
3
2
2 2
5g
g +g
12
2
2
2

2
19g
g g 2
g
12 4
2
2
2
g2
15g +13g
2
2
2
2

2
g
5g +8g
2
2
6g 2
1 H 2g 2
g 2 2
2g 2
L 1
g 2
g 2
L 2
g2
2
g 2
4
2
2
5g2 g12

2
g6
5g 2
6
2
g3
g 2
L 3
L 4
L 5
L 6
L 7
L 8
L 9
g 2
g 2
4
2
3g
4
5g 2
6
g 2
12
2
19g4
2
g2 g 2
g2
2
g g 2
g 2
2
3g 2 3g 2 g 2 4g 2
g 2 +2g 2
g 2 + 2g 2
2
2 2
3g 2
18g 2g
2

2
7g 13g 2
5g2
4
g 2
g 2
2
2
2
g 2g
2
g 2
2g 2
3g 2 g 2
g2
2
5g6
g2
g2
2
g4
2
2
7g +2g
2
g2 + g 2
7g 2
2
10
g2
12
13g 2
6
g2
6
g2
3
3g 2 g 2
17g 2
4
L 10
g 2
g 2
7g4
2g 2 6g 2
2
g4
2
3g 2 7g6

2
2
3g
59g
4
12
4g 2 3g 2
14g 2 + 9g 2
Table 10

1
2
Divergences (part B) contributed from Tr ln(1 X 21
V ; X 2 )/2 up to linear terms of i . The unit is 1/8
0
9
2 2
g 4 3g 2g
v 2 L W/Z
2g 2 g 2
2g 2 + 4g 2
2g 2 g 2
2g 2 g 2 + 32 g 4
2g 2 g 2 + 2g 4
g 4
v 2 L 0
g2
3g 2 + 3g 2
g2 g 2
g 2 g 2 + 34 g 4
g 2 g 2
g2
2
g4
3g 2 g 2
2
4
134
where = 1 20 and d W = W ieA W . The terms with O4Z,W (i ) contribute to

O(p 6 ) and higher order operators and do not contribute to the functions of chiral coefficients
given in Eqs. (56)(69).
Appendix B. Gauge fixing terms and ghost terms
In order to cast the quantum fluctuations of vector boson fields into the minimal form of
Eq. (A.7), we make a special choice of the gauge fixing parameters. In the background field
method, the covariant gauge fixing terms are given as

2
GA
LGF,A =
(B.1)
A + fAZ Z ifAW W W + W + W ,
2

2
GZ
Z + fZ Z + ifZW W W + W + W ,
LGF,Z =
(B.2)
2

+
+ ifW Z Z W + ipW Z W + Z + ipW A W + A
LGF,W = GW d W + + fW W

d W + fW ifW Z Z W + ipW Z W Z ipW A W A .
(B.3)
W
These gauge fixing terms explicitly guarantee the UEM (1) symmetry. Gauge fixing parameters
can be uniquely determined as follows by requiring that the quadratic terms to have the minimal
compact form given in Eq. (A.7):
(1) The gauge parameters related with kinetic terms of the propagators of vector bosons are
given as:
GA = C1 ,
GZ = C3
GW = 1.
(B.4)
C22
C1
(B.5)
(B.6)
(2) The gauge parameter related with the kinetic mixing between the quantum fields A and Z is
given as
fAZ =
C2
.
C12
(B.7)
and W + (W ) are given as

(3) The gauge parameters related with the couplings between A (Z)
e
,
fAW =
(B.8)
GA
1 C7
e C2
,
+
fZW =
(B.9)
GZ 2
GZ C1
C5
,
pW A =
(B.10)
2
C6
,
pW Z =
(B.11)
2
C7
.
fW Z =
(B.12)
2
These gauge parameters guarantee that the covariant differential operator of vector sector
has a Hermitian form.
135
(4) The gauge parameters related with the mixing between the quantum vector bosons and the
Goldstone bosons are determined as
gZ v
fZ =
(B.13)
,
GZ 2
1 gv
.
fW =
(B.14)
GW 2
The parameter fZ takes into account the diagonalization and the normalization of the Z
vector boson and the Z Goldstone boson.
Here, the parameters Ci are defined as
g2g 2
(21 + 8 ),
gZ2

gg
C2 = 2 1 g 2 g 2 + 8 g 2 ,
gZ

g2
C3 = 1 + 2 21 g 2 8 g 2 ,
gZ

gg
C5 = 2
1 (1 + 8 2 3 9 )g 2 ,
gZ

g2
1 (8 3 9 )g 2 + (1 2 )g 2 ,
C6 = 2
gZ

g2
1 + 3 gZ2 .
C7 = 2
gZ
C1 = 1
(B.15)
(B.16)
(B.17)
(B.18)
(B.19)
(B.20)
From the gauge fixing terms given in Eqs. (B.1)(B.3), the determinant of the ghost terms are
found as

F
F
det
(B.21)
= det D D +
,

where iD is non-Hermitian with
D = Cgh + ,
(B.22)
where the matrix Cgh is
GA
G
A fAZ
GZ
0
Cgh =
0
0
0
0
and the matrix is
0
0
=
ipW A W +
ipW A W
0
0
1
0
0
0
,
0
1
(B.23)
i GA fAW W i GA fAW W +
+
i GZ fZW W
i GZ fZW W
(B.24)
.
ipW Z W + ieA + ifW Z Z
0
ipW Z W
0
ieA ifW Z Z
Here we observe that Hermiticity of this determinant is broken by both the kinetic parameter Cgh
and the gauge potential terms by the chiral coefficients i . The parameter Cgh is determined
0
0
136
by the chiral coefficients 1 and 8 , which induce the mixing between photon and the Z boson.
In , Hermiticity is broken by the chiral coefficients 2 , 3 , and 9 . On the other hand, D =
+ is Hermitian and is given as
ieW +
0
0
ieW
ieW +
i ggZ W
0
2
i ggZ W +
2
i ggZ W
ieW
2
ieA + i ggZ Z
The term (F / )(/) is

0
0
0
0
0
0
2 2
gZ v
G
4
Z
0
0
g2 v2
4
0
0
0
i ggZ W +
0
(B.25)
ieA i ggZ Z
(B.26)
2 2
g 4v
We can make the ghost determinant to be Hermitian, by using the fact that
b
a
M a b )
det(M ab F
F
a
det
=
.
b
det(M ab ) det(M a b )
(B.27)
This identity is justified if det M ab and det M a b do not vanish or go to infinity. In the ghost
term, the degree of freedom to choose the matrices M ab and M a b reflects the fact that there are
two types of real ghost fields we can introduce, which is labeled as v type and u type ghosts,
respectively.
By properly choosing the matrices M ab and M a b , we can reorganize the ghost determinant as

F b
mass
2
det M ab a M a b = Dgh Dgh + gh
(B.28)
+ gh
,
mass is the same as that of the vector bosons. There

where Dgh is Hermitian. The mass matrix gh
are several ways to adjust the matrices M ab and M a b in order to make the Feynman rules of the
2 Hermitian, up to linear terms
ghost sector well defined. Although we cannot make the term gh
of i , we find our results given in Table 8 is independent of the procedures to hermitize the ghost
determinant.
References
[1] S. Weinberg, Phys. Rev. D 13 (1976) 974;
S. Weinberg, Phys. Rev. D 19 (1979) 1277;
L. Susskind, Phys. Rev. D 20 (1979) 2619.
[2] C.T. Hill, E.H. Simmons, Phys. Rep. 381 (2003) 235;
C.T. Hill, E.H. Simmons, Phys. Rep. 390 (2004) 553, Erratum.
[3] S. Dutta, K. Hagiwara, Q.S. Yan, hep-ph/0603038, unpublished;
Q.S. Yan, hep-ph/0703189, talk presented at SCGT2006 workshop, Nagoya, Japan, November 2006.
[4] J. Gasser, H. Leutwyler, Ann. Phys. 158 (1984) 142.
[5] J. Gasser, H. Leutwyler, Nucl. Phys. B 250 (1985) 465.
[6] M. Harada, K. Yamawaki, Phys. Rep. 381 (2003) 1, hep-ph/0302103.
[7] T. Appelquist, C.W. Bernard, Phys. Rev. D 22 (1980) 200;
T. Appelquist, C.W. Bernard, Phys. Rev. D 23 (1981) 425.
[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]
137
A.C. Longhitano, Phys. Rev. D 22 (1980) 1166.

A.C. Longhitano, Nucl. Phys. B 188 (1981) 118.
T. Appelquist, G.H. Wu, Phys. Rev. D 48 (1993) 3235, hep-ph/9304240.
J.A. Bagger, A.F. Falk, M. Swartz, Phys. Rev. Lett. 84 (2000) 1385, hep-ph/9908327.
A. Heister, et al., ALEPH Collaboration, Eur. Phys. J. C 21 (2001) 423, hep-ex/0104034.
G. Abbiendi, et al., OPAL Collaboration, Eur. Phys. J. C 33 (2004) 463, hep-ex/0308067.
P. Achard, et al., L3 Collaboration, Phys. Lett. B 586 (2004) 151, hep-ex/0402036.
S. Schael, et al., ALEPH Collaboration, Phys. Lett. B 614 (2005) 7.
V.M. Abazov, et al., D Collaboration, hep-ex/0504019.
H. Georgi, Annu. Rev. Nucl. Part. Sci. 43 (1993) 209.
M.J. Herrero, E. Ruiz Morales, Nucl. Phys. B 418 (1994) 431, hep-ph/9308276;
M.J. Herrero, E. Ruiz Morales, Nucl. Phys. B 437 (1995) 319, hep-ph/9411207;
S. Dittmaier, C. Grosse-Knetter, Nucl. Phys. B 459 (1996) 497, hep-ph/9505266.
R. Barbieri, et al., Nucl. Phys. B 703 (2004) 127.
P. Sikivie, L. Susskind, M. Voloshin, V. Zakharov, Nucl. Phys. B 173 (1980) 189.
M.E. Peskin, T. Takeuchi, Phys. Rev. Lett. 65 (1990) 964;
M.E. Peskin, T. Takeuchi, Phys. Rev. D 46 (1992) 381.
M.E. Peskin, J.D. Wells, Phys. Rev. D 64 (2001) 093003.
W.M. Yao, et al., Particle Data Group, J. Phys. G 33 (2006) 1.
ALEPH Collaboration, Phys. Rep. 427 (2006) 257, hep-ex/0509008.
K. Hagiwara, S. Matsumoto, D. Haidt, C.S. Kim, Z. Phys. C 64 (1994) 559, hep-ph/9409380;
K. Hagiwara, S. Matsumoto, D. Haidt, C.S. Kim, Z. Phys. C 68 (1995) 352, Erratum;
K. Hagiwara, D. Haidt, S. Matsumoto, Eur. Phys. J. C 2 (1998) 95, hep-ph/9706331;
K. Hagiwara, Annu. Rev. Nucl. Part. Sci. 48 (1998) 463.
D.Y. Bardin, P. Christova, M. Jack, L. Kalinovskaya, A. Olchevski, S. Riemann, T. Riemann, Comput. Phys. Commun. 133 (2001) 229, hep-ph/9908433;
A.B. Arbuzov, et al., Comput. Phys. Commun. 174 (2006) 728, hep-ph/0507146.
E. Brubaker, et al., Tevatron Electroweak Working Group, hep-ex/0608032.
K. Hagiwara, R.D. Peccei, D. Zeppenfeld, K. Hikasa, Nucl. Phys. B 282 (1987) 253.
U. Baur, D. Zeppenfeld, Phys. Lett. B 201 (1988) 383.
LEPEWWG/TGC/2005-01, WWW access at http://www.cern.ch/LEPEWWG/lepww/tgc.
J.M. Cornwall, D.N. Levin, G. Tiktopoulos, Phys. Rev. Lett. 30 (1973) 1268;
J.M. Cornwall, D.N. Levin, G. Tiktopoulos, Phys. Rev. Lett. 31 (1973) 572, Erratum;
J.M. Cornwall, D.N. Levin, G. Tiktopoulos, Phys. Rev. D 10 (1974) 1145;
J.M. Cornwall, D.N. Levin, G. Tiktopoulos, Phys. Rev. D 11 (1975) 972, Erratum;
C.H. Llewellyn Smith, Phys. Lett. B 46 (1973) 233;
S.D. Joglekar, Ann. Phys. 83 (1974) 427;
B.W. Lee, C. Quigg, H.B. Thacker, Phys. Rev. D 16 (1977) 1519;
B.W. Lee, C. Quigg, H.B. Thacker, Phys. Rev. Lett. 38 (1977) 883.
J. Distler, B. Grinstein, R.A. Porto, I.Z. Rothstein, Phys. Rev. Lett. 98 (2007) 041601, hep-ph/0604255.
Q.S. Yan, D.S. Du, Phys. Rev. D 69 (2004) 085006;
S. Dutta, K. Hagiwara, Q.S. Yan, Nuc. Phys. B 704 (2005) 75.
M. Tanabashi, private communication.
R.S. Chivukula, S. Matsuzaki, E.H. Simmons, M. Tanabashi, hep-ph/0702218.
L.F. Abbott, Acta Phys. Pol. B 13 (1982) 33.
D.V. Vassilevich, Phys. Rep. 388 (2003) 279.
Electroweak corrections to W -boson hadroproduction

at finite transverse momentum
W. Hollik a , T. Kasprzik a , B.A. Kniehl b,
a Max-Planck-Institut fr Physik (Werner-Heisenberg-Institut), Fhringer Ring 6, 80805 Munich, Germany
b II. Institut fr Theoretische Physik, Universitt Hamburg, Luruper Chaussee 149, 22761 Hamburg, Germany
Received 17 July 2007; accepted 19 September 2007

Abstract
We calculate the full one-loop electroweak radiative corrections to the cross section of single W -boson inclusive hadroproduction at finite transverse momentum (pT ). This includes the O() corrections to W + j
production, the O(s ) corrections to W + production, and the tree-level contribution from W + j photoproduction with one direct or resolved photon in the initial state. We present the integrated cross section
as a function of a minimum-pT cut as well as the pT distribution for the experimental conditions at the
Fermilab Tevatron and the CERN LHC and estimate the theoretical uncertainties.
PACS: 12.15.Lk; 12.38.Bx; 13.85.Fb; 13.85.Qk
Keywords: W boson; Hadroproduction; Electroweak corrections
1. Introduction
The hadroproduction of single W bosons via the DrellYan process in p p collisions at the
CERN Sp pS
led to the discovery of this particle in 1983 [1]. Nowadays, this process serves as a
standard candle to calibrate and monitor the luminosity of hadronic collisions, since its cross section is rather sizeable and W bosons are straightforward to identify experimentally thanks to their
simple and distinct decay signature. The quality of the luminosity determination is thus limited
by the precision to which this cross section is predicted theoretically. It is, therefore, mandatory to
E-mail address: kniehl@desy.de (B.A. Kniehl).

doi:10.1016/j.nuclphysb.2007.09.013
W. Hollik et al. / Nuclear Physics B 790 (2008) 138159
139
calculate higher-order radiative corrections. At present, they are known at next-to-leading order
(NLO) [2] and next-to-next-to-leading order (NNLO) [3] in quantum chromodynamics (QCD) as
well as at NLO [4] in the electroweak sector of the Standard Model (SM) of elementary particle
physics.
In order for the W boson to acquire finite transverse momentum (pT ), it must be produced in
association with one or more other particles. To lowest order (LO) in QCD, the additional particle
is a gluon (g), quark (q), or antiquark (q),
materialising as a hadron jet (j ). The corresponding
partonic subprocesses are of O(s ). Their cross sections are presently known at NLO [5] and
NNLO [6] in QCD, i.e., at O(s2 ) and O(s3 ), respectively. Very recently, also the one-loop
electroweak corrections, at O( 2 s ), were considered [7]. This is also the topic of the present
paper. However, as explained below, we actually study somewhat different cross section observables and arrange for our results to be manifestly infrared (IR) safe by avoiding kinematic cuts
that destroy the inclusiveness of massless quanta. In fact, the situation is complicated by the
circumstance that both photons and gluons can appear as bremsstrahlung.
The observable we thus wish to investigate is the differential cross section for the inclusive
hadroproduction of single W bosons with finite pT at O( 2 s ). Specifically, we concentrate on
the pT distribution and the integrated cross section as a function of a minimum-pT cut, leaving
the distributions in other observables, such as rapidity, for future work. At LO, the system X
recoiling against the W boson is purely hadronic, while at O( 2 s ) it can include a photon ( ).
We thus also have to consider W + production, whose LO partonic cross sections are of O( 2 ),
because its NLO QCD correction contributes at the very order we are aiming at. In fact, the real
radiative corrections to W + j and W + production receive contributions from a common set
of 2 3 partonic subprocesses. In the former (latter) case, IR singularities from the radiation of
soft photons (gluons) cancel against similar contributions from virtual photons (gluons) by the
BlochNordsieck theorem [8]. Similarly, the IR singularities from collinear final-state radiation
(FSR) cancel against similar contributions from the virtual corrections by the KinoshitaLee
Nauenberg theorem [9]. The residual IR singularities from collinear initial-state radiation (ISR)
are factorised and absorbed at O() (O(s )) into the parton distribution functions (PDFs). This
procedure leads to manifestly IR-safe cross section observables and translates into unique and
simple event selection criteria on the experimental side.
By contrast, the notion of electroweak NLO corrections to W + j production comprises a
conceptual problem. In fact, in the treatment of final-state collinear singularities caused by the
parallel emission of a photon from an outgoing (anti)quark line, one is led to introduce a cut
in an appropriate separation variable. Within the collinear phase space region thus defined, the
(anti)quarkphoton system is effectively treated as one particle whose momentum is identified
with that of j and thus subject to an acceptance cut in transverse momentum, pT (j ) > pTmin (j ),
to ensure the experimental observation of j . This includes phase space configurations where the
photon essentially carries all the momentum, while the (anti)quark can, in principle, be arbitrarily soft. This will not generate any soft IR singularities. However, since (anti)quark and gluon
jets can, in general, not be distinguished experimentally on an event-by-event basis, the same
recombination procedure needs to be applied to a gluonphoton system in the final state as well.
This time, a soft gluon will inevitably produce an IR singularity, which can only be canceled by
the NLO QCD corrections to W + production, so that one falls back to the symmetric procedure outlined in the preceding paragraph. Formally, this soft-gluon singularity can be avoided by
applying the pTmin (j ) cut just to the transverse momentum of the gluon, even if it is accompanied
by a collinear photon. However, such a prescription is purely academic and quite unsuitable for
experimental implementation because (anti)quark and gluon jets are treated on different footings.
140
By crossing external lines, the LO partonic subprocesses of W + hadroproduction can be

converted to those of W + j photoproduction with one incoming photon participating directly in
the hard scattering (direct photoproduction). The emission of photons off the proton can happen
either elastically or inelastically, i.e., the proton stays intact or is destroyed, respectively. In both
cases, an appropriate PDF can be evaluated in the WeizsckerWilliams approximation [1012].
Since they are of O(), these direct photoproduction contributions are of O( 3 ). Incoming photons can participate in the hard scattering also through their quark and gluon content, leading
to resolved photoproduction. The contributions from direct and resolved photoproductions are
formally of the same order in the perturbative expansion. This may be understood by observing
that the PDFs of the photon have a leading behaviour proportional to ln(M 2 /2QCD ) /s ,
where M is the factorisation scale and QCD is the asymptotic scale parameter of QCD. Although photoproduction contributions are parametrically suppressed by a factor of /s relative
to the O( 2 s ) corrections discussed above, we shall include them in our analysis because they
turn out to be quite sizeable in an extensive region of phase space.
This paper is organised as follows. In Section 2, we list the partonic cross sections at LO
and explain how to evaluate the hadronic cross section from them. In Section 3, we discuss
in detail the structure of the NLO corrections. In Section
4, we present our numerical results
for p p collisions with

centre-of-mass (c.m.) energy S = 1.96 TeV at the Fermilab Tevatron
and pp collisions with S = 14 TeV at the CERN Large Hadron Collider (LHC).
2. Conventions and LO results
We consider the hadronic process
A(pA ) + B(pB ) W (p) + X,
(2.1)
where the four-momentum assignments are indicated in parentheses. We work in the collinear
parton model of QCD [13] with nf = 5 massless quark flavours q = u, d, s, c, b, neglect the
masses of the incoming hadrons, A and B, and impose the acceptance cut pT > pTcut on the
transverse momentum pT of the W boson. (We assign masses to the partons , g, q, q only to
regulate soft and collinear IR singularities in intermediate steps of our calculation.)
Specifically, denoting u1 = u, u2 = c, d1 = d, d2 = s, and d3 = b, the relevant partonic subprocesses include
ui + dj W + + g,
+
ui + g W + dj ,
dj + g W + + u i ,
(2.2)
(2.3)
(2.4)
at O(s ),
ui + dj W + + ,
+
ui + W + dj ,
dj + W + + u i ,
at
O( 2 ),
(2.5)
(2.6)
(2.7)
and
ui + dj W + + g + ,
+
ui + g W + dj + ,
dj + g W + + u i + ,
(2.8)
(2.9)
(2.10)
141
at O( 2 s ). The partonic subprocesses involving a W boson emerge through charge conjugation. Processes (2.2)(2.5) must be treated also at one loop, O( 2 s ). Processes (2.6) and (2.7)
contribute to direct photoproduction and processes (2.2)(2.4) to resolved photoproduction.
Since photon emission off protons happens at O(), it is sufficient to deal with photoproduction at tree level. In summary, we calculate the cross section of process (2.1) at NLO as the
sum
Wj
ABW X = 0
Wj
where 0
+ 0
Wj
Wj
+ O() + O(s ) + 0
+ Wj ,
(2.11)
W
Wj
and O() are due to processes (2.2)(2.4) at tree level and one loop, 0
Wj
and O(s )
are due to process (2.5) at tree level and one loop, 0

is due to processes (2.8)(2.10) at tree
Wj
level, and is due to processes (2.6) and (2.7) via direct photoproduction and due to processes
(2.2)(2.4) via resolved photoproduction, both at tree-level.
The minimum-pT cut is necessary to stay away from the regions of phase space that are
sensitive to the collinear IR singularities due to the q /g +q , q /g + q , /g q + q ,
and /g q + q splittings, which are present already at LO. Here, an asterisk marks a virtual
parton. The cross section ABW X of the hadronic process (2.1) is related to the cross sections
abW c(d) of the partonic subprocesses,

a(pa ) + b(pb ) W (p) + c(pc ) +d(pd ) ,
(2.12)
where a, b, c, d = , g, q, q and pa = xa pA , pb = xb pB with scaling parameters xa , xb , as the
incoherent sum
ABW X
S, pT > pTcut
1

d LAB
abW c(d) s, pT > pTcut ,
ab ( )
(2.13)
a,b,c(d) 0
where S = (pA + pB )2 and s = (pa + pb )2 = S are the hadronic and partonic c.m. energies,
respectively, = xa xb , and
1
LAB
ab ( ) =

dxa
fa/A xa , M 2 fb/B
, M2
xa
xa
(2.14)
is the parton luminosity defined in terms of the PDFs fa/A (xa , M 2 ), fb/B (xb , M 2 ). Here, M de2 , we have
notes the factorisation mass scale. Introducing the short-hand notation w = MW
2
cut
pT + w + (pTcut )2
0 =
(2.15)
.
S
In order to obtain abW c(d) , we have to evaluate the transition matrix elements T abW c(d)
of processes (2.12), square them, average them over the initial-state spins and colours, and sum
them over the final-state ones, which leads to |T abW c(d) |2 . To the order of our calculation,
T abW cd is calculated at tree level, while T abW c may receive also one-loop contributions,
T abW c = T0abW c + T1abW c , so that

T abW c 2 = T abW c 2 + 2 Re T abW c T abW c .
0
0
1
(2.16)
142
Then we have to integrate over the partonic phase spaces imposing the minimum-pT cut. In
the following two subsections, we describe how this can be conveniently done for the two- and
three-particle final states, respectively.
Since we are dealing with charged-current interactions of quarks, ui and dj , the Cabibbo
KobayashiMaskawa quark mixing matrix Vij appears. At tree level, the cross sections of
processes (2.2)(2.10) contain the overall factor |Vij |2 , and a part of the one-loop corrections
is proportional to Vij Vij Vi j Vi j , where ui and dj are virtual quarks. Since we neglect all
down-quark masses, we can
sum over the indices of the virtual and outgoing down quarks to
trigger the unitarity relation 3j =1 Vij Vi j = ii . In the case of incoming down quarks, we can
absorb the residual appearances of |Vij |2 into a redefinition of their PDFs, as [7]
3

2
fdi /A x, M =
|Vij |2 fdj /A x, M 2 ,
(2.17)
j =1
and similarly for down antiquarks. Therefore, it is sufficient to calculate the partonic cross sections for the flavour-diagonal case, with Vij = ij .
2.1. Two-particle final state
If parton d is absent in process (2.12), we supplement s by two more Mandelstam variables,
t = (pa p)2 and u = (pb p)2 . Four-momentum conservation implies that s + t + u = w, and
we have pT2 = tu/s. The partonic cross section entering Eq. (2.13) is evaluated as

abW c

cut
s, pT > pT
max
p
T
dpT
pTcut
d abW c
,
dpT
(2.18)
where pTmax = (s w)/(2 s ) and

pT
d abW c
T abW c 2 + (t u).

=
dpT
8s (s w)2 4spT2
(2.19)
For the readers convenience, we list the differential cross sections of processes (2.2)(2.7),
in the conventional form
d abW c
1 abW c 2
(2.20)
T
,
=
dt
16s 2
at LO. The Feynman diagrams contributing to processes (2.2) and (2.5) are displayed in Figs. 1(a)
and 1(b), respectively. We have
d udW
dt
+g
d udW
dt
2s s 2 + w 2 2tu
,
9sw2
s 2 tu
2 udW

+g
3t
=
,
1+
12s
s w
dt
(2.21)
where sw = sin w is the sine of the weak-mixing angle. Since the W -boson mass sets the
renormalisation scale of the couplings, it is natural to adopt the definition of Sommerfelds fine-
143
(a)
(b)
Fig. 1. Tree-level diagrams of (a) process (2.2) and (b) process (2.5). The tree-level diagrams of processes (2.3), (2.4),
(2.6), and (2.7) emerge through crossing.
structure constant in terms of Fermis constant GF ,
2
GF sw2 w.
=
(2.22)
The implementation of this renormalisation scheme at one loop is explained in Section 3.1. The
cross sections of processes (2.3), (2.4), (2.6), and (2.7) may be obtained from Eq. (2.21) by
exploiting crossing symmetries, as

+
+
d ugW d
d udW g
3
s2
,
= s2
dt
8
dt
su

+
+
d dgW u
d ugW d
,
= s2
s2
dt
dt
st

+
+
d u W d
d udW
,
= 3 s 2
s2
dt
dt
su

+
+
d W u
u W d
2 d
2 d
.
= s
s
(2.23)
dt
dt
st
2.1.1. Three-particle final states
If parton d is present in process (2.12), then the partonic cross section entering Eq. (2.13) may
be obtained through a four-fold phase-space integration along the lines of Ref. [14]. We work in
the partonic c.m. frame and choose our coordinate system so that pa points along the z direction
and pd lies in the xy plane. We denote the polar angle of pd by and the azimuthal angle of pc
by . As the first three independent variables, we select pd0 , , and , which take the values
0 < pd0 <
sw
,
2 s
0 < < ,
0 < < 2.
(2.24)
In the case of process (2.8), which contains two massless gauge bosons in the final state, it is
convenient to take the fourth variable to be pc0 , with values

1
w
1
w
(2.25)
s 2pd0 < pc0 <
s
.
2
2
s
s 2pd0
144
We then have

0
0
dp dp d cos d
d4 udW
c
+ g
pT >pTcut

1 udW
+ g 2
pT pTcut .
T
4
8(2)
(2.26)
On the other hand, in the case of processes (2.9) and (2.10), which only contain one massless
gauge boson in the final state, it is more useful to choose the fourth variable to be the angle
enclosed between pc and pd , with values
0 < < .
(2.27)
We then have

0
dpd d cos d d pT >pTcut

pd0 [ s( s 2pd0 ) w]
T ugW + d 2 pT p cut ,
=
T
2
0
4
2
16(2) [ s 2pd sin (/2)]
d4 ugW
+ d
(2.28)
and similarly for process (2.10). In order to implement the minimum-pT cut, pT needs to be
expressed in terms of the integration variables, which is conveniently done with the help of
Eqs. (5.40) and (5.42) of Ref. [14] and starting from

2
2
pT = pc1 + pd1 + pc2 + pd2 .
(2.29)
3. NLO results
Wj
Wj
of
We now describe the calculation of the NLO contributions O() , O(s ) , and 0
Eq. (2.11) in some detail.
We employ the following tools. We generate the relevant Feynman diagrams using the symbolic program package FeynArts [15], carry out the spin and colour sums using the program
package FormCalc [16], and perform the PassarinoVeltman reduction of the tensor one-loop
integrals [17] using the program package FeynCalc [18]. Subsequently, we implement the analytical results in a Fortran program. We evaluate the standard scalar one-loop integrals contained
in the purely weak corrections using the program package LoopTools [19], which incorporates
the program library FF [20]. The numerical integrations are performed using the program package Cuba [21], which provides several different integration routines and is, therefore, also well
suited for cross checks.
3.1. Virtual electroweak corrections to W + j production
The virtual electroweak corrections of O() to processes (2.2)(2.4) arise from self-energy,
triangle, box, and counterterm diagrams. They are shown for process (2.2) in Figs. 25, respectively.
Wj
Evaluating the transition matrix element TO() from these loop diagrams, we encounter both
ultraviolet (UV) and IR singularities, which need to be regularised and removed. As usual, we use
dimensional regularisation, with D = 4 2 spacetime dimensions and t Hooft mass scale ,
to extract the UV singularities as single poles in . These are removed by renormalising the
Wj
parameters and wave functions of the LO transition matrix element T0 , which leads to the
Fig. 2. O() self-energy diagrams of process (2.2).
Fig. 3. O() triangle diagrams of process (2.2).
145
146
Fig. 4. O() box diagrams of process (2.2).
Fig. 5. O() counterterm diagrams of process (2.2).
counterterm contribution (see Fig. 5),

Wj
Wj Wj
CT .
TCT = T0
(3.1)
Wj
Owing to the renormalisability of the SM, the UV singularities in TO() cancel, and the physical
limit 0 can be reached smoothly.
The electroweak on-shell renormalisation scheme uses the fine-structure constant defined
in the Thomson limit and the physical particle masses as basic parameters. In order to avoid
the appearance of large logarithms induced by the running of to the electroweak scale MW in
Wj
TO() , it is useful to replace by GF in the set of basic parameters, by substituting
1
GF =
,
2
2sw w 1 r
(3.2)
where r [22] contains those radiative corrections to the muon lifetime which the SM introduces
on top of those derived in the QED-improved Fermi model. In the electroweak on-shell scheme
thus modified, we have
Wj
CT = Ze

sw 1 L
+ Zuu + ZdLd + ZW r ,
sw
2
(3.3)
147
where the renormalisation constants read [23]

T
2 )
T (M 2 )
2
W W (MW
ZZ
1 cw
sw
Z
= 2 Re
,
2
sw
2 sw
MW
MZ2
T (q 2 )
T (0)
1 AA
sw AZ

Ze =
,

2
2
cw MZ2
q
q 2 =0
T (q 2 )
W
W

ZW = Re
2 2,
q 2
q =M
W
ZqLq = Re qLq

2
2
L 2
2
2
R
S
mq mq 2 Re q q q + q q q + 2q q q
.
q
q 2 =m2q
(3.4)
T , T , T , and T are the transverse parts of the respective electroweak gaugeHere, W

W
ZZ
AA
AZ
boson self-energies and mixing amplitudes, qLq , qRq , and qSq are the left-handed, right2 = 1 s2 .
handed, and scalar parts of the quark self-energy, and cw
w
The IR singularities can be of soft or collinear type. The loop diagrams involving virtual
photons interchanged between external lines are plagued by soft IR singularities. Owing to the
BlochNordsieck theorem [8], they cancel against similar singularities arising from the real emission of soft photons, to be discussed in Section 3.3. The loop diagrams involving external quark
or antiquark lines that split into virtual photons and quarks generate collinear IR singularities.
Such singularities also arise from the real emission of collinear photons off external quark or
antiquark lines, as will be explained in Section 3.3. Thanks to the KinoshitaLeeNauenberg
theorem [9], collinear IR singularities from FSR are completely canceled in the sum of real
and virtual corrections provided that the final state is treated inclusively enough. On the other
hand, collinear IR singularities from ISR survive and have to be absorbed into the quark and
antiquark PDFs. For consistency, the splitting functions in the evolution equations of the PDFs
then need to be complemented by their O() terms. IR singularities also arise from the wavefunction renormalisations in Eq. (3.4). We choose to regularise the IR singularities by assigning
infinitesimal masses, , mu , and md , to the photon, the light up-type quarks, and the down-type
quarks, respectively. This is convenient because the standard scalar one-loop integrals C0 and D0
that emerge after the tensor reduction [17] are well established for this regularisation prescription [24]. Although the purely weak loop corrections are altogether devoid of IR singularities,
terms logarithmic in mu and md are generated by the tensor reduction. However, these artificial
IR singularities cancel among themselves.
We emphasise that, in the treatment of both the virtual and real corrections, terms depending
on , mu , and md are extracted analytically and their cancellation is established manifestly, so
that the expressions used for the numerical analysis do not contain these IR regulators.
3.2. Virtual QCD corrections to W + production

The virtual QCD corrections of O(s ) to process (2.5) arise from the self-energy, triangle,
and box diagrams shown in Fig. 6 and the counterterm contribution,
W
W W
CT .
TCT = T0
(3.5)
148
Fig. 6. O(s ) self-energy, triangle, and box diagrams of process (2.5).
The latter only receives contributions from the gluon-induced wave-function renormalisation of
the external quark lines,
W
1 g
g
Zuu + Zd d ,
2
g
Zq q

g,V
= q q m2q 2m2q
CT =
where
(3.6)

g,V 2
g,S 2
q + q q q
.
q 2 q q
q 2 =m2q
(3.7)
g,V
Because parity is conserved within QCD, the quark self-energy has just one vector part q q =
g,L
g,R
q q = q q . Up to terms that vanish in the limit mq 0, we have

g
Zq q =

m2q
m2q
s CF 1
E + ln(4) ln 2 2 ln 2 + 4 + O( ),
4
(3.8)
where CF = (Nc2 1)/(2Nc ) = 4/3 for Nc = 3 quark colours, E is the EulerMascheroni

constant, and now represents an infinitesimal gluon mass.
3.3. Real corrections due to W + j + production
The tree-level diagrams for process (2.8) are shown in Fig. 7. They contribute at the same
time to the electromagnetic bremsstrahlung in process (2.2) and to the QCD bremsstrahlung in
process (2.5), which complicates the treatment of the electroweak corrections to W +j associated
production, as explained in the Introduction. The diagrams contributing to the electromagnetic
bremsstrahlung in processes (2.3) and (2.4) emerge from Fig. 7 by crossing the gluon with the u
and d quarks, respectively.
149
Fig. 7. Tree-level diagrams of process (2.8). The tree-level diagrams of processes (2.9) and (2.10) emerge through
crossing.
When the cross sections of processes (2.8)(2.10) are integrated over their three-particle phase
spaces, one encounters IR singularities of both soft and collinear types. The former stem from
the emission of soft photons and gluons and cancel against similar contributions from the virtual
corrections owing to the BlockNordsieck theorem [8], as explained in Section 3.1. The latter
arise when a massless gauge boson is collinearly emitted from an external massless fermion
line or when a massless gauge boson splits into two collinear massless fermions. Specifically, in
process (2.8), the photon or the gluon can be emitted collinearly from the incoming ui and dj
quarks; in process (2.9), the photon can be emitted collinearly from the incoming ui quark or the
outgoing dj quark, and the gluon can split into a collinear dj dj quark pair; and in process (2.10),
the photon can be emitted collinearly from the incoming dj quark or the outgoing u i quark,
and the gluon can split into a collinear ui u i quark pair. As already mentioned in Section 3.1,
the collinear IR singularities from FSR are canceled by the virtual corrections according to the
KinoshitaLeeNauenberg theorem [9] if the considered process is treated inclusively enough.
By contrast, those from ISR survive and have to be absorbed into the PDFs.
Due to the minimum-pT cut, the photon and the gluon cannot be soft simultaneously because
one of them has to balance the transverse momentum of the W boson. By the same token, there
can only be one collinear situation at a time. However, soft and collinear singularities do overlap,
and care needs to be exercised to avoid double counting.
For consistency, also the IR singularities in the real corrections need to be regularised by
the photon and gluon mass and the light-quark masses mu and md introduced in Sections 3.1
and 3.2. As already mentioned in Section 3.1, their cancellation is achieved analytically, so that
the expressions underlying the numerical analysis are free of them.
As in Refs. [4,25,26], we employ the method of phase space slicing [27] to separate the soft
and collinear regions of the phase space from the one where the momenta are hard and noncollinear, so that the partonic cross section can be written as
Wj
Wj
Wj
d Wj = soft + d coll + d hard .
(3.9)
For definiteness, let us assume that parton d in process (2.12) is the soft or collinearly emitted
one and that partons a and c are the ones emitting ISR and FSR, respectively. In the notation
introduced
in Section 2.1.1, the soft regions of phase space are then defined by < pd0 < E
(s w)/(2 s ), the collinear ones for ISR and FSR by < and < , respec-
150
tively, and pd0 > E, and the hard and non-collinear one by the rest. In Sections 3.3.1 and 3.3.2,
Wj
Wj
we explain how to evaluate d soft and d coll analytically using appropriate approximations. On
Wj
the other hand, d hard can straightforwardly
with high precision [28].
be evaluated numerically
Since E is to be measured in units of s/2, we define s = 2E/ s. The demarcation parameters s , , and must be chosen judiciously. If they are too small, then the numerical
Wj
phase-space integration performed for d hard becomes unstable; if they are too large, the approxWj
Wj
imations adopted for d soft and d coll become crude. In practice, one varies s , , and
to find the respective stability regions. For the problem considered here, this is easily achieved.
3.3.1. Soft singularities
In the soft phase space regions, T abW cd factorises into T abW c times an eikonal factor that
depends on pd . Squaring T abW cd , performing the spin and colour sums, and integrating over
pd with the constraint < pd0 < E, one has [23,29]
abW cd
abW cd
d soft
(, E) = soft
(, E) d abW c .
(3.10)
In the case of soft electromagnetic and QCD bremsstrahlung in process (2.8), we then obtain
+ g
udW
(, E)

2
Q uu + Q2d dd + W W + 2Qu Qd ud + 2Qu uW + 2Qd dW ,
=
2 u
+ g
s CF
udW
soft
(3.11)
(, E) =
(uu + dd + 2ud ),
2
where Qu = 2/3 and Qd = 1/3 are the fractional electric charges of the u and d quarks,
respectively, and
soft
4(E)2
m2u
+
ln
,
s
2
= uu |mu md ,
uu = ln
dd
ud

2
2
1 4(E)2 m2u m2d
1
2
2 mu
2 md
= ln
ln
+
ln
+
ln
+
,
2
4
s
s
3
2
s2
4(E)2 s + w w
+
ln ,
sw s
2

2
1 4(E)2
wm2u
1
2 mu
2 w
ln
+
ln
+ ln
uW = ln
2
s
s
2
(w t)2 4

2
t
u
+ Li2
+ Li2
+
,
wt
wt
6
dW = uW |tu,mu md .
(3.12)
1
Here, Li2 (x) = 0 dt ln(1 tx)/t is the dilogarithm, and terms that vanish for mu = md = 0
have been omitted.
Furthermore, we find the soft-photon correction factor for process (2.9) to be
W W = ln
ugW + d
(, E)

2
=
Qu uu + Q2d dd + W W + 2Qu Qd ud + 2Qu uW + 2Qd dW ,
2
soft
(3.13)
151
in which two terms of Eq. (3.12) are modified to be

sm2d
1 4(E)2 m2u m2d
1 2 m2u
2
t
2
ln
+
ln
+
ln
+
,
+
Li
ud = ln
2
2
4
s
u
3
2
u2
(s w)2

sm2d
1 4(E)2
wm2u
1 2w
2
w
2
dW = ln
ln
ln
+
ln
Li
1
2
2
s
s
6
2
(s w)2 4
(s w)2
(3.14)
Finally, the soft-photon correction factor for process (2.10) emerges from the one of process (2.9) through the simple replacement
+ u
dgW
soft
ugW + d
(, E) = soft

(, E) m
u md
(3.15)
3.3.2. Collinear singularities

As explained in Section 3.3, collinear singularities arise from three sources: (1) the emission
of a photon or gluon from an incoming quark or antiquark; (2) the splitting of an incoming
gluon into a quarkantiquark pair; and (3) the emission of a photon from an outgoing quark or
Wj
antiquark. The resulting contributions to d coll all factorise into the respective LO cross sections
without radiation and appropriate collinear radiator functions [14,30]. In the case of ISR, this also
involves a convolution with respect to the fraction z of four-momentum that the emitting parton
passes on to the one that enters the hard interaction.
Let parton a in process (2.12) be the emitting quark q and parton d the emitted photon or
gluon. Then we have [14,30]
qbW c{ ,g}
(mq , ) =
d coll
{Q2q , s CF }
1
s
qbW c
dz RqIS (mq , , z) d 0
2
z0
pq zpq
,
(3.16)
where s is introduced to exclude a slice of phase space that is both soft and collinear and is
Wj
already included in soft , z0 = 0 S/s, with 0 being defined in Eq. (2.15), and

s()2
2z
RqIS (mq , , z) = Pqq (z) ln
,
4m2q
1 + z2
(3.17)
with
Pqq (z) =
1 + z2
1z
(3.18)
being the LO q q splitting function [31]. This result readily carries over to the case when
parton a is an antiquark q.
Note that the c.m. frame is boosted along the beam axis by the
collinear emission of the photon or gluon.
Now, let parton a in process (2.12) be a gluon that splits into a q q pair, with q being outgoing
and q entering the residual hard scattering. Then we have [26]
gbW cq
(mq , ) =
d coll
s T F
2
1
qbW
c
,
pq zpq
dz RgIS (mq , , z) d 0
z0
(3.19)
152
where TF = 1/2 and

RgIS (mq , , z) = Pgq (z) ln
s(1 z)2 ()2

+ 2z(1 z),
4m2q
(3.20)
with
Pgq (z) = z2 + (1 z)2
(3.21)
being the LO g q splitting function. This result readily carries over to the case when parton d
is an antiquark q.
Finally, let parton c in process (2.12) be the emitting quark q and parton d the emitted photon.
Then we have [14,30]
abW q
(mq , ) =
d coll
Q2q
1
s
abW q
dz RqFS (mq , , z) d 0
(3.22)
where s = ss /(s w) is again to avoid double counting of phase space regions that are both
soft and collinear, and

(s w)2 ()2
2z
FS
Rq (mq , , z) = Pqq (z) ln
(3.23)
+ 2 ln z
,
4sm2q
1 + z2
with Pqq given in Eq. (3.18). This result readily carries over to the case when parton c is an
antiquark q.
The integral in Eq. (3.22) is not a convolution and can easily be carried out, yielding
1
s
dz RqFS (mq , , z) =
0

3
2
(s w)2 ()2
9
2 ln s
+ 2 ln s + .
ln
2
2
3
2
4smq
(3.24)
Wj
In order to obtain d coll
for one of the processes (2.8)(2.10), all possible collinear emissions

must be taken into account one by one.
While the collinear IR singularities from FSR cancel upon combination with the virtual corrections by the KinoshitaLeeNauenberg theorem [9], those from ISR survive. Since their form
is universal, they can be factorised and absorbed into the PDFs [32]. Adopting the modified
minimal-subtraction (MS) factorisation scheme both for the collinear singularities of relative
orders O() and O(s ), this is achieved by modifying the PDF of quark q inside hadron A as

fq/A x, M 2 fq/A x, M 2

Q2q + s CF

3
M2
ln s +
ln 2 ln2 s ln s + 1
= fq/A x, M 2 1
4
mq
1
s
1

Q2q + s CF
dz
M2
x
1
fq/A
, M2
Pqq (z) ln
z
z
2
(1 z)2 m2q

dz
M2
x
2 s TF
fg/A
,M
Pgq (z) ln 2 ,
z
z
2
mq
(3.25)
where M is the factorisation mass scale, which separates the perturbative and non-perturbative
parts of the hadronic cross section.
153
4. Numerical results
We are now in a position to present our numerical results. We start by specifying our
choice of input. We adopt the values GF = 1.6637 105 GeV2 , MW = 80.403 GeV,
MZ = 91.1876 GeV, and mt = 174.2 GeV recently quoted by the Particle Data Group [33],
take the other nf = 5 quarks to be massless partons, and assume MH = 120 GeV, which is
presently compatible with the direct search limits and the bounds from the electroweak precision
tests [33]. We take the absolute values of the CKM matrix elements to be [33]
|Vud | = 0.9377,
|Vcs | = 0.957,
|Vus | = 0.2257,
|Vcb | = 41.6 10
|Vcd | = 0.230,
3
|Vub | = 4.31 103 .
(4.1)
(n )
Since we are working at LO in QCD, we employ the one-loop formula for s f (). We use the
LO proton PDF set CTEQ6L1 by the Coordinated TheoreticalExperimental Project on QCD
(5)
(CTEQ) [34], with QCD = 165 MeV. In the case of photoproduction, we add the photon spectra
for elastic [10] and inelastic [11,12] scattering elaborated in the WeizsckerWilliams approximation. In the latter case, we use the more recent set by Martin, Roberts, Stirling, and Thorne
(MRSTQED04) [12] as our default, and the set by Glck, Stratmann, and Vogelsang (GSV) [11]
to assess the theoretical uncertainty from this source.We choose the renormalisation and factori-
cut
cut 2
2
sation scales to be = M = mcut
T , where mT = (pT ) + MW is the minimum transverse
mass of the produced W boson and is introduced to estimate the theoretical uncertainty. Unless
otherwise stated, we use the default value = 1.
We consider the total cross sections of p p W + X at the Tevatron (run II) with S =
1.96 TeV and pp W + X at the LHC with S = 14 TeV as functions of pTcut . By numerically
differentiating the latter with respect to pTcut , we also obtain the corresponding pT distributions
as d/dpT = d (pTcut )/dpTcut |pcut =pT . Owing to the baryon symmetry of the initial state, the
T
results for W + and W bosons are identical at the Tevatron, and it is sufficient to study one
of them. By contrast, W + -boson production is favoured at the LHC because the proton most
frequently interacts via a u quark. Therefore, it is necessary to study the production of W + and
W bosons separately at the LHC. We compare the contributions of four different orders: (1) the
LO contribution of O(s ) from processes (2.2)(2.4), where the system X accompanying the
W boson contains a hadron jet; (2) the LO contribution of O( 2 ) from process (2.5), where X
contains a prompt photon; (3) the NLO contribution of O( 2 s ) comprising processes (2.2)
(2.5) at one loop as well as processes (2.8)(2.10) at tree level, where X contains a hadron
jet, a prompt photon, or both; and (4) the LO contributions of O( 3 ) from processes (2.6) and
(2.7) via direct photoproduction and from processes (2.2)(2.4) via resolved photoproduction,
where X contains a hadron jet and, in the case of elastic photoproduction, also the scattered
proton or antiproton. Since we consider inclusive one-particle production, we do not use any
information on the composition of X, i.e., we include all possibilities. In the following, we regard
the sum of contributions (1) and (2) as LO and sum of contributions (1)(4) as NLO unless the
perturbative orders are explicitly specified in terms of coupling constants. We thus define the
correction factor K to be the NLO to LO ratio with this understanding.
Let us now discuss the numerical results and their phenomenological implications in detail.
Specifically, Figs. 8, 9(a), and 10(a) refer to the Tevatron, while Figs. 9(b), 10(b), 11, 12, and 13
refer to the LHC. In Fig. 8(a) the NLO result for the total cross section as a function of pTcut is
compared with the LO contributions of O(s ) and O( 2 ) as well as with the photoproduction
154
(a)
(b)
Fig. 8. (a) Total cross section as a function of pTcut and (b) pT distribution of p p W + + X for S = 1.96 TeV
(Tevatron run II). The NLO results are compared with those of orders O(s ), O( 2 ), and O( 3 ) via photoproduction.
(a)
(b)
Fig. 9. NLO corrections

(K 1), with and without the photoproduction contributions, to the
total cross sections of
(a) pp W + + X for S = 1.96 TeV (Tevatron run II) and of (b) pp W + X for S = 14 TeV (LHC) as
functions pTcut . For comparison, also the contributions due to elastic and inelastic photoproductions normalised to the LO
results are shown. In the latter case, the evaluation is also performed with the GSV PDFs.
contribution of O( 3 ). The O(s ) and O( 2 ) results exhibit very similar line shapes, but the
normalisation of the latter is suppressed by a factor of about 500. This may be qualitatively
understood from the partonic cross section formulae in Eq. (2.21) and by noticing that the O(s )
contributions from the Compton-like processes (2.3) and (2.4), which have no counterparts in
O( 2 ), are significantly enhanced by the gluon PDF. As a consequence, the LO result is almost
entirely exhausted by the O(s ) contribution.
(a)
155
(b)
Fig. 10. Total cross sections of (a) p p W + + X for S = 1.96 TeV and pTcut = 20 GeV (Tevatron run II) and of
(b) pp W + X for S = 14 TeV and pTcut = 200 GeV (LHC) as functions of normalised to their default values
for = 1. The NLO results are compared with those of orders O(s ), O( 2 ), and O( 3 ) via photoproduction.
(a)
(b)
Fig. 11. (a) Total cross section as a function of pTcut and (b) pT distribution of pp W + + X for S = 14 TeV (LHC).
The NLO results are compared with those of orders O(s ), O( 2 ), and O( 3 ) via photoproduction.
The inclusion of the NLO correction leads to a moderate reduction in cross section, which
increases in magnitude with pTcut , reaching about 4% for pTcut = 200 GeV, as may be seen from
Fig. 9(a), where the K factor is depicted.
In Fig. 8(a), also the photoproduction contribution is shown. As explained above, we have to
distinguish between elastic and inelastic scatterings off the proton on the one hand, and between
direct and resolved photons on the other hand, so that, altogether, we have four different contributions, which all formally contribute at O( 3 ). The resolved-photon contributions turn out to
be small against the direct-photon ones and are, therefore, not included in Fig. 9(a). As for the
combined direct-photoproduction contribution, we observe from Fig. 8(a), that, except for small
values of pTcut , it overshoots the O( 2 ) contribution, although it is formally suppressed by one
power of ! Detailed inspection reveals that this unexpected enhancement can be traced to the
156
(a)
(b)
Fig. 12. (a) Total cross section as a function of pTcut and (b) pT distribution of pp W + X for S = 14 TeV (LHC).
The NLO results are compared with those of orders O(s ), O( 2 ), and O( 3 ) via photoproduction.
Fig. 13. Ratios of the respective results for W + and W bosons shown in Figs. 11(a) and 12(a).
direct-photoproduction diagram involving the triple-gauge-boson coupling

and the space-like W
boson exchange, which significantly contributes at large values of s. In fact, for a fixed value
of pTcut , the total cross sections of processes (2.6) and (2.7) have an asymptotic large-s behaviour
2
proportional to 1/(mcut
T ) , while those of processes (2.2)(2.5) behave as ln s/s. Consequently,
photoproduction appreciably contributes to the K factor, as is apparent from Fig. 9(a), which also
shows the photoproduction to LO ratios for elastic and inelastic scatterings. The freedom in the
choice of the inelastic photon content of the proton is likely to be the largest source of theoretical
uncertainty in the photoproduction cross section. In order to get an idea of this uncertainty, we
display in Fig. 9(a) also the inelastic-photoproduction to LO ratio evaluated with the GSV photon
spectrum for inelastic scattering. The result is roughly a factor of two smaller than our default
prediction based on the MRSTQED04 spectrum.
In Fig. 10(a), we examine the theoretical uncertainties in the O(s ), O( 2 ), NLO, and photoproduction results due to the freedom in setting the renormalisation and factorisation scales by
exhibiting their dependencies relative to their default values at = 1. The dependencies of
157
the O( 2 ) and direct-photoproduction results stem solely from the factorisation scale M and are
rather feeble, while those of the O(s ) and resolved-photoproduction results are also linked to
the renormalisation scale of s () and are more pronounced, but still not dramatic. The scale
variation of the LO result amounts to less than 15% for 1/2 < < 2. It is only slightly reduced
by the inclusion of the NLO correction. This is expected because the NLO result is still linear in
s (), so that the dependence of s () is not compensated yet.
In Fig. 8(b), the analysis of Fig. 8(a) is repeated for the pT distribution. We observe that the
line shapes and relative normalisations of the various distributions are very similar to those in
Fig. 8(a) and the same comments apply.
Turning to the LHC, we can essentially repeat the above discussion for the Tevatron, except
that we have to take into account the difference between W + and W boson production. Thus,
Fig. 8 has two LHC counterparts, Figs. 11 and 12, for the W + and W bosons, respectively. To
illustrate this difference more explicitly, we show in Fig. 13 the W + to W ratios of the respective results from Figs. 11 and 12. For simplicity, Figs. 9(b) and 10(b), the LHC counterparts of
Figs. 9(a) and 10(a), refer to the averages of the results for W + and W bosons. In the following, we only focus on those features which are specific for the LHC. From Figs. 11 and 12, we
observe that the gaps between the O(s ) and O( 2 ) results are increased by about a factor of
two, to reach three orders of magnitude. This is mainly because the Compton-like processes (2.3)
and (2.4) benefit from the extended dominance of the gluon PDF at small values of x. Furthermore, the photoproduction contributions now significantly exceed the O( 2 ) ones throughout
the entire pTcut and pT ranges. From Fig. 13, we see that the W + to W ratios take values in
excess of unity, as expected, and strongly increase with increasing values of pTcut . Comparing
Figs. 9(a) and 9(b), we find that the K factors are significantly amplified as one passes from
the Tevatron to the LHC. This is due to the fact that the Sudakov logarithms,
which originate
from triangle and box diagrams, become quite sizeable at the large values of s and pT that
can be reached at the LHC. This issue was already dwelled on in Ref. [7], to which we refer the
interested reader. Finally, comparing Figs. 10(a) and 10(b), we conclude that the dependence
is generally somewhat smaller at the LHC.
5. Conclusions
We studied the effect of electroweak radiative corrections at first order on the cross section
of the inclusive hadroproduction of single W bosons with finite values of pT , putting special
emphasis on the notion of infrared-save observables with a democratic treatment of hadron jets
initiated by (anti)quarks and gluons. This is indispensable because, as a matter of principle, a
collinear gluonphoton system cannot be distinguished from a single gluon with the same momentum, so that a minimum-transverse-momentum cut on the gluon is an inadequate tool to
prevent a soft-gluon singularity. This led us to include the O(s ) correction to W + production along with the O() correction to W + j production, both contributing at absolute order
O( 2 s ). We also considered the contribution from events where one of the colliding hadrons
interacts via a real photon, which is of absolute order O( 3 ). The hadron can then either stay
intact (elastic scattering) or be destroyed (inelastic scattering), and the photon can participate in
the hard scattering directly (direct photoproduction) or via its quark and gluon content (resolved
photoproduction), so that four combinations are possible.
We extracted the UV singularities using dimensional regularisation and removed them by
renormalisation in the on-shell scheme. We regularised the soft and collinear IR singularities
by means of infinitesimal photon, gluon, and quark masses, , mu , and md , respectively. We
158
used the phase-space slicing method, with cuts s , , and on the scaled photon and gluon
energies and on the separation angles in the initial and final states, respectively, to isolate the soft
and collinear singularities within the corrections from real particle radiation. We achieved the
cancellation of , mu , and md analytically and ensured that the numerical results are insensitive
to variations of s , , and within wide ranges about their selected values.
We presented theoretical predictions for the total cross sections
with a minimum-pT cut and
in
p
p
collisions
with
S = 1.96 TeV at run II at the
for the pT distributions to be measured
Tevatron and in pp collisions with S = 14 TeV at the LHC, and estimated the theoretical
uncertainties from the scale setting ambiguities. We found that considerably less than 1% of all
W + X events contain a prompt photon. The corrections considered turned out to be negative and
to increase in magnitude with the value of pT . While the reduction is moderate at the Tevatron,
reaching about 4% at pT = 200 GeV, it can be quite sizeable at the LHC, of order 30%
at pT = 2 TeV, which is due to the well-known enhancement by Sudakov logarithms. It is an
interesting new finding that the photoproduction contribution is considerably larger than expected
from the formal order of couplings. In fact, it compensates an appreciable part of the reduction
due to the O( 2 s ) correction.
Acknowledgements
We are grateful to Stefan Dittmaier for helpful theoretical and practical advice, to Gustav
Kramer for useful advice regarding phase space slicing, and to Thomas Hahn, Max Huber, and
Frank Fugel for beneficial discussions. The work of B.A.K. was supported in part by the German
Federal Ministry for Education and Research BMBF through Grant No. 05 HT6GUA.
References
[1] UA1 Collaboration, G. Arnison, et al., Phys. Lett. B 122 (1983) 103;
UA2 Collaboration, M. Banner, et al., Phys. Lett. B 122 (1983) 476.
[2] J. Kubar-Andre, F.E. Paige, Phys. Rev. D 19 (1979) 221;
K. Harada, T. Kaneko, N. Sakai, Nucl. Phys. B 155 (1979) 169;
K. Harada, T. Kaneko, N. Sakai, Nucl. Phys. B 165 (1980) 545, Erratum;
G. Altarelli, R.K. Ellis, G. Martinelli, Nucl. Phys. B 157 (1979) 461;
P. Aurenche, J. Lindfors, Nucl. Phys. B 185 (1981) 274.
[3] R. Hamberg, W.L. van Neerven, T. Matsuura, Nucl. Phys. B 359 (1991) 343;
R. Hamberg, W.L. van Neerven, T. Matsuura, Nucl. Phys. B 644 (2002) 403, Erratum;
R.V. Harlander, W.B. Kilgore, Phys. Rev. Lett. 88 (2002) 201801.
[4] U. Baur, S. Keller, D. Wackeroth, Phys. Rev. D 59 (1999) 013002;
S. Dittmaier, M. Krmer, Phys. Rev. D 65 (2002) 073007.
[5] R.K. Ellis, G. Martinelli, R. Petronzio, Nucl. Phys. B 211 (1983) 106;
P.B. Arnold, M.H. Reno, Nucl. Phys. B 319 (1989) 37;
P.B. Arnold, M.H. Reno, Nucl. Phys. B 330 (1990) 284, Erratum;
R.J. Gonsalves, J. Pawlowski, C.-F. Wai, Phys. Rev. D 40 (1989) 2245;
F.T. Brandt, G. Kramer, S.L. Nyeo, Int. J. Mod. Phys. A 6 (1991) 3973;
W.T. Giele, E.W.N. Glover, D.A. Kosower, Nucl. Phys. B 403 (1993) 633;
L.J. Dixon, Z. Kunszt, A. Signer, Nucl. Phys. B 531 (1998) 3.
[6] C. Anastasiou, L.J. Dixon, K. Melnikov, F. Petriello, Phys. Rev. D 69 (2004) 094008;
K. Melnikov, F. Petriello, Phys. Rev. Lett. 96 (2006) 231803;
K. Melnikov, F. Petriello, Phys. Rev. D 74 (2006) 114017.
[7] J.H. Khn, A. Kulesza, S. Pozzorini, M. Schulze, Phys. Lett. B 651 (2007) 160.
[8] F. Bloch, A. Nordsieck, Phys. Rev. 52 (1937) 54.
[9] T. Kinoshita, J. Math. Phys. 3 (1962) 650;
T.D. Lee, M. Nauenberg, Phys. Rev. 133 (1964) B1549.
[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]
159
B.A. Kniehl, Phys. Lett. B 254 (1991) 267.

M. Glck, M. Stratmann, W. Vogelsang, Phys. Lett. B 343 (1995) 399.
A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Eur. Phys. J. C 39 (2005) 155.
J.D. Bjorken, E.A. Paschos, Phys. Rev. 185 (1969) 1975;
R.P. Feynman, Phys. Rev. Lett. 23 (1969) 1415.
S. Dittmaier, PhD thesis, Wrzburg, 1993.
J. Kblbeck, M. Bhm, A. Denner, Comput. Phys. Commun. 60 (1990) 165;
T. Hahn, Comput. Phys. Commun. 140 (2001) 418.
T. Hahn, M. Perez-Victoria, Comput. Phys. Commun. 118 (1999) 153.
G. Passarino, M. Veltman, Nucl. Phys. B 160 (1979) 151.
R. Mertig, M. Bhm, A. Denner, Comput. Phys. Commun. 64 (1991) 345.
T. Hahn, Acta Phys. Pol. B 30 (1999) 3469;
T. Hahn, Nucl. Phys. B (Proc. Suppl.) 89 (2000) 231;
T. Hahn, Nucl. Phys. B (Proc. Suppl.) 157 (2006) 236.
G.J. van Oldenborgh, Comput. Phys. Commun. 66 (1991) 1.
A. Sirlin, Phys. Rev. D 22 (1980) 971.
M. Bhm, H. Spiesberger, W. Hollik, Fortschr. Phys. 34 (1986) 687;
W.F.L. Hollik, Fortschr. Phys. 38 (1990) 165;
A. Denner, Fortschr. Phys. 41 (1993) 307.
W. Beenakker, A. Denner, Nucl. Phys. B 338 (1990) 349;
S. Dittmaier, Nucl. Phys. B 675 (2003) 447.
M.L. Ciccolini, S. Dittmaier, M. Krmer, Phys. Rev. D 68 (2003) 073003;
K.-P.O. Diener, S. Dittmaier, W. Hollik, Phys. Rev. D 69 (2004) 073005.
K.-P.O. Diener, S. Dittmaier, W. Hollik, Phys. Rev. D 72 (2005) 093002.
K. Fabricius, I. Schmitt, G. Kramer, G. Schierholz, Z. Phys. C 11 (1981) 315;
G. Kramer, B. Lampe, Fortschr. Phys. 37 (1989) 161.
N.M. Monyonko, J.H. Reid, M.A. Samuel, G. Tupper, Z. Phys. C 29 (1985) 381;
F.K. Diakonos, O. Korakianitis, C.G. Papadopoulos, C. Philippides, W.J. Stirling, Phys. Lett. B 303 (1993) 177.
G. t Hooft, M. Veltman, Nucl. Phys. B 153 (1997) 365.
R. Kleiss, Z. Phys. C 33 (1987) 433.
G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298.
J.C. Collins, D.E. Soper, G. Sterman, Adv. Ser. Dir. High Energy Phys. 5 (1988) 1.
Particle Data Group, W.-M. Yao, et al., J. Phys. G 33 (2006) 1.
CTEQ Collaboration, J. Pumplin, D.R. Stump, J. Huston, H.-L. Lai, P. Nadolsky, W.-K. Tung, JHEP 0207 (2002)
012.
Tau anomalous magnetic moment form factor

at super B/flavor factories
J. Bernabu a,b , G.A. Gonzlez-Sprinberg c , J. Papavassiliou a,b ,
J. Vidal a,b,
a Departament de Fsica Terica Universitat de Valncia, E-46100 Burjassot, Valncia, Spain
b IFIC, Centre Mixt Universitat de Valncia-CSIC, Valncia, Spain
c Instituto de Fsica, Facultad de Ciencias, Universidad de la Repblica, Igu 4225, 11400 Montevideo, Uruguay
Received 17 July 2007; accepted 3 September 2007

Abstract
The proposed high-luminosity B/flavor factories offer new opportunities for the improved determination
of the fundamental physical parameters of standard heavy leptons. Compared to the electron or the muon
case, the magnetic properties of the lepton are largely unexplored. We show that the electromagnetic
properties of the , and in particular its magnetic form factor, may be measured competitively in these
facilities, using unpolarized or polarized electron beams. Various observables of the s produced on top of
the resonances, such as cross-section and normal polarization for unpolarized electrons or longitudinal
and transverse asymmetries for polarized beams, can be combined in order to increase the sensitivity on
the magnetic moment form factor. In the case of polarized electrons, we identify a special combination
of transverse and longitudinal polarizations able to disentangle this anomalous magnetic form factor
from both the charge form factor and the interference with the Z-mediating amplitude. For an integrated
luminosity of 15 1018 b1 one could achieve a sensitivity of about 106 , which is several orders of
magnitude below any other existing high- or low-energy bound on the magnetic moment. Thus one may
obtain a QED test of this fundamental quantity to a few % precision.
* Corresponding author at: Departament de Fsica Terica Universitat de Valncia, E-46100 Burjassot, Valncia, Spain.
E-mail address: jorge.vidal@uv.es (J. Vidal).

doi:10.1016/j.nuclphysb.2007.09.001
J. Bernabu et al. / Nuclear Physics B 790 (2008) 160174
161
1. Introduction
Magnetic properties of elementary particles are of paramount importance both to theoretical
and experimental high energy physics. The electron anomalous magnetic moment [1] is measured
with the highest precision available in physics nowadays:
ge 2
(1)
= (1159.6521810 0.0000007) 106
2
and the best determination of the fine structure constant is obtained from this electron property. The theoretical predictions in Quantum Electrodynamics (QED), obtained by Schwinger
in 1948 [2], was the first one-loop computation in quantum field theories to be confronted with
experiment:
= 0.00116.
ae =
(2)
2
Actually, precision measurements of the anomalous magnetic moments do not only test QED
(up to 4-loops), but the electroweak and hadronic interactions as well. In fact, since anomalous
magnetic moments are chirality flipping quantities, weak quantum corrections to the muon anomalous magnetic moment are enhanced by a factor of the order of m2 /m2e 40 000 compared to
that of the electron; the measured value is [3]
ae = e /B 1 =
g 2
(3)
= (1165.92080 0.00054 0.00033) 106 .
2
In addition, new physics models [4], and especially those furnishing mass-enhancements comparable to that of the Standard Model (SM), can be constrained from these extremely precise
measurements, nowadays a subject of intense activity [5].
In comparison with these values, our experimental knowledge on the magnetic moment of the
lepton is rather poor. While the electron and muon anomalous magnetic moments are known
with more than seven figures, the PDG limit for the lepton magnetic moment anomaly is [1]:
a = /(eh /2m ) 1 =
0.052 < a < 0.013 (95% C.L.).
(4)
These numbers are more than one order of magnitude bigger than the first-order QED contribution, given in Eq. (2). Recent computations have re-analyzed both weak and hadronic corrections
for the lepton magnetic anomaly within the SM, with an excellent agreement with previous
computations [6], aSM = 1177.21(5) 106 . The higher-order QED contributions are at the
level of 1% compared to the one-loop result of Schwinger, while hadronic and weak corrections
are at the level of 0.001% and 0.04%, respectively.
The experimental determination of the anomalous magnetic moment of the fast-decaying is
very different from that of the stable or relatively long-lived electron and muon, simply because
one does not have the time to measure its interaction with an external electromagnetic field.
Instead, the magnetic information is carried by the cross-section or partial widths for the pair
production, together with spin matrices or angular distributions of the decay products. For
example, the PDG bounds in Eq. (4) where obtained by the DELPHI Collaboration [7] from the
LEP2 data for the total cross-section for the reaction e+ e e+ e + , assuming that any
deviation from the tree level SM prediction was exclusively due to magnetic anomaly [8].
The magnetic anomaly, together with analogous quantities related to the weak magnetism,
i.e., the magnetic coupling with the Z, have been already investigated experimentally [7,9]. The
contributions to the magnetic and weak magnetic anomalies from physics beyond the SM have
162
also been studied in the context of both low- and high-energy physics [10]. It is important to
emphasize at this point that, strictly speaking, the magnetic moment anomaly is defined with all
three fields entering into the interaction vertex on their mass-shell. However, in several of the
aforementioned experiments the kinematics are such that the s or the photon are in fact offshell; therefore, what one actually measures is the corresponding form-factor (for some value of
the momentum transfer) rather than a itself. This is usually accomplished under the additional
assumption that, of all possible form factors appearing in the off-shell vertex, the one corresponding to the magnetic moment gives the dominant contribution. Such experiments furnish
bounds on the contributions to the magnetic moments from physics beyond the SM (since the
scale of new physics is rather high, these latter contributions are practically on-shell). This
is the point of view adopted in [10], where the most stringent model-independent limit for the
magnetic properties is obtained:
New Phys.
0.007 < a
< 0.005 (95% C.L.).
(5)
The one-loop SM contribution to the magnetic form factor was computed in [11] long time
ago, for arbitrary values of the photon off-shellness (q 2 ) and with the charged fermions onshell. The form factor obtained depends on the electroweak gauge-fixing parameter [12], and
becomes gauge-independent only in the limit q 2 0. On the other hand, the pure QED corrections to the off-shell form factor are automatically gauge-independent, for any value of q 2 .
LEP has been the main source of data on -pair production until the advent of the B factories
and their upgrades. In the near future they are expected to be superseded by several orders of
magnitude, thanks to the high-luminosity super B factories, where 1011 1012 pairs will be produced [13]. Motivated by these possibilities, CP-odd spin correlations have already been studied
in [14], for low-energy physics. In addition, polarized beams are also being considered [13];
their implementation would allow the possibility of defining and measuring new observables, related to the linear polarization, not yet considered in the literature with respect to the magnetic
properties.
In this paper we propose new observables in order to explore the poorly known magnetic
properties of the lepton, using the highest statistics facilities available nowadays and in the
near future. Cross-sections and asymmetries for resonant pair-production on top of the
resonances are studied. In particular we show that some of these observables, with the same discrete symmetry transformations as the magnetic moment, allow one to measure these properties
with a precision up to 106 . We compute the contribution of the magnetic form factor to the
cross-section and the normal linear polarizationfor unpolarized electron beamsand both
transverse and longitudinal polarizations for polarized e beams.
The fact that in this class of experiments one will be measuring the magnetic form-factor
rather than the anomaly provides a unique opportunity to observe strong flavor-dependent effects,
encoded in the momentum-dependence of the form factor, in the context of pure QED. Indeed,
whereas the Schwinger correction (the leading order value of the magnetic form factor at q 2 = 0)
is universal (i.e., independent of the fermion masses), the running of the form factors depends
strongly on the mass of the fermion interacting with the photon. Thus, whereas the corresponding
electron form factor practically vanishes at the values of q 2 that we consider ( resonance), the
form factor drops only to about one quarter of its initial value, because the heavy mass slows
down the running considerably. In addition, for q 2 > 4m2 the form factor develops an imaginary
part, which, as we will demonstrate, can also be experimentally measured.
The paper is organized as follows: in Section 2 the magnetic form factors are defined, and
their one-loop QED prediction (real and imaginary parts) reported. In Section 3 the pair pro-
163
duction at super B factories is studied, and expressions for the differential cross-section and the
normal asymmetry are derived. In Section 4 polarized beams observables are considered, with
particular emphasis on the transverse and longitudinal asymmetries. In Section 5 the advantages
of operating at the resonances are discussed, and the role of the weak contributions, especially
that of the Z boson, considered. Finally, in Section 6 we estimate the sensitivity expected for
the anomalous magnetic moment form factor obtained from these observables, and present our
conclusions and final remarks.
2. f f vertex form factors
The most general Lorentz invariant structure describing the interaction of a vector boson V
with two on-shell fermions f f can be written in terms of six form factors:
f (p )f(p+ )|J (0)|0

1
1
= eu(p
) (F1 + F4 5 ) +
(iF2 + F3 5 ) q +
(iF5 + F6 5 )q v(p+ ),
2mf
2mf
(6)
where q = p+ + p . Since the two fermions are on-shell the form factors Fi appearing in Eq. (6)
are functions of q 2 and m2f only.
In addition, if the current J is conserved, we must have

2

F5 = 0,
q
q2
4m2
F5 +
F6 2mf F4 5 = 0
i
(7)
F6 = q 2f F4
2mf
2mf
so that the final expression for the gauge invariant f f vertex reduces to:
f (p )f(p+ )|J (0)|0

1
= eu(p
) F1 +
(iF2 + F3 5 ) q + q 2 q q/ 5 FA v(p+ ).
2mf
(8)
In this expression, F1 parametrizes the vectorial part of the electromagnetic current

(F1 (0) = 1), FA = F4 /q 2 is the so-called anapole moment, while F2 and F3 parametrize the
usual magnetic and electric dipole moments, respectively, i.e.,
e
df =
F3 (0).
af = F2 (0),
(9)
2mf
The electric dipole moment is a CP violating magnitude, whose value vanishes in the SM up
to three loops for leptons. Observables able to disentangle it from the rest have been studied in
[14,15] using techniques similar to those presented in the present paper; therefore, it is not going
to be considered here. The P-odd anapole moment differs from zero due to weak corrections. In
cross-sections its contribution will be suppressed by factors of q 2 /MZ2 compared to the leading
QED corrections, so that its determination will remain below the sensitivity of the proposed
observables. As emphasized already in the introduction, and as is clear from Eq. (9), F2 (q 2 )
coincides with the anomaly a only at q 2 = 0. In super B factories the squared center-of-mass
energy s = q 2 (10 GeV)2 , and therefore F2 (q 2 ) is no longer the magnetic anomaly.
When attempting to extract the value of F2 from scattering experiments (as opposed to using,
say, a background magnetic field) one encounters additional complications due to the contributions of various other Feynman graphs, not related to the magnetic form factor. For example, in
164
the case of e+ e + that we will consider, one receives contributions not only from the
usual s-channel one-loop vertex corrections but also from box diagrams. The contributions of the
latter may interfere in the experimental determination of what we call F2 (q 2 ), i.e., the magnetic
part coming only from the vertex, and should be somehow subtracted out. This may be done
either by computing the box contributions and subtracting them from the cross-section, or by
performing the measurement in a kinematic region where the boxes happen to be numerically
subleading. The strategy we propose in this paper for eliminating the contamination from the
boxes is to measure the observables on top of the resonances; in this kinematic regime the
(non-resonant) box diagrams are numerically negligible, and only one loop corrections to the
f f vertex are relevant.
The direct computation of the magnetic part of the standard one-loop QED vertex yields [16]
2

2m 1
1+
log
i , for q 2 = s > 4m2 ,
F2 (s) =
(10)
2
s
1
where is the fine structure constant and = (1 4m2 /s)1/2 is the velocity of the . For
M 10 GeV,

F2 M2 = (2.65 2.45i) 104 .
(11)
Evidently, at this energy the real and imaginary parts are of the same order.
Note that the above expression for F2 (s) is gauge-independent, despite being an off-shell amplitude. This fact may be easily verified through an explicit calculation of the vertex diagram;
more generally, the gauge-independence of F2 (s) may be understood in terms of the way the
gauge-cancellations organize themselves in the QED S-matrix elements. Specifically, the vacuum polarization is gauge-independent by itself; the gauge-dependence of the direct box cancels
exactly against that of the crossed box; the gauge-dependence of the vertex correction can therefore cancel only against the fermion self-energy graphs renormalizing the external (on-shell)
fermions. The latter however are proportional to . Therefore the contribution of the vertex
proportional to q must be individually gauge-independent.1
3. e+ e + at super B factories
In this section we first consider the -pair production in e+ e collisions through direct
exchange (diagrams (a) and (b) in Fig. 1). Next, we will show that the basic results of this section
still hold for resonant production.
The spin-independent differential cross-section for pair-production can be written as:
2

2
d 0
=
2 2 sin2
F1 (s)
+ 4 Re F1 (s)F2 (s) .
d cos
8s
Note that, at one loop, we have the identity

Re F1 (s)F2 (s) = Re F2 (s) .
(12)
(13)
The angle is determined in the center-of-mass (CM) frame by the outgoing and the
incoming e momenta. As can be seen from Eq. (12), a precise measurement of the angle
1 This simple organization of the gauge cancellations becomes much more intricate in the case of non-Abelian contributions; nonetheless, a gauge-independent of-shell F2 (s) can still be defined, see [12].
(a)
165
(b)
Fig. 1. Diagrams: (a) direct exchange, (b) F2 in exchange.
allows to fit the differential cross-section and obtain a measurement for the Re{F2 (s)} form
factor. In B/flavor factories, this procedure requires that the production plane and direction
of flight are fully reconstructed. In Ref. [17] it is shown that this can be achieved if both s
decay semileptonically. Following these ideas, the differential cross-section for e+ e
+ (
s+ ) (
s ) h+ h can then be written as [18]:
2

2
d
=
2 2 sin2
F1 (s)
+ 4 Re F2 (s)
d cos
2s

Br h Br + h+ .
(14)
The real part of F2 can be measured by a fit to Eq. (14), with a sensitivity that will be given
basically by the statistical error, smeared by the precision in the determination of the angle of
the outgoing . Results presented in Table 1 of Section 6 only consider statistical errors. The
integrated cross-section from Eq. (14) will provide, to leading order in (F1 = 1, F2 = 0), the
normalization for all the asymmetries considered in this paper.
As can be seen from Eq. (14), the differential cross-section does not depend on the imaginary
part of F2 , as expected. The imaginary part of F2 is a T-odd, C- and P-even quantity; therefore,
a suitable observable to look for its effects will be the normal (to the scattering plane) polarization
of the outgoing , as we will show in what follows.
3.1. Normal asymmetry
In order to have sensitivity to the polarization, one has to measure the angular distribution
of the decaying particles. In fact, in the cross-section all information on the imaginary part of F2
is carried out by the linear spin-terms:
d S
2
=
(s + s+ )y Y+ ,
d cos
4s
(15)
where

Y+ = 2 (cos sin ) Im F2 (s) ,
(16)
and = s/2m is the dilation factor.

We work in the CM frame of reference and the orientation of the coordinate system is the same
x , s y , s z ).
as in Ref. [14]. The s vectors are the spin vectors in the rest system, s = (0, s

Polarization along the directions x, y, z corresponds to what is called transverse, normal, and
longitudinal polarizations, respectively. Eq. (16) shows that the contribution of the chirality flipping F2 to the normal polarization is enhanced by a factor of with respect to other possible
non-chirality flipping contributions.
166
The polarization of the final fermion ( ) can be studied by looking at the angular distribution of its decay products. This again requires that only semileptonic decays must be
considered [17]. The cross-section is [18]:

d e+ e + (
s+ ) (
s ) h+ h

dh+ dh
=
4d e+ e + n
+ n
Br + h+ Br h ,
4
4
(17)
with

q
n
=
(18)
= sin cos , sin sin , cos .
q |
)
The and angles are the azimuthal and polar angles of the produced hadrons h (q
in the rest frame (the * means that the quantity is given in the rest frame) and h is the
polarization analyzer.
To preserve the normal polarization term in the cross-section of Eq. (17) one has to define a
particular integration over the angular variables defined before. Indeed, the usual integration over
the complete range of the variables d erases all the information on the Y+ term in the
cross-section, so we must perform an asymmetric forwardbackward (FB) integration on the
angle. This can be done by defining
1

0

d
d
d(cos )
FB (
s+ , s
) 2
d(cos )
d
d
1

3 (s + s+ )y Im F2 (s) ,
6s
which retains the Im{F2 } term. Then, the cross-section in Eq. (17) can be written as
d 4 FB =

dh+ dh
2 2 +
Br h+ Br h
3s
4
4

n y + n+ y 3 Im F2 (s) .
(19)
(20)
Integrating over as many kinematic variables as possible ( ), without erasing the information
on F2 , we finally find that the differential cross-section can be written as

dFB
2 +
=
Br h+ Br h ( ) 3 Im F2 (s) sin .
d
12s
(21)
To get an observable sensitive to the relevant signal, we must now define the azimuthal normal
asymmetry as:
A
N =

L R
1
2
Im
F
(s)
,
=
2
2(3 2 )
(22)
where
L
2

dFB
,
d
d
0

dFB
= L .
d
(23)
167
4. Polarized e beams
For polarized electrons, available at super B factories, an alternative procedure for measuring F2 would be to consider the longitudinal and transverse polarizations of the outgoing s.
Because F1 and Re{F2 } have the same properties under C, P and T symmetries, any single
observable sensitive to one will also carry information on the other. Therefore, the extraction
of the chirality flipping Re{F2 } requires two independent observables, where F1 and F2 enter
with different coefficients; that would allow to express F2 as a linear combination of the two
measured observables. Following the notation and procedure of Refs. [10,15,19], our aim is to
build observables that are linear in F2 .
Re{F2 } is even under T, C and P, while the longitudinal and transverse (to the scattering plane)
polarizations of each are the only components of the spin matrix that are even under T and C,
but odd under P. For this reason, an observable sensitive to F2 will need an additional P-odd
contribution coming, in our case, from longitudinally polarized electrons.
The linear spin-dependent part of the differential cross-section for pair production, with
polarized electrons with helicity , can be written as

d S
2
=
(s + s+ )y Y+ + (s + s+ )x X+ + (s + s+ )z Z+ ,
d cos
8s
(24)
with

1
X+ = sin |F1 |2 + 2 2 2 Re{F2 } ,

Z+ = cos |F1 |2 + 2 Re{F2 } ,
(25)
and Y+ as defined in Eq. (16). Notice: (i) the combination of the two form factors is different for
the transverse and longitudinal polarization terms; (ii) these two terms have different angular
dependence.
As can be seen from Eq. (25), Re{F2 } contributes linearly to the longitudinal and transverse
polarization. It may again be observed that, due to the fact that F2 is a chirality-flip form
factor, its contribution to the transverse polarization is enhanced by the factor 2 with respect to
the chirality-non-flipping factor F1 . This a very important fact for our purposes, because it will
allow the accurate determination of F2 .
4.1. Transverse asymmetry
Following a procedure similar to the one presented in Section 3.1, it can be seen that the
integration over the variables d erases from Eq. (25) all information on the Z+ and Y+
terms. Then, the differential cross-section for the process e+ e |Pol + (
s+ ) (
s )
h+ h is given by

dh+ dh
22 +
Br h+ Br h
d 4 S
=
2s
4
4

1

2
2 2
|F1 | + 2 Re{F2 } .
n x + n+ x
(26)
168
Subtracting the cross-sections for different helicities,2

1
d2 S
Pol(e ) d 4
=1 d 4
=1 ,
(27)
2
and integrating over as many kinematic variables as possible, without erasing the information on
the F2 term, we get

2 2 +
d S
Br h+ Br h
d Pol(e )
16s

1
|F1 |2 + 2 2 2 Re{F2 } ( ) cos .
(28)
To get an observable sensitive to the relevant signal define the azimuthal transverse asymmetry
as
A
T =

R |Pol L |Pol
3
=
|F1 |2 + 2 2 2 Re{F2 } ,
8(3 2 )
(29)
where
L
Pol
3/2

d
/2

d S
d
Pol(e )

= Br + h+ Br h

()2 1
|F1 |2 + 2 2 2 Re F2 ,
8s
S

/2
d
= L
Pol .
R Pol
d Pol(e )

(30)
(31)
/2
It is clear from Eq. (29) that in order to separate out Re{F2 }, we need to remove the contribution of F1 . This can be done in two ways. The first is to use that |F1 |2 = 1 at tree-level,
and assume that any additional contribution to F1 will be of the same order as F2 ; since F2 is a
chirality flipping quantity, it will be enhanced by a factor 2 with respect to the additional F1 .
Under these assumptions, a measurement of the AT asymmetry (subtracted with the tree level
value |F1 |2 = 1) will translate into a Re{F2 } measurement. The second way is to consider a new
observable, relating |F1 |2 and Re{F2 }, and combine the two measurements to extract the value
of Re{F2 }. This can be done by defining a longitudinal asymmetry (AL ) as follows.
4.2. Longitudinal asymmetry
From Eqs. (24) and (25) it can be seen that, as happened with Eq. (19) for the normal asymmetry, an asymmetrical (FB) integration on the angle will select the longitudinal term of the
cross-section

2
2 2
S
(s + s+ )z Z+ + (s + s+ )y Im{F2 } ,
FB (
s+ , s
)
(32)
4s
3
2 This subtraction eliminates also higher order absorptive parts that may be present. See Ref. [15].
169
where Z + = Z+ / cos . Following a similar procedure as in the previous paragraph, and after
subtracting for different helicities as was done in Eq. (27), but integrating over the azimuthal
angles instead of the polar ones, we end up with the final expression for the asymmetrical
(FB) differential cross-section for polarized electrons:

dFB
2 +
=
Br h+ Br h
d(cos ) Pol(e )
2s

|F1 |2 + 2 Re{F2 } ( ) cos .
(33)
Then, we define the longitudinal asymmetry as
A
L =

FB
(+)|Pol FB
()|Pol
3
2
|
+
2
Re{F
}
,
=
|F
1
2
4(3 2 )
(34)
where
FB
(+)
Pol
1

d cos
dFB
d(cos )
Pol(e )

2

= Br + h+ Br h
|F1 |2 + 2 Re{F2 } ,
4s
0
S
dFB
FB
()
Pol d cos
= FB
(+)
Pol .
d(cos ) Pol(e )
(35)
(36)
Combining Eq. (29) and Eq. (34) one can determine the real part of F2 (s). Specifically,

8(3 2 ) 1

Re F2 (s) =
(37)
A
A
.
T
2 L
3 2
5. Observables on the resonance
As explained in the introduction, our aim is to measure the observables on the top of the
peak where the pair-production is mediated by the resonance. The leading diagrams for
the process e+ e + are shown in Fig. 2. Given that we are interested in pairs
produced by the decays of the resonances, we can use (1S), (2S) and (3S), since their
decay rates into pairs have been measured and are sizeable. We assume that only the resonant
diagrams (c) and (d) of Fig. 2 dominate the process on the peaks, so no contribution from
box diagrams has to be considered. As discussed in Ref. [14], the pair-production at the
peak introduces the same polarization matrix terms as the direct production with a exchange
(diagrams (a) and (b), Fig. 1) that we have calculated in Section 3. The only difference is an
overall factor |H (s)|2 which is responsible for the enhancement of the cross-section at resonant
energies; the pure resonant amplitude is given by
4Q2b |F (M2 )|2

3
= i Br e+ e .
H M2 =
2
i M
(38)
At the peak, the interference of diagrams (a) and (d), plus the interference of diagrams (b)
and (c), shown in Figs. 1 and 2, is exactly zero, and so is the interference of diagrams (a) and (c).
Finally, the only contributions proportional to the F2 come from the interference of diagrams (c)
and (d), while diagram (c) squared gives the leading contribution to the cross-section.
170
(c)
(d)
Fig. 2. Diagrams: (c) production, (d) F2 in production.
(e)
(f)
Fig. 3. and Z interchange on production.
(g)
Fig. 4. Non-resonant Z interchange.
5.1. Z contribution on the resonance

It is important to notice that the observables previously defined receive contributions also from
the standard Z interference, computed at the resonance. The process has been studied in detail
in Ref. [20]; the dominant (resonant) contribution will come from the interference of diagram (c)
in Fig. 2 with diagrams (e) and (f) in Fig. 3.
Following a procedure similar to that explained in detail in [14], one can find the following
relations among the amplitudes of diagrams (c), (e) and (f) with the non-resonant -mediated
diagram (a) and Z-mediated diagram (g) of Fig. 4:

Tc = Ta H M2 ,
Td = Tb H M2 ,

2 Qe
Te = Tg (ae 0, ve vb ) H M
,
Qb

Qe

.
Tf = Tg (a 0, v vb ) H M2
(39)
Qb
From these relations, the resonant Z interference can be extracted from the purely Z
interference. This was studied in Ref. [15] and the additional contributions to the spin-averaged
cross-section and to its spin-dependent part are
2
2 Qe
2
2 2

2
d 0
H M
=
vb
PZ M
M MZ2 M0Z ,
2
d
Qb
8 (2sw cw )
Z
S
2

2
2 Qe
2
2
d
H M
=
vb PZ M ,
d
Qb
8(2sw cw )2
171
(40)

Z
Z
Z MZ (s + s+ )y Y+Z + M2 MZ2 (s + s+ )x X+
,
+ (s + s+ )z Z+
(41)
where
2
2 2

PZ M
= M M 2 2 + 2 M 2 1 ,
Z
Z Z
1 2
1
1
1
vb = + sw2 ,
Qb =
v = + 2sw2 ,
,
Qe = 1,
a= ,
2
2
2 3
3

M0Z = a 2 cos + 2 2 sin2 2v 2 2 sin2 ,

1
Z
=
4v a(2 + cos ) sin ,
X+

Z
Z+
(42)
Y+Z = a sin .
= 4v cos a 1 + cos2 + 2 cos ,
To obtain the contribution of the Z-interference to both the cross-section and the normal asymmetry for unpolarized electrons, one must average over helicities in Eqs. (40) and (41). Then,

d 0
Z
2
2
2
2 2 sin2
=
H M
2
d
16 M

2
Qe vb 2 2
4v
(43)
M M MZ2
PZ M2
,
2
(2s c ) Qb
w w

Z

2
2
2 Qe

2

H M
= a
vb
PZ M2
M2 MZ2
2
d
Qb
8(2sw cw )

1
(s + s+ )x (2 + cos )

2
+ (s + s+ )z 1 + cos + 2 cos .
d S
(44)
Eq. (43) shows that the Z contribution will only enter in the determination of F1 , with the
suppression factor 1.576 103 . Furthermore, it is controlled by the small neutral vector
coupling v to leptons. This implies that the angular distribution is the same as that for |F1 |2 and
the measurement of Re{F2 } from the cross-section is not modified.
Similarly, Eq. (44) shows that there is no contribution from the Z-interference to the normal
asymmetry, and the extraction of the Im{F2 } value from this observable through Eq. (22) is again
not modified.
For polarized electrons, subtracting the cross-sections for different helicities and integrating,
as was done in the previous section for the transverse asymmetry, one obtains the new contribu-
172
tion from Z-interference to Eqs. (30) and (31) as
L Pol
R Pol

2 2
2
2 +
PZ M Br h+ Br h
2
(2sw cw )

2

2 Qe
v
M2 MZ2
H M2
vb
1 a
,
Qb
3
4

2 2
2
2 +
PZ M Br h+ Br h
2
(2sw cw )

2

2 Qe
v
M2 MZ2
H M2
vb
1 a
,
Qb
3
4
(45)
(46)
so that the azimuthal transverse asymmetry, A

T of Eq. (29), reads
A
T =
3(2 2 )
8(3 2 )

4vb vM2 (M2 MZ2 )
2
2
P M
.
2 (2sw cw )2 (2 2 )Qb
(47)
Similarly, the Z contribution to the asymmetric cross-sections of (35) and (36) for the longitudinal polarization of s with polarized electrons is
Z
FB (+)
Pol
Z
FB ()
Pol

2
2
2 +
PZ M Br h+ Br h
2
(2sw cw )

2 Q

vb [a v ],
M2 MZ2
H M2
Qb

2
2
2 +
PZ M Br h+ Br h
2
(2sw cw )

2 Q
M2 MZ2
H M2
vb [a v ],
Qb
(48)
(49)
and the corresponding contribution to the longitudinal asymmetry, A

L of Eq. (34), is given by
A
L =
3
4(3 2 )

4vb vM2 (M2 MZ2 )
2
2
P M
.
(2sw cw )2 Qb
(50)
At energies, the values of the and factors are 3.17 104 and 3.15 103 , respectively.
Contrary to what happens with the non-resonant contribution (which is two orders of magnitude
smaller) this -mediated Z contribution must be taken into account when extracting F2 from
longitudinal and transverse asymmetries for polarized electrons. Nevertheless, as the Z
interference proceeds through the vector neutral current coupling to leptons, the structure of this
amplitude is like the one for the contribution of the charge form factor F1 . As a consequence,
the same combination (37) of the two asymmetries (AT 2

AL ) that is able to cancel the con2
tribution of |F1 | automatically cancels the contribution of the Z interference too. We have thus
shown that Re{F2 } can be separated out from other contributions without any ambiguities.
173
Table 1
Sensitivity of the F2 measurement at the energy (ab = attobarn = 1018 b)
Experiment
Observable
Cross-section
Re{F2 }
Normal asymmetry
Im{F2 }
Transverse and longitudinal

asymmetry combined*
Re{F2 }
BaBar + Belle 2ab1
4.6 106
2.1 105
1.0 105
Super B/flavor factory

(1 yr running) 15ab1
1.7 106
7.8 106
3.7 106
Super B/flavor factory

(5 yr running) 15ab1
7.5 107
3.5 106
1.7 106
* Polarized electrons required.
6. Precision of the F2 (M2 ) measurement and conclusions

We can now estimate the precision that can be achieved on the determination of F2 using the
observables defined before. For our numerical analysis we assume a set of integrated luminosities
for high statistics B/flavor factories. We also consider the or (i.e., h1 , h2 = , )
decay channels for the traced , while we sum up over and hadronic decay channels
for the non-traced .
In Table 1, we show the sensitivity that can be achieved for the magnetic moment form factor
F2 in different scenarios: BaBar + Belle at 2ab1 , B/flavor factory, 1 yr running (15ab1 ) and
5 yr running (75ab1 ). The results presented in Table 1 only consider statistical errors. Almost
all the defined observables show similar accuracy in the determination of F2 , but only the normal
asymmetry is sensitive to its imaginary part. Sensitivities coming from the cross-section require
an accurate determination of the angle of the outgoing for a large variety of angles.
To summarize, we have shown that low energy data makes possible a determination of the
lepton QED effects on the measurement of the anomalous magnetic form factor F2 at the
energies. A fit to the cross-section and a measurement of the normal polarization of the outgoing will determine the real and the imaginary part of F2 up to a precision of 106 . Compared
with the QED prediction of Eq. (11), we see that a positive signal appears and it can be tested
to the percent level. The Z interference will not affect the determination of F2 . Polarized
electron beams also open the possibility to determine the value of Re{F2 } by looking at the transverse and longitudinal polarizations of a single . The precision is again of the order 106 . We
have identified the precise combination of the transverse and longitudinal polarizations for
polarized electrons which is able to disentangle the anomalous magnetic moment form factor
from the contributions of both the charge form factor and, at the same time, the resonant Z
interference.
We conclude that the measurement of these sets of observables at the upcoming super B
factories should furnish a high accuracy determination of the rather poorly known magnetic
properties of the lepton.
Acknowledgements
This work has been supported by CONICYT-PDT-54/94-Uruguay, by MEC and FEDER, under the grants FPA2005-00711 and FPA2005-01678, and by Generalitat Valenciana under the
174
grant ACOMP07-093. The work of J.P. is financed by the Fundacin General UV. J.B. acknowledges CERN Theory Unit for hospitality. J.V. also thanks Dr. J. Salt (IFIC) for helpful
conversations.
References
[1] W.-M. Yao, et al., J. Phys. G 33 (2006) 1.
[2] J. Schwinger, Phys. Rev. 73 (1948) 416;
J. Schwinger, Phys. Rev. 76 (1949) 790.
[3] G.W. Bennett, et al., Muon g 2 Collaboration, Phys. Rev. Lett. 89 (2002) 101804;
G.W. Bennett, et al., Muon g 2 Collaboration, Phys. Rev. Lett. 89 (2002) 129903, Erratum.
[4] T. Kinoshita, W.J. Marciano, in: T. Kinoshita (Ed.), Quantum Electrodynamics, World Scientific, Singapore, 1990,
pp. 419478.
[5] M. Passera, Nucl. Phys. B (Proc. Suppl.) 162 (2006) 242;
A. Czarnecki, W.J. Marciano, Phys. Rev. D 64 (2001) 013014.
[6] M.A. Samuel, G.w. Li, R. Mendel, Phys. Rev. Lett. 67 (1991) 668;
M.A. Samuel, G.w. Li, R. Mendel, Phys. Rev. Lett. 69 (1992) 995, Erratum;
S. Eidelman, M. Passera, Mod. Phys. Lett. A 22 (2007) 159.
[7] J. Abdallah, et al., DELPHI Collaboration, Eur. Phys. J. C 35 (2004) 159.
[8] F. Cornet, J.I. Illana, Phys. Rev. D 53 (1996) 1181.
[9] A. Heister, et al., ALEPH Collaboration, Eur. Phys. J. C 30 (2003) 291.
[10] G.A. Gonzlez-Sprinberg, A. Santamaria, J. Vidal, Nucl. Phys. B 582 (2000) 3.
[11] K. Fujikawa, B.W. Lee, A.I. Sanda, Phys. Rev. D 6 (1972) 2923.
[12] Gauge-independent off-shell form factors in non-Abelian gauge theories can in fact be defined, by resorting to the
pinch technique, see, for example J. Papavassiliou, Phys. Rev. D 41 (1990) 3179;
J. Papavassiliou, C. Parrinello, Phys. Rev. D 50 (1994) 3059;
J. Bernabu, J. Papavassiliou, J. Vidal, Phys. Rev. Lett. 89 (2002) 101802;
J. Bernabu, J. Papavassiliou, J. Vidal, Phys. Rev. Lett. 89 (2002) 229902, Erratum.
[13] SuperB Conceptual Design Report, http://www.pi.infn.it/SuperB.
[14] J. Bernabu, G.A. Gonzlez-Sprinberg, J. Vidal, Nucl. Phys. B 701 (2004) 87.
[15] J. Bernabu, G.A. Gonzlez-Sprinberg, J. Vidal, Nucl. Phys. B 763 (2007) 283.
[16] C. Itzykson, J.B. Zuber, Quantum Field Theory, International Series in Pure and Applied Physics, McGrawHill,
New York, USA, 1980, 705 pp.
[17] J.H. Kuhn, Phys. Lett. B 313 (1993) 458.
[18] Y.S. Tsai, Phys. Rev. D 4 (1971) 2821.
[19] J. Bernabu, G.A. Gonzlez-Sprinberg, M. Tung, J. Vidal, Nucl. Phys. B 436 (1995) 474.
[20] R. Koniuk, R. Leroux, N. Isgur, Phys. Rev. D 17 (1978) 2915;
J. Bernabu, P. Pascual, Phys. Lett. B 87 (1979) 69;
J. Bernabu, P. Pascual, Nucl. Phys. B 172 (1980) 93;
J. Bernabu, F.J. Botella, O. Vives, Eur. Phys. J. C 7 (1999) 205.
Momentum fluctuations of heavy quarks

in the gauge-string duality
Steven S. Gubser
Joseph Henry Laboratories, Princeton University, Princeton, NJ 08544, USA
Received 21 August 2007; accepted 10 September 2007
Abstract
Using the gauge-string duality, I compute two-point functions of the force acting on an external quark
moving through a finite temperature bath of N = 4 super-YangMills theory. I comment on the possible
relevance of the string theory calculations to heavy quarks propagating through a quarkgluon plasma.
1. Introduction
Recent string theory computations [1,2] of the drag force on a heavy quark, and a related
calculation [3] of momentum diffusion for non-relativistic heavy quarks, have raised the tantalizing prospect that the gauge-string duality might help us understand the dynamics of charm and
bottom quarks propagating through the quarkgluon plasma (QGP) produced at the relativistic
heavy ion collider (RHIC). Earlier work in a somewhat similar spirit includes [4]. Independently,
a proposal was made in [5] for extracting the jet-quenching parameter q for N = 4 gauge theory
from a Wilson loop calculation amenable to solution through techniques of classical string theory.1 Subsequent work includes extensions to non-conformal theories [912], non-zero chemical
potentials [1318], and other deformations of N = 4 gauge theory [19]; studies of directional
emission [20,21] and the relation to the magnetic string tension [22]; and calculations of drag
on particles carrying higher representations of the gauge group [23]. The venue for the string
theory calculations is the gauge-string duality [2426] (for reviews see [2729]), in particular
E-mail address: ssgubser@princeton.edu.
1 There exists some debate in the literature regarding the calculation of [5]. Arguments from both sides of this debate
can be found in [68].

doi:10.1016/j.nuclphysb.2007.09.017
176
S.S. Gubser / Nuclear Physics B 790 (2008) 175199
the computation of Wilson loops through dual classical configurations of fundamental strings in
anti-de Sitter space, as first considered in [30,31].
The drag force calculations [13] result in a relaxation time tD for charm quarks that is in the
right ballpark for comparison with data on the nuclear modification factor RAA and the elliptic
flow parameter v2 for heavy quarks, as reported most recently by the STAR and PHENIX collaborations in [32,33]. According to the prescription for comparing N = 4 gauge theory and QCD
advocated in [34], the string theory prediction is tD = 2.1 1 fm/c for charm quarks. This value
is lower than typical values tD 4.5 fm/c used in one of the more successful phenomenological
theories [35] of heavy-quark transport. But one should bear in mind that many steps separate the
drag force calculation from predictions of RAA and v2 for heavy quarks. Prominent among these
steps is a description of heavy quark propagation through the QGP which includes fluctuations
in the force on the quark. The goal of this paper is to calculate the two-point functions of these
fluctuations using the gauge-string duality.
The organization of the rest of this paper is as follows. Section 2 sets the stage by reviewing
the diffusion of momentum of a heavy quark in a Langevin formalism. Section 3 sets up the main
calculation by finding the quadratic action and equations of motion for fluctuations of the trailing
string. Section 4 shows how to translate appropriate solutions of these equations of motion into
the Greens functions of interest. Section 5 shows how to treat the zero-temperature case analytically. Section 6 presents a numerical study of the two-point function of primary interest. Section 7
includes a discussion of how the string theory results might be applied to understanding heavy
quarks propagating through a real-world quarkgluon plasma. This discussion suffers from the
usual difficulties of relating two significantly different theories, namely N = 4 super-YangMills
and QCD. Some significant technical issues are postponed to three appendices.
2. Momentum diffusion in a Langevin description
Following [36] one may attempt to describe the propagation of a heavy quark through a thermal bath in terms of Langevin dynamics (see also, for example, [37,38]):
dpi
= D (p)pi + FiL + FiT ,
dt
L

Fi (t1 )FjL (t2 ) = L (p)p i p j (t1 t2 ),
T

Fi (t1 )FjT (t2 ) = T (p)(ij p i p j )(t1 t2 ),
(1)
where p i = pi /p is the unit 3-vector in the direction of the momentum p.

The string theory prediction for the drag force (ignoring issues of modified dispersion relations raised in [1]) amounts
to
T2
D =
(2)
gY2 M N ,
2
m
independent of p, where m is the mass of the heavy quark. It is often assumed (modulo a subtlety
having to do with how one discretizes the Langevin equations in the process of passing to a
FokkerPlanck description) that L and D are constrained by the Einstein relation

L = 2T ED = gY2 M N T 3 ,
(3)
where
=
1
1 v2
(4)
177
is the standard Lorentz factor for the heavy quark (and I have again ignored the possibility of a
modified dispersion relation). The relation (3) emerges from requiring that a heavy quark propagating according to Eq. (1) should eventually equilibrate to a thermal distribution eE/T , so it is
a consistency condition for the Langevin approach. No such relation is available for T , but for
non-relativistic quarks isotropy demands T = L . Indeed, the result

2T 2
L = T =
(5)
= gY2 M N T 3 for v 1,
D
was obtained through direct calculation in [3] and through use of the Einstein relation in [1].
The Langevin approach (1) is hardly the only one in use for describing quark dynamics in the
QGP. Another prominent paradigm hinges on radiative energy loss [3941]. Central interest in
this description attaches to the BDMPS jet-quenching parameter q.
The jet-quenching parameter
as I prefer to define it is
2
/.
qT = p
(6)
2 is the average transverse momentum acquired by a parton after it has traveled a disHere p
tance , measured in the rest frame of the plasma.2 The path length is supposed to be taken
large enough so that short-range correlations are washed out, but short enough so that the quark
is only slightly deflected from its original trajectory. A definition of q that is preferred in some
works, for example [5,7], refers to a partially light-like Wilson loop. To distinguish between the
q of these works and the definition I prefer, I will employ a subscripted T as in (6). Note that
the definition of qT does not require a strict light-like limit: it can be evaluated at any v. Also,
does not require us to commit to the BDMPS formalism. For example, consider a Langevin description of momentum broadening where a heavy quark travels initially in the x 1 direction and
is acted on by stochastic transverse forces satisfying

dpi
(7)
Fi (t1 )Fj (t2 ) = ij KT (t1 t2 ),
= Fi ,
dt
where KT (t) is an integrable function and i, j run over values 2, 3. If one assumes p (0) = 0,
then at sufficiently late times t (meaning times larger than the characteristic time-scales of KT (t))
one has

p (t)2 = ij pi (t)pj (t) =
t
t
dt2 2KT (t1 t2 ) 2T t,
dt1
0
(8)
where by definition

T =
dtKT (t).
(9)
Comparing with (6), one extracts

qT =
2T
.
v
(10)
2 I would use p in place of p except for the fact that p is usually reserved for momentum perpendicular to the
T
T
beampipe. If the parton is traveling in the x 1 direction, then p is the projection of the momentum onto the x 2 x 3 plane.
178
For a heavy quark with a known initial momentum, one may also consider

qL = (pL )2 /
(11)
where pL is the deviation of the longitudinal momentum of a heavy quark from some average
value. As with qT , the definition (11) does not commit us to a specific formalism; however, a
Langevin analysis precisely analogous to the one leading to (10) gives
L
qL =
(12)
,
v
where

L =

dt F1 (t)F1 (0)
(13)
and F1 is the fluctuating part of the longitudinal force.

The AdS/CFT calculations that are the focus of Sections 36 lead to force correlators from
which qL and qT may be extracted, essentially as in (9), (10), (12), (13). The results are

qL = gY2 M N T 3 5/2 /v.

qT = 2 gY2 M N T 3 /v,
(14)
The result (14) is larger by a factor 1/(1 v 2 )3/4 than expected from (3). So Langevin dynamics
does not capture the whole story. The method for obtaining (14) is to numerically compute a
symmetrized Wightman two-point function of oscillations of the trailing string solution of [1,2]
and identify the zero-frequency component of this two-point function with L or T as appearing
in (9) and (13). The expressions in (14) are obtained as a fit to the low-frequency limit of the
numerics, but they are probably exact statements about the Greens functions in question.3
3. Fluctuations of the trailing string
The drag force prediction from the gauge-string duality,

d p
p
gY2 M N T 2 ,
= D p =
dt
2
m
arises from a classical string configuration [1,2] in the AdS5 -Schwarzschild background,

2 dy 2
zH
L2
1
2
2
2
h dt + d x +
, h 1 y 4 , zH =
,
ds = 2
h
T
zH y
(15)
(16)
where T is the Hawking temperature. The string configuration, for p pointing in the x 1 direction,
is described by

1y
vz
H
1
1
1
x = x0 (t, y) vt +
(17)
tan y + log
.
2
1+y
The fluctuating part of the force on the quark, which is located at the y = 0 end of the string (i.e.,
on the boundary of AdS5 -Schwarzschild) should be computable in terms of linearized fluctuations
3 I thank C. Herzog for first suggesting to me the comparison with the non-relativistic results (5).
179
around the solution (17). More specifically, if

x 1 = x01 (t, y) + x 1 (t, y),
x 2 = x 2 (t, y),
then the NambuGoto action,

1
2
d
det g ,
SNG =
2
x 3 = x 3 (t, y),
g = G x x ,
(18)
(19)
may be expanded to quadratic order in x i to obtain

yS2
L2 1
dt
dy
+
dt dy P x 1
2 zH
y2

dt dy GL x 1 x 1 +
GT x i x i ,
2
SNG =
(20)
i=2,3
where
P =
GT = yS4 GL

zH /y 2 1 y 4
,
1
4 4
zH 1y yS
2
L2 /zH
1 y 2 (1y 4 )2
=
v2
2 yS2
2
L2 /zH
v
2 yS2
1y 4
(21)
v2
1y 4
y 4 yS4
y 2 zH
and I have introduced

4
yS 1 v 2 .
(22)
(23)
If I switch from = (t, y) to some other parametrization of the worldsheet, then P , GL ,
and
GT would transform not as tensors but as tensor densities: that is, they include a factor of
det g .
Because P = 0 and GL is proportional to GT , all three x i (t, y) obey the same equation
of motion, namely

GT = 0,
(24)
where = x i , i = 1, 2, 3. Plugging an expansion

(t, y) =
d
0 () (, t, y),
2
(, t, y) = eit (, y)
(25)
into (24) leads to a radial equation for :

s(y)y2 + t (, y)y + u(, y) (, y) = 0,
where
(26)
180
2

s(y) = y 1 y 4 yS4 y 4 ,

t (, y) = 2 1 y 4 1 y 8 v 2 1 y 4 + iy 3 zH ,

u(, y) = yzH 1 y 4 zH + v 2 y 4 (4iy + zH ) .
(27)
The radial equation (26) has regular singular points at the zeroes of s(y): y = 0, y = , and y =
yS where is any fourth root of unity. The most interesting of these is y = yS , corresponding to
some intermediate point on the string. What is special about this point is that it is the location of
a horizon of the induced metric on the worldsheet. An intuitive way to see this is that a point on
the trailing string with y held fixed follows a timelike trajectory if y < yS and a spacelike one if
y > yS .4 So the region y < yS corresponds to the exterior of the black hole on the worldsheet,
and y > yS corresponds to the interior. No signal from the interior can propagate classically to
the exterior. Instead, signals in the interior region must by causality travel down the string. But
there should be some Hawking radiation from the worldsheet horizon upward toward y = 0, and
it is natural to suppose that it relates to the momentum diffusion. I will not make this connection
directly, but the two-point functions that I will compute are related to both classical absorption
and spontaneous emission by the worldsheet horizon.
Although (26) does not seem to be explicitly solvable, certain limits of it are tractable. For
example, to next-to-leading order in small y, two independent solutions are
Def (, y) = 1 +
2
2 zH
2yS4

y2 + O y4 ,
VEV (, y) = y 3
2
2 zH
10yS4

y5 + O y6 ,
and to leading order in small positive yS y, two independent solutions are

+ (, y) = (yS y)izH /2yS 1 + O(yS y) .
(, y) = 1 + O(yS y),
(28)
(29)
Of the two solutions in (29), + (, y) corresponds to an outgoing wave, because the phase
increases as one goes to smaller values of y. The standard horizon boundary condition, then, is
to disallow this solution. To justify this statement completely, one should show that suppressing
this solution amounts to stipulating that (, t, y) should depend only on the infalling coordinate
at the horizon. This is indeed true, because, to leading order close to the horizon, the infalling
and outgoing coordinates are
u = t,
u+ = t
zH
log(yS y).
2yS
(30)
A justification of (30) is postponed to Appendix A.

Consider now a set of wave-functions R (, t, y) = eit R (, y), defined so that
H
() (, y).
R (, y) = Def (, y) + CR ()VEV (, y) = C
(31)
In (31), Def , VEV , and are regarded as exact solutions of the radial equation (26), specified by their asymptotics at y = 0 or y = yS . In the absence of an analytical solution to (27),
H () must be determined through some approximation scheme or
the quantities CR () and C
through numerics. The wave-functions R (, t, y) have several useful properties:
4 A closely related observation [1,6,42,43] is that there are no-drag configurations of mesons represented as strings
with both ends attached to branes in AdS5 -Schwarzschild provided the string between the quarks hangs no lower than
y = yS .
181
1. R (, t, y) is a solution of the wave equation (24) for .

2. R (, t, 0) = eit .
3. R (, t, y) = R (, t, y). To see this, note that complex conjugation is equivalent to
sending in the radial equation (26), and that the boundary condition at the horizon is
preserved by complex conjugation. This property guarantees that (t, y) is real everywhere
provided 0 () = 0 ().
4. Momentum correlators from the trailing string
As in the previous section, let be one of the x i . Let O be the operator dual to , and
let G = GL or GT , as appropriate. The retarded, time-ordered, and symmetrized Wightman

two-point functions are defined as

GF (t) = i T O(t)O(0) ,
GR (t) = i(t) O(t), O(0) ,

1
G(t) = O(t)O(0) + O(0)O(t) ,
(32)
2
where in general
tr eH Q
,
(33)
tr eH
and H is the Hamiltonian of the gauge theory.
In analogy to the recipe of [44], the proposal for extracting the retarded Greens function from
the wave-functions R (, t, y) is
Q =
GR () = R (, t, y)G y R (, t, y)|y=0 = G y log R (, t, y)|y=0 ,
(34)
where the notation |y=0 means to evaluate at y = 0 after taking the derivatives. The difference
between (34) and (3.15) of [44] is the use of R rather than R .5 Using R would amount
to restricting the sum over to = y, and this does not make sense in light of worldsheet
reparametrization invariance. If the number current associated with R (, t, y) is defined as
J =
G R R ,
2i
(35)
then
Im GR () = J y ,
(36)
and because J is conserved, the right-hand side of (36) can be evaluated at any y. The number
flux is positive at the horizon for > 0 as a consequence of the infalling boundary conditions,
so one finds from (36) that
Im GR () < 0 for > 0.
(37)
This is the correct sign for describing dissipative dynamics. If R had been used instead of R ,
the connection with the conserved number current would be broken. In [44] a restriction was
5 Actually there seems to be one further difference: an overall sign. Possibly I have misunderstood the notation
used [44]. Anyway, I claim that the sign in (34) is the right one.
182
made to consider only diagonal metrics, so the question of including dependence of the wavefunction on coordinates other than the radial one never arose. (The context was slightly different:
instead of G , the metric of interest was the line element of the bulk spacetime.) But (34) is
clearly in the spirit of [44] and its antecedent [25]: note for instance the similarity with (27)
of [25].
The quadratic part of the on-shell action reduces to

1
Son-shell =
(38)
0 ()0 () Re GR (),
2
where to obtain the right-hand side from the quadratic part of SNG I used an integration by
parts, the equations of motion, the formula (34), and the property GR () = GR () , which is
obvious from property 3 at the end of Section 3.
Thus, just as in the examples treated in [44], the real part of GR () may be extracted using a
conventional AdS/CFT formulation in which the on-shell action is the generating functional for
Greens functions. But it turns out that the imaginary part will be of greater interest, due to its
connection with the dissipative dynamics. The relation between Im GR () and G(),
Im GR (),
G() = coth
(39)
2T yS
is modified from the usual one in that T yS appears in place of T . A justification of (39), following [45], is postponed to Appendix B.
Here is a somewhat informal argument for identifying O(t) with minus the force on the quark
in the direction that selects. To be precise, let = x 2 (t, y) : then the claim is O(t) = p 2 (t).
Consider the action

Sq (t) = dt L( 0 )
(40)
for an external quark propagating through the thermal medium along a path
x 1 (t) = vt,
x 2 (t) = 0 (t).
(41)
I have excluded explicit dependence on 0 and t from L because the medium is assumed to be
translationally invariant and static. Note that translational invariance does not imply conservation
of transverse momentum: Sq is evaluated under the path integral and includes couplings to the
gauge theory. For small deviations 0 from a straight path,

L
Sq dt 0
(42)
= dt 0 p = dt 0 p .
0
Of course, p = p2 . Altogether, the generating functional for the two-point functions of interest
is

Z[0 ] = exp i dt 0 p 2 ,
(43)
whence the desired conclusion O = p 2 . More formally, p2 (t) should be regarded as the operator which generates a local displacement at time t of the path of the Wilson loop for the external
quark. A similar line of argument applies when = x 1 , and of course x 3 is equivalent to x 2 .
Other Greens functions than GR () could be computed by using other wave-functions than
R (, t, y) in (34), provided they satisfy the three properties at the end of Section 3. For example, wave-functions A (, t, y) = eit A (, y) with A proportional to + should lead to the
advanced Greens function.
183
5. Zero-temperature Greens functions

As a check on the validity of the prescription (34), and as a way of understand the T
behavior in the T = 0 case, it is useful to detour to a consideration of the zero temperature limit,
where the background metric is just

L2
2
2
2
dt
+
d
x

+
dz
z2
and the unperturbed solution is a string that hangs straight down:
ds 2 =
x 1 = vt,
x 2 = x 3 = 0.
(44)
(45)
I will use worldsheet coordinates = (t, z). An expansion entirely analogous to (20) can be
performed, only now the integration measure is dt dz, and

L2 /z2

1
0
GT = 1 v 2 GL =
.
P = 0,
(46)
0 1/ 2
2
The wave-functions satisfying infalling boundary conditions at the degenerate horizon at z =
were found analytically for v = 0 in [46] and can be adapted immediately to the case of interest:
R (, t, z) = eit ei z/yS (1 i z).
2
(47)
The retarded Greens functions are

GR
T () =
1 R
iL2 2 3
G
()
=
,
2
2 L
(48)
where I have dropped a divergent term proportional to 2 whose Fourier transform is a contact
term supported at t = 0. The associated symmetrized Wightman functions are
L2 2 3
L2 4 3
|| ,
GL () =
|| .
(49)

2
2
In preparation for finding the real-time forms GT (t) and GL (t), consider the general power-law
Fourier integral:
GT () = (sgn ) Im GR
T () =
eit
dt 2 =
|t|
dt
2 cos t
= 2||21 (1 2) sin .
|t|2
(50)
Expanding around = 2 gives

dt
eit
= ||3 + (analytic in ).
6
t4
(51)
There is an analytic continuation implicit in the result (51): whereas the integral (50) can be
evaluated for 0 < < 1/2 with no regularization, the integral in (51) is highly divergent. Regularization schemes differ in how they prescribe the analytic terms on the right-hand side of (51),
corresponding to contact terms in real time, supported at t = 0. Employing (51) and ignoring
contact terms, one obtains
GT (t) =
1
3L2 2 1
G
(t)
=
.
L
2
2 t 4
(52)
184
To me the most surprising feature of the final result (14) in the case of non-zero temperature
is the strong dependence of the magnitude of longitudinal fluctuations on . One can see already
4
in (48) and (52) a very strong dependence: GR
L for fixed or t . In this zero-temperature
setting, the factors of should be capable of being understood just in terms of Lorentz invariance.
As a step in this direction, consider the following toy model: N = 4 super-YangMills with gauge
group U (1). The action for the heavy quark is

S = ds m|x|
+ A x + |x|
(53)
I XI ,
where dots denote d/ds and the parameter s need not be the proper time . The I form a
unit vector in R6 , and they specify properties of the quark: its scalar charges. For an external
quark, one specifies the I as functions of s along with the trajectory x (s). The only case I will
consider (which is also the case considered in [13]) is the one where the I are constant. The
classical equation of motion following from (53) is
mu = W + I X I u ,
(54)
where I have now specialized to the case where s is the proper time , and

X I
+ u u .
x
The velocity vector u = dx /d satisfies
W F u + I
u u = 1,
= diag{1, 1, 1, 1}.
(55)
(56)
Let us now promote F (t, x) and X I (t, x) to operator-valued fields and consider the symmetrized Wightman two-point function of W ( ). Because of conformal invariance and the
obvious identity W u = 0, the answer can only be

+ u u
1
(57)
W ( )W (0) + W (0)W ( ) = CW
2
4
for some constant CW .
Whether W is the entire force acting on the quark depends on whether the last term in (54)
is considered a force or a modification of the momentum. This is immaterial for external quarks
because their mass is infinite and their trajectory can be prescribed to have constant u . So for
the purposes of comparing (57) to (52), this term does not matter. What does matter is that, in
deriving (52), the force was defined in terms of a t derivative, not a derivative:
GW
( )
Fi =
dpi
1 dpi
=
= Wi .
dt
d
(58)
Using (57) and (58), and assuming as usual that the quark moves in the x 1 direction, one obtains

1
1
4 CW
GL (t) = F1 (t)F1 (0) + F1 (0)F1 (t) = 2 GW
(59)
,
11 (t/ ) =
2
t4
and similarly GT (t) = 2 CW /t 4 . Thus there is perfect agreement with (52) if one formally
makes the replacement CW 3L2 / 2 .
The simple-minded analysis leading to (59) of course makes no prediction about the scaling
of the low-frequency parts of GL () and GT () with when the temperature is non-zero. But
what drives the relation GL () = 2 GT () is the corresponding relation GL = 2 GT , and this

holds whether T is zero or not.
185
6. Numerical results
Starting from (22), (31), (34), and (39), a short calculation shows that

2 Im CR () + X ,
GT () = Y coth
GR
T () = Y 2CR () + iX ,
2T yS
where
2
X v 2 2 zH ,
3
L2
3
=

3
3
4 zH
4 zH
3
gY2 M N .
(60)
(61)
The longitudinal Greens functions are always 2 times the transverse ones. Only GT () will be
evaluated explicitly here. To do so, it helps to note that the substitutions

1 + y 2 tan1 y izH /4
(, y) =
(w, Y ),
e
1y
zH
y
w=
Y= ,
(62)
yS
yS
lead to a simplified form of the equation of motion (26):

Y 1 Y 4 Y2 2 1 + Y 4 iwY 3 Y + w 2 Y (w, Y ) = 0.
(63)
An infalling solution R (, y), satisfying the asymptotics (31), corresponds to a solution

R (w, Y ) satisfying

w2 Y 2
+ cR (w)Y 3 + O Y 4 for small Y,
2

H
(w) 1 + O(1 Y ) for Y close to 1.
R (w, Y ) = c
R (w, Y ) = 1 +
From (60), (62), and (64) it is straightforward to show that

gY2 M N
3
2i
R
2cR (w) + w
GT () =
3
yS
3
4zH
and that
GT () =
where
gT (w) =

1
3
2 zH
gY2 M N
yS
w
w
coth
2
2
gT (w),

3 Im cR (w)
1+
.
w
(64)
(65)
(66)
(67)
To calculate cR (w), and thereby gT (w), I employed numerical integration, implemented with
Mathematicas NDSolve, matched to high-order series expansions for Y close to 0 and 1. Asymptotic forms at large and small w were extracted by fitting to the numerics:

gT (w) = 1 + 1.26w 2 + O w 4 for small w,

1
for large w.
gT (w) = |w|3 1 4 + O w 8
(68)
2
w
186
Fig. 1. The thick red lines come from numerical evaluations of gT (w) and the most accurate form of gT () I was able
approx
approx
to obtain. The thin black lines are the analytic approximations gT
(w) and gT
(). The thin dash-dot line in the
approx
gT () plot is the residual, gT () gT
(). The thick dashed line is an approximation to gT () obtained by subtracting
off only the leading |w|3 behavior from gT (w) and numerically Fourier transforming the remainder out to W = 20. (For
interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Graphs of gT (w) and its Fourier transform,

gT ()
dw iw
gT (w),
e
2
(69)
are shown in Fig. 1. By visual inspection, the scale for the characteristic structures in gT () is
1.5.
Because of the leading |w|3 behavior in gT (w) at large w, the real space form gT () is highly
divergent at small . This leading divergence is precisely the behavior observed in the zerotemperature Greens function in Section 5. Contact terms supported at = 0 restore integrability
at the origin. A simple strategy for finding gT () is to subtract the |w|3 term from gT (w) and
perform a numerical Fourier transform on the remainder, cut off at some maximum frequency W .
The trouble with this strategy is that the subleading 1/|w| behavior in gT (w) leads to errors of
order log(W ||) at small . Subtracting this subleading behavior as well leads to a remainder
which is absolutely integrable at large w but not at small w. The simplest form I could find
that agrees with the large w asymptotics in (68) to the order shown and leads to an everywhere
absolutely integrable remainder is

1
1
approx
3
(w) = |w| 1 + 2
1 2
,
gT
2
0 w 2
20 w 2

3 ||/0
2
approx
() = 4 e
1 + 2 , where 0 = (3/8)1/4 .
gT
(70)

40
The intuition of Debye screening suggests that gT () decays exponentially at large , similarly
approx
(). As explained in [44], singularities of the retarded Greens function in the comto gT
plex plane correspond to quasi-normal modes of the corresponding field on the gravity side
of the AdS/CFT duality. In the present context, the quasi-normal modes in question are fluctuations which are purely infalling at the worldsheet horizon and behave as y 3 for small y. Such
fluctuations occur at values of where CR () diverges. Equivalently, in terms of w = zH /yS ,
the quasi-normal modes arise at poles of cR (w). A preliminary numerical analysis yielded the
following approximate values of the first two quasi-normal modes wn :
w1 = 2.620 2.302i,
w2 = 4.650 4.316i.
(71)
187
For each wn , wn is also a quasi-normal frequency. (Quasi-normal modes of the trailing string
were also considered in [1] in the v 0 limit, although the boundary conditions were different.)
A standard expectation is that 1 1/ Im w1 0.44 sets a scale for exponential decay of the
retarded Greens function. This is considerably smaller than ; note however that 1 relates only
to asymptotic statements at large , whereas is an approximate characterization of the main
structures visible in gT (). (Also note that because | Im w1 | > 2, the singularity of gT (w) in the
complex w plane closest to the real axis is not w1 but 2i coming from the factor coth(w/2),
so a decay gT () e2|| is expected rather than e||/1 .)
7. Discussion
One should be able to extract the transport coefficients L , T , qT , and qL from the lowfrequency limit of the symmetrized Wightman functions:
v
T = qT = lim GT (),
0
2
L = v qL = lim GL ().
(72)
The logic supporting (72) is a treatment of momentum diffusion as in Eqs. (7)(10), only
with stochastic averages replaced by the symmetrized Wightman functions. Using (66) and
limw0 gT (w) = 1, one finds the results quoted in (14):

2T
L
qT =
(73)
qL =
= 2 gY2 M N T 3 /v,
= gY2 M N T 3 5/2 /v.
v
v
Because of the dramatic failure of the Einstein relation (3), it does not make sense to plug the
trailing string predictions for D , L , and T into a Langevin description of a finite mass quark:
it would not equilibrate to a MaxwellBoltzman distribution, due to the largeness of L as compared to D at highly relativistic speeds. Perhaps what is needed is a stochastic treatment of the
quasi-normal modes of the trailing string, the lowest of whose frequencies were found in (71).
Because strongly coupled N = 4 gauge theory is different in significant respects from QCD,
there is considerable uncertainty in how one should translate results like (2) and (72) into
quantitative predictions for QCD. To characterize this uncertainty, consider the following two
comparison schemes:
An obvious comparison scheme is to equate the temperature and the YangMills coupling.
I will assume T = 250 MeV and s = 0.5, which are in a representative range for RHIC
physics. To summarize:
obvious scheme: TN =4 = TQCD = 250 MeV,
gY2 M N = 12s = 6.
(74)
An alternative comparison scheme was proposed in [34]. The first part of this prescription
is to equate energy density instead of temperature. Thus in any formula (for example (73)) in
N = 4 into which T enters, one eliminates it in favor of a power of the energy density before
comparing to QCD. This is approximately equivalent to setting TN =4 = TQCD /31/4 . The
second part of the prescription is to determine gY2 M N by matching string theory results for
the force between a static quark and anti-quark to lattice results. This match is conspicuously
imperfect, but at TQCD = 250 MeV a range 3.5 < gY2 M N < 8 was found by comparing static
potentials at r 0.25 fm, with gY2 M N = 5.5 suggested as a typical value. To summarize:
alternative scheme: 31/4 TN =4 = TQCD = 250 MeV,
gY2 M N = 5.5.
(75)
188
Combining (73) with either (74) or (75) leads to the following estimates of qT for a charm
quark propagating through a real QGP:

1
obvious: qT 2.2 GeV2 /fm
= 5.9 GeV2 /fm for pc = 10 GeV/c,
4
v 1 v2

1
alternative: qT 0.51 GeV2 /fm
= 1.4 GeV2 /fm for pc = 10 GeV/c,
4
v 1 v2
(76)
and for qL :

obvious: qL 1.1 GeV2 /fm
1
= 150 GeV2 /fm for pc = 10 GeV/c,
v(1 v 2 )5/4

1
alternative: qL 0.26 GeV2 /fm
= 36 GeV2 /fm for pc = 10 GeV/c,
v(1 v 2 )5/4
(77)
where I have set mc = 1.4 GeV. The choice pc = 10 GeV/c corresponds roughly to pT =
5 GeV for non-photonic electrons. The values of qT and qL scale with temperature as
(TQCD /250 MeV)3 in both the obvious and alternative prescriptions. Based on the large
values for qL in (77), especially using the obvious comparison prescription, a conservative
conclusion is simply that longitudinal fluctuations in the force are strong and merit a more thorough investigation.
There appears to be some tension between the results of the present work and those of [5,7].
We insert a color source in N = 4 gauge theory in the same way, namely
by letting a fundamental
string end on the boundary. This is why our results have the same gY2 M N scaling. But the
functional dependence as well as the magnitude of our results differ significantly: in [5,7] it was
found for partons in the adjoint representation that

3/2 (3/4)
gY2 M N T 3
qLRW =
(78)
(5/4)
in the light-like limit, v 1. Combining (78) with either (74) or (75) leads to
obvious: qLRW 2.6 GeV2 /fm,
alternative: qLRW 0.61 GeV2 /fm.
(79)
Again it is worth noting the sensitive dependence (TQCD /250 MeV)3 of qLRW . It is also important
to recall that qLRW is defined differently from qT , in terms of a partially light-like Wilson loop
rather than momentum broadening.
As a consistency check on the comparison of qT to a transport quantity used in modeling a
real-world quarkgluon plasma, one should inquire whether the path lengths of real charm quarks
are longer than the typical time scale t of force-force correlators. An estimate of this time scale
can be read off from the quantity discussed following (69) as t = zH /yS . Combining this
with either (74) or (75) leads to

obvious: t = (1.0 fm/c) E/10 GeV,

alternative: t = (1.3 fm/c) E/10 GeV.
(80)
Clearly these times are short compared to the lifetime of the QGP, which is roughly 6 fm/c
in central collisions. But it is more relevant to compare to the mean path length of the charm
189
quarks which escape from the QGP and are detected. This is presumably bounded below by
the corresponding quantity for hard partons without a flavor tag, estimated in [47] as mean =
2.5 fm. Perhaps an even more relevant comparison is the relaxation time tD = 1/D for charm
quarks as calculated via the trailing string picture. This quantity was estimated in [34] as tD =
0.6 fm/c using the obvious prescription and 2.1 fm/c using the alternative prescription. The
comparison with tD is the most stringent:

t
obvious:
= 1.7 E/10 GeV,
tD

t
= 0.63 E/10 GeV.
alternative:
(81)
tD
The conclusion from (81) is that, depending somewhat on assumptions, time correlations in the
fluctuating forces on a charm quark probably matter, especially at higher energies. It would
be useful to inquire what happens when the quark mass is made finite explicitly in the string
construction by inserting a flavor brane as in [1,48].
For b quarks, whose mass is roughly 4.8GeV, time correlations are less important for the
energies attainable at RHIC: t scales as 1/ m while tD scales as m, so for fixed energy, the
ratio t /tD is decreased by a factor of (mb /mc )3/2 6.3.
Correlation times may also help clarify the relation with [5,7]: t diverges as v 1, so if the
path length is held fixed as this limit is taken, it conflicts strongly with the 0 limit in (72).
It is tempting to speculate that somehow cutting off the integral (9) at finite time would allow
one to define a q-like
quantity with a finite v 1 limit.

A characteristic feature of radiative descriptions of energy loss following [3941] is the scaling of E not as path length but as 2 . The BrodskyHoyer inequality [49],
1 2
E < q
(82)
,
2
derived from uncertainty principle considerations, is saturated to within a factor of order unity
in the BDMPS formalism. This is in contrast to a Langevin description including drag force,
where E . As a rough comparison between a radiative description and the trailing string
description, it is interesting to ask above what critical path-length c the inequality (82) is satisfied if one extracts the left-hand side from the drag force (ignoring longitudinal fluctuations in
the force) and sets q = qT . The answer is
v2
v2
c =
(83)
.
=
2T
2T yS
For pc = 10 GeV, one finds c = 1.0 fm using the obvious scheme and c = 1.4 fm using
the alternative scheme. In principle one can distinguish between predictions of radiative and
Langevin descriptions of energy loss by making independent measurements of energy loss and
momentum broadening. It has already been suggested [50] in the context of a Langevin description that two-point correlators in azimuthal angle for DD pairs could provide some insight into
the drag force. Further investigation of the potential uses of heavy quark two-point correlators to
distinguish between competing models of energy loss would clearly be of interest.
The trailing string makes the distinctive prediction that, for heavy quarks,
both the average
drag force and fluctuations are enhanced by powers of the Lorentz factor 1/ 1 v 2 for v close
to 1. This clearly implies that heavy quarks feel dissipative effects more strongly when they are
more energetic. So the expectation from the trailing string picture is that RAA as a function of pT
190
for non-photonic electrons has a more negative slope than predicted by treatments such as [51] in
which q is held fixed (i.e., independent of pT ). It is therefore gratifying to observe in Fig. 3 of [33]
a persistently negative slope of RAA over a wide range of momenta, 1.5 GeV/c < pT < 7 GeV/c.
The predictions of [51] include a weaker dependence of RAA on pT over this range.6 The predictions of [38] are also rather flat in this interval, despite a dependence of D (p) on p that is only
slowly decreasing in the range of interest. The predictions of [35], which, like [38], are based
on a Langevin approach, have a significantly larger momentum range in which RAA is strongly
decreasing, and D (p) as presented in the closely related work [37] is again slowly decreasing in
the range of interest. Only three points, with pT roughly between 4.8 and 6.5 GeV/c, lie below
the predictions of [35]; the same three, plus perhaps one other at pT 4.2 GeV/c, lie below the
predictions of [51]. In summary, the comparison of RAA as reported in [33] with existing phenomenological models reinforces the view that dissipative effects increase with the momentum
of the heavy quark; but without some more systematic study, it is hard to tell whether the trailing
string picture does a better job of explaining the data than other approaches.
Results from the STAR collaboration [32] agree fairly well with [51] if the contribution from
bottom quarks is removed. But in light of the comparisons of pp eX PHENIX data [52]
with fixed-order-plus-next-to-leading-log calculations [53], it seems ad hoc to neglect bottom
contributions completely. The authors of [51] cogently warn of the difficulty of disentangling the
contributions of charm and bottom in the absence of vertex reconstruction. In any case, the STAR
data are at least consistent with a more negative slope for RAA as a function of pT than either the
c + b or c only curves from [51].
Acknowledgements
I thank C. Herzog, H. Liu, J. Maldacena, G. Michalogiorgakis, C. Nayak, K. Rajagopal,
U. Wiedemann, and especially J. Casalderrey-Solana and D. Teaney for useful discussions.
I am grateful to B. Zajc for pointing out reference [50] to me, for correspondence regarding
the electrons typical momentum fraction in D-meson decays, and for comments regarding the
interesting pT dependence of RAA for non-photonic electrons as reported in [33]. I am indebted
to J. Friess, G. Michalogiorgakis, and S. Pufu for collaboration on work that led to the results
appearing in Appendix A. This work was supported in part by the Department of Energy under
Grant No. DE-FG02-91ER40671, and by the Sloan Foundation.
Appendix A. Infalling and outgoing coordinates
A crucial property of the wave-functions R (, t, y) appearing in (25) is that they depend
only on the infalling coordinate close to the horizon. The purpose of this appendix is to demonstrate that the infalling and outgoing
coordinates are quoted correctly in (30), to leading order
4
in small yS y, where as usual yS = 1 v 2 . Actually, I will do somewhat more by explicitly
constructing coordinates in which the worldsheet metric is seen to be locally conformal to R1,1
(as any two-dimensional metric with + signature must be).
6 In referring to the predictions of one or another theoretical study, I am relying upon the portrayal of those predictions
in [33] or [32]. This means, in particular, a choice q = 14 GeV2 /fm in the calculations of [51].
191
It is convenient to start by parametrizing the worldsheet using x 1 and y rather than t and y.
The worldsheet metric is easily seen to be

L2
h 1 2 2zH y 2 1
2
2
2
ds2 = g d d = 2
(A.1)
dx dy + zH dy ,
1 2 dx
v
v
zH y 2
where as usual h = 1 y 4 . The manipulations to cast the metric (A.1) in a conformally flat form
are relatively simple because g depends on y but not x 1 :
2

L2
L2
zH y 2 /v
y 4 /v 2
h
2
2
1
dy
dy + 2
1 2
dx
ds2 = 2 1
y
1 h/v 2
v
1 h/v 2
zH y 2
2
L2 1 v 2 2
L2
2
h
v
dq
dy
2 y2
y 2 h v2
v 2 zH

= 2 dq 2 + d2 ,
(A.2)
where
2 =
L2 h v 2
2
y 2 v 2 zH
and I have defined new coordinates

zH y 2 v
dy
1
2
q = x + dy
,
= vzH 1 v
.
h v2
h v2
(A.3)
(A.4)
Because h v 2 = yS4 y 4 , the integrals in (A.4) diverge at the horizon. So one must define q
and piecewise in two regions:

1+y/yS
zH v
S
qI = x 1 + 4iy
(log 1iy/y
1+iy/yS + i log 1y/yS )
S
(A.5)
for 0 < y < yS ;
1+y/yS
zH v
S
( log 1iy/y
I = 4iy
1+iy/yS + i log 1y/yS )
S

1iy/yS
1+y/yS
zH v
qII = x 1 + 4iy
(log
+
i
log
)
1+iy/y
1+y/y
S
S
S
for yS < y < 1.
(A.6)
1+y/yS
zH v
S
(log 1iy/y
II = 4iy
1+iy/yS i log 1+y/yS )
S
The second form of the metric in (A.2) is analogous to the standard form of the Schwarzschild
metric. In region I (outside the horizon) y is the spacelike variable and q is the timelike variable.
In region II (inside the horizon) y is timelike and q is spacelike.
Using x 1 = vt + (y), one may extract expressions for qI in terms of t and y rather than
1
x and y; I , on the other hand, is a function only of y. To construct infalling and outgoing
coordinates in region I, one simply forms
q = q I I .
Expanding in small yS y, one finds

zH
q = vt + ,
q+ = v t
log(yS y) + ,
2yS
(A.7)
(A.8)
where the omitted terms comprise constant terms (i.e., independent of t and y) and terms involving positive powers of yS y. Up to the overall factor of v, the expressions in (A.8) are precisely
the ones given in (30).
192
Appendix B. Relations between real-time thermal propagators

The usual relations among the Greens functions defined in (32) may be summarized as follows7 :
Im GR (),
GF () = Re GR () + i coth
2T
G() = Im GF () = coth
(B.1)
Im GR ().
2T
The first equality in (B.1) usually can be derived by inserting complete sets of states; the second
is immediate from the definitions; and the third is an obvious corollary of the first. However, the
usual derivation of the first equality relies crucially on having a fully equilibrated system, where
Q = tr(eH Q)/ tr eH for any operator Q, and the heavy quark is far from equilibrated.
It turns out that a very simple modification
of (B.1) suffices: one must make the replacement
4
T T yS in (B.1), where yS = 1 v 2 . This applies equally for transverse and longitudinal
fluctuations, and I will not need to distinguish between them in this appendix.
A derivation of the modified form of (B.1) may be carried out using the SchwingerKeldysh
formalism, as adapted to AdS/CFT in [54] following earlier work [5557]. For simplicity, let us
set zH = 1, so T = 1/ . First define Kruskal coordinates U and V such that
UV =
y 1 2 tan1 y
,
e
y+1
V
= e4t .
U
(B.2)
To make the most efficient use of these relations, it helps to note the following points:
1. One may extend the second relation in (B.2) over the entire Penrose diagram by assigning
t an imaginary part which is piecewise constant but changes by an odd-integer multiple of
i/4 as U or V switches sign. A consistent way of doing so is exhibited in Fig. 2(a).
2. The first relation in (B.2) defines a monotonically increasing map from U V (1, e ) to
y (0, ), and U V = 0 maps to y = 1.
3. As a result of the previous two points, the relations (B.2) define a one-to-one map (U, V )
(t, y), where the domain of (U, V ) is points in R2 subject only to the restriction 1 < U V <
e , and the range of (t, y) has four disjoint pieces corresponding to the four regions (RFLP)
in Fig. 2(a). The R piece, for example, comprises points in R2 subject only to the restriction
0 < y < 1.
The trailing string solution (17) may be recast in the form
v
log V + v tan1 y.
(B.3)
2
(Recall that zH = 1.) In this form it is clear that the trailing string is entirely non-singular at the
horizon between the R and F regions. The singularity at V = 0, corresponding to Re t ,
is genuine. In the spirit of the SchwingerKeldysh formalism, one prepares a state of the system
at Re t = and propagates it first forward along Im t = 0 and then back along Im t = /2.
x1 =
7 I am indebted to J. Casalderrey-Solana and D. Teaney for pointing out that the second equality in Eq. (25) of the
original version of this paper was in error, and for private communications related to the contents of this appendix. The
treatment in this appendix follows in its essential points that of [45].
193
Fig. 2. (a) The Penrose diagram for AdS5 -Schwarzschild with conventions on the phase of t specified. Increasing the
real part of t corresponds to moving upward on the R boundary and downward on the L boundary. Increasing y means
moving to the left in R and to the right in L. (b) The odd-numbered regions are narrow slices of the Penrose diagram
corresponding to series expansions around y = 0, yS , and 1. The even-numbered regions fill in between these series
The corresponding regions 1 to 6 are the reflections
expansions. Note that 3 and 3 are disjoint, but 5 is a subset of 5.
of 1 to 6 through the point U = V = 0.
Correspondingly, one may use (B.3) in the R and F regions and

v
log(V ) + v tan1 y iv/2
(B.4)
2
in the L and P regions. The two solutions B.3 and B.4 can be matched at V = 0 by continuing
V through the lower half-planethat is, for Im V slightly negative. Such a continuation is consistent with the conventions laid out in Fig. 2 because in the L region one may rewrite (B.4) in
terms of t and y in precisely the original form (17).
Let us now consider perturbation wave-functions for oscillations of the string in a slightly
more general fashion than in Section 3. The index will run over values in a rather large set:
x1 =
4, 5, 5,
6}.
{1, 2, 3, 3,
(B.5)
These values correspond to different regions of the Penrose diagram, as indicated in Fig. 2(b). In
each of the odd-numbered regions, which are all narrow slices near some definite value of y, one
may expand the general solution to the wave equation for to leading order as follows:

1 = A1 + B1 y 3 eit ,
1 = A1 + B1 y 3 eit ,

3 = A3 + B3 (yS y)i/2yS eit ,
3 = A3 + B3 (yS y)i/2yS eit ,

3 = A3 + B3 (y yS )i/2yS eit ,
3 = A3 + B3 (y yS )i/2yS eit ,

5 = A5 + B5 (1 y) (1 y)i/4 eit ,

5 = A5 + B5 (1 y) (1 y)i/4 eit ,

5 = A5 + B5 (1 y) (V )i/2 .
5 = A5 + B5 (1 y) V i/2 ,
(B.6)
For brevity I have dropped subleading terms from the A-type solutions in (B.6) even when they
are bigger than the B-type solutions. There is no loss of generality in assuming eit dependence
because the entire spacetime has a timelike Killing vector which is /t in each patch.
194
The expressions in (B.6) involve imaginary powers of quantities which are real and positive.
Such expressions are unambiguously defined: x i ei log x . To specify globally one must
prescribe via some analytic continuation how the pairs (3 , 3 ), (3 , 3 ), and (5 , 5 ) fit
together. Also one must solve the wave equation for in the intermediate regions 2, 4, either
through some approximation or via numerics.8 The result of such prescriptions and approximate
solutions can be cast in terms of 2 2 matrices:

A
v = M v , where v =
(B.7)
.
B
Here and are in the set (B.5), excluding the even numbered values. Note that M for each
and is a 2 2 matrix: and label the matrix, not its components. An obvious property of
the M is M = M1
.
The properties of the matrices M that I will require are:
M1,3 = M13 ,
M3,5
= M35
,

1
0
1
0
,
M3,3 =
,
M33 =
0 e/2yS
0 e/2yS
/2 i/4
/2 i/4

e
e
1 0
/2 1
e
,
M5,5 =
M55 =
0 1
0
2
2

/2 1 0
,
M5,
5 = e
0 1

1
1
13 ,
M
M13 D =
CR () CR () iX
(B.8)
(B.9)

0
, (B.10)
1
(B.11)
(B.12)
where D is some diagonal matrix, CR () is the same quantity as introduced in (28), and X =
2v 2 zH /3(1 v 2 ) as in (61). I will justify (B.8)(B.12) after showing how to obtain the first
equation in (B.1) from them.
To calculate SchwingerKeldysh propagators G (), with and equal to 1, one should
specify A1 () and A1 () and calculate the response functions, B1 () and B1 (). To this
end one requires the matrix
M1,1 = M1,3 M3,3 M3,5
M5,5 M5,
5 M55
M53 M33
M31

2
0
13 1 /y
1 .
= M13 M1
M31 = M
M
13
S
33
0 e
(B.13)
The second equality in (B.13) uses (B.8)(B.11); the third equality also uses (B.12) and the
fact that D commutes with the diagonal matrix M33 . Using the relation v1 = M1,1 v1 and the
prescribed values of A1 , one may extract B1 . For example, if A1 = 0, then
B1 () = CF ()A1 (),
where CF () =
CR ()e/yS CR () + iX
.
e/yS 1
(B.14)
8 One could also find numerical solutions in region 6 and perhaps match to some expansion near the future and past
spacelike singularities. This would perhaps be of interest in an exploration of the physics of the black hole singularity,
and it could also be of some relevance in computing higher point functions of operators pertaining to the heavy quark,
but such matters are beyond the scope of the current work.
195
A rephrasing of the result (B.14) is that the wave-function F (, t, y) suitable for computing
the time-ordered propagator GF () = G11 () has leading behavior

GF () = 1 + CF ()y 3 eit
(B.15)
near the boundary of the R region. Then, in precise analogy to (34),
GF () = G y log F (, t, y)|y=0

CR ()e/yS CR ()
= Y iX coth
+2
,
2yS
e/yS 1
where Y is defined as in (61). Comparing with (60), one sees immediately that
Im GR ().
GF () = Re GR () + i coth
2yS
(B.16)
(B.17)
Restoring factors of zH amounts to replacing by 1/T in (B.17), so this is the desired modification of the first equation in (B.1).
Now let us return to the justification of the formulas (B.8)(B.12). The main issue is continuing from 3 to 3 and from 5 to 5 . An argument was made in [54], and extended
to the present context in [45], that the correct way to do this is to continue in the lower half
of the complex V plane and the upper half of the complex U plane when horizon singularities
are encountered. By so doing one imposes infalling conditions on positive frequency modes and
outgoing conditions on negative frequency modes. Formally one may simply send U U + i
and V V i for small positive . In R this corresponds to sending y y + i, and in L
it corresponds to sending y y i. Consider the consequences for matching the oscillatory
which takes place wholly within the R region:
solution from 3 to 3,
lim B3 (yS y i)i/2yS = lim B3 (y yS + i)i/2yS ,
yyS
yyS +
(B.18)
where the first limit is from below and the second is from above. From (B.18) one obtains
B3 = e/2yS B3 , whence the lower right entry in M33 . The opposite behavior arises in M3,3
because y y i in the L region. In crossing the V = 0 line, one finds (V i)i/2 =
e/2 (V + i)i/2 . This completes a justification of (B.9) and (B.11). The other steps can
be justified more straightforwardly, as follows:
(B.8) is a consequence of the wave equation for being the same in regions 2 and 4.
The non-zero imaginary part of t does not affect the equation of motion.
(B.10) is a consequence of applying the rule x i = ei log x only when x is positive. The
details are somewhat intricate, so I will trace the steps for the A-type solution from 5
to 5:
A5 (1 y)i/4 eit = A5 (1 y)i/4 ei Re t e/2
/2 i/4
e
= A5 (V )i/2
e/2
2
= A5 (V )i/2 ,
(B.19)
where in the second equality I used the leading behavior of V near y = 1 in the L region:
(V )2 = (1 y)e4t
e/2
e/2
= (1 y)e4 Re t
.
2
2
(B.20)
196
The first and last expressions in (B.20) are in a form suitable to be raised to the imaginary
power i/4.
(B.12) is a consequence of the behavior of purely infalling and purely outgoing solutions of
the wave equation for . An infalling solution is one with B3 = 0, and its behavior the R
13 . An outgoing
boundary, (28), corresponds to v1 given by the first column of the matrix M
solution should have behavior at the R boundary corresponding to v1 given by the second
13 . It so happens that if R (, t, y) = eit R (, y) is the infalling wavecolumn of M
function constructed in Section 3, then
A (, t, y) = eit f (y)i/2 f (y/yS )i/2yS R (, y)
(B.21)
is a purely outgoing solution, where

f (y)
1 + y 2 tan1 y
.
e
1y
Expanding (B.21) for small y gives

A (, t, y) = 1 + CR () iX y 3 eit ,
(B.22)
(B.23)
as desired.
Appendix C. Speed limits on single quarks
The string theory calculations presented in this paper pertain to external quarks: pointlike
objects in the fundamental representation which have infinite mass associated with their nearfield Coulombic color-electric field. Comparisons with QCD are thus better justified for c and b
quarks, whose mass is well above the typical temperature of the QGP, than for light quarks.
If the velocity of an external quark is too close to unity, there is a new reason to be wary
of trailing string computations: the horizon on the worldsheet is very close to the boundary,
where in a holographic representation of real QCD (assuming there is one) space presumably
becomes highly curved, and the simplest calculations based on the NambuGoto action in AdS5 Schwarzschild may experience significant corrections. Let us try to estimate when problems of
this sort might start to arise. It was suggested already in [24] to associate an energy scale

2
2
L /zH 1 T gY M N
=
(C.1)
=
2 y
2
y
with a radius y in AdS5 -Schwarzschild. This is justified by observing that a static string dangling
from y into the horizon has mass

L2 /zH 1
mstatic =
(C.2)
1
for y 1.
2 y
Now suppose one identifies a scale fail at which AdS/CFT techniques (at least those based on
supergravity and classical strings) start to fail. From (C.1) one extracts a corresponding yfail , and
if the worldsheet horizon has yS < yfail there may be significant corrections to the trailing string
results. Setting yS = yfail and using (23), one finds that the Lorentz factor of the heavy quark that
the trailing string purports to describe is

fail 2
4
.
fail = 2
(C.3)
T
gY M N
197
The trouble with (C.3) is that gY2 M N , fail and T all incorporate considerable uncertainties.
To get an idea of the range of plausible values for fail , let us consider the two prescriptions (74) and (75). I do not have a systematic way of determining fail , but perhaps a reasonable
range to consider is fail = 0.81.6 GeV. Taking fail = 1.2 GeV as a representative value, one
finds

2
fail
,
obvious: fail = 4.9
1.2 GeV

2
fail
alternative: fail = 29
(C.4)
.
1.2 GeV
Because of the quadratic dependence on both fail and T (not to mention the choice of comparison scheme) fail remains substantially uncertain. Evidently, if the obvious prescription is
used, there is some doubt cast on stringy predictions for charm quark when pc 6.7 GeV/c,
corresponding to pT 3.4 GeV. In the alternative scheme, there is less reason to worry about
stringy corrections near the upper end of the trailing string.
Another thing can go wrong if one wants to represent a finite mass quark as a string ending on a D7-brane, as in [1,48]. Given mc for the charm quark, one can use (C.2) to obtain a
corresponding position
yc =
1
1 + 2mc /T
(C.5)
gY2 M N
of the D7-brane. If yS < yc , then there is no horizon on the trailing string: the boundary of the
worldsheet is spacelike. I regard this as a pathology which probably invalidates the trailing string
picture for Lorentz factors > c 1/yc2 . Estimates of c suffer from the same ambiguities as
fail , as discussed above. Using mc = 1.4 GeV and either (74) or (75), one arrives at the estimates
obvious: c = 13,
alternative: c = 53.
(C.6)
These values should again be regarded as incorporating considerable uncertainties: for example,
it is puzzling that the horizon causes the mass of a quark to decrease from its zero-temperature
value, but this decrease is what makes the values in (C.6) higher than those derived from (C.4)
with fail mc . The main lesson to draw from (C.6) is that charms mass is high enough to
avoid threatening the existence of the worldsheet horizon for the momenta accessible at RHIC;
but attempts to treat light quarks in terms of the trailing string construction are perilous indeed.9
References
[1] C.P. Herzog, A. Karch, P. Kovtun, C. Kozcaz, L.G. Yaffe, Energy loss of a heavy quark moving through N = 4
supersymmetric YangMills plasma, hep-th/0605158.
[2] S.S. Gubser, Drag force in AdS/CFT, Phys. Rev. D 74 (2006) 126005, hep-th/0605182.
[3] J. Casalderrey-Solana, D. Teaney, Heavy quark diffusion in strongly coupled N = 4 YangMills, hep-ph/0605199.
9 For example, in the alternative comparison scheme, an up quark whose mass in the medium is assumed to be
m = 300 MeV would be described in terms of a D7-brane with yu = 0.43, so u = 5.5 is the maximum Lorentz factor. It
is hardly fair to ignore corrections to relativistic dispersion relations in this context, but if one does so, the result is a total
energy of 1.7 GeV. As an upper limit on allowed energies for partons described by trailing strings, this is fairly anemic.
198
[4] S.-J. Sin, I. Zahed, Holography of radiation and jet quenching, Phys. Lett. B 608 (2005) 265, hep-th/0407215.
[5] H. Liu, K. Rajagopal, U.A. Wiedemann, Calculating the jet quenching parameter from AdS/CFT, hep-ph/0605178.
[6] M. Chernicoff, J.A. Garcia, A. Guijosa, The energy of a moving quark-antiquark pair in an N = 4 SYM plasma,
JHEP 0609 (2006) 068, hep-th/0607089.
[7] H. Liu, K. Rajagopal, U.A. Wiedemann, Wilson loops in heavy ion collisions and their calculation in AdS/CFT,
hep-ph/0612168.
[8] P.C. Argyres, M. Edalati, J.F. Vazquez-Poritz, Spacelike strings and jet quenching from a Wilson loop, hep-th/
0612157.
[9] A. Buchel, On jet quenching parameters in strongly coupled non-conformal gauge theories, Phys. Rev. D 74 (2006)
046006, hep-th/0605178.
[10] E. Nakano, S. Teraguchi, W.-Y. Wen, Drag force, jet quenching, and AdS/QCD, hep-ph/0608274.
[11] P. Talavera, Drag force in a string model dual to large-N QCD, hep-th/0610179.
[12] Y.-h. Gao, W.-s. Xu, D.-f. Zeng, Jet quenching parameters of SakaiSugimoto model, hep-th/0611217.
[13] C.P. Herzog, Energy loss of heavy quarks from asymptotically AdS geometries, JHEP 0609 (2006) 032, hepth/0605191.
[14] E. Caceres, A. Guijosa, Drag force in charged N = 4 SYM plasma, JHEP 0611 (2006) 077, hep-th/0605235.
[15] F.-L. Lin, T. Matsuo, Jet quenching parameter in medium with chemical potential from AdS/CFT, Phys. Lett. B 641
(2006) 45, hep-th/0606136.
[16] S.D. Avramis, K. Sfetsos, Supergravity and the jet quenching parameter in the presence of R-charge densities,
hep-th/0606190.
[17] N. Armesto, J.D. Edelstein, J. Mas, Jet quenching at finite t Hooft coupling and chemical potential from AdS/CFT,
JHEP 0609 (2006) 039, hep-ph/0606245.
[18] E. Caceres, A. Guijosa, On drag forces and jet quenching in strongly coupled plasmas, hep-th/0606134.
[19] J.F. Vazquez-Poritz, Enhancing the jet quenching parameter from marginal deformations, hep-th/0605296.
[20] J.J. Friess, S.S. Gubser, G. Michalogiorgakis, Dissipation from a heavy quark moving through N = 4 super-Yang
Mills plasma, JHEP 0609 (2006) 072, hep-th/0605292.
[21] J.J. Friess, S.S. Gubser, G. Michalogiorgakis, S.S. Pufu, The stress tensor of a quark moving through N = 4 thermal
plasma, hep-th/0607022.
[22] S.-J. Sin, I. Zahed, Amperes law and energy loss in AdS/CFT duality, hep-ph/0606049.
[23] M. Chernicoff, A. Guijosa, Energy loss of gluons, baryons and k-quarks in an N = 4 SYM plasma, hep-th/0611155.
(1998) 231, hep-th/9711200.
[25] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string theory, Phys. Lett.
B 428 (1998) 105, hep-th/9802109.
[26] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[27] O. Aharony, S.S. Gubser, J.M. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity, Phys.
Rep. 323 (2000) 183, hep-th/9905111.
[28] E. DHoker, D.Z. Freedman, Supersymmetric gauge theories and the AdS/CFT correspondence, hep-th/0201253.
[29] I.R. Klebanov, TASI lectures: Introduction to the AdS/CFT correspondence, hep-th/0009139.
[30] S.-J. Rey, J.-T. Yee, Macroscopic strings as heavy quarks in large N gauge theory and anti-de Sitter supergravity,
Eur. Phys. J. C 22 (2001) 379, hep-th/9803001.
[31] J.M. Maldacena, Wilson loops in large N field theories, Phys. Rev. Lett. 80 (1998) 4859, hep-th/9803002.
[32] STAR Collaboration, B.I. Abelev, et al., Transverse momentum and centrality dependence of high-p(T) non
photonic electron suppression in Au + Au collisions at sN N = 200 GeV, nucl-ex/0607012.
[33] PHENIX Collaboration, A. Adare, et al., Energy loss and flow of heavy quarks in Au + Au collisions at sN N =
200 GeV, nucl-ex/0611018.
[34] S.S. Gubser, Comparing the drag force on heavy quarks in N = 4 super-YangMills theory and QCD, hepth/0611272.
[35] H. van Hees, V. Greco, R. Rapp, Heavy-quark probes of the quarkgluon plasma at RHIC, Phys. Rev. C 73 (2006)
034913, nucl-th/0508055.
[36] B. Svetitsky, Diffusion of charmed quark in the quarkgluon plasma, Phys. Rev. D 37 (1988) 24842491.
[37] H. van Hees, R. Rapp, Thermalization of heavy quarks in the quarkgluon plasma, Phys. Rev. C 71 (2005) 034907,
nucl-th/0412015.
[38] G.D. Moore, D. Teaney, How much do heavy quarks thermalize in a heavy ion collision? Phys. Rev. C 71 (2005)
064904, hep-ph/0412346.
[39] R. Baier, Y.L. Dokshitzer, A.H. Mueller, S. Peigne, D. Schiff, Radiative energy loss of high energy quarks and
gluons in a finite-volume quarkgluon plasma, Nucl. Phys. B 483 (1997) 291, hep-ph/9607355.
199
[40] R. Baier, Y.L. Dokshitzer, A.H. Mueller, S. Peigne, D. Schiff, Radiative energy loss and p(T)-broadening of high
energy partons in nuclei, Nucl. Phys. B 484 (1997) 265, hep-ph/9608322.
[41] B.G. Zakharov, Radiative energy loss of high energy quarks in finite-size nuclear matter and quarkgluon plasma,
JETP Lett. 65 (1997) 615, hep-ph/9704255.
[42] K. Peeters, J. Sonnenschein, M. Zamaklar, Holographic melting and related properties of mesons in a quarkgluon
plasma, Phys. Rev. D 74 (2006) 106008, hep-th/0606195.
[43] H. Liu, K. Rajagopal, U.A. Wiedemann, An AdS/CFT calculation of screening in a hot wind, hep-ph/0607062.
[44] D.T. Son, A.O. Starinets, Minkowski-space correlators in AdS/CFT correspondence: Recipe and applications,
JHEP 0209 (2002) 042, hep-th/0205051.
[45] J. Casalderrey-Solana, D. Teaney, Transverse momentum broadening of a fast quark in a N = 4 YangMills plasma,
hep-th/0701123.
[46] C.G. Callan, A. Guijosa, Undulating strings and gauge theory waves, Nucl. Phys. B 565 (2000) 157, hep-th/9906153.
[47] A. Dainese, C. Loizides, G. Paic, Leading-particle suppression in high energy nucleusnucleus collisions, Eur. Phys.
J. C 38 (2005) 461, hep-ph/0406201.
[48] A. Karch, E. Katz, Adding flavor to AdS/CFT, JHEP 0206 (2002) 043, hep-th/0205236.
[49] S.J. Brodsky, P. Hoyer, A Bound on the energy loss of partons in nuclei, Phys. Lett. B 298 (1993) 165, hep-ph/
9210262.
[50] K. Schweda, et al., D anti-D correlations as a sensitive probe for thermalization in high-energy nuclear collisions,
nucl-ex/0610043.
[51] N. Armesto, M. Cacciari, A. Dainese, C.A. Salgado, U.A. Wiedemann, How sensitive are high-p(T ) electron spectra
at RHIC to heavy quark energy loss? Phys. Lett. B 637 (2006) 362, hep-ph/0511257.
[52] PHENIX Collaboration, A. Adare, et al., Measurement of high-p(T ) single electrons from heavy-flavor decays in
p + p collisions at s = 200 GeV, hep-ex/0609010.

[53] M. Cacciari, P. Nason, R. Vogt, QCD predictions for charm and bottom production at RHIC, Phys. Rev. Lett. 95
(2005) 122001, hep-ph/0502203.
[54] C.P. Herzog, D.T. Son, SchwingerKeldysh propagators from AdS/CFT correspondence, JHEP 0303 (2003) 046,
hep-th/0212072.
[55] G.T. Horowitz, D. Marolf, A new approach to string cosmology, JHEP 9807 (1998) 014, hep-th/9805207.
[56] J.M. Maldacena, Eternal black holes in anti-de Sitter, JHEP 0304 (2003) 021, hep-th/0106112.
[57] V. Balasubramanian, P. Kraus, A.E. Lawrence, S.P. Trivedi, Holographic probes of anti-de Sitter spacetimes, Phys.
Rev. D 59 (1999) 104021, hep-th/9808017.
Single top production at HERA in the Standard Model

and its minimal supersymmetric extension
W. Hollik a , A. Httmann b,1 , B.A. Kniehl b,
a Max-Planck-Institut fr Physik (Werner-Heisenberg-Institut), Fhringer Ring 6, 80805 Munich, Germany
b II. Institut fr Theoretische Physik, Universitt Hamburg, Luruper Chaussee 149, 22761 Hamburg, Germany
Received 7 June 2006; received in revised form 23 July 2007; accepted 12 September 2007
Abstract
The H1 Collaboration at the DESY electronproton collider HERA has observed, in photoproduction and
neutral-current deep-inelastic scattering, an unexpected excess of events with isolated leptons and missing
transverse momentum, especially at large values of hadronic transverse momentuma signature typical for
single top-quark production. This observation is being substantiated in the HERA II run. Motivated by this,
we evaluate the cross section of single top-quark photo- and electroproduction both in the Standard Model
and its minimal supersymmetric extension, considering both minimal and non-minimal flavour-violation
scenarios in the latter case.
PACS: 12.60.Jv; 13.60.Hb; 13.85.Ni; 14.65.Ha
1. Introduction
Searches for single top-quark production via the neutral current (NC),
e p e t + X,
(1)
have been performed by the H1 Collaboration [1] and the ZEUS Collaboration [2] at the DESY
electronproton collider HERA. The H1 Collaboration found several events, leading to a cross
section of = 0.29+0.15
0.14 pb. Alternatively, assuming that the observed events are due to a statis* Corresponding author.
E-mail address: kniehl@desy.de (B.A. Kniehl).

1 Present address: Deutsches Elektronen-Synchrotron (DESY), Notkestr. 85, 22607 Hamburg, Germany.
doi:10.1016/j.nuclphysb.2007.09.004
201
tical fluctuation, upper limits of 0.55 pb on and of 0.27 on the anomalous tu coupling tu
were established at the 95% confidence level (CL). On the other hand, the ZEUS Collaboration
found no evidence for top-quark production and was able to place upper bounds of 0.225 pb on
and of 0.174 on tu at 95% CL.
Single top-quark production via the charged current (CC) is possible at the tree level, but its
cross section is less than 1 fb [3]. Furthermore, such events can be separated experimentally due
to the absence of a scattered electron or positron in the final state.
In this work, we calculate the cross section of process (1) in the Standard Model (SM) and
its minimal supersymmetric (SUSY) extension (MSSM), considering both scenarios with minimal flavour violation (MFV) and non-minimal flavour violation (NMFV). The bulk of the cross
section is due to photoproduction and electromagnetic deep-inelastic scattering (DIS), while the
contribution due to the exchange of a virtual Z boson is greatly suppressed by its mass. Since
photonic interactions cannot change flavour and the top quark does not appear as a parton in the
proton, we are dealing here with a loop-induced process. In the SM and the MFV MSSM, its
cross section is further suppressed by the smallness of the contributing elements of the Cabibbo
KobayashiMaskawa (CKM) matrix. In more general MSSM scenarios, misalignment between
the quark and squark sectors can appear, and the CKM matrix is no longer the only source of
flavour violation. Thus, the flavour-changing (FC) couplings are not Cabibbo suppressed, and
sizeable contributions to FC NC processes can occur. On the other hand, there are strong experimental bounds on squark mixing involving the first generation, coming from data on K 0 K 0 and
D 0 D 0 mixing [4]. If the squark mixing involving the first generation is neglected completely,
there are no new vertices for incoming up-quarks compared to the MFV MSSM. Thus, every
new contribution in this scenario is suppressed by the parton distribution function (PDF) of the
charm quark in the proton. Because the SM cross section for NC single top-quark production is
highly suppressed, every detected event is an indication of physics beyond the SM.
This paper is organised as follows. In Section 2, we describe the analytical calculation of the
SM cross section. In Section 3, we outline the theoretical framework of FC interactions in the
MSSM. The numerical analysis is presented in Section 4. Our conclusions are summarised in
Section 5.
2. SM cross section
In this section, the analytical calculation of the SM cross section is described. As indicated
in Fig. 1, we denote the four-momenta of the incoming proton, parton (up or charm quark), and
electron by P , p, and k, respectively, and those of the outgoing top quark and electron by p
and k , respectively. As for the centre-of-mass (CM) energy and the top-quark mass, we have
S = (P + k)2 and m2t = (p )2 . We neglect the masses of the proton, the incoming quarks, and the
electron, so that P 2 = p 2 = k 2 = k 2 = 0. The four-momentum of the exchanged photon is given
by q = k k , and, as usual, we introduce the virtuality variable Q2 = q 2 > 0. The variable
y = (q P )/(k P ) measures the relative electron energy loss in the proton rest frame.
2.1. Electroproduction
Single top-quark production in NC DIS occurs via the partonic subprocess
e q e t,
(2)
202
Fig. 1. Schematic representation of the hadronic process (1) explaining the four-momentum assignments.
Fig. 2. Feynman diagrams of partonic subprocess (2) in the SM. Here, G is the charged Goldstone boson and di with
i = 1, 2, 3 are the down-type quarks.
where q = u, c. The Feynman diagrams contributing in the SM to process (2) with q = u are
depicted in Fig. 2. The ones for q = c are similar. The amplitude of this processes was also
calculated in Ref. [5]. There appears at least one off-diagonal element of the CKM matrix in
each term. If the contributions of the inner-quark flavours to a single Feynman diagram are added
up and their masses are neglected, the amplitude of this Feynman diagram vanishes due to the
203
unitarity of the CKM matrix. Thus, we cannot neglect the inner-quark masses here. Although our
choice mc = 0 < md appears unphysical at first sight, it is inconsequential in practice. In fact, we
verified that the use of a realistic value of mc affects our numerical results only insignificantly.
A detailed proof of hard-scattering factorisation with the inclusion of heavy-quark masses may
be found in Ref. [6].
Our calculation proceeds along the lines of Ref. [7]. The differential cross section of
process (2) reads

d
1
(3)
=
|M|2 yS Q2 m2t ,
2
dQ dy part 16 S
where , defined as p = P , is the fraction of the proton momentum passed on to the incoming
quark and |M|2 is the squared amplitude averaged (summed) over the spin and colour degrees
of freedom of the initial-state (final-state) particles. The kinematically allowed ranges of Q2 and
y are
Q2cut < Q2 < ymax S m2t ,
Q2 + m2t
< y < ymax ,
S
(4)
where Q2cut defines the demarcation between photoproduction and electroproduction and ymax is
an experimental acceptance cut. In Ref. [1], values for Q2cut and ymax are not specified. In our
numerical analysis, we employ the typical values Q2cut = 4 GeV2 and ymax = 0.95. Our numerical results are insensitive to the precise choice of ymax as long it is close to unity. On the other
hand, the dependence on Q2cut approximately cancels out in the combination of photoproduction
and electroproduction, as we explicitly verify.
In the parton model of QCD, the hadronic cross section of process (1) is obtained by convoluting the partonic cross section of process (2) with the appropriate PDF Fq (, F ), where F
is the factorisation scale, and summing over q = u, c, as
d
dQ2 dy

=
hadr
1

q=u,c

d Fq (, F )
d
dQ2 dy

.
(5)
part

There are two candidate mass scales for F ,namely Q2 and mt , and the optimal choice is
likely to lie somewhere in between, at F = ( Q2 + mt )/2 say. At any rate, we have F mc ,
so that charm is an active quark flavour in the initial state, contributing at full strength via its
PDF.
The expression for |M|2 of process (2) may be decomposed into a hadronic tensor H and
a leptonic tensor L , as
|M|2 =
e2
L H ,
Q4
(6)
where e is the positron charge. We have

H =
1
H H ,
2
spins
(7)
204
where
H = u(p )(F1 PL + F2 PR + F3 p PL
+F4 p PR + F5 p PL + F6 p PR )u(p),
(8)
with helicity projectors PL,R = (1 5 )/2 and form factors F1 , . . . , F6 , which follow from
the explicit evaluation of the Feynman diagrams shown in Fig. 2 and their counterparts for an
incoming charm quark. The leptonic tensor may be decomposed into transverse and longitudinal
components, as
L =

Q2

1 + (1 y)2 T 4(1 y) L ,
y2
(9)
where
4Q2
2

p q + p q
2 p p + 2
2
Q + m2t
+ mt

1
2Q2
2Q2
L = 2
p
+
q
p
+
q
.
Q Q2 + m2t
Q2 + m2t
T = g +
Q2
,
(10)
To obtain the transversal and longitudinal parts of the cross section, the hadron tensor is contracted with the transversal and longitudinal parts of the lepton tensor, respectively.
We generate and evaluate the Feynman diagrams in Fig. 2, with the virtual-photon leg amputated, with the help of the program packages FeynArts [8,9] and FormCalc [9,10]. We work in
t HooftFeynman gauge and use dimensional regularisation to extract the ultraviolet (UV) divergences. We perform the PassarinoVeltman reduction and calculate the squared amplitude (6)
using the program package FeynCalc [11]. For the numerical evaluation of the standard scalar
one-loop integrals, we employ the program package LoopTools [10,12]. We perform the numerical integration with the aid of the program package Cuba [13]. As a check, we also calculate the
amplitude of process (2) with the help of the program package FeynCalc and by hand. All three
independent calculations are found to lead to the same result. Furthermore, we verify the cancellation of the UV divergences, current conservation, and the reality of the squared amplitude.
2.2. Photoproduction
In the photoproduction limit, the virtual photon is considered as real, with Q2 = 0, so that the
longitudinal part of the cross section vanishes. In turn, its energy distribution is described in the
WeizsckerWilliams approximation by the electron-to-photon splitting function

1 + (1 y)2 Q2cut
1
1
f (y) =
(11)
ln 2 + 2ym2e
,
2
y
Qmin
Q2cut Q2min
where = e2 /(4) is Sommerfelds fine-structure constant, me is the electron mass, and Q2min =
y 2 m2e /(1 y) corresponds to the kinematic lower bound. Thus, the cross section of process (2)
in photoproduction is given by

d
= f (y) (y),
(12)
dy part
205
with
=

|M |2 yS m2t ,
yS
(13)
where M is the amplitude of q t . The kinematically allowed range of y is

m2t
< y < ymax .
S
(14)
The hadronic cross section is again obtained by convoluting the partonic cross section for a given
incoming quark with the corresponding PDF and summing over the incoming-quark flavours.
Here, the only candidate mass scale for F is of order mt , and it is plausible to choose F = mt /2
so that there is a smooth transition between photoproduction and electroproduction. Again, we
have F mc , so that the use of a charm PDF is justified.
3. Minimal and non-minimal flavour violation in the MSSM
In the MSSM, there are two sources of FC phenomena [14]. The first one is due to flavour
mixing in the quark sector, just as in the SM. It is produced by the different flavour rotations
in the up- and down-quark sectors, and its strength is driven by the off-diagonal CKM matrix
elements. This mixing produces FC electroweak (EW) interaction terms involving CCs, now also
involving charged Higgs bosons, and SUSY EW interaction terms of the charginoquarksquark
type. Thus, the SM Feynman diagrams of Fig. 2 are supplemented by those shown in Fig. 3, in
which either charged Higgs bosons and down-type quarks or charginos and down-type squarks
circulate in the loops. In the MFV MSSM, this is the only source of FC phenomena beyond the
SM.
The second source of FC phenomena, which is present in the NMFV MSSM, is due to the
possible misalignment between the rotations that diagonalise the quark and squark sectors. When
the squark mass matrix is expressed in the basis where the squark fields are parallel to the quark
fields (the super CKM basis), it is in general non-diagonal in flavour space. This quarksquark
misalignment produces new FC terms in NC as well as in CC interactions. In the SUSY QCD
sector, the FC interaction terms involve NCs of the gluinoquarksquark type. In the case of
process (2), this gives rise to the additional Feynman diagrams shown in the first row of Fig. 4.
In the SUSY EW sector, the FC interaction terms involve NCs of the neutralinoquarksquark
and the charginoquarksquark type. The first type appears exclusively due to quarksquark
misalignment, as in the SUSY-QCD case, whereas the second type receives contributions from
both sources, quarksquark misalignment and CKM mixing. The additional Feynman diagrams
involving neutralinoquarksquark interactions are displayed in the second row of Fig. 4.
In order to simplify our analysis in the NMFV MSSM, we take the CKM matrix to be diagonal, so that SM contributions (cf. Fig. 2) and the genuine MFV-MSSM contributions, i.e.,
the charged-Higgsquark contributions and the part of the charginosquark contributions due to
CKM mixing (cf. Fig. 3), are zero. We are then left with the gluinosquark and neutralinosquark
contributions (cf. Fig. 4) and the residual charginosquark contributions.
Furthermore, we assume that the non-CKM squark mixing is significant only for transitions
between the second- and third-generation squarks, and that there is only leftleft (LL) mixing,
given by an ansatz similar as in Ref. [15], where it is proportional to the product of the SUSY
206
Fig. 3. Additional Feynman diagrams of partonic subprocess (2) arising in the MFV MSSM. Here, dis with i = 1, 2, 3
and s = 1, 2 are the down-type squarks, and i with i = 1, 2 are the charginos.
Fig. 4. Additional Feynman diagrams of partonic subprocess (2) arising in the NMFV MSSM. Here, g is the gluino, u a
with a = 1, . . . , 6 are the up-type squarks, and i0 with i = 1, . . . , 4 are the neutralinos.
masses involved. This assumption is theoretically well motivated by the flavour-off-diagonal

squark squared-mass entries that are radiatively induced via the evolution from high energies
down to the EW scale according to the renormalisation group equations (RGEs) [16]. These
RGEs predict that the FC LL entries scale with the square of the soft-SUSY-breaking masses,
in contrast with the leftright (LR) or rightleft (RL) and the rightright (RR) entries, which
scale with one or zero powers, respectively. Thus, the hierarchy LL LR, RL RR is usually assumed. The same estimates also indicate that the LL entry for the mixing between the
second- and third-generation squarks is the dominant one due to the larger quark-mass factors
involved. On the other hand, the LR and RL entries are experimentally more constrained, mainly
by b s data [17]. With the previous assumption, the squark squared-mass matrices in the
207
(u L , cL , tL , u R , cR , tR ) and (dL , sL , bL , dR , sR , bR ) bases can be written as follows:
2
ML,u
0
2
Mu =
mu Xu
0
0
2
ML,d
0
2
Md =
md Xd
0
0
0
2
ML,c
t
LL ML,c ML,t
0
mc Xc
0
0
2
ML,s
bLL ML,s ML,b
0
ms Xs
0
0
tLL ML,c ML,t
2
ML,t
0
0
mt Xt
0
bLL ML,s ML,b
2
ML,b
0
0
mb Xb
mu Xu
0
0
2
MR,u
0
0
0
mc Xc
0
0
2
MR,c
0
md Xd
0
0
2
MR,d
0
0
0
ms Xs
0
0
2
MR,s
0
0
0
mt Xt
,
0
(15)
2
MR,t
0
0
mb Xb
,
0
0
2
MR,b
(16)
where
q

2
2
2
2
2
ML,q
= MQ,q
+ mq + cos(2)MZ T3 Qq sw ,
2
MR,q
= MU2 ,q + m2q + cos(2)Qq sw2 MZ2
(q = u, c, t),
2
2
2
2
2
MR,q
= MD,q
+ mq + cos(2)Qq sw MZ
(q = d, s, b),
Xq = Aq (tan )2T3 ,
(17)
s sectors, respecand tLL and bLL measure the squark flavour mixing strengths in the tc and b
q
tively. As for the SM parameters, mq , T3 , and Qq are the mass, weak isospin, and electric charge
of quark q; MZ is the Z-boson mass; and sw = sin w is the sine of the weak mixing angle w . As
for the MSSM parameters, tan = v2 /v1 is the ratio of the vacuum expectation values of the two
Higgs doublets; is the Higgshiggsino mass parameter; Aq are the trilinear Higgssfermion
are the scalar masses. Owing to SU(2)L invariance, we
couplings; and MQ,q
, MU ,q , and MD,q
have MQ,u
= MQ,d
, MQ,c
= MQ,s
, and MQ,t
= MQ,b
. Further MSSM input parameters include
the EW gaugino masses M1 and M2 , the gluino mass M3 , and the mass MA0 of the CP-odd
neutral Higgs boson A0 .
In order to reduce the NMFV-MSSM parameter space, we make the following simplifying
s sectors coincide and
assumptions. We assume that the flavour mixing strengths in the tc and b
t
b
put = LL = LL for a simpler notation. Obviously, the choice = 0 represents the case of zero
squark flavour mixing. We assume that the various trilinear Higgssfermion couplings coincide
and write A0 = Au = Ac = At = Ad = As = Ab . We assume that the scalar masses coincide
. As for the
thus defining the common SUSY mass scale M0 = MQ,q
= MU ,{u,c,t} = MD,{d,s,b}
2
2
2
gaugino masses, we impose the GUT relation M1 = (5/3)(sw /cw )M2 , where cw = 1 sw2 , while
we treat the gluino mass parameter M3 as independent. We are thus left with eight independent
MSSM parameters, namely, tan , MA0 , M0 , M2 , M3 , A0 , , and .
208
4. Numerical analysis
In this section, we present our numerical results. We adopt the SM parameters from Ref. [18]
and the effective masses of the down-type quarks from Ref. [19]:
= 1/137.03599911,
s = 0.1176,
(h c) = 0.389379323 GeV mb,

2
MW = 80.403 GeV,
md = 0.041 GeV,
me = 0.510998918 MeV,
MZ = 91.1876 GeV,
ms = 0.15 GeV,
mt = 172.7 GeV,
mb = 4.5 GeV.
(18)
Here, mt and mb correspond to pole masses, while md and ms were determined so that their insertion in the perturbative formula for the vacuum polarisation function of the photon reproduces
the result extracted from the total cross section of hadron production in e+ e annihilation, measured as a function of the CM energy, using a subtracted dispersion relation. This set of quark
masses is especially appropriate for quantitative studies in electroweak physics and is frequently
employed in the literature. It is used here for definiteness and convenience. As a matter of principle, the precise definition of quark mass is not yet fixed in an analysis of leading order (LO) in
QCD like ours. In fact, the freedom of choice of quark-mass definition contributes to the theoretical uncertainty. We employ the standard complex parametrisation of the CKM matrix in terms
of three angles 12 , 23 , 13 and a phase 13 ,

c12 c13
s12 c13
s13 ei13
i
i
13
13
V = s12 c23 c12 s23 s13 e
(19)
,
c12 c23 s12 s23 s13 e
s23 c13
s12 s23 c12 c23 s13 ei13 c12 s23 s12 c23 s13 ei13
c23 c13
where cij = cos ij and sij = sin ij , and adopt from Ref. [18] the values
s12 = 0.2272,
s23 = 0.0422,
s13 = 0.0040,
13 = 1.00.
(20)
As for the proton PDFs, we employ the LO set CTEQ6L1 [20] by the Coordinated Theoretical
Experimental
Project on QCD (CTEQ) Collaboration. We choose the factorisation scale to be

F = ( Q2 + mt )/2, with the understanding that Q2 = 0 in the case of photoproduction. At
HERA II, electrons or positrons of energy Ee = 27.6 GeV
collide with protons of energy Ep =
920 GeV in the laboratory frame, yielding a CM energy of S = 319 GeV.
4.1. Standard Model
Figs. 5 and 6 refer to the SM. Fig. 5 shows the total cross sectionof process (1) as well as its
photoproduction and electroproduction components as functions of S. We observe that, for our
choice of Q2cut , the contribution due to photoproduction is approximately twice as large as the one
due to electroproduction. At the CM energy of HERA, the cross section is of order 1010 fb and
thus many orders of magnitude too small to be measurable. A possible future electronproton
supercollider that uses the HERA proton beam with energy Ep = 920 GeV and the electron
beam of the international linear e+ e collider (ILC) with energy Ee = 500 GeV would have a
CM energy of 1357 GeV. At this energy, the cross section is of order 108 fb and likewise not
measurable.
Fig. 6 shows the Q2 distribution of the electroproduction cross section as well as its transversal
and longitudinal parts. We observe that the transversal part makes the major contribution.
Fig. 5. Total cross section and its photoproduction and electroproduction parts in the SM as functions of
209
S.
Fig. 6. Q2 distribution of the cross section and its transversal and longitudinal parts in the SM under HERA experimental
conditions.
4.2. MSSM with minimal flavour violation

For definiteness, we assume that SUSY is broken according to the mSUGRA scenario of a
Grand Unified Theory (GUT). We assign the following default values to the mSUGRA input
parameters at the GUT scale:
tan = 56,
m0 = 1.25 TeV,
A0 = 260 GeV,
sign() = +1,
m1/2 = 140 GeV,

(21)
210
Fig. 7. Total cross section and its photoproduction and electroproduction parts in the MFV MSSM as functions of
S.
where m0 is the universal scalar mass, m1/2 is the universal gaugino mass, and A0 is the universal
trilinear Higgssfermion coupling. These values approximately maximise the cross section of
process (1) and are in accordance with the experimental bounds from b s decay and on the
masses of the various SUSY particles [18]. We calculate the MSSM mass spectrum with the help
of the program package SuSpect [21].
Figs. 710 all show
the total cross section of process (1) in the MFV MSSM. Specifically,
Fig. 7 displays the S dependence, also separately for the photoproduction and electroproduction contributions, while Figs. 810 exhibit, for HERA experimental conditions, the dependencies on tan , m1/2 , and A0 , respectively, also separately for the charged-Higgsboson and
chargino contributions.
From Fig. 7 we observe that the cross section is of order 105 fb at HERA energy and of order
3
10 fb for the future electronproton supercollider mentioned above. Both values are too small
to yield measurable results.
From Fig. 8 we learn that, as tan approaches its upper limit, the cross section strongly
increases and is mainly generated by the loop diagrams involving charged Higgs bosons. For
small values of tan , the SM contribution (1010 fb) is dominant. Negative interference effects
between the charged-Higgs, chargino, and SM contributions can be seen at large values of tan .
From Fig. 9 we see that, as m1/2 approaches its lower limit, the cross section strongly increases
and is essentially made up by the charged-Higgsboson contribution alone. The latter is dominant
throughout the whole m1/2 range, but there are negative interference effects for all values of m1/2 .
The significant suppression of the chargino contribution for tan = 56 familiar from Fig. 8 is
actually present for all values of m1/2 .
As is evident from Fig. 10, the cross section is largest for A0 = 260 GeV and falls off by
one (two) orders of magnitude as A0 reaches 1 TeV (1 TeV). However, the variation with A0
is less significant than those with tan and m1/2 . For tan = 56, the chargino contribution is
several orders of magnitude smaller than the charged-Higgs one and almost independent of A0 .
We conclude that, in the MFV MSSM, the mSUGRA scenario characterised by the input
parameter values specified in Eq. (21) approximately maximises the cross section of process (1),
211
Fig. 8. Total cross section and its Higgs and chargino parts in the MFV MSSM as functions of tan under HERA
experimental conditions.
Fig. 9. Total cross section and its Higgs and chargino parts in the MFV MSSM as functions of m1/2 under HERA
which still comes out much below the threshold of observability at HERA and a future electron
proton supercollider.
4.3. MSSM with non-minimal flavour violation
Prior to presenting our NMFV-MSSM results, we explain our choice of input parameters.
Scanning the eight-dimensional parameter space defined at the end of Section 3, we find that the
following assignments, which we henceforth take as default, approximately maximise the cross
212
Fig. 10. Total cross section and its Higgs and chargino parts in the MFV MSSM as functions of A0 under HERA
section of process (1):

tan = 8,
MA0 = 170 GeV,
M3 = 195 GeV,
A0 = 950 GeV,
M0 = 475 GeV,
= 345 GeV,
M2 = 655 GeV,
= 0.73.
(22)
These parameters are in accordance with the lower bounds on the squark masses of 100 GeV
and with the experimental lower bounds for the masses of the other SUSY particles [18]. They
are also in accordance with the experimental bounds from b s decay [18]. We calculate the
MSSM spectrum using the program package FeynHiggs [22].
Figs. 1114 all show
the total cross section of process (1) in the NMFV MSSM. Specifically,
Fig. 11 displays the S dependence, also separately for the photoproduction and electroproduction contributions, while Figs. 1214 exhibit, for HERA experimental conditions, the dependencies on , M0 , and M3 , respectively. In Figs. 12 and 13, also the contributions from loops
involving gluinos, charginos, and neutralinos are shown separately.
From Fig. 11 we read off values of order 104 fb and 101 fb for the cross sections at HERA
and the future electronproton supercollider mentioned above, respectively, which is discouraging in the case of HERA and challenging for the electronproton supercollider, depending on its
luminosity.
From Fig. 12 we observe that the cross section strongly increases with , by five orders of
magnitude as runs from 0 to 0.73. For > 0.73, the mass of the lightest squark is less than
the lower limit of 100 GeV. The cross section is almost exhausted by the gluino contribution.
This may be understood by observing that the corresponding Feynman diagrams are enhanced
by a factor of s / relative to those of the chargino and neutralino contributions. The neutralino
contribution exhibits a dependence similar to the full cross section, but is more than four
orders of magnitude smaller. The chargino contribution oscillates about a mean value of 1010 fb
in the range considered. For 0.3 ( 0.3), it overshoots (undershoots) the neutralino
contribution.
Fig. 11. Total cross section and its photoproduction and electroproduction parts in the NMFV MSSM as functions of
213
S.
Fig. 12. Total cross section and its gluino, chargino, and neutralino parts in the NMFV MSSM as functions of under
HERA experimental conditions.
From Fig. 13 we observe that the cross section strongly decreases with increasing value of M0 ,
by more than three orders of magnitude as M0 runs from 475 GeV to 2 TeV. The dominant
role of the gluino contribution observed in Fig. 12 attenuates in the large-M0 regime, where the
chargino contribution gains influence. Nevertheless, there is a clear hierarchy among the gluino,
chargino, and neutralino contributions for M0 700 GeV, the latter one being least important.
For M0 700 GeV, the neutralino contribution exceeds the chargino one. The chargino and
neutralino contributions exhibit minima at 650 GeV and 850 GeV, respectively.
214
Fig. 13. Total cross section and its gluino, chargino, and neutralino parts in the NMFV MSSM as functions of M0 under
HERA experimental conditions.
Fig. 14. Total cross section in the NMFV MSSM as function of M3 under HERA experimental conditions.
From Fig. 14 we learn that the cross section decreases by about two orders of magnitude as
M3 runs from 195 GeV, the Tevatron search limit [18], to 2 TeV.
5. Conclusion
In this paper, the photoproduction and photonic electroproduction of single top quarks in
electronproton scattering was analysed for the first time. The analysis was performed at one
loop in the SM as well as the MSSM with minimal and non-minimal flavour mixing. In all three
215
models, the cross section turned out to be too small to be measurable at HERA. The physics at
HERA remains interesting because, in the light of our results, the single-top-quark-like events
seen by the H1 Collaboration might be a sign of physics not only beyond the SM, but also beyond
the MSSM with conserved R parity.
Acknowledgements
We are grateful to John Collins for useful communications regarding Ref. [6]. The work of
B.A.K. was supported in part by the German Research Foundation DFG through the Collaborative Research Center No. 676 Particles, Strings and the Early Universethe Structure of
Matter and SpaceTime and by the German Federal Ministry for Education and Research BMBF
through Grant No. 05 HT6GUA.
References
[1] H1 Collaboration, A. Aktas, et al., Eur. Phys. J. C 33 (2004) 9;
D.M. South, on behalf of the H1 Collaboration, in: Proceedings of the 14th International Workshop on Deep Inelastic
Scattering (DIS 2006), Tsukuba, Japan, 2024 April 2006, World Scientific, Singapore, 2007, p. 325.
[2] ZEUS Collaboration, S. Chekanov, et al., Phys. Lett. B 559 (2003) 153;
M. Corradi, on behalf of the ZEUS Collaboration, in: Proceedings of the 14th International Workshop on Deep
Inelastic Scattering (DIS 2006), Tsukuba, Japan, 2024 April 2006, World Scientific, Singapore, 2007, p. 321.
[3] T. Stelzer, Z. Sullivan, S. Willenbrock, Phys. Rev. D 56 (1997) 5919;
S. Moretti, K. Odagiri, Phys. Rev. D 57 (1998) 3040.
[4] F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321;
M. Misiak, S. Pokorski, J. Rosiek, Adv. Ser. Direct. High Energy Phys. 15 (1998) 795.
[5] N.G. Deshpande, G. Eilam, Phys. Rev. D 26 (1982) 2463;
S.-P. Chia, G. Rajagopal, Phys. Lett. B 156 (1985) 405;
J.M. Soares, A. Barroso, Phys. Rev. D 39 (1989) 1973;
A. Barroso, Phys. Rev. D 42 (1990) 901;
C.-H. Chang, X.-Q. Li, J.-X. Wang, M.-Z. Yang, Phys. Lett. B 313 (1993) 389.
[6] J.C. Collins, Phys. Rev. D 58 (1998) 094002.
[7] B.A. Kniehl, L. Zwirner, Nucl. Phys. B 621 (2002) 337.
[8] J. Kblbeck, M. Bhm, A. Denner, Comput. Phys. Commun. 60 (1990) 165;
[9] T. Hahn, C. Schappacher, Comput. Phys. Commun. 143 (2002) 54.
[10] T. Hahn, M. Perez-Victoria, Comput. Phys. Commun. 118 (1999) 153.
[11] R. Mertig, M. Bhm, A. Denner, Comput. Phys. Commun. 64 (1991) 345.
[12] T. Hahn, Acta Phys. Pol. B 30 (1999) 3469;
T. Hahn, Nucl. Phys. B (Proc. Suppl.) 89 (2000) 231;
T. Hahn, Nucl. Phys. B (Proc. Suppl.) 157 (2006) 236.
[13] T. Hahn, Comput. Phys. Commun. 168 (2005) 78.
[14] A.M. Curiel, M.J. Herrero, W. Hollik, F. Merz, S. Pearanda, Phys. Rev. D 69 (2004) 075009;
S. Heinemeyer, W. Hollik, F. Merz, S. Pearanda, Eur. Phys. J. C 37 (2004) 481.
[15] T.P. Cheng, M. Sher, Phys. Rev. D 35 (1987) 3484.
[16] K. Hikasa, M. Kobayashi, Phys. Rev. D 36 (1987) 724;
P. Brax, C.A. Savoy, Nucl. Phys. B 447 (1995) 227.
[17] M. Ciuchini, E. Franco, A. Masiero, L. Silvestrini, Phys. Rev. D 67 (2003) 075016;
M. Ciuchini, E. Franco, A. Masiero, L. Silvestrini, Phys. Rev. D 68 (2003) 079901, Erratum.
[18] Particle Data Group, W.-M. Yao, et al., J. Phys. G 33 (2006) 1.
[19] A. Denner, Fortschr. Phys. 41 (1993) 307.
[20] J. Pumplin, D.R. Stump, J. Huston, H.L. Lai, P. Nadolsky, W.-K. Tung, JHEP 0207 (2002) 012.
[21] A. Djouadi, J.-L. Kneur, G. Moultaka, Comput. Phys. Commum. 176 (2007) 426.
[22] S. Heinemeyer, W. Hollik, G. Weiglein, Comput. Phys. Commun. 124 (2000) 76.
YangMills theory in 2 + 1 dimensions:

Coupling of matter fields and string-breaking effects
Abhishek Agarwal a , Dimitra Karabali b , V.P. Nair a,
a Physics Department, City College of the CUNY, New York, NY 10031, USA
b Department of Physics and Astronomy, Lehman College of the CUNY, Bronx, NY 10468, USA
Received 21 May 2007; received in revised form 12 September 2007; accepted 14 September 2007
Abstract
We explore further the Hamiltonian formulation of YangMills theory in 2 + 1 dimensions in terms of
gauge-invariant matrix variables. Coupling to scalar matter fields is discussed in terms of gauge-invariant
fields. We analyze how the screening of adjoint (and other screenable) representations can arise in this
formalism. A Schrdinger equation is then derived for the gluelump states which are the daughter states
when an adjoint string breaks. A variational solution of this Schrdinger equation leads to an analytic
estimate of the string-breaking energy which is within 8.8% of the latest lattice estimates.
PACS: 11.10.Kk; 11.15.Me; 12.38.Lg
Keywords: Adjoint screening; String-breaking
1. Introduction
YangMills gauge theories in two spatial dimensions, especially their nonperturbative properties, are interesting for many reasons. First of all, they can be a model for the more realistic,
but also more complicated, (3 + 1)-dimensional theories. They have nontrivial dynamical content and propagating degrees of freedom in contrast to YangMills theories in 1 + 1 dimensions,
yet they are more amenable to mathematical analysis with some of the more recent techniques
E-mail addresses: abhishek@sci.ccny.cuny.edu (A. Agarwal), dimitra.karabali@lehman.cuny.edu (D. Karabali),

vpn@sci.ccny.cuny.edu (V.P. Nair).
doi:10.1016/j.nuclphysb.2007.09.007
A. Agarwal et al. / Nuclear Physics B 790 (2008) 216239
217
than their (3 + 1)-dimensional counterparts. Beyond being a testing ground for nonperturbative
techniques, they are also of direct relevance in, at least, one physical context, namely, chromodynamics at high temperatures. (Recall that in an imaginary-time formalism, all Matsubara modes
are suppressed at high temperatures, except for the zero modes of bosonic fields, leading to a
dimensionally reduced theory [1].) YangMills theories in three (or 2 + 1) dimensions are also
of interest from the point of view of gauge-gravity duality, particularly in the context of the recent proposal for string and gravity duals of the maximally supersymmetric YangMills theory
in three dimensions [2]. Nonperturbative techniques for strong coupling analyses can also help
elucidate some of the details of the gauge-gravity duality.
A few years ago, a Hamiltonian analysis of pure YangMills theory in 2 + 1 dimensions was
developed based on a gauge-invariant matrix parametrization of the gauge potentials [3,4]. This
led to the calculation of a wave functional for the vacuum state of the theory [5]. (For reviews,
and other approaches and related papers, see [68].) The vacuum expectation value of the Wilson
loop operator could then be evaluated using this wave functional and gave the result

WR (C) = exp[R AC ],
(1)
where R refers to the representation for the Wilson loop, and AC is the area of the loop C. The
string tension R was given as
R = e4 (cA cR /4),
(2)
where e is the coupling constant, and cR , cA denote the quadratic Casimir values for the representation R and for the adjoint representation, respectively. The area law (1) is consistent with
confinement of charged representations. The calculated value of the string tension is in very good
agreement with lattice estimates [9,10]. More recently, motivated by our Hamiltonian analysis,
there has been an attempt to estimate glueball masses [11]. We should note however that the
vacuum wave function used in [11], which is somewhat conjectural, agrees with ours, derived
in [5], only in the high and low energy limits.1 An interesting variant of the Hamiltonian analysis is an anisotropic formulation with two different couplings corresponding to the two spatial
directions [8]. The theory is solvable in the extreme anisotropic limit; however, the approach to
the isotropic limit is not entirely clear. In any case, from the results quoted earlier, it is clear that
it is worth pursuing and elaborating on our Hamiltonian approach.
The analysis was done in what may be characterized as a continuum strong coupling analysis. (This is different from the strong coupling analysis on the lattice.) For YangMills theory in
two spatial dimensions, e2 has the dimensions of mass and the expansion parameter is e2 /k or
k/e2 , where k is a typical momentum scale. Modes of momenta much smaller than e2 have to
be treated nonperturbatively, and can be analyzed in a k/e2 expansion, while modes of momenta
much larger than e2 can be treated perturbatively in an e2 /k expansion. Continuum strong coupling refers to an expansion treating modes of small momenta as the dominant contributions to
any physical quantity. The nature of the expansion brings up many questions immediately. For
example, the area law (1) and the string tension (2) are obtained for any representation, including
representations of zero N -ality (say, for the gauge group SU(N )). But these representations can
be screened. Interpreting the Wilson loop as the propagation of two heavy external charges, we
1 As we emphasize in Appendix A, in carrying out the regularization of the Hamiltonian and the local operators on
which it acts, the order in which the regularization parameter of the Hamiltonian and the point-splitting for the operator
are taken to zero is important in preserving important symmetries such as Lorentz invariance. We expect that this is one
of the reasons for the discrepancies between the two wave functionals.
218
see that a gluon can bind to the external charge producing a color singlet, if the external charges
are in representations of zero N -ality. (The color singlet state produced by binding gluons to the
external charge is generally referred to as a gluelump.) Thus, when the separation of the external
charges becomes large, the energy in the string of gauge fields connecting these charges becomes
large and it is energetically favorable to pair-produce gluons which bind to the external charges,
neutralizing their color charge. There is no potential between the gluelumps, since they are colorneutral, and so we do not expect an area law for the Wilson loop for these representations. The
process of the state of the two heavy charges connected by the string making a transition to the
state of two gluelumps is called string-breaking [12,13]. In our previous analysis, we did not see
string-breaking. There is no inconsistency, since we were using the continuum strong coupling
expansion, and string-breaking would not be seen to the order we calculated, but we need to
understand how screening can arise as we improve on this approximation.
Closely related to this is the question of whether the formula (2) for the string tension can
be exact in any sense. Are there corrections to the formula even for representations of nonzero
N -ality? There are arguments in the literature, based on the 1/N -expansion, that the ratios of
string tensions for different representations should deviate from the ratios of the Casimir invariants [14]. The breaking of the string for a screenable representation is a transition from a
single color singlet operator to two color singlet operators. As such, the transition matrix element is suppressed at large N . So, one possibility is that the formula for the string tension, for
the nonscreenable representations, is exact in the large N limit. This question was numerically
analyzed in detail recently by Teper and Bringoltz [10]. They find that, while the formula (2) for
string tension differs from the lattice result by only about 0.88% as N , the deviations are
still statistically significant. It is therefore important to understand and calculate possible corrections. The approximation of the wave functional for modes of small momenta has a simple form,
a Gaussian function in the appropriate variables; this simple form was used in evaluating the vacuum expectation value (1). While the closeness of the agreement between the lattice results and
the values given by (2) shows that the Gaussian approximation is a very good first step, the wave
functional has non-Gaussian terms and, therefore, it has the potential for explaining corrections
to the formula for the string tension. This is an issue that deserves further analysis.
In this paper, we will address the problem of screening of representations of zero N -ality,
deferring the issue of corrections to the string tension to a later publication. As mentioned earlier,
screening is due to the formation of color singlet gluelump states, where gluons which are paircreated from the vacuum bind to the external charges. Here we will argue for the possibility of
such a bound state from the higher non-Gaussian terms of the wave functional. We shall then
construct the gluelump state, treating the external charge as a heavy scalar in the adjoint (or
some other screenable) representation. A Schrdinger-type eigenvalue equation will be obtained,
within certain approximations explained in the text. We then carry out a variational estimate of
the gluelump energy. This gives us the estimate for the string-breaking energy which we compare
to the value obtained from lattice simulations [13].
For carrying out this analysis, we need to develop a gauge-invariant Hamiltonian approach for
scalar fields coupled to the YangMills fields. This will be the subject of the next section. This
formalism is also the first step in extending the continuum strong coupling analysis of the gauge
theory to the case with matter fields added, and as such, it is of interest beyond understanding
the physics of screening. For instance, it can lead to an analysis of the supersymmetric theory,
once we have further extended this to include fermion fields. We may note that there has been a
recent proposal about the gravity dual for (2 + 1)-dimensional YangMills theory with sixteen
supercharges [2]. The gauge theory in question can be regarded as the dimensional reduction of
219
N = 4 supersymmetric YangMills theory from R S 3 to R S 2 . A prediction for the leading

strong coupling contribution to the masses of the operators built out of the scalars fields has
also been computed from the string theory side [2]. Since the gauge-invariant framework that we
shall develop in the paper naturally leads to a strong coupling expansion for the gauge theory in
the continuum, there is the potential for obtaining similar results directly in the strong coupling
regime of the gauge theory and for the possible confirmation of some details of the gauge-gravity
duality.
We also note that three-dimensional YangMills theory with matter fields in the adjoint representation naturally arises at finite temperatures in an effective theory of deconfinement [15].
The matter fields in this case are the Polyakov loops around the imaginary time direction. They
take values in the gauge group; nevertheless some of our analysis can be adapted to analyze that
theory. We consider this paper again as a prelude to such an analysis.
In Section 2, we construct the Hamiltonian for the (2 + 1)-dimensional YangMills theory
coupled to adjoint scalars in terms of gauge-invariant variables. In Section 3, we give qualitative
arguments for how screening could arise in the Hamiltonian approach. In Section 4, we then use
the gauge-invariant framework outlined in Section 2 to work out the action of the Hamiltonian
for gluelump states of very short spatial extension, appropriate to an extreme strong coupling
approximation. This gives a lower bound on the string-breaking energy. The extension of this to
gluelump states of finite spatial extent is given in Section 5 where a Schrdinger-type eigenvalue
equation is obtained. In Section 6, based on this eigenvalue equation, we obtain an estimate
of the string-breaking energy and compare it with available lattice data. There is a short summary/discussion section and the paper concludes with Appendix A with some comments on how
regularization has to be carried out for various terms in the Hamiltonian.
2. Scalars coupled to the YM field
In this section, we shall set up the essentials of coupling a real scalar field to YangMills theory in two spatial dimensions. We will consider an SU(N ) gauge theory and the scalar field will
be in the adjoint representation of SU(N ), although various formulae can be easily generalized
to any representation.
The action for the theory is

1

1
2 M 22 .
d 3 x Tr (D0 )2 (D)
S = 2 d 3 x Tr F 2 +
(3)
2
2e
The corresponding Hamiltonian is given by

2 + M 22
(E 2 + B 2 ) 2 + (D)
+
2
2e2
a a

2
2
1
M 2a a
B B
a
a
=
+
(
D)
e2 a a + a a +
+
2(D)
.
2

2
2e2
A A
H=
d 2x
(4)
In writing this expression, we have made the gauge choice A0 = 0. A = 12 (A1 + iA2 ), A =
1
2 (A1 iA2 ) are the two complex spacial components of the gauge potential. D, D are the
holomorphic, antiholomorphic covariant derivatives
D = + [A, ],
=
+ [A,
].
D
(5)
220
In our conventions,
= a t a ,
A = iAa t a ,
(6)
where the SU(N ) generators are normalized such that

1
a b
Tr t a t b = ab ,
(7)
t , t = if abc t c .
2
We would like now to rewrite the Hamiltonian in terms of gauge-invariant variables following
the approach used in [35] for pure YangMills theory. As before, we use the parametrization
.
A = M 1 M
A = MM 1 ,
(8)
If the gauge transformations take values in the Lie group G, M is a complex matrix taking values
in GC , the complexification of G. For the case of G = SU(N ), which is what we shall consider
in this paper, M SU(N )C = SL(N, C).
Time-independent gauge transformations are realized as left rotations on the matrix M,
M(
x ) g(
x )M(
x ),
g(
x ) SU(N ).
(9)
As a result, the theory can be entirely rewritten in terms of the gauge-invariant variables
H = M M,
= M M 1 .
(10)
Before we derive the expression of the full Hamiltonian (4) in terms of these gauge-invariant
variables, it is useful to first review the case of the pure YangMills theory. The simplification
of the Hamiltonian in terms of the variables M, M is as follows. Introduce the right-translation
operator pa for M and the left-translation operator p a for M by

pa (
x ), M(
y ) = M(
y )(ita ) (2) (
x y),

(2)
p a (
(11)
x ), M (
y ) = (ita )M (
y ) (
x y).
It was shown in [4] that in terms of these, the kinetic energy operator T may be written as

2
e2
e2
T =
=
rs (
u, v)p r (
u)ps (
v ),
2
2
Aa A a
u,v

(12)
u, v) = Gar (
x , u)Kab (
x )Gbs (
x , v),
rs (
x
where Kab = 2 Tr(ta H tb H 1 ) is the adjoint representative of H , and

1
2
1 (y, y)
ma ,
ma e|x y | / K(x, y)K
(x y)

1
2
x , y) =
y)
ma .
ma e|x y | / K 1 (y, x)K(y,
Gma (
(x y)
Gma (
x , y) =
(13)
These are the regularized versions of the corresponding Greens functions

x , y) =
G(
1
,
(x y)
G(
x , y) =
1
.
(x y)
(14)
221
The parameter controlling the regularization, , acts as a short-distance cut-off. Expression (12)
is to be used on functionals where the point-separation of various factors is much larger than .
(See Appendix A for more details on regularization.)
As discussed in [4] there is an ambiguity in the parametrization (8): (M, M ) and (M V (z),
where V , V are, respectively, antiholomorphic
V (z)M ) give the same gauge potentials A, A,
and holomorphic in the complex coordinates z = x1 + ix2 and z = x1 ix2 . As a result, all
physical observables in the theory should satisfy the so-called holomorphic invariance
H V (z)H V (z).
(15)
The particular choice of regularization used in (13) respects this holomorphic invariance.
When the theory is rewritten in terms of gauge-invariant variables, pa , p a can be realized as
translation operators on H = M M,

x ), H (
y ) = H (
y )(ita ) (2) (
x y),
pa (

(2)
p a (
(16)
x ), H (
y ) = (ita )H (
y ) (
x y).
In [3,4], we have argued that the wave functions can be taken to be functionals of the current
J = (cA /)H H 1 , where cA is the quadratic Casimir invariant defined by cA ab = f amn f bmn .
The action of the operators p, p on the current J is expressed in terms of the commutation
relations

cA
v ), Ja (z) = i Kas (z)z (z v),
ps (

p r (
(17)
u), Jb (w)
= i(Dw )br (w
u),
where Dw ab = cA w ab + ifabc Jc (w).
Using (12) and (17), the action of T on wavefunctions of
the form (J ) can be expressed as

e 2 cA
TYM (J ) =
(18)
a (z) a
ab (z, w)

+
(J ),
2
J (z)
J a (z) J b (w)

z
z,w
where

1
z) = z rs (w,
z) Ksa
(z),
ra (w,

a (z) = ifarm rm (
u, z)
,
uz
ab (z, w)
= Dw br ra (w,
z).
(19)
The above expressions have been computed in [4] where we found that for small , T can be
further simplified as

Ja (z)
TYM (J ) = m
+
Dw G(z, w)
(J ) + O( ), (20)
ab
Ja (z)
Ja (w)
Jb (z)
where m = e2 cA /2 .
As for the potential energy term, it is trivially checked that we can write

a a
2 B B
a J
a :.
d 2 x :J
VYM = d x
=
mcA
2e2
(21)
222
The regularized form of this expression is
VYM =
mcA

cA dim G
1
(
x , y; )Ja (
x ) K(x, y)K
(y, y)
ab Jb (
y)
,
2 2
x,y
(
x , y; ) =
|
x
y |2
(22)
where (
x , y; ) is a
regularized -function, is the parameter of regularization, and we should
x , y) = [K(x, y)K
1 (y, y)]
ab in (22) is a
take the limit where 1/e2 . The operator U ab (
ab
s at different points. U (
holomorphic Wilson line connecting the two J
x , y) is such that the
regularized expression for VYM satisfies holomorphic invariance. In considering the application
of T on this, and on other expressions in the rest of this paper, the correct procedure is to preserve as we take these quantities to zero. This is what was used in [4,5]; it is also briefly
explained in Appendix A.
Since T as defined in (12) is valid only for separations much larger than , we need to use
to define the potential energy term. Correspondingly, we can define the kinetic term also
with a scale by the equation [4]
e2
T () = T + log(2 /)Q,
2

r (
p r (
u) i J
u) ps (
v ).
Q = (
u, v; )Krs (u, v)
(23)
The understanding here is that ;

taken to act on functionals of J with
this operator is to be
a point-separation much larger than , with 0 and 1/e2 .
2
We now turn to the case with the scalar field added. Evidently the term AA
does not change
and we can still use the expression (12). The full kinetic energy operator T is now

1
2
1
2
e2 a a
2
2
a a
A A

2
e
1
2
=
u, v)p r (
u)ps (
v)
,
rs (
2
2
a a
T =
(24)
where is the gauge-invariant field defined in (10). The action of p, p on is given by

pa (
x ), b (
y ) = 0,

x ), b (
y ) = fabc c (
y ) (2) (
x y).
p a (
(25)
The gauge-invariant wave function, (J, ), is now a functional of both gauge-invariant variables J and . In terms of its action on such functionals T can be written as

T = TYM + im
z,w
cd (w,
z)f abc a (w)

b (w)
J d (z) 2
z) are given in (18), (19).

where TYM and cd (w,
2
a a
(26)
223
Similarly the full potential energy term can be expressed in terms of gauge-invariant fields as

M 2a a
BaBa
a
a
+
2(D)
(
D)
+
2
2e2

2
a
M a
2 a
= VYM +
(D)a +
.
cA
2

V=
(27)
For purposes of easy reference, we shall now collect various formulae and write the final form
of the Hamiltonian for the YangMills field coupled to adjoint scalars as

2
1
rs (
u, v)p r (
u)ps (
v)
2
a a

M 2 a a
2 a
+ VYM +
(D)a +
cA
2

a:
a J
=m
Ja (z)
:J
z) ba
+
Dw (w,
+
Ja (z)
Ja (z) Jb (w)

mcA
+ im cd (w,
z)f abc a (w)

b
d
(w)
J (z)
H=
e2
2
z,w
2
+
a a

M 2 a a
2 a
(D)a +
.
cA
2
(28)
It is evident from our derivation that this formula for the Hamiltonian is trivially generalizable to
the case of several scalar fields.
3. Screening: A preliminary qualitative analysis
The next logical step is the application of the gauge-invariant formalism of the previous
section to the gluelump state. But before doing so, it is useful to go over some qualitative considerations of how screening of the adjoint and other representations of zero N -ality could arise in
our Hamiltonian approach. As discussed in [5], the computation of the expectation value of the
Wilson line in pure YangMills theory is of the form

WR (C) =

d(A/G ) eS(H ) WR (C)

d(H ) e2cA Swzw (H ) eS(H ) WR (C),
(29)
where d(A/G ) = d(H ) e2cA Swzw (H ) is the gauge-invariant volume element for YangMills
fields and eS(H ) is given in terms of the vacuum wave function as eS(H ) = 0 0 . Swzw (H )
is the WessZuminoWitten action for the hermitian field H and d(H ) is the Haar measure
for H viewed as an element of SL(N, C)/SU(N ). Explicitly,
1
Swzw (H ) =
2

1 + i
Tr H H
12

Tr H 1 H H 1 H H 1 H .
(30)
224
The expression for the wave function 0 shows that S(H ) has the form

4 2
1
a
a
S(H ) = 2 2 J
J
e cA
(m + m2 2 )

2fabc f (3) (
x , y, z)Ja (
x )Jb (
y )Jc (z) + O J 4 .
(31)
x , y, z) has been given in [5].

The function f (3) (
Eq. (29) shows that expectation values can be considered as averages in a two-dimensional
field theory with the action S(H ). In evaluating the string tension and obtaining the formula in
terms of the Casimir values of the representations, we have basically used the low-momentum
limit of the quadratic terms (terms which are quadratic in the current J ) in this functional. These
terms correspond to

1
S(H ) = 2 d 2 x Fija Fija ,
(32)
4g
where g 2 = me2 = e4 cA /2 . In this approximation, the expectation value reduces to the average calculated using a Euclidean two-dimensional YangMills theory with coupling constant g.
There are many ways to evaluate these averages, one simple way is to go back to the original
variables, the As, and calculate, for example, in a gauge A1 = 0. In this gauge, the commutator term in Fij is absent and we have a Gaussian functional integral. This leads directly to the
formula for the string tension.
As mentioned in the introduction, the values we find are in very good agreement with the
lattice analysis [9,10], although the lattice results are accurate enough to show a statistically
significant deviation, even at large N , where the agreement is the best. Also, it has been argued,
on the basis of the 1/N -expansions, that deviations from the Casimir scaling for string tensions of
different representations are possible [14]. Further, we know, on general theoretical grounds, that
the Wilson loops in representations of zero N -ality can be screened and will not show the area
law behavior for large enough loops. Therefore, we turn to considerations of possible corrections
to this result. The corrections can arise from the higher terms involving three or more powers
of the current J . The string tension is defined by the limit of large Wilson loops. Therefore we
may think of integrating out the high momentum modes in the two-dimensional theory defined
by S(H ) to obtain an effective low-momentum action Seff (H ) and then using this to calculate
the average of the Wilson loop. It is then easy to see that there are two types of corrections
possible. The first set would correspond to a corrected quadratic term, due to terms which modify
the propagator for the currents. This is equivalent to a modified coupling g in the effective
two-dimensional YangMills theory. Since this is at the level of the action and we have not yet
introduced the Wilson line, there is no dependence on the representation of the Wilson line; it
depends only on N and numerical factors. Therefore, these corrections can affect the value of R ,
but the ratio R /F is unaffected.
The second set of corrections involves the J 3 and higher vertices with the currents connected
to the Wilson line via propagators. These cannot be reduced to propagator corrections and hence
we can get representation-dependent modifications. In particular, screening effects have to come
from such terms. In principle, all the J 3 and higher vertices can contribute. To elucidate the
nature of these terms, consider the contributions from the cubic vertex [5]. Written in terms of the
original variables, the cubic term is of the form Tr(D 1 B[B, DB]).
(Since we have phrased the
calculation of the string tension in terms of the As, we use the same variables for this argument.)
225
Fig. 1. A set of terms whose sum can be viewed as the propagation kernel of a bound state (shown as thick gray line).
we see that
Expanding D 1 in powers of A,
D 1 = 1 1 A 1 + .
In the A1 = 0 gauge, the cubic term is thus of the form
a

(3)
c f abc .
Sint 1 1 A 1 + 1 A2 1 Ab2 1 A
2
(33)
(34)
Because of the overall Euclidean invariance, the Wilson line can be visualized as the propagation
of two heavy charges. Consider then the insertion of the interaction term (34). The contractions
of the As from the expansion of D 1 will produce a sequence of terms which is effectively
like the iteration of the kernel in a BetheSalpeter equation for a bound state. In other words,
we get terms which can be interpreted as the propagation kernel of a possible bound state of the
gluons with the external charge of the Wilson line. For the nonscreenable representations, it is
not meaningful to think of the bound state in isolation since it is not a color singlet. But for the
screenable representations, a color singlet bound state can be formed. The picture which emerges
is one where gluons are created from the vacuum and then bind to the external charges to form
color singlets. This can lead to breaking of the single string connecting the external charges into
two such singlet bound states (see Fig. 1).
As for the quantitative calculation of these corrections, we know from the lattice data that they
are small, except for the screening effects. It is, of course, important to calculate the corrections
and provide an analytic justification for why they are small. The calculation of these corrections
and the analytic justification for why they are small will be reported in another paper. Here we
will pursue the analysis of string-breaking for screenable representations. There are two related
issues for the analysis of this question. There is a transition matrix element between the state of
the single string connecting the external charges and the two singlet bound states. This matrix
element will describe the process of the string-breaking, and, from general arguments, it should
be suppressed by powers of N , in a large N expansion. The calculation of this matrix element
is quite involved and we will not carry this out here. However, since the energy at which the
string breaks corresponds to twice the energy of the singlet bound state, we can calculate this
string-breaking energy by calculating the energy of the singlet bound state of a gluon with the
external heavy charge (whose propagation is one side of the Wilson line). This bound state has
generally been referred to as the gluelump. In our language, it corresponds to a wave function of
the form

ab b
a (
x , y)J
x ) K(x, y)K
1 (y, y)
(
y )0
G = d 2 x d 2 y f (
O G 0 .
(35)
226
Here J represents the gauge-invariant gluon which has a mass m. In Ref. [4], we have discussed
a constituent picture for the formation of bound states, such as glueballs, in terms of elementary
constituents of mass m corresponding to J . The gluelump, as described by (35), is consistent
with that picture. The function f (
x , y) is then the wave function of the constituents in the bound
state.
We can now move onto a more specific and quantitative estimate of the bound state energy
rather than the general qualitative arguments given above. For this, the strategy will be to act
with the Hamiltonian on the state (35) and obtain a two-particle Schrdinger-like equation for
the function f (
x , y) characterizing the bound state. The details of this analysis are given in the
next section.
4. The action of the Hamiltonian on G
We will first consider the (continuum) strong coupling limit where we use the state (35) with
the vacuum wave function 0 = exp( 12 S(H )) approximated by

VYM M
0 exp
d 2x 2
2m
2

1
M
d 2x 2 .
= exp 2 d 2 x F 2
2
8g
(36)
We have used the perturbative vacuum for the -field. In the large M limit, because M 2 D,
are suppressed by powers of 1/M.
corrections due to the interaction term in D
With the approximation of 0 as in (36), the action of the Hamiltonian on G can be written
as2
HG = MG + OG HYM 0 + 0 (T OG )

e2
rs (
u, v) p r (
u), VYM ps (
v ), OG 0
4m

e2
rs (
u, v) ps (
v ), VYM p r (
u), OG 0 .
4m
(37)
Since 0 is the vacuum for the YangMills part, HYM 0 = 0. For the third term and the last
term on the right-hand side, it should be kept in mind that p r has nontrivial action on as given
in (25).
In evaluating (T OG ), it is useful to compare it with the action of the YangMills kinetic term,
which we have evaluated elsewhere [4]. Consider

ab b
a (
(
2 = d 2 x d 2 y f (
(38)
x , y)J
x ) K(x, y)K
1 (y, y)
y ).
J
, namely J
b (
Notice that OG is obtained from this by replacing the last J
y ), by b (
y ). The
action of T on the state (38) can be written as the sum of four terms as given below.
2 We take the Hamiltonian to be normally ordered for the scalar fields so that there is no zero-point energy contribution
in (37).
T 2 = I + II + III + IV,

b
e2
a (
(
rs (
u, v) p r (
u), ps (
v ), J
x )U ab (
x , y) J
y )f (
x , y),
I=
2

e2
a (
b (
rs (
II =
u, v) ps (
v ), J
x )U ab (
x , y) p r (
u), J
y ) f (
x , y),
2

e2
a (
b (
rs (
III =
u, v) p r (
u), J
x )U ab (
x , y) ps (
v ), J
y ) f (
x , y),
2

e2
a (
b (
rs (
IV =
u, v)J
x )U ab (
x , y) p r (
u), ps (
v ), J
y ) f (
x , y),
2
227
(39)
x , y) = [K(x, y)K

1 (y, y)]
ab . We will
where U ab (
consider the evaluation of these terms for
2
small separations |
x y| 1/e , with |
x y| . (This means, of course, that the width of f
which controls the average value of |
x y| is small compared to 1/e2 .) Term I then leads to a
and the Coulomb potential V(
mass m for J
x , y). Terms II and III contain a normal-ordering
s. The other terms arising from these two are negligible for
term, with the p, p acting on the J
very small separations; these involve at least one power of |
x y| and three or more J s. Finally
(
the last term gives m for the final J
y ). Thus we find

a (
b (
T 2 =
(40)
2m + V(
x , y) f (
x , y):J
x )U ab (
x , y)J
y ): + O |
x y| .
x,y
The Coulomb potential in (40) is of the form

V(
x , y) = m Ein |
x y|2 /2 m log |
x y|2 /2 ,
(41)
where
z
Ein(z) =

dt
1 et .
t
(42)
In comparing this with T OG , we see that we will get terms similar to I and II, but the terms
(
analogous to III and IV will be absent when J
y ) is replaced by (
y ) because of (25). In
particular we find

e2
a (
I =
rs (
u, v) p r (
u), ps (
v ), J
x )U ab (
x , y) b (
y )f (
x , y)
2

a (
m + V(
x , y) f (
x , y)J
x )U ab (
x , y) b (
y ) + O |
x y| .
=

(43)
x,y
The normal ordering contribution from the term similar to II is actually zero, due to color
contractions.

e2
a (
II =
rs (
u, v) ps (
v ), J
x ) U ab (
x , y) p r (
u), b (
y)
2

a (
(44)
x ) ps (
v ), U ab (
x , y) p r (
u), b (
y ) f (
x , y).
+ J
228
Using the commutation relations (17) we find that the first term in II can be written as

e2
a (
rs (
u, v) ps (
v ), J
x ) U ab (
x , y) p r (
u), b (
y ) f (
x , y)
II (a) =
2

= m z ra (w,
z) K(z, w)K
1 (w, w)
ab f brs s (w)f
(z, w).

(45)
z,w
Explicit calculation of ra in [4] shows that (see Eq. (A.3) in Appendix A)

e/2
z) =
K(w, w)K
1 (z, w)
z (w,
2

(z w)A (z w)
B FAB (w)
+ O( ) ,
+
(46)
A,B1
where = |z w|
2 / . Inserting (46) into (45) we find that only the first term in (46) might
contribute a finite 0 -order term, but this is actually zero due to color contraction.
/2

e

II (a) = m
1 (w)
ra K(z, w)K
ab f brs s (w)f
(z, w)

K(w)K
1 (z, w)
2
z,w
= 0.
(47)
The second term in (44) can be similarly evaluated

e2
a (
rs (
u, v)J
x ) ps (
v ), U ab (
x , y) p r (
u), b (
y ) f (
x , y)
II (b) =
2

e2
a (
=
x ) rs (
y ; x, y)
rs (
y ; y) f rbc f sml K al (x, y)K
bm (y, y)
J
2
c (
y )f (
x , y).
(48)
In writing (48) we used the fact that

v ), U ab (
x , y) = f sml K al (x, y)K
bm (y, y)
(v x) (v y) (v y).
ps (
(49)
Explicit calculation shows that

rs (
y ; x, y)
rs (
y ; y) =
1 (x y)n n
y ).
y Krs (
nn!
(50)
n=1
Inserting (50) into (48) shows that the contribution of II (b) is, indeed, negligible for very small
separations. So we find that

a (
m + V(
x , y) f (
x , y)J
x )U ab (
x , y) b (
y ) + O |
x y| .
(T OG ) =
(51)
x,y
Finally, the remaining terms in (37) can be simplified using

4
.
DJ
[T , VYM ] = 2mVYM +
cA
J
(52)
229
The term 2mVYM is due to both p and p acting on VYM ; this is not needed for the terms under
consideration. We then find

e2
rs (u, v) p r (u), VYM ps (v), OG + ps (v), VYM p r (u), OG
4m

2
a (
(53)
=
x )U ab (
x , y) b (
y ): + O |
x y| .
x , y):J
x f (
2m
x,y
The 0 limit of (20), which has been used to simplify the right-hand side of (52), is adequate
for the calculation of the right-hand side of (53). Notice that the resulting factor of 2 /2m,
along with the mass m in (51), will reproduce the first two terms of the relativistic combination
m2 2 . The remainder is of order |

x y|.
Collecting various terms and results, we get

x2
a (
HG =
x )U ab (
x , y) b (
y )0
+ V(
x , y) f (
x , y) J
M +m
2m
x,y

(54)
+ O |
x y| .
So far, we have not considered the addition of Q as in (23). As mentioned earlier, the Coulomb
x y|2 /2 ). From the definition of Q,
potential in (51) is of the form Ein(|
x y|2 /2 ) log(|

a (
x )U ab (
x , y) b (
y )f (
x , y)
Q J

a (
p r (
u), J
x ) ps (
v ), U ab (x, y) b (
y )f (
x , y)
= (
u, v, )Krs (u, v)

cA
a (
x )U ab (x, y) b (
y )f (
x , y),
J
(55)
=
where, in the first step, we have indicated the term which gives a nonzero result among
the various terms generated by the action of Q. We have evaluated the action of Q on
b (
a (
x )U ab (
x , y)J
y ) before [4]; the nonzero term has the same structure as in (55), except
J
b . Thus, the result in (55) is exactly of the same form as obtained
for the replacement of b by J
a
ab
b
for J (
x )U (
x , y)J (
y ).
The addition of (e2 /2) log(2 /)Q to the definition of T thus leads to the result

x2
a (
HG =
x )U ab (
x , y) b (
y )0
+ V(
x , y, ) f (
x , y) J
M +m
2m
x,y

+ O |
x y| ,
(56)
x y|2 /). (Notice that if
where V(
x , y, ) = m(Ein(|
x y|2 /2 ) + log(2 /)) m log(|
f (
x , y) is of the form (
x , y, ), the Coulomb potential vanishes.) Thus, within the approximations used for this continuum strong coupling expansion, we see that the state G becomes
an eigenstate of the Hamiltonian H with eigenvalue E if f (
x , y) obeys the eigenvalue equation

2
M + m x + V(
(57)
x , y, ) f (
x , y) = Ef (
x , y).
2m
This equation shows that there is a constituent picture for the gluelump. The gauge-invariant
gluon of mass m binds to the external charge via the Coulomb potential. Since we have arrived
230
at this equation keeping only the F 2 -term in (YangMills part of) log 0 , this is appropriate for
spatial momenta small compared to m. The kinetic energy for the gluon is thus the nonrelativistic
expression. As mentioned before, this is the beginning of a series which sums up to
m2 2 [5]. Further, we find only the Coulomb potential. The key question is whether these
approximations are adequate. If we are interested in a prediction in the continuum strong coupling limit, it is adequate to keep the nonrelativistic kinetic energy term. We cannot say, a priori,
whether this is sufficient for comparison with lattice simulations, since a direct correspondence
of the kinematic regimes in the two approaches is not obtained. However, experience with the
comparison of the string tension and glueball masses suggests that this is a good starting point.
The question of the potential energy V(
x , y) is another matter. From the action of TYM on
(
J
x )U (
x , y)(
y ), we get the Coulomb limit of the potential. When effects due to the vacuum
wave function are included, this potential must tend to the linear potential at large distances. We
know this is so, from the existence of a nonzero string tension. However, a direct calculation is
difficult and we shall present a somewhat indirect argument in the next section. We will also use
a slightly different definition of G which is more convenient. On the lattice side, the potential
energy is seen to rise linearly with distance, with a value of the slope or string tension which is in
agreement with our calculations, up to a point, and then it flattens out, indicative of the breaking
of the string. Thus, just at the point where the string breaks, the constituents of the singlet state
(the external charge and the gluon) are separated far enough to be in the regime of the linear
potential. For a sensible comparison with lattice data, we must understand and incorporate the
linear potential. We take up this task in the next section.
However, already at this stage, it is worth noting that, if we consider the strong binding limit
(when the separation |
x y| is negligible) and also the strong coupling limit where 2 m2 ,
the energy in (57) can be approximated by M + m. The resulting value 2(E0 M) = 2m may be
taken as a lower bound on the string-breaking energy.
5. The argument for the linear potential
The strategy we will consider for extracting the linear potential is as follows. We consider a
state of the form

x , y) a (
x )W ab (
x , y) b (
y )|,
|
= f (
(58)
x,y
x
where W (
x , y) is the open Wilson line connecting points x and y, W (
x , y) = P exp( y A)
and and are heavy fields. | specifies the state of the gauge field and can be expressed as
| = O |0, where O is constructed from the gauge fields only.
The matrix element of eH for states of the form (58) can be related to the expectation
value of the Wilson loop; the latter can be expressed in terms of the area law and the linear
potential extracted from it. We use two dissimilar fields and to avoid the possibility of
mutual annihilation during the (Euclidean) time-evolution.
The action for the theory is taken as

S = d 3 x (D)a (D)a + (D)a (D)a M 2 a a + a a + SYM .
(59)
The Hamiltonian has the form
H = HYM + Hscalar + Hint ,
(60)
231
where the last term involves interactions between and A and between and A. This Hamiltonian commutes with N , the number of -particles and N , the number of -particles, separately. Thus matrix elements of the Hamiltonian are block-diagonal and we can restrict attention
to the N = N = 1 block consistently. (A real adjoint scalar particle can, in principle, annihilate with another such particle. This would be an added complication which is irrelevant for
our approach to the linear potential. So we avoid it by choosing two distinct complex scalars to
construct the state. This is not the minimal way of doing this. A minimal way would involve one
complex scalar, but the final results are the same, so we use the more convenient choice of two
complex scalars.)
First consider the normalization of states of the form (58). For this, the states are at equal time.
We can use (
x ) (
y )| = (
x y)|, since O has no s in it. We then find

x , y)f (
x , y)|.
|

= f (
(61)
x,y
The evaluation of | can also be simplified if we approximate the vacuum wave function by
the expression (36). We then find

1
2
2
| = d(A/G ) exp 2 d x F O O .
(62)
4g
We can choose the gauge A1 = 0, whereupon the action is quadratic in A2 with a propagator

a

dk2 ik2 (x2 y2 ) 1
A2 (
(63)
x )Ab2 (
y ) = ab (x1 y1 )
.
e
2
k22
The actual evaluation of (62) can then be done by Wick contractions using (63).
is another state similar to |,
H |,
where |
with
We now turn to the matrix element |e
O and a function h(
x , y) replacing O and f (
x , y), respectively.
H |
|e

= h (
x , y )f (
x , y)|(
y )W (
y , x )(
x )eH (x)W (
x , y) (y)|

x , y )f (
x , y)0| O (
y )W (
y , x )(
x ) (
x )W (
x , y) (
y )O |0.
= h (
(64)
The set of fields at the left end has been shifted by in Euclidean time. Since the fields are
heavy, we can write, for large M, ( ) = eH eH eM (0), so that (, x ) (0, x)
eM (2) (
x x ). We can then simplify the matrix element as

H
|e
(65)
|
= h (
x , y)f (
x , y)e2M 0|O ( )W (C)O (0)|0,

where O (0) = O ( = 0) and W (C) is the Wilson loop W (C) = Tr P exp( C A). The path C
is a rectangle connecting the points (
x , 0), (
y , 0), (
y , ) and (
x , ). The evaluation of the remaining expectation value in (65) can be understood using Wick contractions. There are two types of
contractions involved:
(1) Wick contractions within W (C) and separately within O O .
(2) Wick contractions between W (C) and O O .
232
We neglect the second type of contractions; we comment on this later. In this approximation,
we have
0|[O ] W (C)O (0)|0 0|O ( )O (0)|00|W (C)|0
= e |x y | 0|O ( )O (0)|0
= e |x y | |eHYM |.
(66)
Once we approximate as in the first line of this equation, the expectation value of W (C) can
be evaluated, by Euclidean invariance, as an equal-time average. Also, notice that only eHYM
appears in the last line. Thus, within the approximations we have made,

| h W eH f W | = h f e2M |x y | |eHYM |.
(67)
We may rewrite part of this integral as

h (
x , y)f (
x , y)e |x y |
x,y

h W
= 0|
=0
e |x y |

f W

=0
|0.
(68)
Using this result, we finally get

(2M+ |
x
y |)
f W | = |e
f W eHYM |. (69)
| h W e
is completely arbitrary within the N = N = 1 block, we conclude, by taking terms
Since |
linear in ,

x y|
f W | + f W HYM O |0. (70)
H f W | = 2M + |
Some comments on the approximations used in this calculation are in order at this point.
(1) We have approximated the vacuum wave function by Eq. (36). This is adequate for the linear
potential, but there can be corrections due to the higher terms (in powers of momenta and
powers of J ) in the exponent. We hope to address the issue of higher order terms elsewhere.
(2) Wick contractions between W (C) and O O have been neglected. These can lead to structures different from W (C) on the right-hand side and hence to nondiagonal pieces for the
first term in (70). These are small if the linear potential dominates. The lattice data suggest
that this is the case, as explained earlier. Further, since W (C) has no color charges, these
contractions will be suppressed by powers of 1/N .
(3) We need large M limit to avoid virtual and real processes with additional numbers of and
particles. Such processes could also modify the YangMills vacuum, if we do not take the
large M limit. (This is a quenched approximation for these particles.)
Consider now the case of the gluelump. To get the wave function
gluelump, we can take
ofa the
a , where a = M ak k .
Eq. (70) with O = 1 and apply on this equation the operator J
a M ak . Further the action of on the vacuum wave
This has the effect of replacing by J
233
function for the heavy scalars gives = M 1 . Thus we find

a (
W 0 = J
x ) M ak (
x )W kl (
x , y)M 1lb (
y ) b (
y )0
J
a (
J
x )W ab (
x , y) b (
y )0 .
(71)
x , y) may be written as

The operator W ab (

x
x , y) = P exp
W ab (
ab
KK 1
(72)
It is an open Wilson line for the holomorphic component of the connection KK 1 . (The re as a complex gauge transform (by
sult (72) follows from the fact that we may think of (A, A)
1
1
the matrix M ) of (H H , 0), i.e., (A, A) = (H H , 0)M , and because W kl (

x , y, Ag ) =
1
kl
g
1
1
g (
x )W (
x , y)g(
y ) for A = g Ag + g g.)
As in the case of W , W depends on the curve chosen from y to x . For a short separation
|
x y|, we can write
= U (
x , y)
W (
x , y) = K(x, y)K
1 (y, y)
(73)
a W ab b agrees with the gluelump state used in Section 4. For a longer

so that the functional J
separation, there are differences. Using the composition property of path-ordered exponentials,
we find, for a straight-line path,
1 (y, y)
W (y + 2 , y) = K(y + 2 , y + )K 1 (y + , y + )K(y + , y)K

K(y + 2 , y)K
1 (y, y)
+ KK 1 +

= U (y + 2 , y)
+ KK 1 + .
(74)
Thus the Wilson line W will differ from U (

x , y) = K(x, y)K
1 (y, y)
by terms involving
(KK 1 ) and higher derivatives of it when the series is continued. Such terms evidently correspond to deformations of the path.
a (
x )U ab (
x , y) b (
y)
The gluelump state can be defined, at finite separations, by either J
a
ab
b
(
or J
x )W (
x , y) (
y ). Which is more advantageous will depend on which leads to a simpler equation within the approximations we are using. The emergence of Eq. (70) suggests that
a (
x )W ab (
x , y) b (
y ) is a simpler functional to use, and so we redefine the gluelump as
J

a (
G = d 2 x d 2 y f (
(75)
x , y)J
x )W ab (
x , y) b (
y )0 .
Using (71), we then find

f W 0
H f J W 0 = H, J

W 0 .
+ 2M + |
x y|
f J
(76)
is a local operator. This must be defined, in a regularized way, by an expression

The operator J
x , y) is replaced by (
x , y, ); then we must take the small limit. In
of the form OG where f (
234
this case, we can use the result (56); the Coulomb potential is zero and we can write

0
H, J

u2 a
= (
u, v, ) M + m
u) U ab (
u, v)b (
v )0 + ,
J (
2m
(77)
. The omitted terms

u, v) is used because of the point-splitting needed
to define J
where U ab (

and the other left free so that it
in (77) correspond
to
terms
with
either
p
or
p
acting
on
J

can act on f W in (76); i.e., they are of the form
e2
2

f ps (
rs (
u, v) p r (
u), J
v ), W

f p r (
+ ps (
v ), J
u), W 0

e2
f ps (
=
u, v) p r (
u), J
v ), W
rs (
2

+ ps (
v ), J f p r (
u), W 0 .
(78)

a ][ps (
u, v)[p r (
u), J
v ), W ab ] b
One of the terms on the right-hand side, namely, e2 /2 rs (
is the type of term which led to the Coulomb potential in the calculation of Section 4. So one
might wonder whether this can generate a Coulomb term in addition to the linear potential.
m (
This is not the case. In fact, the Coulomb potential is due to the term farm J
x u) in the
a
commutator
[
p
(
u
),
J
(
x
)]
=
i
D
(
x
u

).
Notice
that
in
(78)
we
have
the
combination
r
ar

] and, since [p r (
[p r (
u), J
u), a (
x )] = farm m (
x u) as well, the relevant term is actually
zero. (This too has to be understood with point-splitting, but this particular result is basically the
same as in the unregulated calculation.) We will neglect the remaining terms in (78), which lead
to new operator structures. These terms lead to operator structures with more powers of J . The
gluelump state we consider can have overlap with states which have the same quantum numbers
W . The fact that the Hamiltonian acting on J
W can evolve it into other structures is
as J
indicative of this possibility of mixing. However, as is well known in quantum mechanics, the
effect of such nondiagonal terms in H comes with energy denominators and can be taken to be
small if the differences between the energies of the lowest gluelump state (which is what we are
interested in) and higher states are large enough. We expect this to be the case at large enough
coupling, since the differences must go like m. But ultimately it is to be justified a posteriori.3
With this understanding of various terms, combining the results in (76) and (77), we finally
have, as in (77) is taken to be very small,

x2
a (
M +m
x )W ab (
x , y) b (
y )0 + .
HG =
(79)
+ |
x y| f (
x , y)J
2m
The approximations involved in this equation are given by the comments after Eq. (70) and by
the comments about the omitted terms in (77), (78).
3 The transition amplitude from the unbroken string state to the gluelumps is seen to be very small from lattice data,
even for SU(2) [12,13]. This is another indication that the off-diagonal elements mentioned above might be small.
Eq. (79) shows that the eigenvalue equation we need for the gluelump states is

x2
M +m
+ |
x y| f (
x , y) = Ef (
x , y).
2m
235
(80)
By our general arguments, the string-breaking energy is given by V = 2(E0 M) where E0 is

the ground state energy eigenvalue of Eq. (80). We now turn to the determination of this value.
6. A variational estimate for string-breaking
Since we are interested in the ground state eigenvalue, the simplest estimate is given by a
variational calculation. Removing the center of mass motion (which is zero as M ) the
energy functional corresponding to (80) is

|f |2
1
d 2x
+ |
x ||f |2 ,
E =M +m+
N
2m

(81)
N = d 2 x |f |2 .
We take a variational ansatz of the form f = exp(|x| ), where both and are regarded as
variational parameters. The minimum of E(, ) with respect to is found to be at = ,
(3)/

(2m ) (3/) /3
2
=
(82)
.
2
A formula for the energy eigenvalue E( , ) as a function of can now be obtained:
(2+1)/ (3)/

2m (3/) 1/3
2
2
E( , ) = M + m +
2m (2/)
2

(3)/

2m (3/)
1/ 2 2
8m (3/) + 8
.
2
(83)
The string tension for a representation R is given by (2); for the present calculation, we need the
adjoint string tension, A = m2 . Using this value of the string tension, the formula for E( , )
can be minimized numerically. The minimum is obtained at = 1.752. This gives E0 M =
E( , ) M = 3.958m, corresponding to a string-breaking energy of Vcal = 2(E( , )
M) = 7.916m.
To check the robustness of the variational result, we have also tried to estimate the stringbreaking energy using a more general variational ansatz of the form
e|x|
.
(1 + |x|)
f=
(84)
Such an ansatz is suggested by the large x solution of the two-dimensional Schrdinger equation
with a linear potential, whose solutions fall off, at large |x|, as
Ai((2m )1/3 |x|) exp[ 23 ((2m )1/3 |x|)3/2 ]
|x|3/4
|x|
(85)
where Ai is the Airy function. However, a numerical minimization of E(, , ), with respect
to the three variational parameters , , , shows that the lowest estimate for E is provided by
= 0, i.e., by the exponential ansatz reported above.
236
We are now in a position to compare this with lattice estimates. The latest data on this is for
SU(2), given by Kratochvila and Forcrand [13]. The string-breaking energy is given by Vlat
8.68m. This arises as follows. Vlat is computed in terms of the lattice spacing a as Vlat
2.063a 1 . In the same set-up, the fundamental
string tension is given by F = 0.0625a 2 . This
can be used to write Vlat 2.06m cF /0.0625cA 8.68m. For our analytic calculation, we
then have |Vlat Vcal |/Vlat 8.76%.
It is interesting also to make a similar estimate for the 0++ glueball mass, at least as a further
check on the approximations used. If we use our picture of two constituents of mass m each,
interacting via the linear potential, the eigenvalue equation is

2m
y2
x2
+ |
x y| f (
x , y) = Ef (
x , y).
2m 2m
(86)
The only difference, compared to what we have done for the gluelump, is that the reduced mass
should now be 12 m. This leads to
M0++ = 5.73m.
(87)
The difference with the lattice estimate of 5.17m [9] is 10.83%. In this case, of course, since
both constituents have mass m M, our 1/m-expansion and the nonrelativistic approximation
are less reliable.
7. Discussion
A Hamiltonian approach to YangMills theory in 2 + 1 dimensions was introduced several
years ago by two of the authors [35]. In this paper we have extended the formalism to include matter fields, specifically scalar fields in the adjoint representation. First of all, this may
be viewed as a first step towards analyzing supersymmetric theories. Secondly, it may be used
for analyzing screening and string-breaking. On general group-theoretic grounds we should expect screening of Wilson loops in the adjoint and other representations of zero N -ality. When
the length of a string in one of the screenable representations is increased, it breaks at some
point with the formation of two color singlet states, each of which is referred to as a gluelump.
We obtained a Schrdinger equation for a gluelump state, the variational solution of which then
gave an analytic estimate of the string-breaking energy. This is within 8.8% of the latest lattice
, of
estimates. The emerging picture is that of a bound state between a constituent gluon J
mass m, and a heavy adjoint scalar of mass M, interacting via a linear potential V A r. Some
approximations are necessary for obtaining the Schrdinger equation, since most of the states
under discussion are unstable; the approximations we have used are given in Section 5. It should
also be emphasized that we are calculating the energy at which the string breaks, the transition
amplitude for the breaking is not calculated. (These are two different questions. The transition
amplitude, as is well known, will be suppressed at large N .)
Specifically for the (2 + 1)-dimensional theory, the string-breaking energy has not been analyzed extensively on the lattice except for the group SU(2). We consider this to be an important
question for which more lattice data are needed, for higher groups, and, in particular, the N dependence.
Also, as explained in the text, string-breaking is closely related to the issue of corrections to
the formula (2) for the string tension. This matter will be taken up in a future publication.
237
Acknowledgements
V.P.N. thanks David Gross for discussions at a preliminary stage of this work during a visit
to KITP, Santa Barbara in 2004. We also acknowledge useful discussions with Barak Bringoltz.
A.A. also wishes to thank the Max Planck Institut fur Gravitationsphysik, where part of this
work was carried out, for hospitality. This research was supported in part by the National Science
Foundation grants PHY-0457304 and PHY-0555620 and by PSC-CUNY grants.
Appendix A
Once the regulated Greens functions (13) have been introduced, it is straightforward to evaluate the regularized form of the Hamiltonian and its action on various functionals. However, it is
important to understand and follow the correct order of limits. Here we shall comment briefly on
this issue.
We start by recalling some elementary properties of a covariant integral I (p, ) resulting from
some term in the perturbation expansion of a field theory. Here p denotes external momenta and
is the short-distance cut-off. Once the integration over the loop momenta is done, I (p, ) is of
the form

O (p) + O(p)
+
O .
I (p, ) =
(A.1)
The first set of terms involves negative powers of and represent potentially divergent terms as
becomes very small. (We include = 0, which may be taken as a logarithmic divergence.) O(p)
represents finite terms and the last set of terms vanish as 0. At finite, but small, , the last set
of terms are suppressed by powers of p 2 ; they are negligible for small external momenta. While
the divergent and finite terms are independent of the regulator (apart from the usual ambiguity of
the subtraction points), the O terms are sensitive to the regulator. Of course, as 0, this does
not matter, but notice that for external momenta of the order of 1/2 , these terms can contribute
and constitute regularization ambiguities. The correct procedure is obviously that the result (A.1)
is to be used only for processes with external momenta small compared to 1/2 .
A simple case which highlights this point is the following. A regulator does not have to respect symmetries, say, Lorentz invariance; the standard lattice regularization
is an example. In
a Lorentz-invariant theory, this can lead to Lorentz-violating terms of order p . If we consistently restrict our calculation to processes with momenta p 1/2 , these spurious regulatordependent terms will be negligible.
In a Hamiltonian formulation, we define the regulated kinetic energy operator T ( ), given by
(12), (13). Since it is a functional differential operator, to interpret it properly, we have to consider
its action on different functionals. Consider
a functional ( ) with the average separation of
points between fields in the functional being . When we act on this with T ( ), there are three
types of terms possible.
(1) Terms which diverge as 0. If there are such terms, we must introduce a subtraction
procedure to make T ( ) well defined and eliminate these terms. (The redefinition of T ( )
with the addition of Q in (23) may be viewed as an example of this.)
(2) Terms which are finite as 0. These are physical effects.
(3) Terms which are of order . These can be neglected. However, if we take , or 0
at finite , these terms can give finite results, just as the O terms in (A.1). Exactly as in
238
the previous case, the correct procedure is to realize that the T ( ) can only be used for
functionals with ; otherwise there will be spurious regulator-dependent terms.
Keeping this in mind, all the calculations have to be done with 0 first. A natural question
is then: How do we define a local operator like VYM B 2 ? Note that this is needed for
the Hamiltonian, and further we will need the commutator of T with this operator. From what
was said before, VYM must be defined in a regulated way with an average point-separation of
, with . Then the action of T ( ) on this is correctly obtained. To make the operator
local, we then take 0. In other words, we must take , to zero, keeping . This is the
correct procedure and it will yield unambiguous results. This procedure is even more crucial in a
Hamiltonian framework; otherwise, spurious terms may appear and it may not even be obvious
that they are spurious, since Lorentz symmetry is not manifest.
This procedure was used in [4,5]. For the purposes of this paper, we note that an expression
for ra was obtained in [4] as
ra (w,
z) =

1
1 (u, w)K(
u)K 1 (z, u)
ra e/2
ra K(w)K
(z w)

z),
+
ra (w,
(A.2)

1
/2
1 (u, w)K(
u)K 1 (z, u)
e
z K(w)K
2z
zw

2
+
1 (u, w)K(
u)z K 1 (z, u)
z K(w)K
zw
1
1
2
1
+
(u, w)K(
u)z K (z, u)K(
z) K (z)
K(w)K

(z w)(z w)
ra
2
(A.3)
+O ,
and = (z w)(z w)/ .
If we take to be very small

where u = 12 (z + w), u = 12 (z + w)

compared to the average separation of fields in functionals on which T acts,
z) does
ra (w,
z), namely, expression (A.2) with
not contribute and we can use the simplified form for ra (w,

z) set to zero. In fact we can simplify the first term even further, approximating ra (w,
z)
ra (w,
as

1
ra (w,
z) =
1 (z, w)
ra e/2
ra K(w)K
(z w)

(A.4)
+ terms of higher order in or (z w), (z w)
.

z) =
ra (w,
This is correct and adequate, but

it must
be kept in mind that T so defined acts on operators
with a point-splitting separation , and in taking the small -limit we must preserve this
inequality.
References
[1] See, for example, D. Gross, R. Pisarski, L. Yaffe, Rev. Mod. Phys. 53 (1981) 43 and references therein.
[2] H. Lin, J.M. Maldacena, Phys. Rev. D 74 (2006) 084014, hep-th/0509235.
[3] D. Karabali, V.P. Nair, Nucl. Phys. B 464 (1996) 135;
D. Karabali, V.P. Nair, Phys. Lett. B 379 (1996) 141;
D. Karabali, V.P. Nair, Int. J. Mod. Phys. A 12 (1997) 1161.
239
[4] D. Karabali, C. Kim, V.P. Nair, Nucl. Phys. B 524 (1998) 661.
[5] D. Karabali, C. Kim, V.P. Nair, Phys. Lett. B 434 (1998) 103.
[6] H. Schulz, hep-ph/0008239;
V.P. Nair, Nucl. Phys. B (Proc. Suppl.) 108 (2002) 194, Talk at Lightcone Physics: Particles and Strings, Trento,
September 2001.
[7] There have been a number of other analytic attempts and approaches, some of them related to ours, for YangMills
in 2 + 1 dimensions. Some relevant articles are: M.B. Halpern, Phys. Rev. D 16 (1977) 1798;
M.B. Halpern, Phys. Rev. D 16 (1977) 3515;
M.B. Halpern, Phys. Rev. D 19 (1979) 517;
I. Bars, F. Green, Nucl. Phys. B 148 (1979) 445;
J. Greensite, Nucl. Phys. B 158 (1979) 469;
J. Greensite, Nucl. Phys. B 166 (1980) 113;
D.Z. Freedman, R. Khuri, Phys. Lett. A 192 (1994) 153;
M. Bauer, D.Z. Freedman, Nucl. Phys. B 450 (1995) 209;
F.A. Lunev, Phys. Lett. B 295 (1992) 99;
O. Ganor, J. Sonnenschein, Int. J. Mod. Phys. A 11 (1996) 5701;
S.R. Das, S. Wadia, Phys. Rev. D 53 (1996) 5856;
I.I. Kogan, A. Kovner, Phys. Rev. D 52 (1995) 3719;
I.I. Kogan, A. Kovner, hep-th/0205026;
P. Mansfield, D. Nolland, JHEP 9907 (1999) 028;
P. Mansfield, JHEP 0404 (2004) 059;
S.G. Rajeev, hep-th/0401202;
P. Orland, G. Semenoff, Nucl. Phys. B 576 (2000) 627;
P. Orland, Phys. Rev. D 70 (2004) 045014.
[8] P. Orland, Phys. Rev. D 71 (2005) 054503;
P. Orland, Phys. Rev. D 74 (2006) 085001;
P. Orland, Phys. Rev. D 75 (2007) 025001;
P. Orland, arXiv: 0704.0940 [hep-ph].
[9] M. Teper, Phys. Rev. D 59 (1999) 014512;
B. Lucini, M. Teper, Phys. Rev. D 66 (2002) 097502.
[10] B. Bringoltz, M. Teper, Phys. Lett. B 645 (2007) 383, hep-th/0611286.
[11] R.G. Leigh, D. Minic, A. Yelnikov, Phys. Rev. Lett. 96 (2006) 222001, hep-th/0512111;
R.G. Leigh, D. Minic, A. Yelnikov, hep-th/0604060.
[12] O. Philipsen, H. Wittig, Phys. Lett. B 451 (1999) 146;
P.W. Stephenson, Nucl. Phys. B 550 (1999) 427.
[13] S. Kratochvila, P. de Forcrand, Nucl. Phys. B 671 (2003) 103.
[14] A. Armoni, M. Shifman, Nucl. Phys. B 664 (2003) 233;
A. Armoni, M. Shifman, Nucl. Phys. B 671 (2003) 67.
[15] R.D. Pisarski, Phys. Rev. D 74 (2006) 121703.
Metastable flux configurations and de Sitter spaces

Katrin Becker , Yu-Chieh Chung, Guangyu Guo
Department of Physics, Texas A&M University, College Station, TX 77843, USA
Received 25 June 2007; accepted 10 September 2007
Abstract
We derive stability conditions for the critical points of the no-scale scalar potential governing the dynamics of the complex structure moduli and the axio-dilaton in compactifications of type IIB string theory
on CalabiYau three-folds. We discuss a concrete example of a T 6 orientifold. We then consider the
four-dimensional theory obtained from compactifications of type IIB string theory on non-geometric backgrounds which are mirror to rigid CalabiYau manifolds and show that the complex structure moduli fields
can be stabilized in terms of HRR only, i.e. with no need of orientifold projection. The stabilization of all
the fields at weak coupling, including the axio-dilaton, may require to break supersymmetry in the presence
of HNS flux or corrections to the scalar potential.
Published by Elsevier B.V.
1. Introduction
There are many ways to obtain models of particle phenomenology using string theory. A good
starting point is to construct a model with N = 1 supersymmetry in four dimensions. One can
obtain such models, for example, by compactifying M-theory on G2 -holonomy manifolds, Ftheory on CalabiYau four-folds or type II theories on CalabiYau orientifolds. It is a beautiful
fact that these models have a moduli space of vacuum states. However, concrete predictions can
only be made if the mechanism which picks the vacuum state of string theory can be identified.
By including fluxes as background fields the continuous ambiguity associated with the vacuum
expectation values of the moduli fields is replaced by a discrete freedom associated with the
choice of flux numbers. However, the number of possible vacuum states is still enormous and it
E-mail addresses: kbecker@physics.tamu.edu (K. Becker), ycchung@physics.tamu.edu (Y.-C. Chung),

guangyu@physics.tamu.edu (G. Guo).
0550-3213/$ see front matter Published by Elsevier B.V.
doi:10.1016/j.nuclphysb.2007.09.019
K. Becker et al. / Nuclear Physics B 790 (2008) 240257
241
has been argued to built a whole landscape of solutions. However, most of these string theory
backgrounds have flat directions and the number of solutions with all moduli stabilized is very
limited.
Stabilizing all the scalar fields associated with a CalabiYau compactification of string theory
at weak coupling is a particularly hard problem. In the context of compactifications of type IIB
string theory on a CalabiYau orientifold, for example, one of the fields which is conventionally
stabilized using fluxes is the axio-dilaton. This removes the arbitrariness associated with the vacuum expectation value of this field. However at the same time this means that the string coupling
constant is no longer a parameter whose value we can freely choose but it is determined in terms
of fluxes. Weak coupling can then only be achieved if some flux numbers can be chosen to be
large. But taking such a limit and making the moduli fields heavy is difficult, and it has been conjectured in Ref. [1] that it is actually impossible. This situation may change once supersymmetry
is broken and as a result it is very important to determine the properties of flux configurations
leading to stable critical points of the scalar potential while breaking supersymmetry. This is the
aim of the present paper.
Lets illustrate this idea in the interesting example of the KKLT model [2]. Here complex
structure moduli and the axio-dilaton acquire an expectation value due to perturbative fluxes
while preserving an N = 1 supersymmetry. Non-perturbative corrections to the superpotential
cause the radial modulus to become heavy compared to the AdS cosmological constant while
the masses of the complex structure moduli will generically be of the order to the inverse AdS
length and may not be heavy enough to be considered stabilized [1]. This situation changes once
these vacua are lifted to dS spaces. According to Ref. [2] this can be achieved by assuming the
presence of an anti-D3 brane which contributes a factor
V
1
,
(Im )3
(1.1)
to the scalar potential. Once this contribution is taken into account the potential for the radial
modulus displays a metastable minimum at which the scalar potential takes a positive value.
Moreover, the masses of the moduli are of the order to the AdS scale which can be much larger
than the dS scale and as a result after supersymmetry is broken all the moduli fields can turn out
to be heavy enough.
Adding anti-D3 branes is one way to uplift the potential to positive values. Following the
argument of Ref. [4], it should also be possible to obtain a potential contribution resembling the
one resulting from anti-D3 branes by considering flux configurations for which DI W = 0 for
some I . From the no-scale form of the potential it follows that such a contribution is positive and
its dependence on is precisely equal to the one originating from anti-D3 branes. This makes it a
natural alternative to the KKLT model. Since DI W = 0 the flux can no longer be imaginary selfdual (ISD) but will acquire an imaginary anti-self dual (IASD) component. Requiring that the
scalar potential is critical in the complex structure and axio-dilaton directions imposes conditions
on the fluxes which we will derive in the present paper while the radial modulus is not stabilized.
We then consider the four-dimensional theory obtained from compactifications of type IIB
string theory on backgrounds which are mirror to rigid CalabiYau manifolds, i.e. non-geometric
backgrounds with no Khler structure. In this case the flux induced superpotential does depend
explicitly on all scalar fields, i.e. the complex structure moduli and the axio-dilaton. Mirror symmetry implies that on the type IIB side the Khler potential for the axio-dilaton differs from the
conventional one obtained from dimensional reduction [1]. This fact enables us to find a scalar
potential which stabilizes all the complex structure moduli in terms of RR fluxes only while
242
requiring no orientifold charge. However the axio-dilaton is not fixed and slides off to weak
coupling. The axio-dilaton could be stabilized if HNS is taken into account and supersymmetry
is broken to render the scalar fields heavy enough. Another possibility is to take perturbative
corrections to the Khler potential and non-perturbative corrections to the superpotential into
account [1].
The organization of this paper is as follows. In Section 2 we consider geometric compactifications of type IIB string theory on CalabiYau three-folds. We derive the conditions imposed
on the flux configurations to lead to stable critical points of the scalar potential in the complex
structure and axio-dilaton directions. We explicitly show that the critical points do correspond to
minima of the potential by computing the Hessian matrix. We illustrate the idea in the example of
a torus orientifold. In Section 3 we consider the four-dimensional theory obtained from compactifications of type IIB strings on mirrors of rigid CalabiYau manifolds. We find a scalar potential
which stabilizes all the complex structure moduli in terms of RR fluxes only while requiring no
orientifold charge. We discuss several possibilities to stabilize the axio-dilaton at weak coupling.
We then end with some conclusions and speculations.
2. Type IIB string theory compactified on CalabiYau three-folds
In this section we discuss geometric compactifications of type IIB strings and analyze the
critical points of the scalar potential. To set up the notation we start in Section 2.1 deriving the
form of the scalar potential following closely Ref. [5]. In Section 2.2 we derive the conditions to
obtain a critical point of the potential. In Section 2.3 we explicitly check that the critical points
correspond to minima by computing the Hessian matrix. In Section 2.4 we present a concrete
example.
2.1. The scalar potential
Our starting point is the low-energy effective action of type IIB strings in the ten-dimensional
Einstein frame

2
F(5)
1
M M
GG
SIIB = 2
d 10 x g R
12 Im
4 5!
2(Im )2
210
1
C(4) G G
+ Sloc .
(2.1)
2
Im
8i10
Here the axio-dilaton is written in terms of the RR scalar C(0) and the dilaton according to
= C(0) + ie ,
(2.2)
and
1
1
F(5) = F(5) C(2) HNS + B(2) HRR ,
(2.3)
2
2
where HRR and HNS are the two three-forms with potentials C(2) and B(2) respectively and
G = HRR HNS . The condition that F(5) is self-dual should be imposed by hand. The Bianchi
identity for the five-form field is
2
d F(5) = HNS HRR + 210
T3 3loc .
(2.4)
By integrating the Bianchi identity over the internal manifold M6 , we get

1
HNS HRR + Qloc
3 = 0,
(2)4 2
243
(2.5)
M6
2 T = (2)4 2 . This identity means the sum of the D3

where we have used the relation 210
3
charges from background fields and localized sources vanishes. From Eq. (2.1) one obtain the
four-dimensional scalar potential by dimensional reduction

GG
i
1
6 GG
d
y
g
.
V=
(2.6)
2
2
3
3
Im
Im
2410 (Im )
410 (Im )
M6
M6
By using the flux induced superpotential [3] which is explicitly given by

G ,
W=
(2.7)
M6
and the Khler potential

K = 3 log i( )
log i( ) log i
,
(2.8)
M6
where is the radial modulus, the scalar potential (2.6) can be transformed into the standard
N = 1 supergravity form

V = eK g a b Da W Db W 3|W |2
(2.9)
where a and b label all moduli and the axio-dilaton. Because the superpotential is independent
of the scalar potential takes the no-scale form
V = eK FI F I ,
(2.10)
where I and J label the complex structure moduli and the axio-dilaton. Here and in the following
we will be using the notation of [6]
FI = DI W,
ZI J = DI DJ W,
UI J K = DI DJ DK W,
(2.11)
and indices are raised using the inverse of the Khler metric gI J = I J K.
2.2. Critical points of the scalar potential
The scalar potential will be critical1 in the complex structure and axio-dilaton directions if the
first derivatives vanish, i.e. if

I V = eK ZI J F J + FI W = 0.
(2.12)
1 In Ref. [18] stability conditions in the context of flux compactifications were also discussed.
244
One, but not the most general, solution of this condition is given by flux configurations satisfying
FI = 0. Using the explicit expression for the superpotential we have

1
,
Fi =
(2.13)
G i and F =
G

M6
M6
where i is the basis of harmonic (2, 1) forms and with lower case indices i, j we label the
complex structure moduli only. This implies that the non-vanishing components of G can lie in
the (0, 3) or (2, 1) directions. In other words, G is ISD, G = iG. Moreover, this critical point
is stable because the scalar potential (2.10) is positive semi-definite and at the critical points the
potential vanishes.
In the following we would like to find the most general solution of Eq. (2.12). We start by
rewriting Eq. (2.12) in the form
Z F + Zj F j + F W = 0,
Zi F + Zij F j + Fi W = 0.
Note that
M6 G
Zij = ij k
M6
(2.14)
1
Z i =

i ,
G
Z = 0.
(2.15)
M6
A simple computation (we include the details in Appendices A and B) shows that the first condition in Eq. (2.14) is equivalent to

(2.16)
G G = 0,
M6
while the second condition leads to

+ ij k Ai B j = 0.
(B B k + AA k )
(2.17)
M6
Here we introduced the Hodge decomposition
G = A + Ai i + B i i + B
(2.18)
and ij k are the Yukawa couplings. From here we conclude that the potential for the scalars (2.10)
does not have a critical point for an arbitrary choice of flux. Only if Eqs. (2.16) and (2.17) are
satisfied can we find a critical point in all directions except the size. This is not always possible.
If HNS = 0, for example, then the dilaton cannot be stabilized since the only non-vanishing
contribution to the dilaton potential comes from the overall factor eK . As a result no critical
point exists since Eq. (2.16) is violated.
It is not difficult to see that all flux combinations can lead to critical points of the potential
except if G is given by a combination of the following components
G(3,0) + G(0,3) ,
G(3,0) + G(2,1) ,
G(3,0) + G(0,3) + G(2,1) ,
(2.19)
or their complex conjugates. A flux of the form G(3,0) + G(0,3) , for example, is easily seen to
violate the condition (2.16).
245
Among the possible flux combinations leading to critical points of the scalar potential only
a flux lying in the (2, 1) or (1, 2) directions preserves supersymmetry. The (2, 1) component
obviously preserves supersymmetry, as it satisfies
DI W = D W = 0.
(2.20)
However a flux in the (1, 2) direction also preserves supersymmetry if accompanied by a change
in the sign of the tadpole due to fluxes. The reason for this is that type IIB supergravity in ten
dimensions is invariant under the change of sign of all RR fluxes. Changing the signs of RR
and as a result a flux lying in the (2, 1) direction should lead to the
fields replaces G by G
same physics as a flux in the (1, 2) direction. The (1, 2) component does satisfy the conventional
supergravity constraint DI W = D W = 0, but with a superpotential given by

.
W= G
(2.21)
M6
The derivation of this superpotential will be discussed in Appendix B. The two superpotentials
W and W are related to each other by a CPT transformation. Any other flux components satisfying Eqs. (2.16) and (2.17) will not preserve supersymmetry and lead to a positive cosmological
constant or vanishing cosmological constant if only a (3, 0) (or (0, 3)) component is turned on.
On the other hand, due to the no-scale structure of the potential the radial modulus cannot be
stabilized.
2.3. The Hessian matrix
The no-scale potential is positive definite. As a result solutions which lead to a vanishing
potential at the critical point V are necessarily stable. However, we are interested in solutions
for which V > 0 and as a result we have to check the stability of the solutions. In order to
determine if the critical points are stable we compute the Hessian matrix H . It turns out that it
only has positive eigenvalues which means that the critical points are minima in the complex
structure and axio-dilaton directions. Indeed, the second derivatives of the scalar potential are
given by

I J V = eK UI J K F K + 2ZI J W ,

I J V = eK gI J FK F K RILJK FL F K + 2FI FJ + ZI L Z JK g LK + gI J |W |2 .
(2.22)
The critical points will be stable if
d 2 = H dw dw 0,
(2.23)
where w labels all coordinates, i.e. and label the axio-dilaton, complex structure moduli
and their complex conjugates. Using formulas which are explicitly presented in Appendix A we
obtain

d 2 = eK g Z Z dw dw + g U U dw dw
(2.24)
where U = D D D W and Z = D D W are the generalization of UI J K and ZI J . As a
result the Hessian matrix is positive semi-definite and the critical points correspond to minima.
246
2.4. An example
In this section we describe a concrete example in terms of a type IIB orientifold compactification. This example is closely related to examples discussed in Refs. [4] and [7]. We will be
following their notation. Let x i and y i , for i = 1, 2, 3 be the six real coordinates on T 6 . These coordinates are subjected to the periodic identifications x i x i + 1 and y i y i + 1. The complex
structure is parameterized by complex parameters ij , and
zi = x i + ij y j ,
(2.25)
are global holomorphic coordinates. The explicit orientifold is T 6 /R(1)FL , where R is the
involution which changes the sign of all torus coordinates, R: (x i , y i ) (x i , y i ). The holomorphic three-form is
= dz1 dz2 dz3 ,
(2.26)
and the metric is
ds 2 = dzi d z i .
We choose the following orientation

dx 1 dx 2 dx 3 dy 1 dy 2 dy 3 = 1,
(2.27)
(2.28)
T6
and the basis of H 3 (T 6 , Z):

0 = dx 1 dx 2 dx 3 ,
1
ij = ilm dx l dx m dy j , 1 i, j 3,
2
1
ij
= j lm dy l dy m dx i , 1 i, j 3,
2
0 = dy 1 dy 2 dy 3 ,

which satisfies T 6 I J = IJ . The fluxes can be expanded in this basis
1
HRR = a 0 0 + a ij ij + bij ij + b0 0 ,
(2)2
1
HNS = c0 0 + cij ij + dij ij + d0 0 .
(2)2
(2.29)
(2.30)
Here we take a 0 , a ij , b0 , bij , c0 , cij , d0 , dij to be even integers, so that all the O3-planes are of
the standard type and the issues regarding flux quantization discussed in Ref. [8] can be avoided.
In this case, the total number of O3-planes is 64 and each plane has D3-brane charge 1/4.
For simplicity we only turn on the diagonal components of the flux, so that we can set the offdiagonal components of ij equal to zero at the critical points. This condition can be imposed by
restricting to an enhanced symmetry locus on the moduli space of the T 6 [4]. For example, we
will consider configurations which are symmetric under

R1 : x 1 , x 2 , x 3 , y 1 , y 2 , y 3 x 1 , x 2 , x 3 , y 1 , y 2 , y 3 ,

R2 : x 1 , x 2 , x 3 , y 1 , y 2 , y 3 x 1 , x 2 , x 3 , y 1 , y 2 , y 3 .
(2.31)
247
Only the diagonal components of the complex structure ij , and the three forms 0 , ii , 0 ,
ii are preserved under these symmetries, so that the only non-vanishing flux components are
a 0 , a ii , b0 , bii and c0 , cii , d0 , dii . We are left with 3 non-vanishing complex moduli and the
axio-dilaton.
To use the conditions (2.16) and (2.17) which we derived in Section 2.2, we need to transform
the scalar potential (2.6) into the standard N = 1 supergravity formula (2.9). For tori having a
general complex structure the result is complicated (see for example [15] and [4]). However for
tori with diagonal complex structure, we can express the scalar potential in the form

3

K
ij

g Di W Dj W + g D W D W ,
V =e
(2.32)
i,j =1
with superpotential (2.7) and Khler potential,

K = 3 log i( )
log i( ) log i(1 1 )(2 2 )(3 3 ) ,
(2.33)
where we used i to replace ii . Before we proceed we have one more comment. Generally we
can only set ij = 0 (for i = j ), after computing the first derivative of the scalar potential (2.6),
but on the symmetric locus, the criticality conditions ij V = 0 (for i = j ) are automatically
satisfied. As a result we can set ij = 0 (for i = j ) at the beginning of the computation and only
deal with the conditions ii V = 0. However, when computing the second derivatives we cannot
set ij = 0 before we differentiate, as there are non-vanishing terms of the form 2ij V , which
will disappear if we set ij = 0 (for i = j ) at the beginning.
Next we consider a flux in the (2, 1) + (1, 2) direction, so the conditions (2.17) and (2.16)
take the form
ij k Aj B k = 0 and gi j Ai B j = 0.
(2.34)
Since we are working with a torus we set 123 = 1 and one solution to the above condition is
A3 = B 3 = 0,
A1 B 2 = B 1 A2 ,
A1 B 1
A2 B 2
+
= 0.
(Im 1 )2 (Im 2 )2
(2.35)
For the concrete torus orientifold we are considering the tadpole cancellation condition takes the
form

i
= 32.
GG
(2.36)
2 Im (2)4 2
T6
In the following we will present a concrete solution of Eq. (2.35). For simplicity we redefine
the parameters according to
Ai = 2i Im i Im A i ,
and B i = 2i Im i Im B i
(2.37)
and drop the tilde in the following. The conditions (2.35) and (2.36) can be written as
A1 B 2 = B 1 A2 ,
B 1 B 1 = B 2 B 2 ,
3

A2 A 2 B 2 B 2 Im
Im i = 4
i=1
and the non-vanishing components of HRR and HNS are

a 0 = Im A1 + A2 + B 1 + B 2 ,
(2.38)
248

a 11 = Im A1 1 + A2 1 + B 1 1 + B 2 1 ,

a 22 = Im A1 2 + A2 2 + B 1 2 + B 2 2 ,

a 33 = Im A1 3 + A2 3 + B 1 3 + B 2 3 ,

b0 = Im A1 1 2 3 + A2 1 2 3 + B 1 1 2 3 + B 2 1 2 3 ,

b11 = Im A1 2 3 + A2 2 3 + B 1 2 3 + B 2 2 3 ,

b22 = Im A1 1 3 + A2 1 3 + B 1 1 3 + B 2 1 3 ,

b33 = Im A1 1 2 + A2 1 2 + B 1 1 2 + B 2 1 2 ,

c0 = Im A1 + A2 + B 1 + B 2 ,

c11 = Im A1 1 + A2 1 + B 1 1 + B 2 1 ,

c22 = Im A1 2 + A2 2 + B 1 2 + B 2 2 ,

c33 = Im A1 3 + A2 3 + B 1 3 + B 2 3 ,

d0 = Im A1 1 2 3 + A2 1 2 3 + B 1 1 2 3 + B 2 1 2 3 ,

d11 = Im A1 2 3 + A2 2 3 + B 1 2 3 + B 2 2 3 ,

d22 = Im A1 1 3 + A2 1 3 + B 1 1 3 + B 2 1 3 ,

d33 = Im A1 1 2 + A2 1 2 + B 1 1 2 + B 2 1 2 .
(2.39)
Usually one starts with certain flux numbers and then determines the values of moduli fields.
Here we solve the inverse problem, namely, we start with the value of the moduli and determine
the flux numbers which stabilize the moduli at the given values. To solve Eq. (2.38) using even
flux numbers (2.39) we use the ansatz
Im = 4,
Im 1 = Im 2 = Im3 = 1.
(2.40)
So one solution of Eq. (2.38) is

A1 = 3i,
A2 = 3i,
B 1 = 2 + 2i,
B 2 = 2 + 2i.
(2.41)
From Eq. (2.39), we can explicitly compute the flux numbers and obtain
0 11 22 33
a , a , a , a = (16, 24, 24, 16),
(b0 , b11 , b22 , b33 ) = (16, 0, 0, 16),
0 11 22 33
c , c , c , c = (4, 0, 0, 4),
(d0 , d11 , d22 , d33 ) = (4, 6, 6, 4)
(2.42)
which are all even integrals.

3. Type IIB mirrors of type IIA strings compactified on rigid CalabiYau three-folds
In this section we would like to generalize the previous analysis to type IIB theories which
arise as mirrors of type IIA models compactified on rigid CalabiYau three-folds, i.e. with
h2,1 = 0. On the type IIB side these correspond to models with h1,1 = 0 and consequently are
not ordinary CalabiYau manifolds since a Khler form is missing but can nevertheless be described using conformal field theory techniques. Here we will be interested in the properties of
249
the resulting four-dimensional theories which contain h2,1 + 1 four-dimensional N = 1 chiral

superfields originating from the complex structure moduli and the axio-dilaton. The number of
these fields will in general be reduced if we consider an orientifold projection.
It has been shown in Ref. [1] that for compactifications of type IIB strings on backgrounds
with no Khler structure the Khler potential for the axio-dilaton and the complex structure is

K = 4 log i( ) log i ,
(3.1)
which differs by a subtle factor 4 from the conventional Khler potential for the axio-dilaton.
This unconventional factor 4 has the consequence that supersymmetric flux configurations are
no longer required to be ISD [1]. The Khler potential (3.1) also causes the scalar potential to
display new and interesting properties. In order to illustrate this imagine one considers a real
three-form flux, i.e. a flux configuration with HNS = 0. Then

W = WRR = HRR ,
(3.2)
and the scalar potential can be written in the form

V = eK g i j Di WRR Dj WRR + |WRR |2 ,
(3.3)
which is positive definite and depends non-trivially on the complex structure. If

i V = 0 for i = 1, . . . , h2,1 ,
(3.4)
the potential is critical in all the complex structure directions. So for example, one solution of
Eq. (3.4) is given by
HRR = a( + ),
(3.5)
where a is some real constant. This equation determines the complex structure moduli. Indeed,
it turns out that this is nothing else than the equation defining a rank 1 attractor which is well
known from black hole physics. Eq. (3.5) can, for example, be explicitly solved in the large
complex structure limit as has been shown by Shmakova in Ref. [9] (see also Ref. [10]). These
critical points are stable since the only non-vanishing entries of the Hessian matrix are
2
i j V = 2eK gij
|WRR | .
(3.6)
The scalar potential (3.3) has been studied before in the literature in the context of nonsupersymmetric attractors (for a partial list of references on non-supersymmetric attractors
see [11]). In particular, the critical points of the potential are the solutions of

HRR = 2 Im eKcs W F i i ,
(3.7)
subjected to the constraint
Zij F j + 2Fi W = 0
which can be written as
2Fi W + ij k F j F k = 0.
(3.8)
(3.9)
250
Moreover, these critical points are stable since the Hessian matrix written in terms of2

d 2 = 2eK g Z Z dw dw + F F dw dw ,
(3.10)
is positive definite. In this form the critical points correspond non-supersymmetric attractor
points as described in Ref. [12]. This indicates that within a non-geometric model with h1,1 = 0
the proposal of Ref. [4] leads to an interesting new class of backgrounds in which all the complex
structure moduli can be stabilized in terms of RR fluxes only with no need of negative energy
sources like orientifold planes.
Using the solution (3.5) shows that the potential at the minimum satisfies
V > 0,
(3.11)
if a = 0 so the external space is dS. However, before we can conclude that supersymmetry is
spontaneously broken by the solution (3.5) we should take into account the dependence on the
axio-dilaton arising from the overall factor eK (Im )4 . This factor causes the potential to
slope to zero at infinity so a supersymmetric state is gained back at infinity and as it stands
the theory has no ground state at all. Here (as in [1]) we will simply assume that perturbative
corrections to the Khler potential and non-perturbative corrections to the superpotential could
achieve this stabilization and lead to a metastable ground state.
In order to stabilize the axio-dilaton using perturbative fluxes the only possibility is to use
a non-vanishing HNS flux. By including RR and NS three-form fluxes one obtains a fourdimensional superpotential which does depend non-trivially on all moduli fields. Any geometric
compactification would lead to a superpotential which is independent of the Khler moduli and
consequently the radial modulus would slide off to infinity. As a result even in the absence of
any type of corrections moduli stabilization may be possible within the non-geometric model by
including all possible fluxes. Moreover, in order to obtain moduli fields which are heavy enough
we may have to break supersymmetry [1]. But note that once the NS flux is non-vanishing the
scalar potential is no longer positive definite and it is not obvious that supersymmetry breaking
vacua, and in particular the phenomenologically interesting vacua leading to a positive cosmological constant, exist. As an illustrative toy example lets consider a non-geometric model with
h2,1 = 0, i.e. a model with only one massless scalar field, the axio-dilaton, with a Khler potential

K = 4 log i( ) ,
(3.12)
and a superpotential
W = WRR WNS ,
(3.13)
where WRR and WNS are constants. The condition for unbroken supersymmetry has one solution
only
=

1
Re(W NS WRR ) + 2i Im(W NS WRR ) .
WNS W NS
However, it is not difficult to see that the scalar potential is also critical if

1
i
=
Re(WNS WRR ) Im(WNS WRR ) ,
2
WNS W NS
2 Here the indices , label the complex structure moduli and their complex conjugates.
(3.14)
251
which leads to D W = 0 so that supersymmetry is broken. Moreover, the scalar potential at the
minimum is negative so that the external space is AdS. As a result supersymmetry breaking critical points of the potential do exist even though in this case they lead to an AdS space. However,
it is interesting that a single four-dimensional chiral field with a Khler potential of the form
(3.12) avoids the no-go theorem of Ref. [16] (see also [17]) according to which dS or Minkowski
space vacua with a broken supersymmetry are never possible in a theory with a single chiral field
for any superpotential if the Khler potential is K = n log[i( )] with 1 n 3. As a
result stable dS vacua are no longer excluded. It will be very interesting to see if by considering a
realistic model with a non-vanishing number of complex structure moduli fields stable critical
points of the potential at which supersymmetry is broken can be found. We leave this topic for
future research.
4. Conclusion and speculations
In this paper we have analyzed stability conditions of the no-scale scalar potential determining
the dynamics of the complex structure moduli and the axio-dilaton in geometric flux compactification of type IIB strings to four dimensions. In order to obtain critical points which do not
preserve supersymmetry an essential ingredient is the appearance of IASD flux components. But
not any IASD flux is allowed. Fluxes have to satisfy the conditions (2.16) and (2.17) to stabilize
all fields except the Khler structure moduli.
Searching for the critical points of the scalar potential obtained from compactifications of
type IIB strings on mirrors of rigid CalabiYau three-folds we discovered a fascinating and unexpected analogy to black hole physics and, in particular, to non-supersymmetric attractors. This
mapping was possible because of the peculiarities of the axio-dilaton Khler potential in the nongeometric setting derived recently in Ref. [1]. The similarities between the attractor mechanism
and flux compactifications have been known for some time (see for example Ref. [13]) but an
explicit mapping of the scalar potentials is new. Complex structure moduli stabilization can be
achieved in terms of HRR only, i.e. in terms of a real three-form. It will be interesting to further
explore if lessons learned from black hole physics can be used to discover properties of flux
vacua with a small and positive cosmological constant.
Moreover, compactifications of type IIB strings on mirrors of rigid CalabiYau manifolds
lead to a flux induced superpotential which depends non-trivially on all scalar fields even in the
absence of any non-perturbative effects once the RR and NS three-form fluxes are included. This
leads to the interesting possibility that perturbative fluxes alone may stabilize all moduli fields
once supersymmetry is broken.
Acknowledgements
We would like to thank Melanie Becker, Jason Kumar, Ergin Sezgin, Eva Silverstein, Cumrun
Vafa and Johannes Walcher for valuable discussions and communications. This work was supported in part by NSF grants PHY-0505757 and the University of Texas A&M. K.B. would like
to thank the Galileo Galilei Institute for Theoretical Physics and the CERN theory division for
hospitality and partial financial support during the completion of this work.
252
Appendix A
In this appendix, we present the details of some of the computations presented in this paper.
To set up our notation we start by reviewing a few basic formulas regarding CalabiYau manifolds [14]. On a CalabiYau three-fold, there exists a unique harmonic (3, 0) form , whose first
derivatives satisfy
= Ki + i and
=0
(A.1)
zi
z i
where i is a harmonic (2, 1) form. The Khler potential on the complex structure moduli space
is

Kcs = log i .
(A.2)
As is easy to check
i Kcs = Ki

and
gi j = i j Kcs =
i j
.

(A.3)
One important property of the (3, 0) form is that it is undefined up to multiplication by a

holomorphic function f (z)
f (z).
(A.4)
Under (A.4) the Khler potential transforms as

Kcs Kcs log f (z) log f(z),
(A.5)
which leaves the metric on moduli space invariant. For convenience, we can define a gauge
covariant derivative
i = Di = i + i Kcs ,
(A.6)
and thus under the Khler transformation, it transforms according to Di f Di , i.e. i

f i . One can also generalize the definition of the covariant derivative to other quantities which
transform like
(a,b) f a fb (a,b)
(A.7)
under the Khler transformation. In this case the covariant derivatives take the form
Di (a,b) = (i + ai Kcs ) (a,b) ,
Dj (a,b) = (j + bj Kcs ) (a,b) .
(A.8)
The partial derivatives i and i are to be replaced by ordinary covariant derivatives i , j when
acting on tensors. It is easy to see that under Khler transformations
Di (a,b) f a fb Di (a,b)
and
Dj (a,b) f a fb Dj (a,b) .
(A.9)
We also require
[Di , Dj ] = gi j ,
and Dk gi j = 0.
(A.10)
253
Using the above formulas, we can get the results quoted in the table below
Derivatives of the basis
Spans
Di = i
1
k
Di j =
ij k
Di j = gi j
Di = 0
(3, 0)
(2, 1)
(1, 2)
(A.11)
(0, 3)
where the Yukawa couplings are defined as

ij k = Di Dj Dk .
(A.12)
The above results are the tools needed to compute the first derivative of scalar potential (2.10).
Because the scalar potential is invariant under Khler transformation, i.e. a = b = 0, we can
transform the ordinary derivatives into covariant derivatives

I V = DI V = eK ZI J F J + FI W
(A.13)
with the notation (2.11). To obtain an explicit expression for I V = 0, we need to compute a few
quantities,

F i = Di W =
G i ,
M6
F = D W = W + KW =
ij k
Zij = Di Dj W =

1
Z i = D Di W =

Z = D D W
,
G
M6
G k ,
M6
i ,
G
M6
= F F
+ KF = 0.
As a result the critical condition I V = 0 can be explicitly written as

i G
G
i + G
0,
G

ij k =
k
i
j G = 0.
+ G

G
+ G j G
G
(A.14)
(A.15)
After using the Hodge decomposition for G
G = A + Ai i + B i i + B
the condition (A.15) can be further written in the form

G G = 0,
(B B k + AA k ) + ij k Ai B j = 0
(A.16)
(A.17)
254
which are Eqs. (2.16) and (2.17). To derive these equations, we have used the property that the
harmonic (2, 1) and (0, 3) forms are imaginary self-dual, and the harmonic (1, 2) and (3, 0)
forms are imaginary anti-self-dual on CalabiYau three-fold.
Now we are going to compute the second derivative of scalar potential by noting that
I J V = DI DJ V ,
I J V = DI DJ V
(A.18)
at the critical point I V = 0. After a little algebra, the second derivatives of the scalar potential
(2.10) are

I J V = eK UI J K F K + 2ZI J W ,

I J V = eK UJI K F K + FI FJ + ZI L Z JK g LK + gI J |W |2 ,
(A.19)
where UJI K = DJ DI DK W . The above formula can be easily transformed to (2.22) by using the
identity:
[DI , DJ ]FK = gI J FK + RI JK L FL .
(A.20)
To get expression (2.24), we need to generalize the definition of UI J K and ZI J to

U = D D D W
and U = U ,
(A.21)
and
Z = D D W
and Z = Z ,
(A.22)
where , , and label all coordinates, i.e. the axio-dilaton, complex structure moduli and their
complex conjugates. Using the results quoted in Table (A.11), we have

G

Uij k = Di Dj Dk W =
ij k ,

k ij k
G
,
Uij = Di Dj D W =

1
m F n .
Ukij
=
2 ij k m n
( )
(A.23)
One consequence of Eqs. (A.23) and (A.14) is

F Uij = F k Uij k ,
ZJI = gI J W,
ZJ I = 0,
UK JI = gI J FK ,
U i = U = UK JI = U j = 0.
The above expressions are useful to show the equivalence of (2.24) and (A.19).
(A.24)
255
Appendix B
In this appendix we explicitly show the appearance of the two superpotentials

,
W = G , and W = G
(B.1)
by dimensional reduction of ten-dimensional supergravity theories. Our convention is 01...9 = 1,

and
dx m0 dx mn =
1
m ...m m0 ...mn dx mn+1 dx m9 .
(9 n)! n+1 9
We take the type IIB effective action (2.1) together with the local terms are

+ p
Sloc =
d p+1 Tp G
Cp+1 .
R 4
(B.2)
(B.3)
R 4
To perform the dimensional reduction, we assume that the metric is independent of external
coordinates
ds 2 = e2A(y) dx dx + e2A(y) g mn (y) dy m dy n .
The Einstein equation is

1
2
RMN = k10
TMN gMN T
8
(B.4)
2
S
with TMN =
.
g g MN
The non-compact components of the Einstein equation can be written as

1
1 8A
1 loc
2
2
2
loc
R =
(m ) g + k10 T T g .
|G| e
8 Im
4
8
(B.5)
(B.6)
On the other hand, using the metric (B.4), we obtain

R = e2A 2 Ag
(B.7)
which yields
loc
1
1
1 2 2A m
Tm T .
e
e2A |G|2 + e10A |m |2 + k10
(B.8)
8 Im
4
8
This can also be written in the form

2

1
e2A |G|2 + e6A (m )2 + m e4A
2 e4A =
2 Im
loc
1 2 2A m
+ k10
(B.9)
e Tm T .
2
To compute the equation of motion for C4 we only need to consider a few terms in the action
namely

p
1
C(4) G G
(5) F(5) 1
(B.10)
F
Cp+1 .
+
2
2
Im
2
810
8i10
2 A =
R 4
256
The appearance of extra factor

identity is
1
2
is a consequence of the self-duality of the five form. The Bianchi
GG
2
d F(5) =
T3 3loc .
+ 2k10
2i Im
(B.11)
As F(5) is self-dual, we have

F5 = (1 + ) d dx 0 dx 1 dx 2 dx 3
(B.12)
and the Bianchi identity becomes

i
2
mnp + 2e6A m e4A m + 2k10
T3 3loc .
e2A Gmnp G
12 Im
By summing or subtracting Eqs. (B.9) and (B.13), we get
2 =

2 e4A =

2
1
e2A |G i G|2 + e6A m m e4A
2 Im

loc
m
2 2A 1
+ 2k10
e
T3 3loc .
Tm T
4
(B.13)
(B.14)
The left-hand side of the above equation vanishes when integrated over a compact manifold M6 .
As a result there are two solutions
6 G = iG,
= e4A ,
D3,
with O3,
6 G = +iG,
= +e
with O3, D3.
4A
at the same time.

Notice that we cannot have O3 and D3
Using the results above we can perform the dimensional reduction

d 10 x gR = d 4 x g4 d 6 y g6 8(A)2 e4A .
(B.15)
(B.16)
Taking into account the fact the self-duality of the five-form we get

d 10 x
2
F(5)

=
d 4x
g4

d 6y
e4A
g6
(m )2 .
2
(B.17)
Since = e4A , this term gives the same contribution as the Einstein term
d x g R

= d 4 x g4

= d 4 x g4

4
= d x g4
10
2
F(5)
4
d 6y

g6 (m )2 e4A

g6 (m )2 4m m A

1
6
4A
mnp
4A 2
loc
d y g6
2e 10 T3 3 .
e Gmnp G
12i Im
d 6y
(B.18)
Where we have used the Bianchi identity (B.13). The second term in the last equation of (B.18)
will cancel the first term of Sloc , and the CS term cancels the second term of Sloc . At the end, the
scalar potential is

e4A
1
4
i G).
G 6 (G
Sv = 2
d x g4
2 Im
210
From this expression, we can write the scalar potential in the standard form with

, or W = G .
W = G
257
(B.19)
(B.20)
References
[1] K. Becker, M. Becker, J. Walcher, Runaway in the Landscape, arXiv: 0706.0514 [hep-th].
[2] S. Kachru, R. Kallosh, A. Linde, S. Trivedi, de Sitter vacua in string theory, Phys. Rev. D 68 (2003) 046005,
hep-th/0301240.
[3] S. Gukov, C. Vafa, E. Witten, CFTs from CalabiYau fourfolds, Nucl. Phys. B 584 (2000) 69, hep-th/9906070;
T.R. Taylor, C. Vafa, RR flux on CalabiYau and partial supersymmetry breaking, Phys. Lett. B 474 (2000) 130,
hep-th/9912152;
P. Mayr, On supersymmetry breaking in string theory and its realization in brane, Nucl. Phys. B 593 (2001) 99,
hep-th/0003198.
[4] A. Saltman, E. Silverstein, The scale of the no-scale potential and de Sitter model building, hep-th/0411271.
[5] S.B. Giddings, S. Kachru, J. Polchinski, Hierarchies from fluxes in string compactifications, Phys. Rev. D 66 (2002)
106006, hep-th/0105097.
[6] F. Denef, M.R. Douglas, Distributions of nonsupersymmetric flux vacua, hep-th/0411183.
[7] S. Kachru, M. Schulz, S.P. Trivedi, Moduli stabilization from fluxes in a simple IIB orientifold, hep-th/0201028.
[8] A.R. Frey, J. Polchinski, N = 3 warped compactification, Phys. Rev. D 65 (2002) 126009, hep-th/0201029.
[9] M. Shmakova, CalabiYau black holes, Phys. Rev. D 56 (1997) 540, hep-th/9612076.
[10] G.W. Moore, Arithmetic and attractors, hep-th/9807087.
[11] F. Denef, Supergravity flows and D-brane stability, JHEP 0008 (2000) 050, hep-th/0005049;
K. Goldstein, N. Iizuka, R.P. Jena, S.P. Trivedi, Non-supersymmetric attractors, Phys. Rev. D 72 (2005) 124021,
hep-th/0507096;
K. Saraikin, C. Vafa, Non-supersymmetric black holes and topological strings, hep-th/0703214.
[12] S. Ferrara, G.W. Gibbons, R. Kallosh, Black holes and critical points in moduli space, Nucl. Phys. B 500 (1997) 75,
hep-th/9702103;
R. Kallosh, New attractors, JHEP 0512 (2005) 022, hep-th/0510024.
[13] G.W. Moore, Les Houches lectures on strings and arithmetic, hep-th/0401049.
[14] P. Candelas, X. de la Ossa, Moduli space of CalabiYau Manifolds, Nucl. Phys. B 355 (1991) 455.
[15] L. Andrianopoli, S. Ferrara, M. Trigiante, Fluxes, supersymmetry breaking and gauged supergravity, hepth/0307139.
[16] O. Lebedev, V. Lowen, Y. Mambrini, H.P. Nilles, M. Ratz, Metastable vacua in flux compactifications and their
phenomenology, JHEP 0702 (2007) 063, hep-ph/0612035.
[17] R. Brustein, S.P. de Alwis, Moduli potentials in string compactifications with fluxes: Mapping the discretuum, Phys.
Rev. D 69 (2004) 126006, hep-th/0402088;
M. Gomez-Reino, C.A. Scrucca, Locally stable non-supersymmetric Minkowski vacua in supergravity, JHEP 0605
(2006) 015, hep-th/0602246;
O. Lebedev, H.P. Nilles, M. Ratz, de Sitter vacua from matter superpotentials, Phys. Lett. B 636 (2006) 126, hepth/0603047.
[18] D. Lst, S. Reffert, E. Scheidegger, W. Schulgin, S. Stieberger, Moduli stabilization in type IIB orientifolds (II),
Nucl. Phys. B 766 (2007) 178, hep-th/0609013.
A class of BPS time-dependent 3-charge microstates

from spectral flow
Jon Ford, Stefano Giusto , Ashish Saxena
Department of Physics, University of Toronto, Toronto, Ontario, Canada M5S 1A7
Received 23 May 2007; accepted 17 September 2007
Abstract
We construct an infinite family of asymptotically flat 3-charge solutions carrying D1, D5 and momentum
charges. Generically the solutions also carry two angular momenta. The geometries describe the spectral
flow of all the ground states of the D1D5 CFT. The family is parametrized by four functions describing
the embedding of a closed curve in R4 and an integer n labelling the spectral flow on the left sector. After
giving the general prescription for spectral flowing any of the ground states, we give an explicit example of
the construction. We identify the asymptotic charges of the resulting solution and show the matching with
the corresponding CFT result.
PACS: 11.25.-w; 04.65.+e; 04.70.-s
Keywords: String theory; Supergravity; BPS solutions
1. Introduction
The Mathur conjecture [1,2] provides a gravitational interpretation of the thermodynamical
characteristics of horizons and black hole systems. The conjecture proposes that instead of a
single classical black hole metric, there exists an ensemble of horizon-free geometries, each of
which possesses the same conserved charges and asymptotics as the original black hole. In this
model the horizon and the associated BekensteinHawking entropy should arise due to a statisti* Corresponding author.
E-mail addresses: jford@physics.utoronto.ca (J. Ford), giusto@physics.utoronto.ca (S. Giusto),

ashish@physics.utoronto.ca (A. Saxena).
doi:10.1016/j.nuclphysb.2007.09.008
J. Ford et al. / Nuclear Physics B 790 (2008) 258280
259
cal averaging over the ensemble [3]. Results from string theory, specifically the AdS/CFT duality
[4,5], have made it possible to explicitly find some of these geometries and to verify that they
possess the necessary properties to be classified as gravitational microstates.
So far the strongest evidence in support of the Mathur conjecture has come from the study
of the 2-charge D1D5 system, which consists of D1- and D5-branes wrapped on a five torus.
Although this system does not produce a black hole with a classical horizon of finite area, it
does possess degenerate states and therefore entropy. This entropy can be shown to result from a
stretched horizon generated by quantum corrections [6,7]. In [8] Lunin and Mathur constructed a
family of regular microstate geometries by performing a series of duality operations that mapped
the D1D5 system to a system consisting of a fundamental string with winding and momentum
charges. In this construction the gravity duals were parametrized in terms of a closed curve in R4
that characterizes the profile of the oscillating string. It has been shown that these geometries
are non-singular (with the exception of acceptable orbifold singularities) and horizon-free, and
that this family of microstate geometries accounts for a finite fraction of the entropy of the 2charge black hole [9]. In [10] it was shown that the chiral primaries in the NS sector are related
to these solutions by a spectral flow transformation. The work on the 2-charge system has since
been continued in [11,12], in which the complete set of microstates has been found and a precise
mapping rule between the CFT states and the gravity duals has been proposed.
The 2-charge system provides substantial motivation for the Mathur proposal. Yet another
non-trivial check on the conjecture is provided by considering the 3-charge system obtained by
adding momentum along the D1-branes. This D1D5P system is the prototypical example of
a 5D BPS black hole in string theory. Its naive geometry in the gravity picture is an extremal
ReissnerNordstrom type black hole, which possesses macroscopic entropy and a regular horizon [13]. The D1D5P system has been analyzed extensively in both the gravity and the field
theory pictures [14,15]. In [16] the entropy was computed microscopically by looking at the degrees of freedom in the world volume theory and exact agreement with the BekensteinHawking
entropy was found. Since the 3-charge system possesses all the physically relevant characteristics
of a black hole (i.e. macroscopic entropy and a horizon), finding the microstates of the D1D5P
system would represent a compelling test of validity of the Mathur conjecture. Unfortunately
constructing microstates in the 3-charge case poses a significant technical challenge. The techniques used in [8] cannot be repeated because dualities cannot be used to map the D1D5P
system to a simpler one. Some progress has been made in understanding 3-charge systems as
supertubes in M-theory [1720]. Despite the difficulty, gravity duals for some simple D1D5P
states have been found. For example, in [1] perturbative techniques were used to add a single unit
of momentum charge to a particular Ramond ground state and the resulting solution was shown
to be regular. In [2123] axially-symmetric microstates were constructed by adding momentum
to the simplest members of the 2-charge D1D5 ensemble. Again, these geometries were found
to be horizon-free and to possess the necessary properties of black hole microstates proposed by
the Mathur conjecture. In [25] some non-BPS microstates were found and shown to be smooth.
More general classes of 3- and 4-charge BPS solutions possessing at least one axial isometry have
been found in [26]. However, the solutions mentioned above represent only a small subset of the
3-charge microstates and further work in this area is necessary to understand the microscopic
description of black hole entropy.
In this article we present a new class of time-dependent, BPS gravity duals for the D1D5P
system, consisting of an infinite family of solutions parameterized by an integer and a closed
curve in R4 . This class of solutions represent the spectral flow of all Ramond ground states of
260
the D1D5 field theory. The procedure we use to construct these solutions is summarized by the
following basic steps:
We start with the supergravity solutions dual to the set of Ramond ground states of the
2-charge D1D5 system. We then set the charges equal and express the geometry in the
language of six-dimensional minimal supergravity.
After taking the near-horizon limit of the 2-charge geometries we add momentum by applying a spectral flow transformation on the left sector of the CFT [27,28]. This produces a
3-charge geometry with AdS3 S 3 asymptotics.
We then find an extension of the solution that produces a new asymptotically flat geometry
satisfying the equations of [29].
A similar procedure was used in [21] to construct an axially-symmetric 3-charge solution.
The layout of this paper is as follows. In Section 2 we review the generic form of solutions
in six-dimensional minimal supergravity and introduce the GMR notation [29] and equations of
motion. In Section 3 we give the 2-charge metrics that are dual to the Ramond ground states of the
D1D5 system, re-write them in the GMR notation and reduce to six-dimensions. In Section 4 we
add momentum to the system by applying a spectral flow transformation and re-diagonalize into
the GMR form. In Section 5 we show how the 3-charge near-horizon solution can be extended
to an asymptotically flat solution. In Section 6 we present an explicit example of our results by
constructing a new 3-charge gravity dual corresponding to a profile that breaks one of the two
axial symmetries of S 3 . In Section 7 we explicitly calculate the conserved charges for the special
case geometry and show these results to be in agreement with the dual CFT. Finally, in Section 8
we summarize our results and discuss possible future work.
2. Minimal supergravity in six dimensions
In this section we begin with a brief discussion of minimal supergravity in six dimensions. The
bosonic field content of the theory consists of a graviton g and a self-dual three-form. Solutions
of this theory can be lifted to ten-dimensional type IIB supergravity by adding a flat T 4 and
identifying the three-form as the field strength of the RamondRamond 2-form gauge field. In a
particular duality frame, solutions constructed in this manner carry equal D1 and D5 charges [24].
Conversely, ten-dimensional 2-charge solutions of type IIB theory with the charges set equal (so
that the dilaton is trivial) and then reduced along the T 4 can be embedded in six-dimensional
supergravity.
All supersymmetric solutions of this theory have been classified in [29]. In the following we
review their construction and equations of motion.
2.1. General form of supersymmetric solutions in 6D
In [29] it was shown that all supersymmetric solutions of minimal 6D supergravity can be
written in the following form (GMR form),

F
ds 2 = 2H 1 (dv + ) du + + (dv + ) + H hpq dx p dx q ,
(2.1)
2
G(3) = d6 C (2)

= d6 H 1 (dv + ) (du + ) + (dv + ) G + + 2 + 4 (DH + H ).

(2.2)
261
The metric hpq defines the four-dimensional base space; H , F are 0-forms, , are 1-forms
while G + and are 2-forms on the base space; d6 is the exterior derivative on the six-dimensional
manifold. It is important to note that H , F , , , and hpq will in general possess functional dependence on the v coordinate. A derivative with respect to v is denoted by a dot. All the
functions above are independent of u, hence u

is a null Killing vector of the geometry as required by supersymmetry.
The base metric is also required to possess three almost complex structures satisfying the
quaternionic algebra. In the following we will not distinguish the almost complex structures
from the associated 2-forms. Let J i (i = 1, 2, 3) be a basis of almost complex structures for the
base metric hpq : these are anti-self-dual 2-forms, with respect to hpq , that satisfy

dJ i = v J i .
(2.3)
Here and in the following, d refers to the exterior derivative operator on the base space. One also
defines an anti-self-dual 2-form which, loosely speaking, measures the v-dependence of the
base metric,
=
H ij k i pq j
J
J pq J k .

16
The self-dual 2-form G + is defined as

F
G + = H 1 (D)+ + D
2
(2.4)
(2.5)
where, for any p-form , the operator D is given by

D = d
(2.6)
and
D + 4 D
2
(the operator 4 uses the metric hpq ). Finally introduce the 1-form
(D)+ =
(2.7)
F
DF
(2.8)

.
2
2
With the above definitions the equations of motions for metric (2.1) and the self-dual 3-form
(2.2) can now be written in terms of four-dimensional operators (d, 4 ) on the base space

dJ i = v J i ,
(2.9)
L = +
D = 4 D,
(2.10)

D G = D 4 (DH + H ) ,
(2.11)

+

+
,
d(G + 2) = v G + 2 + 4 (DH + H )
(2.12)

H pq 2
1
h v (H hpq ) + v H hpq v (H hpq ) 2p Lp
4 D(4 L) =
2
4

1 +
4 G + 2 G + + 2 + 2H 1 4 (D) .
(2.13)
2
The first two equations (2.9) and (2.10) guarantee the existence of a Killing spinor so that the
solution is supersymmetric. Eqs. (2.11) and (2.12) are the equations of motion for the field
+
262
strength G(3) , which is just the Bianchi identity dG(3) = 0 since G(3) is self-dual in six dimensions. Eq. (2.13) is the only component of the Einstein equations that is not implied by the
previous equations.
3. 2-Charge geometries
3.1. Writing the D1D5 supergravity solutions in GMR and GH form
We begin with the supergravity solutions dual to the set of Ramond ground states of the D1
D5 system [8]. As discussed in the previous section we will take the D1 and D5 charges to be
equal. In this case the non-trivial part of the geometry can be written in six dimensions as a twodimensional fiber, spanned by coordinates t and y, over a four-dimensional base space, which is
just flat R4 [24]. The metric and the 2-form gauge field are given by [8],

ds 2 = (1 + H )1 (dt A)2 + (dy + B)2 + (1 + H )pq dx p dx q
= 2(1 + H )1 (dv + )(du

+ )
+ (1 + H )pq dx p dx q ,
(3.1)

(2)
1
C = (1 + H ) (dt A) (dy + B) + C
(du + )
= (1 + H )1 (dv + )
(3.2)
+ C
where in order to write the metric in the GMR form we have introduced light-cone coordinates
on the (t, y)-fiber and defined 1-forms, and ,
on the base space as
t +y
t y
u= ,
v= ,
2
2
A+B
AB
= ,
= .
2
2
(3.3)
(3.4)
The metric functions must satisfy the equations

d 4 d H = 0,
d C = 4 d H ,
d 4 dA = 0,
(3.5)
dB = 4 dA.
(3.6)
Note that these equations imply that the field strengths of and are self-dual and anti-self-dual,
respectively.
In [8] these geometries were constructed by performing a series of dualities on the geometry sourced by a fundamental string, multiply-wound on the y circle and carrying momentum
along y. By this construction the family of D1D5 geometries was parameterized by a closed
curve in R4 , which in the FP picture represents the profile of the string carrying momentum.
With the profile function expressed as xi = Fi (v), i = 1, . . . , 4, v [0, L], the solutions of equations (3.5) can be written in terms of the profile as [8],
Q
H =
L
L
dv
0
Q
A=
L
1
,
|x F(v)|2
L
dv
0
Fi
dxi .
|x F(v)|2
(3.7)
(3.8)
263
The parameter L is related to the length of the compactification circle of y and Q is related
to the D5 charge. For details see [2]. The forms C and B are defined implicitly by Eqs. (3.6).
Furthermore, the profile function F(v) is constrained by
F 2 = 1
(3.9)
in order to embed the solutions in minimal supergravity.

3.2. Computing C
The 2-form C appearing in the specification of the LuninMathur geometries may be given
an integral expression as follows. Recall that the defining equation for C is
d C = 4 d H
(3.10)
where
Q
H =
L
L
0
dv
.
|x F(v)|2
(3.11)
The star operation above is to be performed with respect to the flat metric on R4 . If one sets
F(v) = 0 in the expression for H , the dual is easily found to be

Q x12 + x22 x32 x42 12 34 11 33
C =
(3.12)
2 x12 + x22 + x32 + x42 (x12 + x22 )(x32 + x42 )
where
11 = x1 dx1 + x2 dx2 ,
12 = x1 dx2 x2 dx1 ,
33 = x3 dx3 + x4 dx4 ,
34 = x3 dx4 x4 dx3 .
(3.13)
The translational isometry of R4 implies that the solution for a general F(v) is given by replacing x by x F(v) in Eq. (3.12) and integrating over v.
Q
C =
2L
L
0
dv
2
2
|x F|2 (x1 F1 )2 + (x2 F2 )2 (x3 F3 )2 + (x4 F4 )2
(3.14)
where
2 = 12 34 11 33 ,
(3.15)
11 = (x1 F1 ) dx1 + (x2 F2 ) dx2 ,
(3.16)
12 = (x1 F1 ) dx2 (x2 F2 ) dx1 ,
(3.17)
33 = (x3 F3 ) dx3 + (x4 F4 ) dx4 ,
(3.18)
34 = (x3 F3 ) dx4 (x4 F4 ) dx3 .
(3.19)
We will use this procedure in Section 6 to derive an explicit example of a new 3-charge geometry
with a single axial isometry.
264
3.3. Base space in GibbonsHawking form

It will be useful to write the base space R4 in GibbonsHawking (GH) form [30] by changing
from Cartesian to GH coordinates as
,
x2 = 2 r sin sin
,
x1 = 2 r sin cos
2
2
2
2
+
+
,
x4 = 2 r cos sin
.
x3 = 2 r cos cos
(3.20)
2
2
2
2
After applying the above coordinate transformation the base metric in (3.1) becomes
ds42 = pq dx p dx q = V 1 (e)
2 + V ds32
(3.21)
where ds32 = dr 2 + r 2 d 2 + r 2 sin2 d 2 is the flat metric on R3 and e denotes the einbein on
the fiber of GH space
e = d + .
(3.22)
V and are a 0-form and a 1-form respectively on
R3
that satisfy the GH condition
d3 V = 3 d3 .
(3.23)
We denote by d3 the exterior differential on R3 and by 3 the Hodge-star operation with respect
to ds32 . In our case of flat R4 , we have explicitly
1
V = ,
r
= cos d.
(3.24)
The Hodge star operation with respect to the base space ds42 will be denoted as 4 , and we have
chosen the orientation1 such that r = +1.
3.4. Near-horizon limit
The near-horizon limit of metric (3.1) is obtained by dropping the one in 1 + H , that is by
sending 1 + H H . Then, the 2-charge geometry in the near-horizon limit is given by

2
= 2H 1 (dv + )(du
+ )
+ H V 1 (e)
2 + V ds32 ,
dsn.h.
(3.25)
(2)
(du + )
Cn.h.
= H 1 (dv + )
+ C.
(3.26)
4. Spectral flow to 3-charge system

Starting from the near-horizon geometry (3.25), momentum can be added by a spectral flow
operation, which in the gravity picture is the following coordinate change [27,28]
+v
where
=
2 2n
,
Ry
(4.1)
n Z,
1 Note that in Cartesian coordinates this orientation corresponds to

1234 = 1.
(4.2)
265
Ry is the radius of the y circle. Note that, as the generic 2-charge geometry depends on , the
geometry after spectral flow depends on v, via the combination + v. The action of spectral
flow distinguishes from the other coordinates on R4 and this is one of the primary reasons
why it is more convenient to work in GibbonsHawking coordinates. This also motivates the
following decomposition of the 1-forms and
= e 0 + 1
= e0 + 1 ,
(4.3)
3
where the 1-forms 1 and 1 have components only along R . After applying the coordinate
change (4.1) to the metric (3.25) the result can be brought back into the GMR form

F
2
1
dsn.h. = 2H (dv + ) du + + (dv + ) + H h pq dx p dx q

(4.4)
2
with
H
,
H =
1 + 0
H 2
F
,
= 0 2
2
2V (1 + 0 )
,
=
1 + 0
=
H 2 e
0
H 2
+2
1 + 0
V (1 + 0 )
V (1 + 0 )2
(4.5)
and
2 + V ds32 ,
h pq dx p dx q = V 1 (e)
e = d + , V = V (1 + 0 ), = 1 .
(4.6)
Here and in the following all the quantities after spectral flow are to be evaluated at + v
instead of . Since spectral flow on the gravity side is simply a coordinate transformation, the
new metric (4.4) by construction satisfies the GMR equations of motion and possesses the same
AdS3 S 3 asymptotics and supersymmetries as the metric before the spectral flow.
4.1. Notation
Since we will need to make reference to several different metrics (i.e. 2-charge and 3-charge
metrics with flat or AdS asymptotics) there is potential for confusion regarding the domain on
which operators or functions are defined. In an attempt to clarify matters we will make use of the
following notational conventions:
A bar indicates that the specified function or operator refers to the near-horizon 2-charge
metric (3.25) prior to spectral flow. For example 4 refers to the Hodge star operator with
respect to the flat R4 base metric.
A tilde refers to the metric (4.4), i.e. the 3-charge metric after spectral flow with AdS3 S 3
asymptotics. For example, 4 refers to the Hodge star operation with respect to the base
metric (4.6).
Neither a bar nor a tilde indicates that a function, form or operator is to be taken with respect
to the 3-charge metric with flat asymptotics to be constructed in the next section.
266
5. Extending to flat asymptotics

Our aim is to extend the near-horizon geometry of (4.4) to obtain an asymptotically flat 3charge solution. To this end, we should look for a solution where H is replaced by
H = 1 + H .
(5.1)
In order to satisfy the equations of motion, we will also need to modify the other quantities
appearing in the metric (4.4). In general one has
= + ,
(5.2)
= + ,
(5.3)
hpq = h pq + hpq ,
(5.4)
F = F + F.
(5.5)
We need to find (, , F, hpq ) such that the six-dimensional metric defined by H , , , F

and hpq is asymptotically flat and satisfies the equations of motion given in (2.9)(2.13).
5.1. Base metric
We know that by construction the near-horizon geometry satisfies the equations of motion.
We make the ansatz that the base metric hpq of the asymptotically flat geometry equals the base
metric h pq of the near-horizon geometry
hpq dx p dx q = h pq dx p dx q V 1 e2 + V ds32 ,
e = d + ,
(5.6)
with
V = V ,
= .
(5.7)
The main justification of this ansatz is that we know this is what happens in the axially-symmetric
case [21]. The fact that the base metric remains invariant provides a great deal of simplification,
for example the almost complex structures Ji = Ji remain the same and the Hodge star operator
is also invariant. Hence 4 = 4 .
5.2.
Under the assumption that the base metric is unchanged, the simplest way to satisfy Eq. (2.9)
is to assume that is also unchanged
= .
(5.8)
Then Eq. (2.10), the self-duality of the field strength of , is automatically satisfied as well. The
invariance of also greatly simplifies the process of solving for the remaining quantities. In
particular it guarantees that the differential operator D remains invariant.
267
5.3. and G +
In order to find the new and F , let us first look at the quantity G + which is defined in terms
of (D)+ and F in (2.5). First consider Eq. (2.11): as H changes as in (5.1), it is clear that in
order to satisfy this equation we must assume that G + changes
G + = G + + G + .
(5.9)
Similarly, Eq. (2.12) requires that both G + and change. The change of can be derived
directly from its definition (2.4), given the fact that H changes as in (5.1) and the assumption
that the base metric (and thus the complex structures J i ) do not change. For a metric of the
from (5.6), a convenient choice of the (almost) complex structures is
ij k j
dx dx k .
2
Substituting this into (2.4) we find
J i = e dx i V
(5.10)
H
i J i
4V
(5.11)
=
where
= 2 .
= = ( )
(5.12)
Similarly, for the near-horizon geometry one has

H
i J i .
4V
Thus, the change in is given by
=
= =
(e V 3 ).
4V
(5.13)
(5.14)
Since G + appears in the equations of motion in the combination G + + 2 , one might guess
that the change in G + must be related to ; more precisely, we take G + to be the self-dual
version2 of 2 ,
2
(5.15)
(e + V 3 ).
2V
One can now verify that the guess above solves both equations (2.11) and (2.12): using the fact
that the tilded quantities satisfy the equations, it is sufficient to check that the variations of
(2.11) and (2.12) hold, i.e.
G + =
D 4 + D G + = 0,

(G + + 2) = v (G + + 2) + 4 .
(5.16)
2 For a metric of the form ds 2 = V 1 e 2 + V ds 2 , with a 1d fiber and 3d base space, a self-dual form on the full space
4
3
can be constructed out of any 1-form f3d on the base space ds32 as: f(4d) = e f(3d) + V 3 f(3d) . Anti-self-dual forms
are similarly given by f(4d) = e f(3d) V 3 f(3d) .
268
5.4. F and
Having found the form of the new G + , one should determine and F in such a way that
Eq. (2.5) is satisfied.3 One should have
F
D = G + + (H + 1)G + .
(5.17)
2
A relation involving G + , that will be useful in solving (5.17), can be found by looking at the
gauge field. Start with the gauge field of the 2-charge system in the near-horizon region (3.26)
and apply the spectral flow transformation (4.1). Writing the four-dimensional 2-form C as
(D)+ +
C = e C 1 + C 2
(5.18)
where C 1 and C 2 are a 1-form and 2-form along R3 , we find the gauge field of the spectral-flowed
system to be

H
(2)
1
(du + )
e + C 1 C 1 + C.
C = H (dv + )
(5.19)
+ (dv + )
V
On the other hand, for a general supersymmetric solution the 3-form field strength should have
the form given in Eq. (2.2)
(3) = d6 C (2)
G

(du + )
= d6 H 1 (dv + )
(G + + 2)
+ 4 (DH + H ).
+ (dv + )
(5.20)
Using the facts that the GMR ansatz has no dependence on u and for our 3-charge metric dependence on v is directly related to by (4.1), the operator d6 can be re-written as
d6 = d + dv v = d + dv .
Define a 1-form

H
e + C 1 .
=
V
(5.21)
(5.22)
Comparison of (5.19) and (5.20) implies, after some lengthy algebraic manipulations, that

2 H
C 1
+ C 2 .
e + e
D = G + + 2 +
(5.23)
V
1 + 0
If we take the self-dual part of (5.23), and use (5.15), we find

e (V C 1 + 3 C 2 ) + V (V 3 C 1 + C 2 ) .
(D)+ = G + + H G + +
(5.24)
V
The last term on the r.h.s. of (5.24) can be made to vanish by taking the 2-form C to be anti-selfdual, with respect to the flat base metric
C = 4 C
V 3 C 1 + C 2 = 0.
(5.25)
3 Eq. (2.5) only involves (D)+ , and thus determines only up to an anti-self-dual 1-form, i.e. a 1-form such
that (D )+ = 0. In solving Eq. (2.5), we will have to guess the anti-self-dual part of . This guess will be justified
by proving the validity of the Einstein equation (2.13), which explicitly involves (D) .
269
Imposing condition (5.25) fixes the gauge invariance of the C defined by Eq. (3.6). One can show
that such a gauge can always be attained: for example, let C L be the solution of (3.6) in Lorentz
gauge (which is certainly an allowed gauge),
d 4 C L = 0.
(5.26)
C = C L 4 C L
(5.27)
Then
solves both (3.6) and (5.25).

Eq. (5.24) simplifies to
With this C,
(D)+ = G + + H G + .
(5.28)
Now comes quite close to solving the equation for (5.17). For a complete solution we need
to add an additional piece in order to produce the extra G + term that is missing from (5.28). Let
= +
(5.29)
where the 1-form should satisfy

F
(5.30)
D.
2
We know from the solution of [21] that and F vanish in the axially symmetric (i.e. independent) case. We have found only one solution of (5.30) with this property: this solution
has
(D )+ = G +
F = 0
(5.31)
and a that can be constructed as follows. Decompose as

= 2 e
0 + 1 .
(5.32)
If we set 0 = 0, we find, using the expression for G + in Eq. (5.15), that 1 must satisfy
V 1 + 3 (d3 )1 = 1 .
In order to find a solution of (5.33) we introduce a 2-form
struct ) on R4 satisfying
(5.33)
C
(analogous to the C used to con-
d C = 4 .
(5.34)
Such a C is guaranteed to exist if is chosen to satisfy the Lorentz gauge

d 4 = 0.
Moreover take
(5.35)
to be anti-self-dual (with respect to flat metric on
C = 4 C .
R4 )
(5.36)
As we explained above, this is a gauge choice for C that can always be attained. Decompose C
as
C = e C 1 + C 2 = e C 1 V 3 C 1 .
(5.37)
270
Then a satisfying (5.30) is given by

= 2 1 = 2 C 1 .
(5.38)
To summarize, we have found that

H
e + C1 + 2 C 1 ,
=
V
F = 0.
(5.39)
We now have a complete ansatz for the asymptotically flat 3-charge geometry that by construction
satisfies the first four equations of motion (2.9)(2.12). We are left to check that this ansatz
satisfies the Einstein equation (2.13). Since the other four equations of motion only depend on
the self-dual part of , satisfying the Einstein equation provides a very non-trivial validation of
the construction of the new metric functions. Using again the fact that the near-horizon geometry
satisfies (2.13), it is sufficient to consider only the variation of (2.13), which is given by the
following equation
4 D(4 v )
H
1
1
= hpq v2 (H hpq ) + hpq v2 hpq + hpq v2 hpq
2
2
2

1
1
1
+ v hpq v (H hpq ) + v H hpq v hpq + v hpq v hpq 2p (v )p
4
4
4

(G + + 2) 1 4 (G + + 2) (G + + 2)
4 (G + + 2)
2

+ 2H 1 4 (D) .
(5.40)
A straightforward but tedious computation shows that our ansatz satisfies this equation.
5.5. Summary of the asymptotically flat solution
Starting from a 2-charge geometry specified by V , , H , and ,
spectral flow generates a 3 ,
charge geometry with AdS3 S 3 asymptotics, described by V , , H , ,
F given in Eqs. (4.5),
(4.6). The continuation of this geometry to the asymptotically flat region is then given by:

F
2
1
ds = 2H (dv + ) du + + (dv + ) + H V 1 e2 + V ds32
(5.41)
2
where
H = 1 + H ,
V = V ,
= ,
= ,
F = F,

H
e + C1 + 2 C 1 .
=
V
(5.42)
Here C 1 and C 1 are the 1-forms appearing in the decompositions

C = e C 1 + C 2 ,
C = e C 1 + C 2
(5.43)
271
and C and C are 2-forms on R4 satisfying
d C = 4 d H , C = 4 C,

d C = 4 ,
C = 4 C .
(5.44)
6. Special case: A new 3-charge solution

In order to provide a concrete example we have applied the general results of Section 5 to a
specific profile, which leads to a new asymptotically flat, -dependent 3-charge geometry. For
this geometry we have also directly verified that the equations of motion are satisfied.
Consider the following profile function parameterized in Cartesian coordinates,

2v
2v
2v
2v
, cos sin
, sin cos
, sin sin
.
F(v) = a cos cos
(6.1)
2
L
2
L
2
L
2
L
Note that this profile corresponds to an SO(4) rotation of the axially-symmetric solution found
in [8,27,28]. The matrix that accomplishes the necessary rotation can be written as
0
sin 2
0
cos 2
0
cos 2
0
sin 2
=
(6.2)
sin
0
cos 2
0
2
0
sin 2
0
cos 2
where the rotation is parametrized by . It is clear that for = 0 this matrix becomes the identity
and profile (6.1) reduces to the one in [8,27,28] which leads to an axially-symmetric metric. It is
important to note that the spectral flow operation (4.1) singles out a particular coordinate of R4 ,
i.e. , and thus does not commute with SO(4). Consequently our new 3-charge geometry cannot
be produced by a rotation of the geometry of [21,22].
Due to the complexity of the metric functions describing the new 3-charge geometry, we
will present its construction in a series of steps. We start with the well know axially-symmetric
2-charge geometry found in [8,27,28] and then perform the rotation to produce the 2-charge
geometry associated with (6.1). We then write the 1-forms associated with this geometry in GMR
form. Finally, we give the asymptotically flat 3-charge geometry using the results of Sections 4
and 5.
6.1. 2-Charge geometry
We start with the axially symmetric 2-charge geometry defined by [8,27,28],
1
2
2

ds 2 = H (0)
dt A(0) + dy + B (0)

1 2
(0)

2
+H
r(d + cos d ) + ds3
r
(6.3)
where
H (0) =
,
(0)
4rc

r +c
Q
(0)
A = 1 (0) (d d ),
8 c
rc
(6.4)
(6.5)
272
(0)

Q r c
=
1 (d + d )
8 c rc(0)
(6.6)
and

a 2
.
(6.7)
4
We must now apply the rotation matrix (6.2) to the above metric. It can be shown the rotation
induces the following coordinate transformation in the GH frame
rc(0) =
r 2 + c2 + 2rc cos ,
c=
cos = cos cos cos sin sin ,

sin sin
tan =
,
cos cos sin + cos sin
cos sin sin + sin (cos sin + cos cos sin )
.
tan =
cos sin cos sin (sin sin cos cos cos )
(6.8)
(6.9)
(6.10)
After applying the above coordinate transformation to Eqs. (6.4)(6.6), we obtain the 2-charge
geometry associated with the profile (6.1). This rotated geometry is specified by the harmonic
function H and the 1-forms A and B,
Q
,
H =
4rc

r + c cos cos
sin sin
Q
d
d
+
d
,
A= 1
rc
1 cos
1 cos
8 c

cos + cos
sin sin
Q r c
1
d +
d + d
B=
rc
1 + cos
1 + cos
8 c
(6.11)
(6.12)
(6.13)
where
rc =
r 2 + c2 + 2rc cos .
(6.14)
In the next section, where we give the asymptotically flat 3-charge solution, it will be more useful
to have the above expressions in the GMR notation. This will allow us to directly apply the results
of Section 5. From Eqs. (6.12) and (6.13) we can write down the components of the 1-forms
and ,

Q cos (r rc + c cos ) + cos ((rc r) cos c)
0 =
(6.15)
,
4rc
2c(1 cos2 )
r = 0,
(6.16)

Q (rc r c cos ) sin sin
,
=
(6.17)
4rc
2c(1 cos2 )

Q c
=
(6.18)
4rc 2
and
=
r = 0,
Q cos ((r rc ) cos + c) + cos (rc r c cos )

,
4rc
2c(1 cos2 )
(6.19)
(6.20)
Q ((rc r) cos c) sin sin

,
4rc
2c(1 cos2 )
Q r rc
=
.
4rc 2c
=
273
(6.21)
(6.22)
6.2. New asymptotically flat 3-charge geometry

The metric of the asymptotically flat 3-charge geometry has the standard GMR form and can
be written as

F
ds 2 = 2H 1 (dv + ) du + + (dv + ) + H V 1 e2 + V ds32 .
(6.23)
2
In this section we find expressions for the metric functions , , F, H, , V , in terms of the
2-charge fields and ,
which are given explicitly in Eqs. (6.15)(6.22). From Eqs. (4.5) and
(5.42) one can find that H , , and F are given by
Q
,
4rc (1 + 0 )
=
,
(1 + 0 )
H =1+
F = 2
r 2 Q2
16rc2 (1 + 0 )
(6.24)
(6.25)
.
(6.26)
The base metric is completely specified by the three-dimensional 1-form and the scalar function V , which for this case can be written as
1
V = (1 + 0 ),
r
= d d + cos (1 + 0 ) d.
From Eq. (5.42) is given by

H
= e + C1 + 2 C 1 + 0 .
V
(6.27)
(6.28)
(6.29)
Here 0 denotes a constant 1-form, that is left undetermined by the equations of motion, and
will be fixed below by requiring asymptotic flatness. In order to find C 1 and C 1 we must solve
Eqs. (5.43) and (5.44) which we will do in the following. The piece denoted by can be written
in terms of and by using Eq. (4.5)
=

rQ2
(1 + 0 )(d + cos d) .
+
2
2
1 + 0
16rc (1 + 0 )
(6.30)
Asymptotic flatness requires that vanishes at large distances: we note that, for r
1, 0,
however the term proportional to e in behaves as
H
Q
e (d + cos d).
V
4
The term proportional to d , in the above relation, can be cancelled by taking
0 =
Q
d.
4
(6.31)
(6.32)
274
The term proportional to d, on the other hand, has to cancel against C 1 , so that we must require
that

1
1
Q
1
C 1 = O 2 dr + O
(6.33)
d + O
cos d.
r
r
4
r
The 1-form C 1 must vanish at large distances as

1
1
1
d + O
d.
C 1 = O 2 dr + O
r
r
r
(6.34)
6.2.1. Computation of C 1
From the analysis in Section 5 we know that C 1 defines an anti-self-dual 2-form,
C = e C 1 V 3 C 1
(6.35)
where C must satisfy

d C = 4 d H .
(6.36)
The solution for C may be found by using Eq. (3.14). Though an exact computation of the integral
is possible the result is quite messy. It will be sufficient for our purpose to find the asymptotic
behavior of C using (3.14).
6.2.2. Asymptotic form of C 1
In the limit of large r we find following expressions for the coordinate components of C 1 ,
cQ
C 1,r = 2 sin sin sin ,
4r
cQ
C 1, =
sin sin cos ,
4r

Q
c sin
C1, =
cos +
(cos cos sin + cos sin ) .
4
r
(6.37)
(6.38)
(6.39)
We note that the leading order behavior for the components satisfies the asymptotic boundary
condition given in Eq. (6.33).
6.2.3. Computation of C 1
C 1 can be found by considering an anti-self-dual 2-form,
C = e C 1 V 3 C 1
(6.40)
where C must satisfy
d C = 4 .
For the specific geometry under consideration the solution to this equation is

Q (r rc + c cos ) sin sin
r sin2 d sin cos dr .
C 1 =
4rc
2c(1 cos2 )
This C 1 vanishes at infinity as required by Eq. (6.34).
(6.41)
275
7. Charges for the special case and comparison with the CFT
In this section we compute the asymptotic charges associated with the special case geometry
presented in Section 6. We also identify the corresponding state in the dual CFT and show that
the field theory and gravity results are in agreement.
7.1. Dimensional reduction
In order to compute the charges we must first dimensionally reduce to the five-dimensional
Einstein frame. Starting from the general six-dimensional GMR ansatz (2.1) and completing the
squares we have,

F
2
1
ds = 2H (dv + ) du + + (dv + ) + H dsB2
2
1

2
F
F
2
dt + A(t) + H 1 1
= H 1 1
(7.1)
dy + A(y) + H dsB2
2
2
where
+
A(t) = ,
2

F 1 F F
(y)
A = 1
+ + dt .
2
2
2
2
Now the five-dimensional metric in the Einstein frame is

F 2/3 4/3
F 1/3 2/3 2
2
(t) 2
dt + A
H
+ 1
H dsB .
ds5 = 1
2
2
(7.2)
(7.3)
(7.4)
7.2. Mass
In the limit of large r the tt component of the metric is expected to behave as [31]
gtt 1
2G(5) M
3 r
(7.5)
where G(5) is Newtons constant in 5D

G(5) =
G(10)
4 5 g 2 4
.
=
V (2Ry )
V Ry
(7.6)
Here V is the volume of T 4 and g is the string coupling. From metric (7.4) and using the explicit
metric functions given in Eqs. (6.24) and (6.26), we have the following expansion for large r,

F 2/3 4/3
2Q + a 2 n(n + cos )
gtt = 1
(7.7)
H
1 +
.
2
6r
By comparing the above expressions we can read off the mass of the geometry

M=
2Q + a 2 n(n + cos ) .
(5)
4G
(7.8)
276
7.3. Momentum
In order to find the momentum charge consider the behaviour of the t component of A(y) . In
the limit of large r this term behaves as [31]
Qp
.
4r
From Eqs. (7.3) and (6.26) we find
(y)
(y)
At
At
(7.9)
F
a 2 n(n + cos )
.
2
4r
(7.10)
This gives
Qp = a 2 n(n + cos ).
(7.11)
The quantized momentum charge np is given by

Ry
Ry 2
np =
a n(n + cos ) .
Qp =
(5)
(5)
4G
4G
(7.12)
7.4. Angular momentum

The angular momentum charges can be obtained from the five-dimensional U (1) gauge
field A(t) , by considering [32],
2R
y
2R
y

+ G(5)
dy
J cos2 d + J sin2 d
r
2
0
0
(7.13)

4
where the coordinates r , , , are the usual spherical polars for R . These are related to
the GH coordinates introduced in Section 3 by
+
+
= ,
=
,
=
.
r = 2 r,
(7.14)
2
2
2
Then for the special case geometry given in Section 6.2, we find that,

a
Q(1 2n cos ) + 2a 2 n(n + cos ) ,

J =
(7.15)
(5)
16G

a
J =
(7.16)
Q(1 + 2n + cos ) 2a 2 n(n + cos ) .
(5)
16G
1
2Ry
1
dy A(t) =
2Ry
7.5. Boost needed

The above results however lead to a problem because expressions (7.15) and (7.16) are not
the charges expected from the CFT. An analogous problem was noted in [22], where the axiallysymmetric case was studied. In that case it was shown that extending the near-horizon geometry
to the asymptotically flat region, using the -independent analogue of the rules of Section 5,
produces a geometry in which the compact y coordinate has the wrong periodic identification.
By studying the regularity of the metric, it was also shown in [22] that the asymptotically flat
metric can be made smooth by changing the periodic identifications on y; this can be formally
achieved by performing a boost along y. This operation also changes the charges of the solution.
277
In the case of our -dependent geometry we expect a similar phenomenon to happen. The mismatch between the gravity and the CFT charges noted above signals the fact that a similar boost
has to be performed in this case as well. A rigorous derivation of this statement would require
a full singularity analysis which is beyond the scope of this paper. Assuming the necessity of a
boost we show that it leads to a perfect agreement between the gravity and the CFT charges.
The same boost is expected to desingularize the geometry. The coordinate transformation which
produces the required boost is
u ue ,
v ve .
(7.17)
This change of coordinates acts on , and F as

e ,
e ,
F F e2 .
(7.18)
The momentum charge then transforms as

Qp Qp e2 = a 2 e2 n(n + cos ).
(7.19)
If we also redefine the parameter a as

a = ae
,
(7.20)
we see that the momentum charge after boost is

Qp = a 2 n(n + cos ).
(7.21)
The 1-form A(t) transforms as

A(t)
e + e
.
(7.22)
Expanding A(t) above according to (7.13) gives the boosted values of the angular momenta. It
turns out that if we take
e2 =
Q
,
Q + 2Qp
(7.23)
the boosted angular momenta assume the simple form

Qa
(7.24)
(1 2n cos ),
16G(5)
Qa
J =
(7.25)
(1 + 2n + cos ).
16G(5)
We will see in the next subsection that these values match the ones expected from the CFT. It
is important to note that the boost parameter in Eq. (7.23) has the same form as the one found
in [22] for the axially-symmetric case. Though the explicit value of Qp is different in the two
cases, the boost parameter as a function of the charges of the solution is identical. This suggests
that Eq. (7.23) gives the proper boost for the metric generated by an arbitrary profile.
In order to compare with the CFT it is convenient to rewrite the charges in terms of the
microscopic quantities. For this purpose we recall that the D1/D5 charge is quantized as
J =
Q=
(2)4 g 3 n1
= g n5 .
V
(7.26)
278
We also need the relation between the parameters a, Q and the radius Ry of the compact direction y: based on consistency with the = 0 case [22] we conjecture this relation to be
Q
.
(7.27)
a
Note that the relation (7.27) and the boost described above, fix the global identifications of the
geometry. These identifications are crucial to ensure that the metric is free of orbifold singularities. Finally, using (7.26) and (7.27) we find
Ry =
np = n1 n5 n(n + cos ),
n 1 n5
J =
(1 2n cos ),
4
(7.28)
n 1 n5
J =
(1 + 2n + cos ).
4
(7.29)
7.6. Comparison with CFT

The dual CFT is a (4, 4) superconformal theory with central charge
c = 6n1 n5 .
(7.30)
It has an SU(2)L SU(2)R R-symmetry that corresponds to SO(4) rotations on the 4 spatial
non-compact directions of the gravity system. The diagonal generators of SU(2)L SU(2)R , JL
and JR , are connected to the spacetime angular momenta J and J as
J L = J J ,
J R = J + J .
(7.31)
The momentum charge np is given by

np = hL hR
(7.32)
where hL and hR are the left and right conformal dimensions. The geometry we have constructed
thus corresponds to a state with quantum numbers
n 1 n5
n 1 n5
JL =
(7.33)
(2n + cos ),
JR =
,
hL hR = n1 n5 n(n + cos ).
2
2
We will show that these quantum numbers are the ones appropriate for a state obtained by
applying spectral flow, in the left sector, to a Ramond ground state. Spectral flow is an automorphism of the CFT superconformal algebra that acts independently on the left and right sectors
and transforms the quantum numbers as
c
c
hL hL + L JL + L2 , JL JL + L ,
(7.34)
24
12
c
c
hR hR + R JR + R2 , JR JR + R .
(7.35)
24
12
If the parameters L and R are even integers, spectral flow sends the R sector to itself and the
NS sector to itself, if L and R are odd it interchanges the R and NS sectors.
The 2-charge geometry (3.25) is dual to an R ground state with
n 1 n5
n 1 n5
cos ,
JR =
,
hL = hR = 0.
JL =
(7.36)
2
2
If we act on this state by spectral flow with parameters
L = 2n,
R = 0
(7.37)
we reach the R sector state with

n 1 n5
JL =
(2n + cos ),
2
hL = n1 n5 n(n + cos ),
n 1 n5
,
2
hR = 0.
279
JR =
(7.38)
This agrees with the charges found in the gravity picture given in Eq. (7.33).
8. Discussion
In summary, we have presented a new family of asymptotically flat, time-dependent, BPS
solutions of type IIB supergravity carrying D1, D5, and momentum charges. This set of solutions
is parameterized by an integer n, defining the spectral flow of the left sector, and by a closed
curve in R4 . These solutions represent the spectral flow of all Ramond sector ground states of
the D1D5 system. Constructing the solutions from known bound states of the D1D5 system
using an exact symmetry (spectral flow) guarantees that the new 3-charge solutions are also
bound states of the branes. In addition to finding the general family of solutions we have also
explicitly presented a new v-dependent 3-charge geometry. For this geometry we have calculated
the conserved asymptotic charges and shown that the results are in agreement with the dual
conformal field theory.
A few outstanding issues pertaining to the new geometries presented here warrant further
investigation. In the near-horizon limit our solutions are related to smooth geometries of [8]
by a globally defined coordinate transformation, and hence are regular in this limit. However,
it is not a priori guaranteed that smoothness persists even after extending the geometries to the
asymptotically flat region. Potential singularities may only appear at the origin of the base space
and at the location of the string profile. In the axially symmetric case it was shown [22] that
there exists, in both regions, a globally well defined coordinate transformation that brings the
metric into a manifestly smooth form. The smoothness of the geometry depends crucially on the
global identifications imposed on the y coordinate. It would be important to demonstrate that the
periodic identifications on the y coordinate introduced in Section 7.5 lead to a smooth geometry
even in our -dependent case, though the task is expected to be significantly more involved.
A more ambitious goal would be to show that the analogous identifications performed on the
geometries corresponding to a general profile also lead to smooth solutions. The same analysis
would shed some light on the qualitative features of the geometries.
Also, it is worth noting that the solutions presented here were constructed within the framework of six-dimensional minimal supergravity, which requires that the D1 and D5 charges be
equal (Q1 = Q5 ). This condition made it possible to use the results of [29] to find the correct
asymptotically flat extensions of the near-horizon geometry after spectral flow. It would be interesting to find a generalized class of solutions without this restriction.
Another useful extension of the calculations presented here would be to find expressions for
the conserved charges corresponding to the general solution given in Eqs. (5.42)(5.44) and
to then compare the result with the dual conformal field theory. For this agreement to hold,
we expect that the boost introduced in Section 7.5 would be crucial. Furthermore, a detailed
investigation into the relationship between the CFT and this class of gravity duals could possibly
lead to some new insight into the AdS/CFT duality and its relation to black hole microscopics.
In conclusion, we have presented here a large class of 3-charge geometries, parameterized by a
curve in R4 , which generically break all but one (null) isometry of the 6D space, and have known
CFT duals. We believe that these solutions represent a significant step towards the construction
of general 3-charge microstates.
280
Acknowledgements
The authors would like to thank Iosif Bena, Oleg Lunin, Samir Mathur, Amanda Peet, Simon
Ross and Yogesh Srivastava, for many valuable discussions. S.G. and A.S. acknowledge the hospitality of the Aspen Center for Physics during the initial stages of this work. J.F. was supported
by an Ontario Graduate Scholarship. S.G. and A.S. were supported by NSERC.
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]
S.D. Mathur, A. Saxena, Y.K. Srivastava, Nucl. Phys. B 680 (2004) 415, hep-th/0311092.
S.D. Mathur, Fortschr. Phys. 53 (2005) 793, hep-th/0502050.
O. Lunin, S.D. Mathur, Phys. Rev. Lett. 88 (2002) 211303, hep-th/0202072.
J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, Int. J. Theor. Phys. 38 (1999) 1113, hep-th/9711200.
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150;
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
A. Sen, Mod. Phys. Lett. A 10 (1995) 2081, hep-th/9504147;
A.W. Peet, Nucl. Phys. B 456 (1995) 732, hep-th/9506200.
A. Dabholkar, Phys. Rev. Lett. 94 (2005) 241301, hep-th/0409148.
O. Lunin, S.D. Mathur, Nucl. Phys. B 610 (2001) 49, hep-th/0105136;
O. Lunin, S.D. Mathur, Nucl. Phys. B 623 (2002) 342, hep-th/0109154.
O. Lunin, J.M. Maldacena, L. Maoz, hep-th/0212210.
O. Lunin, S.D. Mathur, A. Saxena, Nucl. Phys. B 655 (2003) 185, hep-th/0211292.
M. Taylor, JHEP 0603 (2006) 009, hep-th/0507223.
L.F. Alday, J. de Boer, I. Messamah, Nucl. Phys. B 746 (2006) 29, hep-th/0511246;
L.F. Alday, J. de Boer, I. Messamah, JHEP 0612 (2006) 063, hep-th/0607222;
K. Skenderis, M. Taylor, Phys. Rev. Lett. 98 (2007) 071601, hep-th/0609154;
I. Kanitscheider, K. Skenderis, M. Taylor, JHEP 0704 (2007) 023, hep-th/0611171.
J.C. Breckenridge, R.C. Myers, A.W. Peet, C. Vafa, Phys. Lett. B 391 (1997) 93, hep-th/9602065.
M. Cvetic, D. Youm, Nucl. Phys. B 476 (1996) 118, hep-th/9603100;
D. Youm, Phys. Rep. 316 (1999) 1, hep-th/9710046.
F. Larsen, E.J. Martinec, JHEP 9906 (1999) 019, hep-th/9905064.
A. Strominger, C. Vafa, Phys. Lett. B 379 (1996) 99, hep-th/9601029.
D. Mateos, P.K. Townsend, Phys. Rev. Lett. 87 (2001) 011602, hep-th/0103030.
I. Bena, P. Kraus, Phys. Rev. D 70 (2004) 046003, hep-th/0402144;
I. Bena, Phys. Rev. D 70 (2004) 105018, hep-th/0404073.
S. Giusto, S.D. Mathur, Y.K. Srivastava, Nucl. Phys. B 754 (2006) 233, hep-th/0510235.
D. Bak, K. Kim, N. Ohta, JHEP 0601 (2006) 072, hep-th/0511051.
O. Lunin, JHEP 0404 (2004) 054, hep-th/0404006.
S. Giusto, S.D. Mathur, A. Saxena, Nucl. Phys. B 701 (2004) 357, hep-th/0405017.
S. Giusto, S.D. Mathur, A. Saxena, Nucl. Phys. B 710 (2005) 425, hep-th/0406103.
S. Giusto, S.D. Mathur, Nucl. Phys. B 729 (2005) 203, hep-th/0409067.
V. Jejjala, O. Madden, S.F. Ross, G. Titchener, Phys. Rev. D 71 (2005) 124030, hep-th/0504181.
I. Bena, N.P. Warner, Adv. Theor. Math. Phys. 9 (2005) 667, hep-th/0408106;
I. Bena, P. Kraus, Phys. Rev. D 72 (2005) 025007, hep-th/0503053;
I. Bena, N.P. Warner, Phys. Rev. D 74 (2006) 066001, hep-th/0505166;
P. Berglund, E.G. Gimon, T.S. Levi, JHEP 0606 (2006) 007, hep-th/0505167;
A. Saxena, G. Potvin, S. Giusto, A.W. Peet, JHEP 0604 (2006) 010, hep-th/0509214;
I. Bena, C.W. Wang, N.P. Warner, hep-th/0604110;
V. Balasubramanian, E.G. Gimon, T.S. Levi, hep-th/0606118;
I. Bena, C.W. Wang, N.P. Warner, JHEP 0611 (2006) 042, hep-th/0608217.
V. Balasubramanian, J. de Boer, E. Keski-Vakkuri, S.F. Ross, Phys. Rev. D 64 (2001) 064011, hep-th/0011217.
J.M. Maldacena, L. Maoz, JHEP 0212 (2002) 055, hep-th/0012025.
J.B. Gutowski, D. Martelli, H.S. Reall, Class. Quantum Grav. 20 (2003) 5049, hep-th/0306235.
G.W. Gibbons, S.W. Hawking, Phys. Lett. B 78 (1978) 430.
R.C. Myers, M.J. Perry, Ann. Phys. 172 (1986) 304.
A. Dabholkar, J.P. Gauntlett, J.A. Harvey, D. Waldram, Nucl. Phys. B 474 (1996) 85, hep-th/9511053.
More meta-stable brane configurations

without D6-branes
Changhyun Ahn
Department of Physics, Kyungpook National University, Taegu 702-701, Republic of Korea
Received 6 July 2007; received in revised form 4 September 2007; accepted 19 September 2007
Abstract
We describe the intersecting brane configurations, consisting of NS-branes, D4-branes (and anti-D4branes), in type IIA string theory corresponding to the meta-stable nonsupersymmetric vacua of N = 1
SU(Nc ) SU(Nc ) SU(Nc ) gauge theory with bifundamentals. By adding the orientifold 4-plane to these
brane configurations, we also discuss the meta-stable brane configurations for other gauge theory with
bifundamentals. Furthermore, we study the intersecting brane configurations corresponding to the nonsupersymmetric meta-stable vacua of other gauge theory with bifundamentals, by adding the orientifold 6-plane.
PACS: 11.25.Uv; 11.30.Pb
Keywords: Meta-stable vacua; IIA string theory; Gauge theory; SUSY breaking
1. Introduction
In the standard type IIA brane configuration, the quark masses correspond to the relative
displacement of the D6-branes (0123789) and D4-branes (01236) along the 45 directions geometrically. Then the eigenvalues of quark mass matrix correspond to the positions of D6-branes
in 45 directions. See the review paper [1] for the gauge theory and the brane dynamics. The
Seiberg duality in the classical brane picture can be accomplished by exchanging the locations
of the NS5-brane (012345) and NS5 -brane (012389) along x 6 direction each other.
The geometric misalignment of D4-branes connecting both NS5 -brane and D6-branes in the
magnetic brane configuration can be interpreted as a nontrivial F-term condition in the gauge
E-mail address: ahn@knu.ac.kr.
doi:10.1016/j.nuclphysb.2007.09.023
282
C. Ahn / Nuclear Physics B 790 (2008) 281316
theory with massive flavors. Then the F-term equations can be partially cancelled by both recombination of flavor-D4-branes with the color-D4-branes and then movement of those D4-branes
into the 45 directions. This phenomenon in magnetic brane configuration corresponds to the fact
that some entries in the magnetic dual quarks acquire nonzero vacuum expectation values to minimize the F-term in the dual gauge theory side. Moreover, the remaining flavor-D4-branes that do
not move to 45 directions, connecting to NS5 -brane, can move along 89 directions freely since
D6-branes and NS5 -brane are parallel and this geometric freedom of meson field corresponds
to the classical pseudomoduli space of nonsupersymmetric vacua of the gauge theory.
On the other hand, it is known that the NS-brane configuration in type IIA string theory,
where there exist only two types of NS5-brane and NS5 -brane, preserves N = 2 supersymmetry
in four dimensions [1]. The geometry [2] of the coincident NS5-branes is characterized by the
metric, the dilaton, and the field strength and is useful to construct the DBI action for D4-branes.
In order to break the supersymmetry, one adds D4-branes and anti-D4-branes [3]. By adding
D4-branes suspending between the NS5-brane and the NS5 -brane, and anti-D4-branes (D4branes) suspending between the NS5-brane and the other NS5 -brane, the supersymmetry of this
system is broken [3]. The low energy dynamics can be described by the gauge theory on the D4branes. The brane configuration corresponding to the electric theory with vanishing mass for the
bifundamentals consists of the left NS5 -brane, the middle NS5-brane and the right NS5 -brane
and two sets of D4-branes suspended between two NS5 -branes. The gauge group is a product
of two unitary gauge groups and there exist bifundamentals. For the nonvanishing mass for these
bifundamentals, the relative displacement between the two NS5 -branes along the 45 directions
occurs. By taking the Seiberg dual for one of two gauge group factors with nonvanishing masses
for the bifundamentals, the magnetic dual theory has a cubic superpotential between the dual
quarks and a meson which is nothing but a quadratic term of bifundamentals in an electric theory.
Also the linear term in a meson appears in this magnetic superpotential. Then the F-term equation
for this meson field leads to the supersymmetry breaking. One finds that supersymmetry is broken
classically but is restored quantum mechanically. It turns out the classical nonsupersymmetric
vacuum becomes long-lived state [3].
As the distance between the two NS5 -branes along the 45 directions becomes zero, this brane
configuration with D4- and D4-branes can decay and the geometric misalignment between flavorD4-branes arises, as before. Due to the presence of NS5-brane in this system, there exists an
attractive force between the tilted D4-branes and NS5-brane. The explicit and careful computation of DBI action for these D4-branes in the background created by NS5-brane has been done by
the work of [3] and this effect of the gravitational attraction leads to a curve for tilted D4-branes
rather than a straight line. Then for small displacement of two NS5 -branes, the ground state is
given by this curved brane configuration. As this displacement between two NS5 -branes is
increased, the ground state brane configuration is given by straight brane configuration. The
meta-stable vacua of [4] arise in some region of parameter space. In this description, the dual
quarks are represented by the bifundamentals of product gauge group and the mass term is encoded by the relative displacement of two NS5 -branes in 45 directions, as we mentioned before.
Note that there exist no D6-branes in this brane configuration.1 When one of the NS5 -branes
goes to infinity along the x 6 direction, then the corresponding gauge group becomes a global
symmetry and the theory leads to a standard N = 1 SQCD with fundamentals. In other regions,
1 A replacement of D6-branes with NS5 -brane corresponds to the gauging of the flavor group (global symmetry) of the
gauge theory realized on the D4-branes and this replacement might be useful to construct the phenomenological model
building.
283
a generalization of [4] showing very similar qualitative phenomena in classical string theory
occurs [3].
The focus on the new meta-stable brane configurations by adding an orientifold 4-plane and
an orientifold 6-plane to the above brane configuration studied by [3], along the line of [58], was
given in [9]. When the former was added, no extra NS-branes or D-branes were needed. However,
when the latter was added, the extra NS-branes or D-branes into the above brane configuration
were needed in order to have a product gauge group.
In this paper, we continue to find out new meta-stable brane configurations which contain
four NS-branes or six NS-branes, along the line of [3,9], by starting from the known or new
supersymmetric brane configurations in type IIA string theory. Compared with the previous approaches given by [58], the superpotential in the magnetic theory has very simple form because
there are no D6-branes in the brane configurations and this fact allows us to analyze the metastable vacua easily using the F-term equations and one loop effective potential. But the number
of NS-branes is increased for a given gauge theory with matters since the role of D6-branes is
replaced by NS5 -brane. Some of the meta-stable brane configurations lead to the known metastable brane configuration corresponding to the gauge theory with less gauge group factors in the
literature, by replacing the NS5 -brane with coincident D6-branes. Basically, the gauge group
will be a triple product gauge group for four NS-branes with D4-branes. When we add an orientifold 4-plane into this brane configuration, the gauge group will be a triple product between
symplectic gauge group and an orthogonal gauge group, alternatively, depending on the orientifold 4-plane charge. When we add an orientifold 6-plane into six NS-branes with D4-branes,
one of the gauge group factor will be a symplectic or an orthogonal gauge group depending on
the orientifold 6-plane charge and the other gauge group factor will be unitary.
In Section 2, we describe the type IIA brane configuration corresponding to the electric theory
based on the N = 1 SU(Nc ) SU(Nc ) SU(Nc ) gauge theory with the bifundamentals and deform this theory by adding the mass term for the bifundamental. We construct the three different
dual magnetic theories by taking the Seiberg dual for each gauge group factor. Then we consider
the nonsupersymmetric meta-stable minimum and present the corresponding intersecting brane
configurations of type IIA string theory.
In Section 3, we discuss the type IIA brane configuration, by adding the orientifold 4-plane
to the brane configuration in Section 2, corresponding to the electric theory based on the N = 1
Sp(Nc ) SO(2Nc ) Sp(Nc ) gauge theory with matters and deform this theory by adding the
mass term for the bifundamental. Then we construct the corresponding dual magnetic theories by
taking the Seiberg dual for each gauge group factor. We consider the nonsupersymmetric metastable minimum and present the corresponding intersecting brane configurations of type IIA
string theory. We also comment on the case of N = 1 SO(2Nc ) Sp(Nc ) SO(2Nc ) gauge
theory with matters.
In Section 4, we discuss the type IIA brane configuration corresponding to the electric theory
based on the N = 1 Sp(Nc ) SU(Nc ) SU(Nc ) gauge theory with matters and deform this
theory by adding the mass term for the bifundamental. Then we construct the two different dual
magnetic theories by taking the Seiberg dual for each unitary gauge group factor. We consider
the nonsupersymmetric meta-stable minimum and present the corresponding intersecting brane
configurations of type IIA string theory. Moreover, we also discuss the meta-stable brane configurations corresponding to the electric theory based on the N = 1 SO(Nc ) SU(Nc ) SU(Nc )
gauge theory by changing the orientifold 6-plane charge.
In Section 5, we make some comments for the future directions.
284
2. Meta-stable brane configurations with four NS-branes

2.1. Electric theory
The type IIA brane configuration [10,11] corresponding to N = 1 supersymmetric gauge
theory with gauge group

SU(Nc ) SU Nc SU Nc
(2.1)
and with a field F charged under (Nc , Nc ), a field G charged under (Nc , Nc ), and their conjugates
can be described by the left NS5 -brane (012389), the left NS5L -brane (012345), the
F and G
L
right NS5R -brane (012389), the right NS5R -brane (012345), Nc -, Nc - and Nc -color D4-branes
(01236). The fields F and F correspond to 44 strings connecting the Nc -color D4-branes with
correspond to 44 strings connecting the Nc -color
Nc -color D4-branes while the fields G and G

D4-branes with Nc -color D4-branes.
The left NS5L -brane is located at x 6 = 0 and let us denote the x 6 coordinates for the NS5L brane, the NS5R -brane and the NS5R -brane by x 6 = y1 , y2 , y2 + y3 , respectively. The Nc
D4-branes are suspended between the NS5L -brane and the NS5L -brane, the Nc D4-branes are
suspended between the NS5L -brane and the NS5R -brane, and the Nc D4-branes are suspended
between the NS5R -brane and the NS5R -brane. We draw this brane configuration in Fig. 1(A) for
the vanishing mass case.2
The gauge couplings of SU(Nc ), SU(Nc ) and SU(Nc ) are given by a string coupling constant gs , a string scale s and the x 6 coordinates yi for three NS-branes through
g12 =
gs s
,
y1
g22 =
gs s
,
y2
g32 =
gs s
.
y3
For example, as y3 goes to , the SU(Nc ) gauge group becomes a global symmetry and the
theory leads to SQCD with the gauge group SU(Nc ) SU(Nc ) and Nc flavors in the fundamental
representation.
There is no superpotential in Fig. 1(A) since the NS5L -brane is perpendicular to two NS5 branes and the NS5R -brane is perpendicular to two NS5-branes. Let us deform this theory.
Displacing the two NS5 -branes relative each other in the
v x 4 + ix 5
direction corresponds to turning on a quadratic mass-deformed superpotential for the fields F
and F as follows:
W = mF F m
(2.2)
where the first gauge group indices in F and F are contracted, each second gauge group index
in them is encoded in and the mass m is given by
m=
x
x
= 2 .
2
s
(2.3)
The gauge-singlet for the first dual gauge group is in the adjoint representation for the second dual gauge group, i.e., (1, (Nc Nc )2 1, 1) (1, 1, 1) under the dual gauge group (2.4).
2 There are similar brane configurations in the context of quiver gauge theory [12,13].
285
Fig. 1. The N = 1 supersymmetric electric brane configuration for the gauge group SU(Nc ) SU(Nc ) SU(Nc ) and
with vanishing (A) and nonvanishing (B) mass for the bifundamentals F and F . The Nc
bifundamentals F , F , G and G
D4-branes in (A) are decomposed into (Nc Nc ) D4-branes which are moving to +v direction in (B) and Nc D4-branes
which are recombined with those D4-branes connecting between NS5R -brane and NS5R -brane in (B).
Then the is an (Nc Nc )(Nc Nc ) matrix. The NS5R -brane together with (Nc Nc )-color
D4-branes is moving to the +v direction for fixed other branes during this mass deformation. In
other words, the Nc D4-branes among Nc D4-branes are not participating in the mass deformation. Then the x 5 coordinate( x) of NS5L -brane is equal to zero while the x 5 coordinate
of NS5R -brane is given by x. Giving an expectation value to the meson field corresponds
to recombination of Nc - and Nc -color D4-branes, which will become Nc -color D4-branes in
Fig. 1(A) such that they are suspended between the NS5L -brane and the NS5R -brane and pushing them into the
w x 8 + ix 9
direction. We assume that the number of colors satisfies
Nc Nc Nc .
Now we draw this brane configuration in Fig. 1(B) for nonvanishing mass for the fields F
and F .
2.2. Magnetic theory
By applying the Seiberg dual to the SU(Nc ) factor in (2.1), the two NS5L,R -branes can be
located at the inside of the two NS5-branes, as in Fig. 2. Starting from Fig. 1(A) and moving the
NS5R -brane with (Nc Nc ) D4-branes to the +v direction leading to Fig. 1(B) and interchanging the NS5L -brane and the NS5L -brane, one obtains Fig. 2(A). Before arriving at Fig. 2(A),
there exists an intermediate step where the (Nc Nc ) D4-branes are connecting between the
NS5L -brane and the NS5L -brane, (Nc Nc ) D4-branes connecting between the NS5L -brane
and NS5R -brane, and Nc D4-branes between the NS5L -brane and the NS5R -brane. By introducing Nc D4-branes and Nc anti-D4-branes between the NS5L -brane and NS5L -brane,
reconnecting the former with the Nc D4-branes connecting between NS5L -brane and the NS5L brane (therefore (Nc Nc ) D4-branes) and moving those combined (Nc Nc ) D4-branes to
286
Fig. 2. The N = 1 magnetic brane configuration for the gauge group SU(N c = Nc Nc ) SU(Nc ) SU(Nc ) corresponding to Fig. 1(B) with D4- and D4-branes (A) and with a misalignment between D4-branes (B) when the
NS5 -branes are close to each other. The number of tilted D4-branes in (B) can be written as Nc Nc = (Nc Nc ) N c .
+v direction, one gets the final Fig. 2(A) where we are left with (Nc Nc ) anti-D4-branes
between the NS5L -brane and NS5L -brane.
When two NS5 -branes in Fig. 2(A) are close to each other, then it leads to Fig. 2(B) by
realizing that the number of (Nc Nc ) D4-branes connecting between NS5L -brane and NS5R brane can be rewritten as (Nc Nc ) plus N c . If we ignore Nc D4-branes and NS5R -brane from
Fig. 2(B), then the brane configuration becomes the one in [3].
The dual gauge group is given by

SU N c = Nc Nc SU Nc SU Nc .
(2.4)
c , Nc , 1), a field g charged under (1, Nc , Nc ),
The matter contents are the field f charged under (N
and their conjugates f and g under the dual gauge group (2.4) and the gauge-singlet for
the first dual gauge group in the adjoint representation for the second dual gauge group, i.e.,
(1, (Nc Nc )2 1, 1) (1, 1, 1) under the dual gauge group (2.4).
The cubic superpotential with the mass term (2.2) in the dual theory is given by
Wdual = f f + m .
(2.5)
Here the magnetic fields f and f correspond to 44 strings connecting the N c -color D4-branes
(that are connecting between the NS5L -brane and the NS5R -brane in Fig. 2(B)) with Nc -flavor
D4-branes (that are a combination of three different D4-branes in Fig. 2(B)). Among these Nc flavor D4-branes, only the strings ending on the upper (Nc Nc ) D4-branes and on the tilted
middle (Nc Nc ) D4-branes in Fig. 2(B) enter the cubic superpotential term. Although the
(Nc Nc ) D4-branes in Fig. 2(A) cannot move any directions, the tilted (Nc Nc )-flavor D4branes can move w direction in Fig. 2(B). The remaining upper N c D4-branes are fixed also and
cannot move any direction. Note that there is a decomposition

Nc Nc = Nc Nc + N c .
The brane configuration for zero mass for the bifundamental, which has only a cubic superpotential, can be obtained from Fig. 2(A) by moving the upper NS5R -brane together with
287
(Nc Nc )-color D4-branes into the origin v = 0. Then the number of dual colors for D4-branes
becomes N c between NS5L -brane and NS5L -brane and Nc between two NS5 -branes as well as
Nc D4-branes between NS5R -brane and NS5R -brane. Or starting from Fig. 1(A) and moving
the NS5L -brane to the left all the way past the NS5L -brane, one also obtains the corresponding
magnetic brane configuration for massless case.
The brane configuration in Fig. 2(A) is stable as long as the distance x between the upper
NS5 -brane and the lower NS5 -brane is large, as in [3]. If they are close to each other, then this
brane configuration is unstable to decay and leads to the brane configuration in Fig. 2(B). One
can regard these brane configurations as particular states in the magnetic gauge theory with the
gauge group (2.4) and superpotential (2.5). The (Nc Nc N c )-flavor D4-branes of straight
brane configuration of Fig. 2(B) bend due to the fact that there exists an attractive gravitational
interaction between those flavor D4-branes and NS5L -brane from the DBI action, by following
the procedure of [3], as long as the distance y3 goes to because the presence of an extra
NS5R -brane does not affect the DBI action. For the finite and small y3 , the careful analysis for
DBI action is needed in order to obtain the bending curve connecting two NS5 -branes.
When the upper NS5 -brane (or NS5R -brane) is replaced by coincident (Nc Nc ) D6-branes
and the NS5R is rotated by an angle 2 in the (v, w) plane in Fig. 2(B), this brane configuration
reduces to the one found in [14] where the gauge group was given by SU(nf + nc nc ) SU(nc )
with nf multiplets, nf multiplets, flavor singlets and gauge singlets. Then the present number
(Nc Nc ) corresponds to the nf , the number Nc corresponds to nc and the number Nc corresponds to the nc of [14]. Note that the number of D4-branes touching NS5R -brane in Fig. 2(B)
is equal to (Nc Nc ).
The quantum corrections can be understood for small x by using the low energy field theory
on the branes. The low energy dynamics of the magnetic brane configuration can be described
by the N = 1 supersymmetric gauge theory with gauge group (2.4) and the gauge couplings for
the three gauge group factors are given by
2
g1,mag
=
gs s
,
y1
2
g2,mag
=
gs s
,
(y2 y1 )
2
=
g3,mag
gs s
.
y3
The dual gauge theory has an adjoint of SU(Nc ) and bifundamentals f , f, g and g under the
dual gauge group (2.4) and the superpotential corresponding to Figs. 2(A) and 2(B) is given by
Wdual = h f f h2 ,
2
h2 = g2,mag
, 2 =
x
.
2gs 3s
Then f f is an N c N c matrix where the second gauge group indices for f and f are contracted
with those of while 2 is an (Nc Nc ) (Nc Nc ) matrix. Although the field f itself is
an antifundamental in the second gauge group which is a different representation for the usual
standard quark coming from D6-branes, the product f f has the same representation for the
product of quarks and moreover, the second gauge group indices for the field play the role
of the flavor indices, as in comparison with the brane configuration in the presence of D6-branes
before.
Therefore, the F-term equation, the derivative Wdual with respect to the meson field cannot
be satisfied if the (Nc Nc ) exceeds N c . So the supersymmetry is broken. That is, there exist
three equations from F-term conditions:
f f 2 = 0 and f = 0 = f .
288
Fig. 3. The N = 1 supersymmetric electric brane configuration for the gauge group SU(Nc ) SU(Nc ) SU(Nc ) and
with vanishing (A) which is identical to Fig. 1(A) and nonvanishing (B) mass for the
bifundamentals F, F , G and G
bifundamentals F and F . This deformation is different from the one (2.2) given previously. In (B), the NS5L -brane
together with Nc D4-branes is moving to +v direction.
Then the solutions for these are given by

1N c
f =
,
f = 1N c
0

0 ,
=
0
0
0 1(N N N c )
c
(2.6)
where the zero of f is an (Nc Nc N c ) N c matrix, the zero of f is an N c (Nc
Nc N c ) matrix and the zeros of are N c N c , N c (Nc Nc N c ), and (Nc Nc
N c ) N c matrices. Then one can expand these fields around on a point (2.6), as in [4,1518] and
one arrives at the relevant superpotential up to quadratic order in the fluctuation. At one loop, the
(1)
effective potential Veff for 0 leads to the positive value for m2 implying that these vacua are
0
stable.
2.3. Other magnetic theory-I
Let us consider other magnetic theory for the same undeformed electric theory given in Section 2.1. Here we consider the different mass deformation. By applying the Seiberg dual to the
SU(Nc ) factor in (2.1), the two NS5L,R -branes can be located at the left-hand side of the two
NS5-branes, as in Fig. 4.
By starting from Fig. 3(A) which is the same as Fig. 1(A) and moving the NS5L -brane with
Nc D4-branes to the +v direction leading to Fig. 3(B) and interchanging the NS5L -brane and the
NS5R -brane, one obtains Fig. 4(A). Before arriving at Fig. 4(A), there exists an intermediate step
where the Nc D4-branes are connecting between the NS5L -brane and the NS5R -brane, (Nc
Nc + Nc ) D4-branes are connecting between the NS5R -brane and NS5L -brane, and Nc D4branes are suspended between the NS5L -brane and the NS5R -brane. By moving the combined
Nc D4-branes, obtained from reconnection of those D4-branes between NS5L -brane and the
NS5R -brane and those D4-branes between the NS5R -brane and NS5L -brane (therefore between
the NS5L -brane and the NS5L -brane), to +v direction, one gets the final Fig. 4(A) where we are
left with (Nc Nc ) anti-D4-branes between the NS5R -brane and NS5L -brane. We assume that
289
Fig. 4. The N = 1 magnetic brane configuration for the gauge group SU(Nc ) SU(N c = Nc + Nc Nc ) SU(Nc )
corresponding to Fig. 3(B) with D4- and D4-branes (A) and with a misalignment between D4-branes (B) when the
NS5 -branes are close to each other. The number of tilted D4-branes is equal to Nc Nc = Nc N c in (B).
the number of colors satisfies

Nc + Nc Nc Nc .
When two NS5 -branes in Fig. 4(A) are close to each other, then it leads to Fig. 4(B) by realizing that the number of Nc D4-branes connecting between NS5L -brane and NS5L -brane can be
rewritten as (Nc Nc ) plus N c .
The dual gauge group is

SU(Nc ) SU N c = Nc + Nc Nc SU Nc .
(2.7)
c , Nc ),
c , 1), a field g charged under (1, N
The matter contents are the field f charged under (Nc , N
the second dual gauge group in the adjoint representation for the first dual gauge group, i.e.,
(N2c 1, 1, 1) (1, 1, 1) under the dual gauge group (2.7). Then the is an Nc Nc matrix. All
the Nc D4-branes participate in the mass deformation.
The cubic superpotential with the mass term in the dual theory3 is given by
Wdual = f f + m
(2.8)
where we define as F F , the second gauge group indices in F and F are contracted,
each first gauge group index in them is encoded in and the mass m is given by (2.3) where
x refers to the distance between two NS5 -branes along the x 5 direction in Fig. 4(A). Let us
emphasize that although the which has first gauge group indices looks similar to the previous which has second gauge group indices in (2.2), the group indices are different. Here the
magnetic fields f and f correspond to 44 strings connecting the N c -color D4-branes (that are
connecting between the NS5L -brane and the NS5L -brane in Fig. 4(B)) with Nc -flavor D4-branes
(which are realized as corresponding D4-branes in Fig. 4(A)). Although the Nc D4-branes in
3 One can also construct the mass deformation by rotating NS5 -brane and moving it to +v direction. Then the brane
R
configuration will look like as Fig. 15.
290
Fig. 4(A) cannot move any directions, the tilted (Nc Nc )-flavor D4-branes can move w direction in Fig. 4(B). The remaining upper N c D4-branes are fixed also and cannot move any
direction. Note that there is a decomposition

Nc = Nc Nc + N c .
The brane configuration for zero mass for the bifundamental, which has only a cubic superpotential, can be obtained from Fig. 4(A) by moving the upper NS5 -brane (or NS5L -brane)
together with Nc -color D4-branes into the origin v = 0. Then the number of dual colors for D4branes becomes Nc between two NS5 -branes, N c between NS5R -brane and NS5L -brane and Nc
between NS5L -brane and NS5R -brane. Or starting from Fig. 3(A) and moving the NS5L -brane
to the right all the way past the NS5R -brane, one also obtains the corresponding magnetic brane
configuration for massless case.
NS5 -brane and the lower NS5 -brane is large. If they are close to each other, then this brane configuration is unstable to decay to the brane configuration in Fig. 4(B). One can regard these brane
configurations as particular states in the magnetic gauge theory with the gauge group (2.7) and
superpotential (2.8). The (Nc N c )-flavor D4-branes of straight brane configuration of Fig. 4(B)
bend since there exists an attractive gravitational interaction between those flavor D4-branes and
NS5L -brane from the DBI action, as long as the distance y3 is large because the presence of an
extra NS5R -brane does not affect the DBI action. For the finite and small y3 , the careful analysis
for DBI action is needed in order to obtain the bending curve connecting two NS5 -branes. Or
if y3 goes to zero, then this extra NS5R -brane plays the role of enhancing the strength for the
NS5-branes and will affect both the energy of bending curve, Ecurved , which is proportional to 1l
with l ks where k is the number of NS5-branes and x which depends on both 1l and l [3].
When the upper NS5 -brane (or NS5L -brane) is replaced by coincident Nc (that is equal to
the number of D4-branes touching the NS5L -brane) D6-branes, this brane configuration looks
similar to the one found in [14] where the gauge group was given by SU(nf + nc nc )
SU(nc ) with nf multiplets, nf multiplets, flavor singlets and gauge singlets. Then the present
Nc corresponds to the nf , the number Nc corresponds to nc , and Nc corresponds to the nc
of [14].
The low energy dynamics of the magnetic brane configuration can be described by the N = 1
supersymmetric gauge theory with gauge group (2.7) and the gauge couplings for the three gauge
group factors are given by
2
=
g1,mag
gs s
,
(y1 y2 )
2
=
g2,mag
gs s
,
y2
2
g3,mag
=
gs s
.
(y2 + y3 )
The dual gauge theory has an adjoint of SU(Nc ) and bifundamentals f , f, g and g under
the dual gauge group (2.7) and the superpotential corresponding to Figs. 4(A) and 4(B) is given
by
Wdual = hf f h2 ,
2
h2 = g1,mag
, 2 =
x
.
2gs 3s
Then f f is an N c N c matrix where the first gauge group indices for f and f are contracted
with those of while 2 is an Nc Nc matrix. The product f f has the same representation for
the product of quarks and moreover, the first gauge group indices for the field play the role of
the flavor indices when there are D6-branes before.
291
be satisfied if the Nc exceeds N c . So the supersymmetry is broken. That is, there exist three
equations from F-term conditions: f f 2 = 0 and f = 0 = f. Then the solutions for these
are given by

0
0
1N
c
=
,
f = 1N c 0 ,
f =
(2.9)
0 0 1(Nc N )
0
c
is an (Nc N c ) N c matrix, the zero of

are N c N c , N c (Nc N c ) and (Nc
where the zero of f

f is an N c (Nc N c )
matrix and the zeros of
N c ) N c matrices. Then
one can expand these fields around on a point (2.9), as in [4,15] and one arrives at the relevant
(1)
superpotential up to quadratic order in the fluctuation. At one loop, the effective potential Veff
2
for 0 leads to the positive value for m0 implying that these vacua are stable.
2.4. Other magnetic theory-II
One can think of the following dual gauge group

SU(Nc ) SU Nc SU N c = Nc Nc
(2.10)
by performing the magnetic dual for the last gauge group in (2.1). The electric brane configuration can be given in terms of Fig. 1(A) or modified Fig. 1(A) with an exchange between NS5-brane and NS5 -brane. Then for the latter, the resulting brane configuration is given by NS5L -brane,
NS5L -brane, NS5R -brane, and NS5R -brane from the left to the right in the x 6 direction.
In order to obtain the above dual gauge group, we need to interchange between the NS5R brane and the NS5R -brane, as we did before. One can do this either by following the previous
procedure or by looking at Fig. 2 from the negative w direction which is an opposite viewpoint,
compared with Fig. 2. In other words, we are looking at Fig. 2 from the other side of w. Then
the resulting brane configuration in this case can be obtained by taking a reflection for all the
NS-branes, D4-branes and anti-D4-branes with respect to the NS5L -brane (rotating them to the
left for fixed NS5L -brane) in Figs. 2(A) and 2(B). Then the N = 1 magnetic brane configuration
for the gauge group SU(Nc ) SU(Nc ) SU(N c = Nc Nc ) corresponds to Fig. 5(A ) with
D4- and D4-branes and Fig. 5(B ) with a misalignment between D4-branes when the NS5 branes are close to each other. The number of tilted D4-branes in Fig. 5(B ) can be written as
Nc Nc = (Nc Nc ) N c . We do not present Figs. 5(A ) and 5(B ) here.
Let us consider other magnetic theory for the same electric theory given in Section 2.1 with
Fig. 1(A). By applying the Seiberg dual to the SU(Nc ) factor in (2.1) and interchanging the
NS5R -brane and the NS5R -brane, one obtains Fig. 5(A ). Before arriving at Fig. 5(A ), there
exists an intermediate step where Nc D4-branes between NS5L -brane and the NS5R -brane, the
Nc D4-branes are connecting between the NS5L -brane and the NS5R -brane, and (Nc Nc )
D4-branes are connecting between the NS5R -brane and NS5R -brane. By rotating NS5L -brane
by an angle 2 which will become NS5M -brane, moving it with the (Nc Nc ) D4-branes to
+v direction where we introduce (Nc Nc ) D4-branes and (Nc Nc ) anti-D4-branes between
the NS5R -brane and the NS5R -brane, one gets the final Fig. 5(A ) where we are left with
(Nc Nc ) anti-D4-branes between the NS5-brane and the NS5R -brane. When two NS5 -branes
in Fig. 5(A ) are close to each other, then it leads to Fig. 5(B ) by realizing that the number
of (Nc Nc ) D4-branes connecting between NS5M -brane and NS5-brane can be rewritten as
(Nc Nc ) plus N c .
292
Fig. 5. The N = 1 magnetic brane configuration for the gauge group SU(Nc ) SU(Nc ) SU(N c = Nc Nc ) with D4and D4-branes (A ) and with a misalignment between D4-branes (B ) when the NS5 -branes are close to each other.
The number of tilted D4-branes in (B ) can be written as Nc Nc = (Nc Nc ) N c . The deformation is related to the
bifundamentals G and G.
The brane configuration in Fig. 5(A ) is stable as long as the distance x between the upper
NS5 -brane and the lower NS5 -brane (or NS5R -brane) is large. If they are close to each other,
then this brane configuration is unstable to decay to the brane configuration in Fig. 5(B ). One
can regard these brane configurations as particular states in the magnetic gauge theory with the
gauge group and superpotential. The (Nc Nc N c )-flavor D4-branes of straight brane configuration of Fig. 5(B ) bend since there exists an attractive gravitational interaction between those
flavor D4-branes and NS5-brane from the DBI action. As mentioned in [9], the two NS5 -branes
are located at different side of NS5-brane in Fig. 5(B ) and the DBI action computation for this
bending curve should be taken into account.
The matter contents are the field f charged under (Nc , Nc , 1), a field g charged under
c ) and their conjugates f and g under the dual gauge group (2.10) and the gauge(1, Nc , N
singlet which is in the adjoint representation for the second dual gauge group, in other
words, (1, (Nc Nc )2 1, 1) (1, 1, 1) under the dual gauge group (2.10). Then the is
an (Nc Nc ) (Nc Nc ) matrix. Only (Nc Nc ) D4-branes can participate in the mass deformation.
The cubic superpotential with the mass term is given by
Wdual = g g + m
(2.11)
and the third gauge group indices in G and G

are contracted,
where we define as GG

each second gauge group index in them is encoded in . Here the magnetic fields g and g
correspond to 44 strings connecting the N c -color D4-branes (that are connecting between the
NS5M -brane and the NS5-brane in Fig. 5(B )) with Nc -flavor D4-branes. Among these Nc -flavor
D4-branes, only the strings ending on the upper (Nc Nc ) D4-branes and on the tilted middle
(Nc Nc ) D4-branes in Fig. 5(B ) enter the cubic superpotential term. Although the (Nc Nc )
D4-branes in Fig. 5(A ) cannot move any directions, the tilted (Nc Nc )-flavor D4-branes
can move w direction. The remaining upper N c D4-branes are fixed also and cannot move any
direction. Note that there is a decomposition

293
The brane configuration for zero mass for the bifundamental, which has only a cubic superpotential, can be obtained from Fig. 5(A ) by moving the upper NS5 -brane together with
(Nc Nc )-color D4-branes into the origin v = 0. Then the number of dual colors for D4-branes
becomes Nc between the NS5L -brane and the NS5M -brane, Nc between the NS5M -brane and the
NS5-brane and N c between NS5-brane and NS5R -brane. Or starting from Fig. 1(A) and moving
the NS5R -brane to the right all the way past the NS5R -brane, one also obtains the corresponding
supersymmetric gauge theory with gauge group (2.10) and the gauge couplings for the three
gauge group factors are given by
2
g1,mag
=
gs s
,
y1
2
g2,mag
=
gs s
,
(y2 y3 )
2
=
g3,mag
gs s
.
y3
The dual gauge theory has an adjoint of SU(Nc ) and bifundamentals f, f, g and g under the
dual gauge group (2.10) and the superpotential corresponding to Figs. 5(A ) and 5(B ) is given
by
Wdual = h g g h2 ,
2
h2 = g2,mag
, 2 =
x
.
2gs 3s
Then g g is an N c N c matrix where the second gauge group indices for g and g are contracted
with those of while 2 is an (Nc Nc ) (Nc Nc ) matrix. The product g g has the same
representation for the product of quarks and moreover, the second gauge group indices for the
field play the role of the flavor indices.
. Then the solutions for
three equations from F-term conditions: g g 2 = 0 and g = 0 = g
these are given by

g =
1N
c
0

,
g
= 1N c

0 ,
=
0
0
0 1(N Nc )N
c
where the zero of g is an (Nc Nc N c ) N c matrix, the zero of g
is an N c (Nc

Nc N c ) matrix and the zeros of are N c N c , N c (Nc Nc N c ) and (Nc Nc
N c ) N c matrices. Then one can expand these fields around on a point, as in [4,15] and one
arrives at the relevant superpotential up to quadratic order in the fluctuation. At one loop, the
(1)
effective potential Veff
for 0 leads to the positive value for m2 implying that these vacua are
0
stable.
3. Meta-stable brane configurations with four NS-branes plus O4-plane
In this section, we add an orientifold 4-plane to the previous brane configurations and find out
new meta-stable brane configurations. Or one can realize these brane configurations by inserting
the extra NS-brane and O4-planes into the brane configuration [16].
294
Fig. 6. The N = 1 supersymmetric electric brane configuration for the gauge group Sp(Nc ) SO(2Nc ) Sp(Nc ) and
bifundamentals F and G with vanishing (A) and nonvanishing (B) mass for the bifundamental F . The 2Nc D4-branes
in (A) are decomposed into 2(Nc Nc ) D4-branes which are moving to v direction in Z2 symmetric way in (B) and
2Nc D4-branes which are recombined with those D4-branes connecting between NS5R -brane and NS5R -brane in (B).

The type IIA brane configuration [19] corresponding to N = 1 supersymmetric gauge theory
with gauge group

Sp(Nc ) SO 2Nc Sp Nc
(3.1)
and with a field F charged under (2Nc , 2Nc ), a field G charged under (2Nc , 2Nc ) can be
described by the left NS5L -brane, the left NS5L -brane, the right NS5R -brane, the right NS5R brane, 2Nc -, 2Nc - and 2Nc -color D4-branes as well as an O4 -plane (01236) we should add. The
O4 -planes act as (x 4 , x 5 , x 7 , x 8 , x 9 ) (x 4 , x 5 , x 7 , x 8 , x 9 ) as usual and they have RR
charge 1 playing the role of 1 D4-brane.
We draw this brane configuration in Fig. 6(A) for the vanishing mass case.
There is no superpotential in Fig. 6(A). Let us deform this theory. Displacing the two NS5 branes relative each other in the +v direction corresponds to turning on a quadratic massdeformed superpotential for the field F as follows:
W = mF F m
(3.2)
where the first gauge group indices in F are contracted, each second gauge group index
in F is encoded in and the mass m is given by (2.3). The gauge-singlet for the
first dual gauge group is in the adjoint representation for the second dual gauge group,
i.e., (1, (Nc Nc )(2Nc 2Nc 1), 1) under the dual gauge group (3.3). Then the is a
2(Nc Nc ) 2(Nc Nc ) matrix. The half NS5R -brane [20] together with (Nc Nc )-color
D4-branes is moving to the +v direction (and their mirrors to v direction) for fixed other
branes during this mass deformation. The 2Nc D4-branes among 2Nc D4-branes are not participating in the mass deformation. Then the x 5 coordinate of NS5L -brane is equal to zero while the
x 5 coordinates of half NS5R -brane are given by x.
Giving an expectation value to the meson field corresponds to recombination of 2Nc - and
2Nc -color D4-branes, which will become 2Nc -color D4-branes in Fig. 6(A) such that they are
suspended between the NS5L -brane and the NS5R -brane and pushing them into the w direction.
295
Fig. 7. The N = 1 magnetic brane configuration for the gauge group Sp(N c = Nc Nc 2) SO(2Nc ) Sp(Nc )
NS5 -branes are close to each other. The number of tilted D4-branes in (B) can be written as Nc Nc + 2 =
(Nc Nc ) N c .
We assume that the number of colors satisfies

Nc Nc + 2 Nc .
Now we draw this brane configuration in Fig. 6(B) for nonvanishing mass for the field F .
By applying the Seiberg dual to the Sp(Nc ) factor in (3.1), the two NS5L,R -branes can be
located at the inside of the two NS5-branes, as in Fig. 7. Starting from Fig. 6(B) and interchanging
the NS5L -brane and the NS5L -brane, one obtains Fig. 7(A).
Before arriving at Fig. 7(A), there exists an intermediate step where the 2(Nc Nc 2)
D4-branes are connecting between the NS5L -brane and the NS5L -brane, (Nc Nc ) D4-branes
connecting between the NS5L -brane and NS5R -brane (and their mirrors), and 2Nc D4-branes
between the NS5L -brane and the NS5R -brane. By introducing 2Nc D4-branes and 2Nc
anti-D4-branes between the NS5L -brane and NS5L -brane, reconnecting the former with the
Nc D4-branes connecting between NS5L -brane and the NS5L -brane (therefore (Nc Nc ) D4branes) and moving those combined (Nc Nc ) D4-branes to +v direction (and their mirrors to
v direction), one gets the final Fig. 7(A) where we are left with 2(Nc Nc + 2) anti-D4-branes
between the NS5L -brane and NS5L -brane. When two NS5 -branes in Fig. 7(A) are close to each
other, then it leads to Fig. 7(B) by realizing that the number of (Nc Nc ) D4-branes connecting
between NS5L -brane and NS5R -brane can be rewritten as (Nc Nc + 2) plus N c .

Sp N c = Nc Nc 2 SO 2Nc Sp Nc .
(3.3)
c , 2Nc , 1), a field g charged under
The matter contents are the field f charged under (2N

(1, 2Nc , 2Nc ) under the dual gauge group (3.3) and the gauge-singlet that is in the adjoint
representation for the second dual gauge group, i.e., (1, (Nc Nc )(2Nc 2Nc 1), 1) under the
dual gauge group. That is, the is a 2(Nc Nc ) 2(Nc Nc ) antisymmetric matrix.
296
The cubic superpotential with the mass term (3.2) in the dual theory is given by
Wdual = ff + m .
(3.4)
Here the magnetic field f corresponds to 44 strings connecting the 2N c -color D4-branes (that
are connecting between the NS5L -brane and the NS5R -brane including the mirrors) with 2Nc flavor D4-branes (that is a combination of three different D4-branes including the mirrors in
Fig. 7(B)). Among these 2Nc -flavor D4-branes, only the strings ending on the upper 2(Nc
Nc 2) D4-branes and on the tilted middle 2(Nc Nc + 2) D4-branes including the mirrors in
Fig. 7(B) enter the cubic superpotential term. Although the (Nc Nc ) D4-branes in Fig. 7(A)
cannot move any directions, the tilted 2(Nc Nc + 2)-flavor D4-branes including the mirrors
can move w direction. The remaining upper N c D4-branes (and its mirrors) are fixed also and

Nc Nc = Nc Nc + 2 + N c .
The brane configuration for zero mass for the bifundamental, which has only a cubic superpotential, can be obtained from Fig. 7(A) by moving the upper and lower NS5 -branes together
with (Nc Nc )-color D4-branes into the origin v = 0. Then the number of dual colors for D4branes becomes 2N c between NS5L -brane and NS5L -brane and 2Nc between two NS5 -branes
as well as 2Nc between NS5R -brane and NS5R -brane. Or starting from Fig. 6(A) and moving
the NS5L -brane to the left all the way past the NS5L -brane, one also obtains the corresponding
NS5 -brane and the middle NS5 -brane is large. If they are close to each other, then this brane
configuration is unstable to decay and leads to the brane configuration in Fig. 7(B). One can regard these brane configurations as particular states in the magnetic gauge theory with the gauge
group (3.3) and superpotential (3.4). The upper (Nc Nc N c )-flavor D4-branes of straight
brane configuration of Fig. 7(B) bend due to the fact that there exists an attractive gravitational
interaction between those flavor D4-branes and NS5L -brane from the DBI action, as long as the
distance y3 goes to because the presence of an extra NS5R -brane does not affect the DBI action. For the finite and small y3 , the careful analysis for DBI action is needed in order to obtain the
bending curve connecting two NS5 -branes. Of course, their mirrors, the lower (Nc Nc N c )flavor D4-branes of straight brane configuration of Fig. 7(B) can bend and their trajectories connecting two NS5 -branes should be preserved under the O4-plane, i.e., Z2 symmetric
way.
When the upper and lower half NS5R -branes are replaced by coincident (Nc Nc ) D6-branes
and the NS5R is rotated by an angle 2 in the (v, w) plane in Fig. 7(B), this brane configuration
reduces to the one found in [16] where the gauge group was given by Sp(nf + nc nc 2)
SO(2nc ) with 2nf multiplets, flavor singlet and gauge singlets. Then the present (Nc Nc )
corresponds to the nf , the number Nc corresponds to nc and Nc corresponds to the nc of [16].
However, the gauge group Sp(Nc ) corresponds to the different gauge group SO(2nc ). When
we discuss Section 3.5 and take the Seiberg dual for the middle gauge group, then it becomes
SO(2Nc ) Sp(N c = Nc + Nc Nc 2) SO(2Nc ). Then the Nc corresponds to the nf , the
number Nc corresponds to nc and Nc corresponds to the nc of [16]. If we ignore 2Nc D4-branes
and NS5R -brane from Fig. 7(B), then the brane configuration becomes the one in [6,21].
297
The dual gauge theory has an adjoint of SO(2Nc ) and bifundamentals f and g under the
dual gauge group (3.3) and the superpotential corresponding to Figs. 7(A) and 7(B) is given by
Wdual = h ff h2 ,
2
h2 = g2,mag
, 2 =
x
.
2gs 3s
Then ff is a 2N c 2N c matrix where the second gauge group indices for f are contracted with
those of while 2 is a 2(Nc Nc ) 2(Nc Nc ) matrix. The product ff has the same representation for the product of quarks and moreover, the first gauge group indices for the field
play the role of the flavor indices, as we observed above for the comparison with the brane configuration in the presence of D6-branes.
be satisfied if the 2(Nc Nc ) exceeds 2N c . So the supersymmetry is broken. That is, there exist
two equations from F-term conditions: ff 2 = 0 and f = 0. Then the solutions for these
are given by

0
0
12N c

f =
(3.5)
,
=
0 0 1(N N N c )i2
0
c
where the zero of f is a 2(Nc Nc N c ) 2N c matrix and the zeros of are 2N c 2N c ,
2N c 2(Nc Nc N c ), and 2(Nc Nc N c ) 2N c matrices. Then one can expand these fields
around on a point (3.5), as in [4] and one arrives at the relevant superpotential up to quadratic
(1)
order in the fluctuation. At one loop, the effective potential Veff for 0 leads to the positive value
2
for m implying that these vacua are stable.
0
3.3. Other magnetic theory-I

Let us consider other magnetic theory for the same electric theory given in Section 3.1. By
applying the Seiberg dual to the SO(2Nc ) factor in (3.1), the two NS5L,R -branes can be located
at the left-hand side of the two NS5-branes, as in Fig. 9.
Fig. 8(A) is the same as the one in Fig. 6(A) and one moves half NS5L -brane together with Nc
D4-branes to +v direction (and its mirrors to v direction) and is given by Fig. 8(B). Starting
from Fig. 8(B) and interchanging the NS5L -brane and the NS5R -brane, one obtains Fig. 9(A).
Before arriving at Fig. 9(A), there exists an intermediate step where the Nc D4-branes are connecting between the NS5L -brane and the NS5R -brane (and their mirrors), 2(Nc Nc + Nc + 2)
D4-branes are connecting between the NS5R -brane and NS5L -brane, and 2Nc D4-branes are
suspended between the NS5L -brane and the NS5R -brane. By moving the combined Nc D4branes, obtained from the reconnection of those D4-branes between the NS5L -brane and the
NS5R -brane and those D4-branes between the NS5R -brane and the NS5L -brane (therefore between the NS5L -brane and the NS5L -brane), to +v direction (and their mirrors to v direction),
one gets the final Fig. 9(A) where we are left with 2(Nc Nc 2) anti-D4-branes between the
NS5R -brane and NS5L -brane. We assume that the number of colors satisfies
Nc + Nc Nc 2 Nc .
When two NS5 -branes in Fig. 9(A) are close to each other, then it leads to Fig. 9(B) by realizing
that the number of Nc D4-branes connecting between NS5L -brane and NS5L -brane in Fig. 9(A)
can be rewritten as (Nc Nc 2) plus N c . If we ignore 2Nc D4-branes and NS5R -brane and
298
Fig. 8. The N = 1 supersymmetric electric brane configuration for the gauge group Sp(Nc ) SO(2Nc ) Sp(Nc )
and bifundamentals F and G with vanishing (A) which is the same as Fig. 6(A) and nonvanishing (B) mass for the
bifundamental F . This deformation is different from the previous case (3.2). In (B), the 2Nc D4-branes are moving to
v directions in Z2 symmetric way.
Fig. 9. The N = 1 magnetic brane configuration for the gauge group Sp(Nc ) SO(2N c = 2Nc + 2Nc 2Nc + 4)
Sp(Nc ) corresponding to Fig. 8(B) with D4- and D4-branes (A) and with a misalignment between D4-branes (B) when
the NS5 -branes are close to each other. The number of tilted D4-branes is equal to Nc Nc 2 = Nc N c in (B).
change the O4-plane charge (corresponding to change the symplectic gauge group into the orthogonal gauge group and vice versa) from Fig. 9(B), then the brane configuration becomes the
one in [9].

Sp(Nc ) SO 2N c = 2Nc + 2Nc 2Nc + 4 Sp Nc .
(3.6)
c , 1), a field g charged under

The matter contents are the field f charged under (2Nc , 2N
c , 2Nc ) under the dual gauge group (3.6) and the gauge-singlet for the second dual
(1, 2N
gauge group in the adjoint representation for the first dual gauge group, i.e., (Nc (2Nc + 1), 1, 1)
under the dual gauge group. Then the is a 2Nc 2Nc matrix. All the 2Nc D4-branes are
participating in the mass deformation.
299
The cubic superpotential with the mass term in the dual theory4 is given by
Wdual = ff + m
(3.7)
where we define as F F and the second gauge group indices in F are contracted, each
first gauge group index in them is encoded in . Although the that has first gauge group
indices looks similar to the previous that has second gauge group indices, the group indices
are different. Here the magnetic field f corresponds to 44 strings connecting the 2N c -color
D4-branes (that are connecting between the NS5L -brane and the NS5L -brane in Fig. 9(B)) with
2Nc -flavor D4-branes including the mirrors (which are realized as corresponding D4-branes in
Fig. 9(A)). Although the Nc D4-branes (and its mirrors) in Fig. 9(A) cannot move any directions,
the tilted (Nc Nc 2)-flavor D4-branes (and its mirrors) can move w direction in Fig. 9(B). The
remaining upper N c D4-branes (and its mirrors) are fixed also and cannot move any direction.
Note that there is a decomposition

Nc = Nc Nc 2 + N c .
The brane configuration for zero mass for the bifundamental, which has only a cubic superpotential, can be obtained from Fig. 9(A) by moving the upper and lower NS5 -branes together
with Nc -color D4-branes into the origin v = 0. Then the number of dual colors for D4-branes
becomes 2Nc between two NS5 -branes, 2N c between NS5R -brane and NS5L -brane and 2Nc
between NS5L -brane and NS5R -brane. Or starting from Fig. 8(A) and moving the NS5L -brane
to the right all the way past the NS5R -brane, one also obtains the corresponding magnetic brane
configuration for massless case.
NS5 -brane and the middle NS5 -brane is large. If they are close to each other, then this brane
configuration is unstable to decay and leads to the brane configuration in Fig. 9(B). One can regard these brane configurations as particular states in the magnetic gauge theory with the gauge
group (3.6) and superpotential (3.7). The (Nc N c )-flavor D4-branes of straight brane configuration of Fig. 9(B) bend due to the fact that there exists an attractive gravitational interaction
between those flavor D4-branes and NS5L -brane from the DBI action, as long as the distance y3
goes to because the presence of an extra NS5R -brane does not affect the DBI action. For the
finite and small y3 , the careful analysis for DBI action is needed in order to obtain the bending
curve connecting two NS5 -branes. Or if y3 goes to zero, then this extra NS5R -brane plays the
role of enhancing the strength for the NS5-branes and will affect both the energy of bending
curve, Ecurved , and x [3]. Of course, their mirrors, the lower (Nc N c )-flavor D4-branes of
straight brane configuration of Fig. 9(B) can bend and their trajectories connecting two NS5 branes should be preserved under the O4-plane, i.e., Z2 symmetric way.
When the NS5 -brane (or NS5L -brane) is replaced by coincident Nc D6-branes, this brane
configuration looks similar to the one found in [16] where the gauge group was given by
SO(2nf + 2nc 2nc + 4) Sp(nc ) with 2nf multiplets, flavor singlet and gauge singlets. Then
the present Nc corresponds to the nf , the number Nc corresponds to nc and Nc corresponds to
the nc of [16].
4 One can also construct the mass deformation by rotating NS5 -brane and moving it to v direction, as in previous
R
case in Section 2.3. The brane configuration can be obtained easily.
300
The dual gauge theory has an adjoint of Sp(Nc ) and bifundamentals f and g under the dual
gauge group (3.6) and the superpotential corresponding to Figs. 9(A) and 9(B) is given by
Wdual = hff h2 ,
2
h2 = g1,mag
, 2 =
x
.
2gs 3s
Then ff is a 2N c 2N c matrix where the first gauge group indices for f are contracted with
those of while 2 is a 2Nc 2Nc matrix. The product ff has the same representation for the
product of quarks and moreover, the first gauge group indices for the field play the role of the
flavor indices as we observed above.
be satisfied if the 2Nc exceeds 2N c . So the supersymmetry is broken. That is, there exist two
equations from F-term conditions: ff 2 = 0 and f = 0. Then the solutions for these are
given by

0
0
12N
c
,
=
f =
(3.8)
0 0 12(Nc N )
0
c
where the zero of f is a 2(Nc N c ) 2N c matrix and the zeros of are 2N c 2N c ,
2N c 2(Nc N c ) and 2(Nc N c ) 2N c matrices. Then one can expand these fields around on
a point (3.8), as in [4] and one arrives at the relevant superpotential up to quadratic order in the
(1)
fluctuation. At one loop, the effective potential Veff for 0 leads to the positive value for m20
implying that these vacua are stable.
3.4. Other magnetic theory-II
One can think of the following dual gauge group

Sp(Nc ) SO 2Nc Sp N c = Nc Nc 2
(3.9)
by performing the magnetic dual for the last gauge group in (3.1). The electric brane configuration can be given in terms of Fig. 6(A) or modified Fig. 6(A) with an exchange between
NS5-brane and NS5 -brane. For the latter, the resulting brane configuration is given by NS5L brane, NS5L -brane, NS5R -brane, and NS5R -brane from the left to the right in the x 6 direction.
One can take the magnetic dual either by following the previous procedure or by looking at
Fig. 7 from the negative w direction which is an opposite viewpoint, compared with Fig. 7. In
other words, we are looking at Fig. 7 from the other side of w.
Then the resulting brane configuration in this case can be obtained by taking a reflection
for all the NS-branes, D4-branes and anti-D4-branes with respect to the NS5L -brane (rotating
them to the left for fixed NS5L -brane) in Figs. 7(A) and 7(B). Then the N = 1 magnetic brane
configuration for the gauge group Sp(Nc ) SO(2Nc ) Sp(N c = Nc Nc 2) corresponds to
Fig. 10(A ) with D4- and D4-branes and Fig. 10(B ) with a misalignment between D4-branes
when the NS5 -branes are close to each other. The number of tilted D4-branes in Fig. 10(B )
can be written as Nc Nc + 2 = (Nc Nc ) N c . We do not present Figs. 10(A ) and 10(B )
here.
We turn to the other case. Let us consider other magnetic theory for the same electric theory
given in Section 3.1 with Fig. 6(A). By applying the Seiberg dual to the Sp(Nc ) factor in (3.1)
from Fig. 6(A) and interchanging the NS5R -brane and the NS5R -brane, one obtains Fig. 10(A ).
Before arriving at Fig. 10(A ), there exists an intermediate step where 2Nc D4-branes between
301
Fig. 10. The N = 1 magnetic brane configuration for the gauge group Sp(Nc ) SO(2Nc ) Sp(N c = Nc Nc 2)
with D4- and D4-branes (A ) and with a misalignment between D4-branes (B ) when the NS5 -branes are close to each
other. The number of tilted D4-branes in (B ) can be written as Nc Nc + 2 = (Nc Nc ) N c . The deformation is
related to the bifundamentals G.
NS5L -brane and the NS5L -brane, the 2Nc D4-branes are connecting between the NS5L -brane
and the NS5R -brane, (Nc Nc 2) D4-branes are connecting between the NS5R -brane and
NS5R -brane (and their mirrors). By rotating NS5L -brane by an angle 2 , moving it with the
(Nc Nc ) D4-branes to +v direction where we introduce 2(Nc Nc ) D4-branes and 2(Nc Nc )
anti-D4-branes between the NS5R -brane and the NS5R -brane, one gets the final Fig. 10(A )
where we are left with 2(Nc Nc + 2) anti-D4-branes between the NS5R -brane and the NS5R brane. When two NS5 -branes in Fig. 10(A ) are close to each other, then it leads to Fig. 10(B )
by realizing that the number of (Nc Nc ) D4-branes connecting between NS5M -brane and NS5brane can be rewritten as (Nc Nc + 2) plus N c .
The brane configuration in Fig. 10(A ) is stable as long as the distance x between the upper
NS5 -brane and the middle NS5 -brane (or NS5R -brane) is large. If they are close to each other,
then this brane configuration is unstable to decay to the brane configuration in Fig. 10(B ).
One can regard these brane configurations as particular states in the magnetic gauge theory with
the gauge group and superpotential. The upper (Nc Nc N c )-flavor D4-branes of straight
brane configuration of Fig. 10(B ) bend since there exists an attractive gravitational interaction
between those flavor D4-branes and NS5-brane from the DBI action. As mentioned in [9], the
two NS5 -branes are located at different side of NS5-brane in Fig. 10(B ) and the DBI action
computation for this bending curve should be taken into account. Of course, their mirrors, the
lower (Nc Nc N c )-flavor D4-branes of straight brane configuration of Fig. 10(B ) can bend
and their trajectories connecting two NS5 -branes should be preserved under the O4-plane, i.e.,
Z2 symmetric way.
The matter contents are the field f charged under (2Nc , 2Nc , 1), a field g charged under
c ) under the dual gauge group (3.9) and the gauge-singlet that is in the adjoint
(1, 2Nc , 2N
representation for the second dual gauge group, i.e., (1, (Nc Nc )(2Nc 2Nc 1), 1) under the
dual gauge group. That is, the is a 2(Nc Nc ) 2(Nc Nc ) antisymmetric matrix.
The cubic superpotential with the mass term in the dual theory is given by
Wdual = gg + m
(3.10)
302
where we define as GG and the third gauge group indices in G are contracted, each
second gauge group index in G is encoded in . Although the that has second gauge group
indices looks similar to the previous that has first gauge group indices, the group indices are
different. Here the magnetic field g correspond to 44 strings connecting the 2N c -color D4branes including the mirrors (that are connecting between the NS5M -brane and the NS5-brane
in Fig. 10(B )) with 2Nc -flavor D4-branes. Among these 2Nc -flavor D4-branes, only the strings
ending on the upper 2(Nc Nc 2) D4-branes and on the tilted middle 2(Nc Nc + 2) D4branes in Fig. 10(B ) enter the cubic superpotential term. Although the (Nc Nc ) D4-branes
(and its mirrors) in Fig. 10(A ) cannot move any directions, the tilted (Nc Nc + 2)-flavor D4branes (and its mirrors) can move w direction. The remaining upper and lower N c D4-branes are
fixed also and cannot move any direction. Note that there is a decomposition

Nc Nc = Nc Nc + 2 + N c .
The brane configuration for zero mass for the bifundamental, which has only a cubic superpotential, can be obtained from Fig. 10(A ) by moving the upper and lower NS5 -branes
together with (Nc Nc )-color D4-branes into the origin v = 0. Then the number of dual colors
for D4-branes becomes 2Nc between two NS5 -branes and 2Nc between the NS5M -brane and
the NS5-brane and 2N c between the NS5-brane and the NS5R -brane. Or starting from Fig. 6(A)
and moving the NS5R -brane to the left all the way past the NS5R -brane, one also obtains the
corresponding magnetic brane configuration for massless case.
The dual gauge theory has an adjoint of SO(2Nc ) and bifundamentals f and g under the
dual gauge group (3.9) and the superpotential corresponding to Figs. 10(A ) and 10(B ) is given
by
Wdual = h gg h2 ,
2
h2 = g2,mag
, 2 =
x
.
2gs 3s
Then gg is a 2N c 2N c matrix where the second gauge group indices for g are contracted with
those of while 2 is a 2(Nc Nc ) 2(Nc Nc ) matrix. The product gg has the same representation for the product of quarks and moreover, the first gauge group indices for the field
play the role of the flavor indices.
be satisfied if the 2(Nc Nc ) exceeds 2N c . So the supersymmetry is broken. That is, there exist
two equations from F-term conditions: gg 2 = 0 and g = 0. Then the solutions for these
are given by

g =
12N
c
0

,
=
0
0
0 1(N Nc N )i2
c
(3.11)
where the zero of g is a 2(Nc Nc N c ) 2N c matrix and the zeros of are 2N c 2N c ,
2N c 2(Nc Nc N c ), and 2(Nc Nc N c ) 2N c matrices. Then one can expand these
fields around on a point (3.11), as in [4] and one arrives at the relevant superpotential up to
(1)
quadratic order in the fluctuation. At one loop, the effective potential Veff for 0 leads to the
positive value for m2 implying that these vacua are stable.
0
303
3.5. Other magnetic theories-III

By changing the charges of O4-plane in previous brane configuration of Fig. 6(A), the type IIA
brane configuration is realized by an N = 1 supersymmetric gauge theory with

SO(2Nc ) Sp Nc SO 2Nc
and corresponding matter contents. Then by deforming the theory by mass term and taking the
magnetic dual on each gauge group factor, one gets meta-stable brane configurations. There exists
an N = 1 magnetic supersymmetric gauge theory with SO(2N c = 2Nc 2Nc + 4) Sp(Nc )
SO(2Nc ) with matters which corresponds to Fig. 7 with opposite O4-plane charges. Also there is
an N = 1 magnetic supersymmetric gauge theory with SO(2Nc )Sp(N c = Nc +Nc Nc 2)
SO(2Nc ) with matters which corresponds to Fig. 9 with opposite O4-plane charges. Finally, there
exists an N = 1 magnetic supersymmetric gauge theory with SO(2Nc ) Sp(Nc ) SO(2N c =
2Nc 2Nc + 4) with matters which corresponds to Fig. 10 with opposite O4-plane charges. The
remaining analysis can be done easily without any difficulty.
4. Meta-stable brane configurations with six NS-branes plus O6-plane
In this section, we add an orientifold 6-plane to the previous brane configuration for the product gauge group [10] realized by three NS-branes, together with the extra mirrors for them, and
find out new meta-stable brane configurations. Or one can realize these brane configurations by
inserting the two outer NS-branes into the brane configuration [14,22].
The type IIA brane configuration corresponding to N = 1 supersymmetric gauge theory with
gauge group

Sp(Nc ) SU Nc SU Nc
(4.1)
and with a field F charged under (2Nc , Nc ), a field G charged under (Nc , Nc ), and their conju can be described by the left NS5 -brane, the NS5-brane, the right NS5 -brane
gates F and G
L
R
(and their mirrors), 2Nc -, Nc - and Nc -color D4-branes as well as O6-plane (0123789).5 The
O6 -plane acts as (x 4 , x 5 , x 6 ) (x 4 , x 5 , x 6 ) and has RR charge 4.
Let us place an O6-plane at the origin x 6 = 0 and let us denote the x 6 coordinates for the
NS5L -brane, the NS5-brane and the NS5R -brane by x 6 = y1 , y1 + y2 , y1 + y2 + y3 , respectively.
Their mirrors can be understood similarly. The 2Nc D4-branes are suspended between the NS5L brane and its mirror, the Nc D4-branes are suspended between the NS5L -brane and the NS5brane (and their mirrors), and the Nc D4-branes are suspended between the NS5-brane and the
NS5R -brane (and their mirrors). We draw this brane configuration in Fig. 11(A) for the vanishing
mass for the field G.
5 From now on, when we say about NS-branes (NS5-brane or NS5 -brane), they refer to those in positive region of x 6 .
Their mirrors in the negative region of x 6 are understood with O6-plane while we are taking the brane motion. In other
words, there exist three NS-branes: NS5L -brane, NS5-brane and NS5R -brane from Fig. 11(A).
304
Fig. 11. The N = 1 supersymmetric electric brane configuration for the gauge group Sp(Nc ) SU(Nc ) SU(Nc ) and
with vanishing (A) and nonvanishing (B) mass for the bifundamental G and G.
In (B),
bifundamental F , F , G and G
the NS5R -brane together with Nc D4-branes is moving to +v direction (and their mirrors to v direction).
The gauge couplings of Sp(Nc ), SU(Nc ) and SU(Nc ) are given by a string coupling constant gs , a string scale s and the x 6 coordinates yi for three NS-branes through
g12 =
gs s
,
2y1
g22 =
gs s
,
y2
g32 =
gs s
.
y3
As y3 goes to , the SU(Nc ) gauge group becomes a global symmetry and the theory leads to
SQCD with the gauge group Sp(Nc ) SU(Nc ) and Nc flavors in the fundamental representation.
There is no superpotential in Fig. 11(A). Let us deform this theory. Displacing the two
NS5 -branes relative each other in the +v direction corresponds to turning on a quadratic massdeformed superpotential for the field G as follows:
m
W = mGG
(4.2)
are contracted and the mass m is given by (2.3).

where the second gauge group indices in G and G

The gauge-singlet for the second dual gauge group is in the adjoint representation for the third
dual gauge group, i.e., (1, 1, Nc 2 1) (1, 1, 1) under the dual gauge group (4.3). The is an
Nc Nc matrix. The NS5R -brane together with Nc -color D4-branes is moving to the +v direction for fixed other branes during this mass deformation (and their mirrors to v direction).
Then the x 5 coordinate of NS5L -brane is equal to zero while the x 5 coordinate of NS5R -branes
is given by x. Giving an expectation value to the meson field corresponds to recombination
of Nc - and Nc -color D4-branes, which will become Nc - or Nc -color D4-branes in Fig. 11(A)
such that they are suspended between the NS5L -brane and the NS5R -brane and pushing them
into the w direction. We assume that the number of colors satisfies
2Nc + Nc Nc 2Nc .
Now we draw this brane configuration in Fig. 11(B) for nonvanishing mass for the fields G
The geometry for three NS-branes in Fig. 11(B) is the same as the one given by first three
and G.
NS-branes in Fig. 1(B).
305
Fig. 12. The N = 1 magnetic brane configuration for the gauge group Sp(Nc ) SU(N c = 2Nc + Nc Nc ) SU(Nc )
NS5 -branes are close to each other. The number of tilted D4-branes is equal to Nc 2Nc = Nc N c in (B). The
notation for the anti-D4-branes is used for the bar on the number of those branes in (A).

By applying the Seiberg dual to the SU(Nc ) factor in (4.1), the NS5L,R -branes can be located
at the outside of the two NS5-branes, as in Fig. 12. Starting from Fig. 11(B) and interchanging
the NS5L -brane and the NS5-brane (and their mirrors), one obtains Fig. 12(A).
Before arriving at Fig. 12(A), there exists an intermediate step where the (Nc Nc +2Nc ) D4branes are connecting between the NS5-brane and the NS5L -brane, Nc D4-branes are connecting
between the NS5L -brane and NS5R -brane (and their mirrors) as well as 2Nc D4-branes between
the NS5-brane and its mirror. By reconnecting the Nc D4-branes connecting between the NS5brane and the NS5L -brane with the Nc D4-branes connecting between NS5L -brane and the
NS5R -brane and moving those combined Nc D4-branes to +v direction (and their mirrors to
v direction), one gets the final Fig. 12(A) where we are left with (Nc 2Nc ) anti-D4-branes
between the NS5-brane and NS5L -brane. When two NS5 -branes in Fig. 12(A) are close to each
other, it becomes Fig. 12(B) by realizing that the number of Nc D4-branes connecting between
NS5-brane and NS5R -brane in Fig. 12(A) can be rewritten as (Nc 2Nc ) plus N c . The brane
configuration consisting of NS5-brane and two NS5 -branes in Fig. 12(B) is exactly the same as
those in Fig. 2(B).

Sp(Nc ) SU N c = 2Nc + Nc Nc SU Nc .
(4.3)
c , 1), a field g charged under
The matter contents are the field f charged under (2Nc , N

c , Nc ), and their conjugates f and g under the dual gauge group (4.3) and the gauge(1, N
singlet for the second dual gauge group in the adjoint representation for the third dual gauge
group, i.e., (1, 1, Nc 2 1) (1, 1, 1) under the dual gauge group. Then the is an Nc Nc
matrix.
The cubic superpotential with the mass term (4.2) is given by
Wdual = g g + m .
(4.4)
Here the magnetic fields g and g correspond to 44 strings connecting the Nc -color D4-branes
(that are connecting between the NS5-brane and the NS5R -brane in Fig. 12(B)) with Nc -flavor
306
D4-branes (which are realized as corresponding D4-branes in Fig. 12(A)). Although the Nc D4branes in Fig. 12(A) cannot move any directions, the tilted (Nc 2Nc )-flavor D4-branes can
move w direction in Fig. 12(B) (and its mirrors). The remaining upper N c D4-branes are fixed
also and cannot move any direction. Note that there is a decomposition

Nc = Nc 2Nc + N c .
The brane configuration for zero mass for the bifundamental, which has only a cubic superpotential, can be obtained from Fig. 12(A) by moving the upper NS5 -brane (or NS5R -brane)
together with Nc -color D4-branes into the origin v = 0 (and their mirrors). Then the number
of dual colors for D4-branes becomes 2Nc between the NS5-brane and its mirror, N c between
NS5-brane and NS5L -brane and Nc between NS5L -brane and NS5R -brane. Or starting from
Fig. 11(A) and moving the NS5-brane to the left all the way past the NS5L -brane (and their
mirrors), one also obtains the corresponding magnetic brane configuration for massless case.
NS5 -brane and the lower NS5 -brane is large. If they are close to each other, then this brane
configuration is unstable to decay and leads to the brane configuration in Fig. 12(B). One can
regard these brane configurations as particular states in the magnetic gauge theory with the gauge
group (4.3) and superpotential (4.4).
One can perform similar analysis in our brane configuration since one can take into account
the behavior of parameters geometrically in the presence of O6-plane. Then the upper (Nc N c )flavor D4-branes of straight brane configuration of Fig. 12(B) can bend due to the fact that there
exists an attractive gravitational interaction between those flavor D4-branes and NS5-brane from
the DBI action, by following the procedure of [3], as long as y1 is very large. Then the mirror
of NS5-brane does not affect the flavor D4-branes. On the other hand, if y1 goes to zero, then
the mirror of NS5-brane plays the role of enhancing the strength for the NS5-branes and will
affect both the energy of bending curve, Ecurved , and x. Of course, their mirrors, the lower
(Nc N c )-flavor D4-branes of straight brane configuration of Fig. 12(B) can bend and their trajectories connecting two NS5 -branes should be preserved under the O6-plane, i.e., Z2 symmetric
way.
2
g1,mag
=
gs s
,
2(y1 + y2 )
2
=
g2,mag
gs s
,
y2
2
g3,mag
=
gs s
.
(y3 y2 )
The dual gauge theory has an adjoint of SU(Nc ) and bifundamentals f, f, g and g under
the dual gauge group (4.3) and the superpotential corresponding to Figs. 12(A) and 12(B) is
given by
Wdual = h g g h2 ,
2
h2 = g3,mag
, 2 =
x
.
2gs 3s
Then g g is an N c N c matrix where the third gauge group indices for g and g are contracted
with those of while 2 is an Nc Nc matrix. The product g g has the same representation
for the product of quarks and moreover, the third gauge group indices for the field play the
role of the flavor indices.
When the upper NS5 -brane (or NS5R -brane) is replaced by coincident Nc D6-branes in
Fig. 12(B), this brane configuration looks similar to the one found in [14] where the gauge group
307
Fig. 13. The N = 1 supersymmetric electric brane configuration for the gauge group Sp(Nc ) SU(Nc ) SU(Nc ) and
the bifundamentals with vanishing (A) which is the same as Fig. 11(A) and nonvanishing (B) mass for the bifundamental
This deformation is different from the one in (4.2). The Nc D4-branes in (A) are decomposed into (Nc 2Nc )
G and G.
D4-branes which are moving to +v direction in (B) and 2Nc D4-branes which are recombined with those D4-branes
connecting between NS5L -brane and its mirror in (B).
was given by SU(nf + 2nc nc ) Sp(nc ) with nf multiplets and singlets. Then the present 2Nc
corresponds to the 2nc , Nc corresponds to nc , and Nc corresponds to the nf of [14].
equations from F-term conditions: g g 2 = 0 and g = 0 = g
these are given by

0
0
1N

c
g =
(4.5)
=
,
g
= 1N c 0 ,
0 0 1(N N )
0
c
is an N c (Nc N c ) matrix
where the zero of g is an (Nc N c ) N c matrix, the zero of g
and the zeros of are N c N c , N c (Nc N c ) and (Nc N c ) N c matrices. Then one can
expand these fields around on a point (4.5), as in [4] and one arrives at the relevant superpotential
(1)
up to quadratic order in the fluctuation. At one loop, the effective potential Veff for 0 leads to
2
the positive value for m implying that these vacua are stable.
0
4.3. Other magnetic theory

Let us consider other magnetic theory for the same electric theory given in Section 4.1. By
applying the Seiberg dual to the SU(Nc ) factor in (4.1), the NS5L,R -branes can be located at
the inside of the two NS5-branes, as in Fig. 14. Starting from Fig. 13(B) and interchanging the
NS5-brane and the NS5R -brane (and their mirrors), one obtains Fig. 14(A). The geometry for
three NS-branes in Fig. 13(B) is the same as the one given by first three NS-branes in Fig. 3(B).
Before arriving at Fig. 14(A), there exists an intermediate step where the (Nc 2Nc ) D4branes are connecting between the NS5L -brane and the NS5R -brane, (Nc Nc ) D4-branes are
connecting between the NS5R -brane and NS5-brane (and their mirrors) as well as 2Nc D4-branes
between NS5R -brane and its mirror. By reconnecting the (Nc 2Nc ) D4-branes connecting between the NS5L -brane and the NS5R -brane with the (Nc 2Nc ) D4-branes connecting between
NS5R -brane and the NS5-brane where we introduce 2Nc D4-branes and 2Nc anti-D4-branes
308
Fig. 14. The N = 1 magnetic brane configuration for the gauge group Sp(Nc ) SU(Nc ) SU(N c = Nc Nc )
corresponding to Fig. 13(B) with D4- and D4-branes (A) and with a misalignment between D4-branes (B) when
the NS5 -branes are close to each other. The number of tilted D4-branes in (B) can be written as Nc 2Nc =
(Nc 2Nc ) N c .
and moving those combined D4-branes to +v direction (and their mirrors to v direction), one
gets the final Fig. 14(A) where we are left with (Nc 2Nc ) anti-D4-branes between the NS5R brane and the NS5-brane. We assume that the number of colors satisfies
Nc Nc 2Nc .
realizing that the number of (Nc 2Nc ) D4-branes connecting between NS5L -brane and NS5brane in Fig. 14(A) can be rewritten as (Nc 2Nc ) plus N c . The brane configuration consisting
of NS5-brane and two NS5 -branes in Fig. 14(B) is exactly the same as those in Fig. 4(B).

Sp(Nc ) SU Nc SU N c = Nc Nc .
(4.6)
The matter contents are the field f charged under (2Nc , Nc , 1), a field g charged under
c ) and their conjugates f and g under the dual gauge group (4.6) and the gauge(1, Nc , N
singlet which is in the adjoint representation for the second dual gauge group, in other
words, (1, (Nc 2Nc )2 1, 1) (1, 1, 1) under the dual gauge group (4.6). Then the is
an (Nc 2Nc ) (Nc 2Nc ) matrix. Only (Nc 2Nc ) D4-branes can participate in the mass
deformation.
The cubic superpotential with the mass term is given by
Wdual = g g + m
(4.7)
and the third gauge group indices in G and G

are contracted,
where we define as GG

each second gauge group index in them is encoded in . Although the that has second gauge
group indices looks similar to the previous that has third gauge group indices, the group
indices are different. Here the magnetic fields g and g correspond to 44 strings connecting
the N c -color D4-branes (that are connecting between the NS5L -brane and the NS5-brane in
Fig. 14(B)) with Nc -flavor D4-branes. Among these Nc -flavor D4-branes, only the strings ending
on the upper (Nc Nc ) D4-branes and on the tilted (Nc 2Nc ) D4-branes in Fig. 14(B) enter the
309
cubic superpotential term. Although the (Nc 2Nc ) D4-branes in Fig. 14(A) cannot move any
directions, the tilted (Nc 2Nc )-flavor D4-branes can move w direction. The remaining upper
N c D4-branes are fixed also and cannot move any direction. Note that there is a decomposition

Nc 2Nc = Nc 2Nc + N c .
The brane configuration for zero mass for the bifundamental, which has only a cubic superpotential, can be obtained from Fig. 14(A) by moving the upper NS5 -brane together with
(Nc 2Nc )-color D4-branes into the origin v = 0 (and their mirrors). Then the number of dual
colors for D4-branes becomes 2Nc between the NS5L -brane and its mirror, Nc between the
NS5L -brane and the NS5R -brane and N c between NS5R -brane and NS5-brane. Or starting from
Fig. 13(A) and moving the NS5-brane to the right all the way past the NS5R -brane (and their
mirrors), one also obtains the corresponding magnetic brane configuration for massless case.
NS5 -brane and the lower NS5 -brane is large. If they are close to each other, then this brane configuration is unstable to decay and leads to the brane configuration in Fig. 14(B). One can regard
these brane configurations as particular states in the magnetic gauge theory with the gauge group
(4.6) and superpotential (4.7). Then the upper (Nc 2Nc N c )-flavor D4-branes of straight
brane configuration of Fig. 14(B) can bend due to the fact that there exists an attractive gravitational interaction between those flavor D4-branes and NS5-brane from the DBI action, as long
as y1 is very large. Of course, their mirrors, the lower (Nc 2Nc N c )-flavor D4-branes of
straight brane configuration of Fig. 14(B) can bend and their trajectories connecting two NS5 branes should be preserved under the O6-plane, i.e., Z2 symmetric way.
When the upper NS5 -brane (or NS5L -brane) is replaced by coincident (Nc 2Nc ) D6-branes
in Fig. 14(B), this brane configuration looks similar to the one found in [14] where the gauge
group was given by SU(nf + nc nc ) SO(nc ) with nf multiplets, bifundamentals, and singlets.
Then the present 2Nc corresponds to the nc , (Nc 2Nc ) corresponds to nf , and Nc corresponds
to the nc of [14]. Note that Sp(Nc ) corresponds to SO(nc ). Moreover, there is a meta-stable brane
configuration for the gauge group given by SU(nc ) SU(nf + nc nc ) with fundamentals,
bifundamentals, an antisymmetric flavor, a conjugate symmetric flavor, and singlets where there
are NS5 -brane, O6 -planes, and eight semi infinite D6-branes at x 6 = 0. Then the our 2Nc
corresponds to the nc , the number (Nc 2Nc ) corresponds to nf , and our Nc corresponds to
the nc of [23].
2
g1,mag
=
gs s
,
2y1
2
g2,mag
=
gs s
,
(y2 y3 )
2
=
g3,mag
gs s
.
y3
The dual gauge theory has an adjoint of SU(Nc ) and bifundamentals f , f, g and g under the
dual gauge group (4.6) and the superpotential corresponding to Figs. 14(A) and 14(B) is given
by
Wdual = h g g h2 ,
2
h2 = g2,mag
, 2 =
x
.
2gs 3s
with those of while 2 is an (Nc 2Nc ) (Nc 2Nc ) matrix. The product g g has the same
310
field play the role of the flavor indices, as above.
be satisfied if the (Nc 2Nc ) exceeds N c . So the supersymmetry is broken. That is, there exist
these are given by

0
0
1N
c
g =
(4.8)
=
,
g
= 1N c 0 ,

0 0 1(N 2Nc )N
0
c
c
where the zero of g is an (Nc 2Nc N c ) N c matrix, the zero of g
is an N c (Nc
2Nc N c ) matrix and the zeros of are N c N c , N c (Nc 2Nc N c ) and (Nc 2Nc
N c ) N c matrices. Then one can expand these fields around on a point (4.8), as in [4] and one
arrives at the relevant superpotential up to quadratic order in the fluctuation. At one loop, the
(1)
0
stable.
4.4. Other magnetic theories
In this subsection, we add an orientifold 6-plane with positive charge to the previous brane
configuration for the product gauge group [10] realized by three NS-branes, together with the extra mirrors for them, and find out new meta-stable brane configurations. Or one can realize these
brane configurations by inserting the two outer NS-branes into the brane configuration [14,22].
4.4.1. Electric theory
The type IIA brane configuration corresponding to N = 1 supersymmetric gauge theory with
gauge group

SO(Nc ) SU Nc SU Nc
(4.9)
and with a field F charged under (Nc , Nc ), a field G charged under (Nc , Nc ), and their conjugates
can be described by the left NS5L -brane, the NS5 -brane, the right NS5R -brane (and
F and G
their mirrors), Nc -, Nc - and Nc -color D4-branes as well as O6+ -plane (0123789). The O6+ plane acts as (x 4 , x 5 , x 6 ) (x 4 , x 5 , x 6 ) and has RR charge +4.
Let us place an O6+ -plane at the origin x 6 = 0 and let us denote the x 6 coordinates for the
NS5L -brane, the NS5 -brane and the NS5R -brane by x 6 = y1 , y1 + y2 , y1 + y2 + y3 , respectively.
Their mirrors can be understood similarly. The Nc D4-branes are suspended between the NS5L brane and its mirror, the Nc D4-branes are suspended between the NS5L -brane and the NS5 brane (and their mirrors), and the Nc D4-branes are suspended between the NS5 -brane and the
NS5R -brane (and their mirrors). We assume that the number of colors satisfies
Nc + Nc Nc Nc .
4.4.2. Magnetic theory
By applying the Seiberg dual to the SU(Nc ) factor in (4.9) and interchanging the NS5L -brane
and the NS5 -brane (and their mirrors), one obtains Fig. 15(A).
Before arriving at Fig. 15(A), there exists an intermediate step where the (Nc Nc + Nc )
D4-branes are connecting between the NS5 -brane and the NS5L -brane, Nc D4-branes are connecting between the NS5L -brane and NS5R -brane (and their mirrors) as well as Nc D4-branes
311
Fig. 15. The N = 1 magnetic brane configuration for the gauge group SO(Nc ) SU(N c = Nc + Nc Nc ) SU(Nc )
with D4- and D4-branes (A) and with a misalignment between D4-branes (B) when the NS5 -branes are close to each
other. Note that the number of D4-branes on the gauge group SO(Nc ) is equal to Nc not 2Nc . The number of tilted
D4-branes is equal to Nc Nc = Nc N c in (B). The deformation is related to the bifundamentals G and G.
between NS5 -brane and its mirror. By rotating NS5R -brane by an angle 2 , moving it with Nc
D4-branes to +v direction (and their mirrors to v direction), one gets the final Fig. 15(A) where
we are left with (Nc Nc ) anti-D4-branes between the NS5L -brane and NS5-brane. When two
NS5 -branes in Fig. 15(A) are close to each other, it becomes Fig. 15(B) by realizing that the
number of Nc D4-branes connecting between NS5-brane and NS5R -brane can be rewritten as
(Nc Nc ) plus N c .

SO(Nc ) SU N c = Nc + Nc Nc SU Nc .
(4.10)
c , Nc ),
c , 1), a field g charged under (1, N
The matter contents are the field f charged under (Nc , N
the second dual gauge group in the adjoint representation for the third dual gauge group, i.e.,
(1, 1, Nc 2 1) (1, 1, 1) under the dual gauge group. Then the is an Nc Nc matrix.
The cubic superpotential with the mass term is given by (4.4) where we define as
and the second gauge group indices in G and G
are contracted, each third gauge group
GG
index in them is encoded in . Although the that has third gauge group indices looks
similar to the previous that has second gauge group indices the group indices are different.
Here the magnetic fields g and g correspond to 44 strings connecting the N c -color D4-branes
(that are connecting between the NS5-brane and the NS5R -brane in Fig. 15(B)) with Nc -flavor
D4-branes (which are realized as corresponding D4-branes in Fig. 15(A)). Although the Nc D4branes in Fig. 15(A) cannot move any directions, the tilted (Nc Nc )-flavor D4-branes can move
w direction in Fig. 15(B) (and its mirrors). The remaining upper N c D4-branes are fixed also and

Nc = Nc Nc + N c .
The brane configuration for zero mass for the bifundamental, which has only a cubic superpotential, can be obtained from Fig. 15(A) by moving the upper NS5 -brane together with Nc -color
D4-branes into the origin v = 0 (and their mirrors). Then the number of dual colors for D4-branes
becomes Nc between the NS5L -brane and its mirror, N c between NS5L -brane and NS5-brane
and Nc between NS5-brane and NS5R -brane.
312
The brane configuration in Fig. 15(A) is stable as long as the distance x between the upper NS5 -brane and the lower NS5 -brane is large. If they are close to each other, then this
brane configuration is unstable to decay to the brane configuration in Fig. 15(B). One can regard these brane configurations as particular states in the magnetic gauge theory with the gauge
group and superpotential. The upper (Nc N c )-flavor D4-branes of straight brane configuration
of Fig. 15(B) bend since there exists an attractive gravitational interaction between those flavor
D4-branes and NS5-brane from the DBI action. As mentioned in [9], the two NS5 -branes are
located at different side of NS5-brane in Fig. 15(B) and the DBI action computation for this
gauge group factors are given by the expressions in Section 4.2. The dual gauge theory has an
adjoint of SU(Nc ) and bifundamentals f , f, g and g under the dual gauge group (4.10)
and the superpotential corresponding to Figs. 15(A) and 15(B) is given by the expressions in
Section 4.2. Then g g is an N c N c matrix where the third gauge group indices for g and g
are contracted with those of while 2 is an Nc Nc matrix. The product g g has the same
representation for the product of quarks and moreover, the third gauge group indices for the field
play the role of the flavor indices.
When the upper NS5 -brane (or NS5R -brane) is replaced by coincident Nc D6-branes in
Fig. 15(B), this brane configuration looks similar to the one found in [14] where the gauge group
was given by SU(nf + nc nc ) Sp(nc ) with nf multiplets and singlets. Then the present Nc
corresponds to the nc , Nc corresponds to nc , and Nc corresponds to the nf of [14].
equations from F-term conditions: g g 2 = 0 and g = 0 = g
. Then the solutions for these
are given by the expressions in Section 4.2. Then one can expand these fields around on a point,
as in [4] and one arrives at the relevant superpotential up to quadratic order in the fluctuation. At
(1)
one loop, the effective potential Veff for 0 leads to the positive value for m2 implying that
0
these vacua are stable.
4.4.3. Other magnetic theory
Let us consider other magnetic theory for the same electric theory given in Section 4.4.1. By
applying the Seiberg dual to the SU(Nc ) factor in (4.9) and interchanging the NS5 -brane and
the NS5R -brane (and their mirrors), one obtains Fig. 16(A).
Before arriving at Fig. 16(A), there exists an intermediate step where the Nc D4-branes are
connecting between the NS5L -brane and the NS5R -brane, (Nc Nc ) D4-branes are connecting
between the NS5R -brane and NS5 -brane (and their mirrors) as well as Nc D4-branes between
NS5L -brane and its mirror. By rotating NS5L -brane by an angle 2 , moving it with the (Nc Nc )
D4-branes to +v direction where we introduce (Nc Nc ) D4-branes and (Nc Nc ) anti-D4branes between the NS5R -brane and the NS5 -brane (and their mirrors to v direction), one gets
the final Fig. 16(A) where we are left with (Nc Nc ) anti-D4-branes between the NS5-brane
and the NS5R -brane. We assume that the number of colors satisfies
Nc Nc Nc .
realizing that the number of (Nc Nc ) D4-branes connecting between NS5L -brane and NS5-
313
Fig. 16. The N = 1 magnetic brane configuration for the gauge group SO(Nc ) SU(Nc ) SU(N c = Nc Nc ) with
D4- and D4-branes (A) and with a misalignment between D4-branes (B) when the NS5 -branes are close to each other.
The number of tilted D4-branes in (B) can be written as Nc Nc = (Nc Nc ) N c . The deformation is different from
the previous one.
brane can be rewritten as (Nc Nc ) plus N c . The brane configuration consisting of NS5-brane
and two NS5 -branes in Fig. 16(B) is exactly the same as those in Fig. 5(B ).

SO(Nc ) SU Nc SU N c = Nc Nc .
(4.11)
c )
The matter contents are the field f charged under (Nc , Nc , 1), a field g charged under (1, Nc , N
and their conjugates f and g under the dual gauge group (4.11) and the gauge-singlet which
is in the adjoint representation for the second dual gauge group, in other words, (1, (Nc Nc )2
1, 1) (1, 1, 1) under the dual gauge group (4.11). Then the is an (Nc Nc ) (Nc Nc )
matrix. Only (Nc Nc ) D4-branes are participating in the mass deformation.
The cubic superpotential with the mass term is given by (4.7) where we define as GG
and the third gauge group indices in G and G are contracted, each second gauge group index in
them is encoded in . Although the that has second gauge group indices looks similar to the
previous that has third gauge group indices, the group indices are different. Here the magnetic
fields g and g correspond to 44 strings connecting the N c -color D4-branes (that are connecting
between the NS5L -brane and the NS5-brane in Fig. 16(B)) with Nc -flavor D4-branes. Among
these Nc -flavor D4-branes, only the strings ending on the upper (Nc Nc ) D4-branes and on
the tilted (Nc Nc ) D4-branes in Fig. 16(B) enter the cubic superpotential term. Although the
(Nc Nc ) D4-branes in Fig. 16(A) cannot move any directions, the tilted (Nc Nc )-flavor D4branes can move w direction. The remaining upper N c D4-branes are fixed also and cannot move
any direction. Note that there is a decomposition

The brane configuration for zero mass for the bifundamental, which has only a cubic superpotential, can be obtained from Fig. 16(A) by moving the upper NS5 -brane together with
(Nc Nc )-color D4-branes into the origin v = 0 (and their mirrors). Then the number of dual
colors for D4-branes becomes Nc between the NS5L -brane and its mirror, Nc between the NS5L brane and the NS5-brane and N c between NS5-brane and NS5R -brane.
314
When the upper NS5 -brane (or NS5L -brane) is replaced by coincident (Nc Nc ) D6-branes
in Fig. 16(B), this brane configuration looks similar to the one found in [14] where the gauge
group was given by SU(nf + 2nc nc ) Sp(nc ) with nf multiplets, bifundamentals, and
singlets. Then the present Nc corresponds to the 2nc , (Nc Nc ) corresponds to nf , and Nc
corresponds to the nc of [14]. Note that SO(Nc ) corresponds to Sp(nc ). Moreover, the metastable brane configuration corresponding to gauge group given by SU(nc ) SU(nf + nc nc )
with fundamentals, bifundamentals, a symmetric flavor, a conjugate symmetric flavor, and singlets was given in [23] where there exists NS5-brane on the O6-plane. Then our Nc corresponds
to the nc , our (Nc Nc ) corresponds to nf , and our Nc corresponds to the nc .
NS5 -brane and the lower NS5 -brane is large. If they are close to each other, then this brane
configuration is unstable to decay to the brane configuration in Fig. 16(B). One can regard these
brane configurations as particular states in the magnetic gauge theory with the gauge group and
superpotential. The upper (Nc Nc N c )-flavor D4-branes of straight brane configuration of
Fig. 16(B) bend since there exists an attractive gravitational interaction between those flavor
D4-branes and NS5-brane from the DBI action. As mentioned in [9], the two NS5 -branes are
located at different side of NS5-brane in Fig. 16(B) and the DBI action computation for this
gauge group factors are given by the expressions in Section 4.3. The dual gauge theory has an
adjoint of SU(Nc ) and bifundamentals f, f, g and g under the dual gauge group (4.11) and
the superpotential corresponding to Figs. 16(A) and 16(B) is given by the one in Section 4.3.
with those of while 2 is an (Nc Nc ) (Nc Nc ) matrix. The product g g has the same
field play the role of the flavor indices.
these are given by

0
0
1N

c
0
1
g =
(4.12)
,
=
,
g
=
N c
0 0 1(N Nc )N
0
c
N c )
where the zero of g is an

Nc
matrix, the zero of g

is an N c (Nc
Nc N c ) matrix and the zeros of are N c N c , N c (Nc Nc N c ) and (Nc Nc
N c ) N c matrices. Then one can expand these fields around on a point (4.12), as in [4] and
one arrives at the relevant superpotential up to quadratic order in the fluctuation. At one loop, the
(1)
0
stable.
(Nc
N c
5. Conclusions and outlook

The meta-stable brane configurations we have found are summarized by Figs. 2, 4, 5, 7, 9,
10, 12, 14, 15 and 16. If we replace the NS5 -brane in Figs. 2(B), 7(B) with opposite O4-plane
charge, Fig. 14(B) with opposite O6-plane charge, and Fig. 16(B) with opposite O6-plane charge,
315
with the coincident D6-branes, those brane configurations become nonsupersymmetric minimal
energy brane configurations found in [14], in [16], in [14], and in [14], respectively.
So far, we have considered the cases for even number of NS-branes, i.e., four and six. For
odd cases, i.e., three and five NS-branes, the construction of meta-stable brane configuration has
been done in [9]. So it is natural to ask what happens if there are seven NS-branes. When this
extra seventh NS-brane is located at the O6-plane in Section 4, then the gauge group will be the
same as the one in Section 2, i.e., SU(Nc ) SU(Nc ) SU(Nc ) with different matter contents.
This can be obtained also from the brane configuration of [23] by adding two outer NS-branes.
It would be interesting to find out how the meta-stable brane configurations appear.
Some different directions on the meta-stable vacua are present in recent relevant works [24
33] where some of them are described in the type IIB string theory. It would be very interesting
to find out how the meta-stable brane configurations from type IIA string theory including the
present work are related to those brane configurations from type IIB string theory.
Acknowledgements
I would like to thank D. Kutasov for discussions. I would like to thank Kyungho Oh, who
passed away from cancer, for ongoing collaboration and discussions during the last 10 years
and, in memory of him, I would like to dedicate this work to him. This work was supported
by grant No. R01-2006-000-10965-0 from the Basic Research Program of the Korea Science &
Engineering Foundation.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
A. Giveon, D. Kutasov, Brane dynamics and gauge theory, Rev. Mod. Phys. 71 (1999) 983, hep-th/9802067.
C.G. Callan, J.A. Harvey, A. Strominger, Supersymmetric string solitons, hep-th/9112030.
A. Giveon, D. Kutasov, Gauge symmetry and supersymmetry breaking from intersecting branes, hep-th/0703135.
K. Intriligator, N. Seiberg, D. Shih, Dynamical SUSY breaking in meta-stable vacua, JHEP 0604 (2006) 021, hepth/0602239.
H. Ooguri, Y. Ookouchi, Meta-stable supersymmetry breaking vacua on intersecting branes, Phys. Lett. B 641
(2006) 323, hep-th/0607183.
S. Franco, I. Garcia-Etxebarria, A.M. Uranga, Non-supersymmetric meta-stable vacua from brane configurations,
JHEP 0701 (2007) 085, hep-th/0607218.
I. Bena, E. Gorbatov, S. Hellerman, N. Seiberg, D. Shih, A note on (meta)stable brane configurations in MQCD,
JHEP 0611 (2006) 088, hep-th/0608157.
C. Ahn, Brane configurations for nonsupersymmetric meta-stable vacua in SQCD with adjoint matter, Class. Quantum Grav. 24 (2007) 1359, hep-th/0608160.
C. Ahn, Meta-stable brane configurations by adding an orientifold-plane to GiveonKutasov, arXiv: 0706.0042
[hep-th].
J.H. Brodie, A. Hanany, Type IIA superstrings, chiral symmetry, and N = 1 4D gauge theory dualities, Nucl. Phys.
B 506 (1997) 157, hep-th/9704043.
C. Ahn, R. Tatar, Geometry, D-branes and N = 1 duality in four dimensions with product gauge groups, Phys. Lett.
B 413 (1997) 293, hep-th/9705106.
R. Argurio, M. Bertolini, S. Franco, S. Kachru, Gauge/gravity duality and meta-stable dynamical supersymmetry
breaking, JHEP 0701 (2007) 083, hep-th/0610212.
R. Kitano, H. Ooguri, Y. Ookouchi, Direct mediation of meta-stable supersymmetry breaking, Phys. Rev. D 75
(2007) 045022, hep-ph/0612139.
C. Ahn, Meta-stable brane configuration of product gauge groups, arXiv: 0704.0121 [hep-th].
D. Shih, Spontaneous R-symmetry breaking in ORaifeartaigh models, hep-th/0703196.
C. Ahn, Meta-stable brane configuration and gauged flavor symmetry, hep-th/0703015.
C. Ahn, More on meta-stable brane configuration, Class. Quantum Grav. 24 (2007) 36033616, hep-th/0702038.
C. Ahn, Meta-stable brane configuration with orientifold 6 plane, JHEP 0705 (2007) 053, hep-th/0701145.
316
[19] C. Ahn, K. Oh, R. Tatar, Branes, geometry and N = 1 duality with product gauge groups of SO and Sp, J. Geom.
Phys. 31 (1999) 301, hep-th/9707027.
[20] G. Bertoldi, B. Feng, A. Hanany, The splitting of branes on orientifold planes, JHEP 0204 (2002) 015, hep-th/
0202090.
[21] C. Ahn, M-theory lift of meta-stable brane configuration in symplectic and orthogonal gauge groups, Phys. Lett.
B 647 (2007) 493, hep-th/0610025.
[22] E. Lopez, B. Ormsby, Duality for SU SO and SU Sp via branes, JHEP 9811 (1998) 020, hep-th/9808125.
[23] C. Ahn, Meta-stable brane configurations with five NS5-branes, arXiv: 0705.0056 [hep-th].
[24] J. Marsano, K. Papadodimas, M. Shigemori, Nonsupersymmetric brane/antibrane configurations in type IIA and
M theory, arXiv: 0705.0983 [hep-th].
[25] I. Garcia-Etxebarria, F. Saad, A.M. Uranga, Supersymmetry breaking metastable vacua in runaway quiver gauge
theories, arXiv: 0704.0166 [hep-th].
[26] S. Murthy, On supersymmetry breaking in string theory from gauge theory in a throat, hep-th/0703237.
[27] R. Argurio, M. Bertolini, S. Franco, S. Kachru, Metastable vacua and D-branes at the conifold, hep-th/0703236.
[28] Y.E. Antebi, T. Volansky, Dynamical supersymmetry breaking from simple quivers, hep-th/0703112.
[29] M. Wijnholt, Geometry of particle physics, hep-th/0703047.
[30] J.J. Heckman, J. Seo, C. Vafa, Phase structure of a brane/anti-brane system at large N , hep-th/0702077.
[31] R. Tatar, B. Wetenhall, Metastable vacua, geometrical engineering and MQCD transitions, JHEP 0702 (2007) 020,
hep-th/0611303.
[32] H. Verlinde, On metastable branes and a new type of magnetic monopole, hep-th/0611069.
[33] M. Aganagic, C. Beem, J. Seo, C. Vafa, Geometrically induced metastability and holography, hep-th/0610249.
Differences between charged-current

coefficient functions
S. Moch a , M. Rogal a , A. Vogt b,
a Deutsches Elektronensynchrotron DESY, Platanenallee 6, D-15738 Zeuthen, Germany
b Department of Mathematical Sciences, University of Liverpool, Liverpool L69 3BX, United Kingdom
Received 30 August 2007; accepted 14 September 2007

Abstract
Second- and third-order results are presented for the structure functions of charged-current deep-inelastic
scattering in the framework of massless perturbative QCD. We write down the two-loop differences between the corresponding crossing-even and -odd coefficient functions, including those for the longitudinal
structure function not covered in the literature so far. At three loops we compute the lowest five moments of
these differences for all three structure functions and provide approximate expressions in Bjorken-x space.
Also calculated is the related third-order coefficient-function correction to the Gottfried sum rule. We confirm the conjectured suppression of these quantities if the number of colours is large. Finally we derive the
second- and third-order QCD contributions to the PaschosWolfenstein ratio used for the determination of
the weak mixing angle from neutrinonucleon deep-inelastic scattering. These contributions are found to
be small.
PACS: 12.38.-t; 13.60.Hb; 12.38.Bx
Keywords: Quantum chromodynamics; Deep-inelastic scattering; Precision perturbative corrections
1. Introduction
Structure functions in deep-inelastic scattering (DIS) are among the most extensively measured observables. Today the combined data from fixed-target experiments and the HERA collider spans about four orders of magnitude in both Bjorken-x and the scale Q2 = q 2 given by
E-mail address: andreas.vogt@liverpool.ac.uk (A. Vogt).

doi:10.1016/j.nuclphysb.2007.09.022
318
S. Moch et al. / Nuclear Physics B 790 (2008) 317335
the momentum q of the exchanged electroweak gauge boson [1]. In this article we focus on the
W -exchange charged-current (CC) case, see Refs. [24] and [58] for recent measurements in
neutrino DIS and at HERA. With six structure functions, F2W , F3W and FLW , this case has a far
richer structure than, for example, electromagnetic DIS with only two independent observables,
F2 and FL .
More detailed measurements are required to fully exploit the resulting potential, for instance
at a future neutrino factory, see Ref. [9], and the LHeC, the proposed high-luminosity electronproton collider at the LHC [10]. Already now, however, charged-current DIS provides important
information on the parton structure of the proton, e.g., its flavour decomposition and the valencequark distributions. Moreover, present results are also sensitive to electroweak parameters of the
Standard Model such as sin2 W , see Ref. [11], and the space-like W -boson propagator [12]. As
discussed, for example, in Refs. [1316], a reliable determination of sin2 W from neutrino DIS
requires a detailed understanding of non-perturbative and perturbative QCD effects.
The perturbative calculations for the unpolarised structure functions in DIS have almost been
completed to the next-to-next-to-leading order (NNLO) of massless QCD. These results include
the splitting functions, controlling the scale evolution of the parton distributions, to the third order
in the strong coupling constant s [17,18], as well as the hard-scattering coefficient functions
for F1 , F2 and F3 to second order in s [1923]. For the longitudinal structure function FL =
F2 2xF1 the third-order coefficient functions are required at NNLO. So far these quantities
have been computed only for electromagnetic (photon-exchange) DIS [24,25]. In fact, it appears
that even the second-order coefficient functions for the charged-current FL have not been fully
presented in the literature.
It is convenient to consider linear combinations of the charged-current structure functions
p p
W
(a = 2, 3, L) for neutrino DIS. For
Fa with simple properties under crossing, such as Fa
all these combinations either the even or odd moments can be calculated in Mellin-N space
in the framework of the operator product expansion (OPE), see Ref. [26]. The results for the
p+ p
can be taken over from
third-order coefficient functions for the even-N combinations F2,L
p+ p
electromagnetic DIS [24,25]. Also the coefficient function for the odd-N based quantity F3
is completely known at three-loop accuracy, with the results only published via compact paramep p
p p
trizations so far [27]. For the remaining combinations F2,L
and F3
, on the other hand,
only the first five odd and even integer moments of the respective coefficient functions have
been calculated to third order in Ref. [28] following the approach of Refs. [2931] based on the
M INCER program [32,33].
The complete results of Refs. [24,25,27] fix all even and odd moments N . Hence already the
present knowledge is sufficient to determine also the lowest five moments of the differences of
corresponding even-N and odd-N coefficient functions and to address a theoretical conjecture
[34] for these quantities. Furthermore these moments facilitate x-space approximations in the
style of, e.g., Ref. [35] which are sufficient for most phenomenological purposes, including the
determination of the third-order QCD corrections to the PaschosWolfenstein relation [36] used
for the extraction of sin2 W from neutrino DIS.
The outline of this article is as follows. In Section 2 we briefly specify our notations and write
(2)
down the complete second-order results ca (x) for the above coefficient-function differences.
We discuss their behaviour at the end points x = 0 and x = 1, and provide compact but accurate
parametrizations for use in numerical applications. We then proceed, in Section 3, to our new
(3)
(3)
results for the five lowest odd moments of c2,L and even moments of c3 , as a byproduct deriving the third-order coefficient-function correction to the Gottfried sum rule. These three-loop
319
moments are presented in a numerical form and employed to construct x-space approximations
valid at x 102 . In Section 4 we address the numerical implications of our results. In particular
we discuss the higher-order QCD corrections to the PaschosWolfenstein relation. Our findings
are finally summarized in Section 5. The lengthy full expressions of the new third-order moments
in terms of fractions and the Riemann -function can be found in Appendix A.
2. The complete second-order results
We define the evenodd differences of the CC coefficient functions Ca for a = 2, 3, L as
p+ p
C2,L = C2,L
p p
C2,L
p p
C3 = C3
p+ p
C3
(2.1)
The signs are chosen such that the differences are always evenodd in the moments N accesp+ p
[27,31] is removed
sible by the OPE [26], and it is understood that the d abc dabc part of C3
before the difference is formed. The non-singlet quantities (2.1) have an expansion in powers
of s ,

Ca =
(2.2)
asl ca(l) ,
l=2
where, as throughout this and the next section, we are have normalized the expansion parameter
as as = s /(4). There are no first-order contributions to these differences, hence the sums start
at l = 2 in Eq. (2.2).
All known DIS coefficient functions in massless perturbative QCD can be expressed in terms
of the harmonic polylogarithms Hm1 ,...,mw (x) with mj = 0, 1. Our notation for these functions
follows Ref. [37] to which the reader is referred for a detailed discussion. For w 3 the harmonic polylogarithms can be expressed in terms of standard polylogarithms; a complete list can
be found in Appendix A of Ref. [23]. A F ORTRAN program for these functions up to weight
w = 4 has been provided in Ref. [38], with an unpublished extension also covering w = 5. In the
remainder of this section we employ the short-hand notation
H0, . . . , 0,1,0, . . . , 0,1,... (x) = H(m+1),(n+1),... (x)

m

n
(2.3)
and additionally suppress the arguments of the harmonic polylogarithms for brevity.
Exact expressions for (moments of) the coefficient functions will be given in terms of the
SU(Nc ) colour factors CA = Nc and CF = (Nc2 1)/(2Nc ), while we use the QCD values
CA = 3 and CF = 4/3 in numerical results. All our results are presented in the MS scheme
for the standard choice r = f = Q of the renormalization and factorization scales.
(2)
The second-order coefficient functions c2(2) and cL
for the evenodd differences of F2,L
read

324
16
164
144 2
(2)
c2 (x) = CF [CF CA /2]
+ 112(1 + x)1 3 + x 1 +
x+
x
5
5
5
5
144
403 + 1363 x + 82 + 562 x + 962 x 2
2 x 3 32H2,0
5
+ 96H2,0 (1 + x)1 + 128H2,0 x 128H1 (1 + x)1 2 + 48H1 2
144H1 2 x + 32H1,1,0 128H1,1,0 (1 + x)1 224H1,1,0 x
16
144
+ 64H1,0 + H1,0 x 2 + 64H1,0 x + 96H1,0 x 2
H1,0 x 3
5
5
320
64H1,0,0 + 160H1,0,0 (1 + x)1 + 160H1,0,0 x + 64H1,2 (1 + x)1

28
16
292
H0 32H0 (1 + x)1 H0 x 1
H0 x
5
5
5
144
+ 32H0 (1 + x)1 2 +
H0 x 2 16H0 2 + 16H0 2 x 16H0,0 64H0,0 x
5
144
96H0,0 x 2 +
H0,0 x 3 + 24H0,0,0 48H0,0,0 (1 + x)1 24H0,0,0 x
5

32H1 + 32H1 x 16H2 16H2 x + 16H3 32H3 (1 + x)1 16H3 x ,
32H1,2 + 32H1,2 x +
(2.4)

64
416
256
96
(2)
cL
(x) = CF [CF CA /2]
x 1
+
x + x 2 + 643 x + 322 x + 642 x 2
5
5
5
5
96
2 x 3 + 64H2,0 x 64H1 2 x 128H1,1,0 x + 64H1,0
5
64
96
+ H1,0 x 2 32H1,0 x 1 + 64H1,0 x + 64H1,0 x 2 H1,0 x 3
5
5
32
64
448
96
+ 64H1,0,0 x + H0 H0 x 1
H 0 x + H0 x 2
5
5
5
5

96
32H0,0 x 64H0,0 x 2 + H0,0 x 3 .
(2.5)
5
(2)
The corresponding quantity c3 for the charged-current structure functions F3 is given by

(2)
(2)
c3 (x) = c2 (x) CF [CF

624 16 1 464
144 2
CA /2]
+ x +
x+
x + 323
5
5
5
5
144
2 x 3 + 32H2,0 + 96H2,0 x
5
32H1 2 96H1 2 x 64H1,1,0 192H1,1,0 x + 64H1,0
+ 963 x 162 + 482 x + 802 x 2
16
144
H1,0 x 2 16H1,0 x 1 + 64H1,0 x + 80H1,0 x 2
H1,0 x 3
5
5
16
112
592
144
+ 32H1,0,0 + 96H1,0,0 x H0 x 1
H0
H0 x +
H0 x 2
5
5
5
5

144
2
3
+ 16H0,0 48H0,0 x 80H0,0 x +
(2.6)
H0,0 x .
5
+
Expressions equivalent to Eqs. (2.4) and (2.6) have first been published in Refs. [20,22], respectively, and were later confirmed in Ref. [23]. To the best of our knowledge, on the other
(2)
hand, the function cL has not been documented in the literature before, see, e.g., Ref. [39] and
references therein. It was however calculated by the authors of Refs. [2022], distributed in a
F ORTRAN package of the two-loop coefficient functions, and employed for the parametrizations
of Ref. [40]. Our expression (2.5) agrees with this unpublished result.
It is instructive to briefly consider the end-point limits of the above results. Suppressing the
ubiquitous factor CF CF A CF [CF CA /2], the small-x behaviour of Eqs. (2.4)(2.6) is
321
(2)
Fig. 1. The oddeven non-singlet differences c2,L (x) of Eqs. (2.4) and (2.5), compared at x 0.8 to the corresponding
even-N coefficient functions calculated in Refs. [19,20,23].
(2)
c2 (x) 4 ln3 x 8 ln2 x (28 162 ) ln x 64 + 82 + 723 + ,

(2)
c3 (x) 4 ln3 x 16 ln2 x + (12 + 162 ) ln x + 44 + 242 + 403 + ,

(2)
cL (x) 32 ln x 48 + .
(2.7)
Thus the evenodd differences are not suppressed with respect to the p + p two-loop nonsinglet coefficient functions for x 0: the same powers of ln x enter Eqs. (2.7) and those
(2)
quantities. At large x, on the other hand, all three functions ca are suppressed by factors
(1 x)2 times logarithms, reading

(2)
c2 (x) = (12 82 )[1 x]CF CF A + O [1 x]2 ,

(2)
c3 (x) = (20 82 )[1 x]CF CF A + O [1 x]2 ,

(2)
cL (x) = (32 162 )[1 x]2 CF CF A + O [1 x]3 .
(2.8)
(2)
(x) (both multiplied by 1 for display purposes) are comThe differences c2(2) (x) and cL
pared to the corresponding even-N p + p coefficient functions in Fig. 1. The quantities (2.4)
and (2.5) are negligible at x 0.1 and at x 0.3, respectively, but indeed comparable to the
even-moment coefficient functions at small x. The corresponding results for F3 are qualitative
(2)
similar to those for F2 , but with c3 (x) small down to x 0.01.
For certain numerical applications, for instance for use with complex-N packages like
Ref. [41], it is convenient to have parametrizations of Eqs. (2.4)(2.6) in terms of elementary
functions. With an error of less than 0.1% these functions can be approximated by
322

(2)
c2 (x) 9.1587 57.70x + 72.29x 2 5.689x 3 xL0 68.804 + 24.40L0

+ 2.958L20 + 0.249L0 + 8/9L20 (2 + L0 ) (1 x),

c3(2) (x) 29.65 + 116.05x 71.74x 2 16.18x 3 + xL0 14.60 + 69.90x

0.378L20 8.560L0 + 8/9L20 (4 + L0 ) (1 x),

(2)
cL (x) 10.663 5.248x 7.500x 2 + 0.823x 3 + xL0 11.10 + 2.225L0

0.128L20 + 64/9L0 (1 x)2 .
(2.9)
Here we have employed the short-hand L0 = ln x and inserted the QCD values of CF and CA .
3. Third-order moments and approximations
Recently the first five odd-integer moments have been computed of the third-order coeffip p
in charged-current DIS, together with the corresponding moments
cient functions for F2,L
p p
N = 2, . . . , 10 for F3
[28]. Unlike previous fixed-N calculations, the complete three-loop
p+ p
[24,25]1 and F3P + P [27] facilitate analytic continuations to these values
results for F2,L
of N . We have performed this continuation using the x-space expressions in terms of harmonic
polylogarithms [37] and the Mellin transformation package provided with version 3 of F ORM
[42]. Thus we are in a position to derive the respective lowest five moments of the hitherto unknown third-order contributions to the evenodd differences (2.1). These moments represent the
main new results of this article. With one exception (see below) the exact SU(Nc ) expressions
are however deferred to Appendix A.
Here we present numerical results for QCD, using the conventions introduced at the beginning
of Section 2, recall especially as s /(4) and the scale choice r = f = Q. In addition nf
denotes the number of effectively massless quark flavours, and we use the notation Ca,N for the
N th moment of Ca (x). The results for F2 and FL read
C2,1 = 4.378539253as2 + as3 (125.2948456 0.6502282123nf ),
C2,3 = 0.138066958as2 + as3 (5.554493975 + 0.1939792023nf ),
C2,5 = 0.032987989as2 + as3 (0.707322026 + 0.0004910378nf ),
C2,7 = 0.013235254as2 + as3 (0.008816536 0.0201069660nf ),
C2,9 = 0.006828983as2 + as3 (0.133159220 0.0200289710nf )
(3.1)
and
CL,1 = 2.138954096as2 + as3 (106.6667685 + 3.294301343nf ),
CL,3 = 0.078259985as2 + as3 (9.239637919 + 0.2718024935nf ),
CL,5 = 0.016892540as2 + as3 (2.548566852 + 0.0650677125nf ),
CL,7 = 0.006263113as2 + as3 (1.075400460 + 0.0251053847nf ),
CL,9 = 0.003001231as2 + as3 (0.560603262 + 0.0122952192nf ).
(3.2)
1 The 3 coefficient functions for this process are those of photon-exchange DIS, but without the contributions of the
s
f l11 flavour classes, see Fig. 1 of Ref. [25], where the two photons couple to different quark loops.
323
The lowest even moments for the structure function F3 are given by
C3,2 = 0.1135841071as2 + as3 (8.386266870 + 0.0605431788nf ),
C3,4 = 0.0683669250as2 + as3 (1.237248886 + 0.0971522112nf ),
C3,6 = 0.0350849853as2 + as3 (1.370404531 + 0.0496762716nf ),
C3,8 = 0.0208455457as2 + as3 (1.052847874 + 0.0282541123nf ),
C3,10 = 0.0137316528as2 + as3 (0.798850682 + 0.0177100327nf ).
(3.3)
The new s3 contributions are rather large if compared to the leading second-order results also
included in Eqs. (3.1)(3.3) with, e.g., as = 1/50 corresponding to s 0.25. Except for the
lowest moment for a = 2, L, on the other hand, the integer-N differences Ca,N are entirely
negligible compared to the p p moments of Refs. [28,31].
Before we turn to the x-space implications of Eqs. (3.1)(3.3), let us briefly discuss some
interesting structural features of our third-order results. For this purpose we consider the exact
SU(Nc ) expression for the lowest moment of c2(3) given by

404720
562784 2 33200
175030 49216
(3)
2
2 +
3
+
2 3
c2,1 = CF CF A
81
27
81
135 2
9

4160
8992 3 1472 2
363896
303377 41350
5
2
3 + CF2 CF A
+
2
3
9
63
3
162
27
81

396824 2 26000
25616
1456 2 56432 3
+
2
2 3 +
5 +

135
9
9
3 3
315 2

39592
1408
30424 2 1792
8786 3056
2 +
3 +
2 3
2
5 .
+ CF CF A nf
81
27
81
9
135
9
(3.4)
(3)
As all other calculated moments of the functions ca (x), this result contains an overall factor
CF A = CF CA /2 = 1/(2Nc ). Hence the third-order evenodd differences are suppressed in
the large-Nc limit as conjectured, to all orders, in Ref. [34] on the basis of two-loop results in
particular for N = 1 Adler and Gottfried sum rules, for a recent discussion see also Ref. [43]. In
fact, up to the additional f l11 contribution absent in charged-current DIS (recall footnote 1),

d abc dabc
1472
256 2 1280
(3)
e.m. c2,1 =
3
2
5
288 + 962 +
Nc
3
5
3
= 33.67693293nf in QCD,
(3.5)
Eq. (3.4) represents the third-order coefficient-function correction to the Gottfried sum rule
(GSR)2 , since the Adler sum rule involving the non-singlet coefficient function C2,1 of the
p p combination does not receive any perturbative or non-perturbative corrections, see,
e.g., Ref. [44].
(l)
Another interesting feature of the functions ca=2,3 in Eq. (2.2) is the presence of -functions
up to weight 2l in the integer moments, e.g., terms up to 23 and 32 occur in the third-order result
2 Note that our overall normalization and expansion parameter differ from those of Ref. [34]. Consequently the corresponding GSR coefficients (3.1), (3.4) and (3.5) are larger by a factor 4l /3 at order sl than in their notation.
324

p p
(3.4). This is in contrast to the natural (OPE-based) moments of Ca

which only include
contributions up to weight 2l 1, see Refs. [2831]. Yet the x-space expressions of all these
quantities consist of harmonic polylogarithms up to weight 2l 1 corresponding to harmonic
sums up to weight 2l. Note also that, in the approach of Refs. [2022], the absence of weight-2l
terms in the natural moments appears to require a cancellation between different diagram classes.
We now return to the numerical moments (3.1)(3.3) and investigate their consequences
for the x-space functions ca(3) (x). We follow an approach successfully used, for instance, in
Ref. [35] when only the coefficient-functions moments of Refs. [2931] were known. Based on
the two-loop end-point behaviour in Eqs. (2.7) and (2.8) we expect small-x terms up to ln5 x
(3)
(3)
and ln3 x in c2,3 (x) and cL (x), respectively, and large-x limits including contributions up to
(3)
(1 x)a ln2 (1 x) with 2,3 = 1 and L = 2. Thus the x-space expressions of ca will be of
the form

ca(3) (x) = (1 x)a
2

m=1
Am ln
(1 x) + casmooth (x) + B1
ln x
1x

+
72
a
Bn lnn x,
n=2
(3.6)
where the functions casmooth (x) are finite for 0 x 1. For moment-based approximations a
simple ansatz is chosen for these functions, and its free parameters are determined from the
available moments together with a reasonably balanced subset of the coefficients Am and Bn .
This ansatz and the choice of the non-vanishing end-point parameters are then varied in order to
(3)
estimate the remaining uncertainties of ca (x). Finally for each value of a two (out of about
50) approximations, denoted below by A and B, are selected which indicate the widths of the
uncertainty bands.
For F2 and FL these functions are, with L0 = ln x, x1 = 1 x and L1 = ln x1 ,

(3)
c2,A (x) = 54.478L21 + 304.6L1 + 691.68x x1 + 179.14L0 0.1826L30

x
+ nf 20.822x 2 282.1 1 +
x1 285.58x + 112.3 3.587L20 L0 ,
2

(3)
c2,B (x) = 13.378L21 + 97.60L1 + 118.12x x1 91.196L20 0.4644L50

x
+ nf 4.522L1 + 447.88 1 +
x1
2

+ (514.02x + 147.05 + 7.386L0 )L0
(3.7)
and

(3)
cL,A
(x) = 495.49x 2 + 906.86 x12 983.23xx1 L0 + 53.706L20 + 5.3059L30

+ nf 29.95x 3 59.087x 2 + 379.91 x12 273.042xL20 + 71.482x1 L0 ,

(3)
cL,B (x) = (78.306L1 + 6.3838x)x12 + 20.809xx1 L0 114.47L20 22.222L30

+ nf 12.532L1 + 141.99x 2 250.62x x12

(153.586x 0.6569)x1 L0 .
(3.8)
325
p+ p
Fig. 2. The exact third-order coefficient functions of the even-N structure functions F2,L
[24,25] for four massless
flavours, and the corresponding odd-moment quantities obtained from these results and the approximations (3.7) and
(3.8) for the evenodd differences.
The corresponding results for F3 read

(3)
c3,A (x) = 3.216L21 + 44.50L1 34.588 x1 + 98.719L20 + 2.6208L50

nf 0.186L1 + 61.102(1 + x) x1

+ 122.51xL0 10.914L20 2.748L30 ,

(3)
c3,B (x) = 46.72L21 + 267.26L1 + 719.49x x1 171.98L0 + 9.470L30

x
+ nf 0.8489L1 + 67.928 1 +
x1
2

+ 97.922xL0 17.070L20 3.132L30 .
(3)
(3.9)
The resulting approximations for the p p

odd-N coefficient functions c2,L (x) are compared
in Fig. 2 to their exact counterparts [24,25] for the even-N non-singlet structure functions. The
third-order evenodd differences remain noticeable to larger values of x than at two loops, e.g.,
up to x 0.3 for F2 and x 0.6 for FL for the four-flavour case shown in the figure. The
(3)
moments N = 1, 3, . . . , 9 constrain c2,L (x) very well at x 0.1, and approximately down to
x 102 .
For some applications, such as the PaschosWolfenstein relation addressed in the next section,
(3)
one needs the second moments of the functions c2,L (x). These quantities can now be determined
approximately from the above x-space results, yielding
326
Fig. 3. Convolution of the six third-order CC coefficient functions for F2,3,L in p + p [24,25,27] and p p
[Eqs. (3.7)(3.9)] DIS with a schematic but typical non-singlet distribution f . All results have been normalized to f (x),
suppressing a large but trivial variation of the absolute convolutions for small and large values of x.
(3)
c2,2 = 20.19 0.39 + (0.691 0.040)nf ,

(3)
= 24.75 0.15 (0.792 0.014)nf .
cL,2
(3.10)
Here the central values are given by the respective averages of the approximations A and B in
Eqs. (3.7) and (3.8) which directly provide the upper and lower limits.
Returning to x-space we recall that uncertainty bands as in Fig. 2 do not directly indicate the
range of applicability of these approximations, since the coefficient functions enter observables
only via smoothening Mellin convolutions with non-perturbative initial distributions. In Fig. 3 we
therefore present the convolutions of all six third-order CC coefficient functions with a characteristic reference distribution. It turns out that the approximations (3.7) and (3.8) of the previous
figure can be sufficient down to values even below x = 103 . The uncertainty of c3(3) (x), on the
other hand, becomes relevant already at larger values, x 102 , as the lowest calculated moment
of this quantity, N = 2, has far less sensitivity to the behaviour at low x.
The three-loop corrections to the non-singlet structure functions are rather small even well
below the x-values shown in the figurerecall our small expansion parameter as : the thirdorder coefficient are smaller by a factor 2.0 103 if the expansion is written in powers of s .
Their sharp rise for x 1 is understood in terms of soft-gluon effects which can be effectively
resummed, if required, to next-to-next-to-next-to-leading logarithmic accuracy [45]. Our even
(3)
odd differences ca (x), on the other hand, are irrelevant at x > 0.1 but have a sizeable impact
at smaller x in particular on the corrections for F2 and FL .
327
Fig. 4. The first two approximations, denoted by LO and NLO, of the differences (2.2) for F2 and FL in charged-current
DIS. The results are shown for representative values of s and nf after convolution with the reference distribution f (x)
also employed in Fig. 3. The dashed curves correspond to the two approximations in Eqs. (3.7) and (3.8) for the new s3
contributions.
4. Applications
(3)
The approximate results for ca (x) facilitate a first assessment of the perturbative stability
of the evenodd differences (2.1). In Fig. 4 we illustrate the known two orders for F2 and FL for
s = 0.25 and nf = 4 massless quark flavours, employing the same reference quark distribution
as in Fig. 3. Obviously our new s3 corrections are important wherever these coefficient-function
differences are non-negligible. On the other hand, our results confirm that these quantities are
very small, and thus relevant only when a high accuracy is required. Presently this condition is
fulfilled only for the determination of the weak mixing angle W from neutrino DIS to which we
therefore turn now.
For this purpose one considers the so-called PaschosWolfenstein relation defined in terms of
a ratio of neutral-current and charged-current cross sections for neutrinonucleon DIS [36],
R =
( N X) ( N X)
.
( N X) ( N + X)
(4.1)
R directly measures sin2 W if the up and down valence quarks in the target carry equal momenta, and if the strange and heavy-quark sea distributions are charge symmetric. At the lowest
order of perturbative QCD one generally finds
RLO
=
1
sin2 W .
2
(4.2)
328
The quantity (4.1) has attracted considerable attention in recent years due to a determination of
sin2 W by the NuTeV collaboration [11]: within the Standard Model their result is at variance
with other measurements of this quantity [1], see also Refs. [1315] for detailed discussions.
Beyond the leading order Eq. (4.2) receives perturbative QCD corrections which involve the
second moments of coefficient functions for the N N neutral- and charged-current structure
functions.3 Armed with the results of Sections 2 and 3 we are now able to finalize the corresponding s2 contribution for massless quarks [14] and to present an accurate numerical result at
order s3 . We denote by q q q the second Mellin moments of the valence distributions of
the flavours q = u, d, s, . . . ,
1
q =
dx x q(x) q(x)
(4.3)
The QCD corrections to R can be expanded in inverse powers of the dominant isoscalar combination u + d of the parton distributionsrecall that the measurements of this ratio are
performed for (almost) isoscalar targets. After inserting the expansion of the MS coefficient
functions in powers of s , the PaschosWolfenstein ratio Eq. (4.1) can be written as

u d + c s 2
2
2
+ gLd gRd
3 gLu gRu
u +d

8 s
2
61
32 2
83
s2 15127 89
2
+ gL gR
+ 2
2 + 3 2
nf
9
1944
81
27
45
162

s3 5175965 356
586
128 2 190
9062
2
+ 3
2
3
2 +
5
n f + n f 3
52488
729
27
405
81
729
3

226 2
1 (3)
1
(3)
+
n c +
c
+ O (u + d )2 + O s4 .
729 f 32 2,2 128 L,2
R = gL2 gR2 +
(4.4)
Here the left- and right-handed weak couplings gLu , gLd , gRu and gRd are related to the weak
mixing angle sin2 W by
2
2
gL2 gLu
+ gLd
=
1
5
sin2 W + sin4 W ,
2
9
2
2
gR2 gRu
+ gRd
=
5 4
sin W .
9
(4.5)
Beyond the tree level, of course, these relations receive electroweak radiative corrections, see,
e.g., Ref. [46]. Eq. (4.4) shows the well-known fact that the relation (4.2) receives corrections if
the parton content of the target includes an isotriplet component, u = d , or a quark sea with
a C-odd component, s = 0 or c = 0. Notice also that perturbative QCD only affects these
corrections.
The exact second-order contribution in Eq. (4.4) differs from the result in Ref. [14] where
(2)
the function cL (x) of Eq. (2.5) was not included. The third-order corrections can now be
completed in a numerical form, using our approximations (3.10) for the second moments of
3 Specifically the ratio R includes, besides all N N CC coefficient functions, the neutral-current quantity C NC
3
which is equal to its charged-current counterpart C3N N at the perturbative orders considered here.
329

(3)
c2,L (x). For nf = 4 flavours (and disregarding electroweak corrections) we obtain

1
u d + c s
1
7 2
2
2
R = sin W +
sin W
1 sin W +
2
u + d
3
2

8 s
1 + 1.689s + (3.661 0.002)s2 + O (u + d )2 + O s4 .

9
(4.6)
The perturbation series in the square brackets appears reasonably well convergent for relevant
values of the strong coupling constant, with the known terms reading, e.g., 1 + 0.42 + 0.23 for
s = 0.25. Thus the s2 and s3 contributions correct the NLO estimate by 65% in this case. On
the other hand, due to the small prefactor of this expansion, the new third-order term increases the
complete curved bracket in Eq. (4.5) by only about 1%, which can therefore by considered as the
new uncertainty of this quantity due to the truncation of the perturbative expansion. Consequently
previous NLO estimates of the effect of, for instance, the (presumably mainly non-perturbative,
see Refs. [4749]) charge asymmetry of the strange sea remain practically unaffected by higherorder corrections to the coefficient functions.
5. Summary
In this article we have presented new results for the coefficient functions of inclusive chargedcurrent DIS in the framework of massless perturbative QCD. We have filled a gap in the two-loop
(2)
and p p
literature by writing down the corresponding difference cL (x) of the p + p
structure functions FL . Our main results are the lowest five (even- or odd-integer) Mellin mo(3)
ments of the third-order corrections ca (x) for all three structure functions Fa=2,3,L and approximations in Bjorken-x space based on these moments which are applicable down to at least
x 102 . As a byproduct we have calculated the related third-order coefficient-function correction to the Gottfried sum rule in photon-exchange DIS.
All our third-order results are proportional to the non-planar colour factor CA 2CF , thus
confirming a conjecture by Broadhurst, Kataev and Maxwell on the 1/Nc2 suppression of these
coefficient-function differences in the limit of a large number of colours Nc . Numerically our s3
corrections prove relevant in particular for F2 and FL wherever the differences of the p + p and
p p coefficient functions are not negligible. We have employed the above results to derive the
second- and third-order QCD corrections to the PaschosWolfenstein ratio R used to determine
the weak mixing angle from neutrino deep-inelastic scattering. The uncertainty due to uncalculated higher-order coefficient functions has been reduced to a level amply sufficient for the
foreseeable future, i.e., 1% for the coefficient-function factor multiplying the quark-distribution
asymmetries.
F ORM files and F ORTRAN subroutines with our results can be obtained from the preprint
server http://arXiv.org by downloading the source of this article. Furthermore they are available
from the authors upon request.
Note added
(3)
(3)
While this article was finalized, the 11th moments of the functions c2 (x) and cL (x)
have been computed [50]. Both results fall into the bands generated by the respective x-space
approximations in Section 3, thus confirming the reliability of these uncertainty estimates.
330
Acknowledgements
We would like to thank S. Alekhin, D. Broadhurst and A. Kataev for stimulating discussions. The work of S.M. and M.R. has been supported by the Helmholtz Gemeinschaft under
contract VH-NG-105 and in part by the Deutsche Forschungsgemeinschaft in Sonderforschungsbereich/Transregio 9. During the final stage of this research A.V. enjoyed the hospitality of the
Instituut-Lorentz of Leiden University.
Appendix A
Here we present the analytic expressions for the Mellin-space coefficient-function differences
(3)
ca,N
which were given numerically in Eqs. (3.1)(3.3). We use the notations and conventions
as specified at the beginning of Section 2 and above Eq. (2.4).
(3)
The first moment of c2 (x) has been written down in Eq. (3.4) above. The remaining known
moments of this quantity are given by

10093427
1472 2 7787113
1805677051 2648
(3)
2
c2,3 = CF CF A
5 +
3

2
466560
9
810
3 3
1944

55336
378838 2 8992 3
+
2 3
2
2
9
45
63

9321697
1456 2 8046059
5165481803 40648
+
5
3 +
+
2
+ CF2 CF A
1399680
9
810
3 3
1944

798328 2 56432 3
49842 3 +
2
2
135
315

405586
139573
20396669 1792
5 +
3
2
+ nf C F C F A
116640
9
405
486

1408
50392 2
+
(A.1)
2 3
2 ,
9
135

149815672
1472 2
18473631996593 17584
5 +
3
3827250000
45
7875
3 3

291199027
330416
2577928 2 8992 3
2 +
2 3
2
2
50625
45
225
63

1270840912
1456 2
47560
16016244428419
+ CF2 CF A
+
5
3 +
3827250000
9
70875
3 3

1321405949
89128
26658224 2 56432 3
+
2
2 3 +
2
202500
15
3375
315 2

6514448
1652773
181199822513 1792
+ n f CF CF A
5 +
3
2
765450000
9
4725
3375

1408
11888 2
+
2 3
,
9
27 2
(3)
c2,5
= CF CF2 A
(A.2)
331
177036089007294328733 27248
65397081433
5 +
3
32934190464000000
63
2646000
1472 2 340303364748629
2563996

2 +
2 3
3 3
46675440000
315
4570738447 2 8992 3
2
330750
63 2

213694072871074531 1821772
438487320707
+ CF2 CF A
+
5
3
45177216000000
315
18522000
1456 2 418808510000479
2071492
3 +
2
2 3
+
3
46675440000
315
6241478743 2 56432 3
2
+
661500
315 2

38079608000704561 1792
22115039
5 +
3
+ n f CF CF A
117622108800000
9
13230

113587875043
1408
2296328 2
(A.3)
2 +
2 3
2 ,
166698000
9
4725

5676515460744370321603 25664
11165079556403
(3)
2
c2,9 = CF CF A
5 +
3
1000376035344000000
63
375070500
1472 2 8178803099431493
1648352

2 +
2 3
3 3
945177660000
189
23488033336 2 8992 3
2
1488375
63 2

32102287673972370020989 1162796
89153747611
+
5
3
+ CF2 CF A
6002256212064000000
189
3087000
1456 2 342078312478997
1332820
+
2
2 3
+
3 3
30005640000
189
3187232017 2 56432 3
2
+
297675
315 2

21832132134852204299 1792
6271692134
5 +
3
+ n f CF CF A
52400649470400000
9
3274425

1931824297943
1408
164116 2
(A.4)
2 +
2 3
.
2250423000
9
315 2
(3)
c2,7
= CF CF2 A
The corresponding lowest five odd-integer moments for the longitudinal structure function read

(3)
cL,1

21977 608
2648
3068
2
5
3
2 4482 3 3362
9
3
9
9

17819 1568
5648
1376
2304 2
2
5 +
3 +
2 + 2882 3 +
+ CF C F A
9
3
9
9
5 2

1366 496
328
224 2
3
2
,
(A.5)
+ n f CF CF A
9
9
9
15 2
= CF CF2 A
332

(3)
cL,3

12350749
52516
47
7544 2
+ 3525 +
3 + 2 + 962 3
19440
45
27
15 2

10152961
16412
1168 2
2
3685
3 2422 + 1442 3 +
+ CF CF A
12960
15
3 2

16757 2936
16
368 2
+
3 2
,
+ n f CF CF A
(A.6)
1620
45
9
15 2
= CF CF2 A

735306721 1888
558244
442783
(3)
5 +
3
2
cL,5 = CF CF2 A
17010000
3
315
675

4160 2
2
+ 4482 3
9

6741265367
736
1285168
51493
5
3 +
2
+ CF2 CF A
10206000
3
945
405

69608 2
2
+ 962 3 +
225

2107157 8816
11992
736 2
+
3
2
2 ,
+ n f CF CF A
255150
105
405
45
(A.7)
(3)
354522585410107
47266403
1095179473
14085 +
3
2
666792000000
23625
945000

147056 2
2
+ 7522 3
375

11388456807174161
176925641
4569363329
1845
3 +
2
+ CF2 CF A
28005264000000
132300
13230000

663878 2
2
+ 722 3 +
2625

369546282989 124282
220747
184 2
(A.8)
+
3
2
2 ,
+ n f CF CF A
50009400000
1575
5250
15
cL,7 = CF CF2 A
(3)
cL,9
1346454911003496947 10528
13247918
5 +
3
1323248724000000
5
6125

325373958827
5184
296736 2
2 +
2 3
208372500
5
875 2

17693872049573089 736
125991917
5
3
+ CF2 CF A
73513818000000
5
99225

10496201057
288
1688888 2
2 +
2 3 +
2
+
23152500
5
7875

23852323249607 1249264
1542176
736 2
+
3
2
2 . (A.9)
+ n f CF CF A
1444021425000
17325
33075
75
= CF CF2 A
333
(3)
Finally the analytic expressions for the moments c3,N in Eq. (3.3) are

650360
1472 2 712328
840949 9344
(3)
c3,2
= CF CF2 A
+
5
3 +
+
2
243
9
81
3 3
243

47920
30416 2 8992 3
2 3 +
2 +
2
9
5
63

580504
1456 2 742390
38416
15979879
+ CF2 CF A
5 +
3

2
4374
9
81
3 3
243

577264 2 56432 3
+ 43682 3
+
135 2
315 2

57128
46112
1408
119522 1792
+ nf CF CF A
+
5
3 +
2
2 3
729
9
81
243
9

40024 2
+
,
135 2

292322783
1472 2
21230721185377 23704
(3)
c3,4 = CF CF2 A
+
5
3 +
4374000000
45
20250
3 3

5644168873
100792
6477802 2 8992 3
+
2
2 3 +
2 +
2
1215000
15
675
63

272933467
1456 2
19991706724601
+ CF2 CF A
52085 +
3
4374000000
20250
3 3

6307524619
253064
2500616 2 56432 3
2 +
2 3
2 +
1215000
45
375
315 2

755894
21942049
15339664501 1792
+ n f CF CF A
+
5
3 +
2
72900000
9
675
60750

1408
53128 2
2 3 +
,
9
135 2
(A.10)
(A.11)

3380925064
1472 2
172761364527374293 21200
(3)
c3,6
= CF CF2 A
+
5
3 +
32162295375000
63
165375
3 3

147865501939
2395856
75351016 2 8992 3
+
2
2 3 +
2 +
2
24310125
315
6125
63

67828543996
1456 2
1810712
313157547783370669
+ CF2 CF A
5 +
3
64324590750000
315
3472875
3 3

3604225183081
667864
1397140016 2 56432 3
2 +
2 3
2 +
486202500
105
165375
315 2

16004944
23420609
503591542653161 1792
+ n f CF CF A
+
5
3 +
2
1837845450000
9
11025
42875

1408
2135888 2
(A.12)
2 3 +
,
9
4725 2
334
45882775286477927067311 22640
38829577931303
+
5
3
8003008282752000000
63
1500282000
1472 2 5677110453154657
7858468
+
+
2
2 3
3 3
756142128000
945
43146817871 2 8992 3
2 +
+
2976750
63 2

126830527574348410837327 5795836
4175977929883
+ CF2 CF A
5 +
3
24009024848256000000
945
166698000
1456 2 194854342276579
6509876
3
2 +
2 3
3
20003760000
945
58805551031 2 56432 3
2 +
5953500
315 2

1026540911
3380190329263337489 1792
+
5
3
+ n f CF CF A
9527390812800000
9
595350

3263620615369
1408
778496 2
+
(A.13)
2
2 3 +
2 ,
4500846000
9
1575

2924815993615556996346598663 210944
(3)
+
5
c3,10 = CF CF2 A
483334682604559632000000
385
3080312718428437
1472 2 13720175530646448109
3 +
+
2
99843767100
3 3
1537594013340000

8455904
3568998808 2 8992 3
2 3 +
2 +
945
218295
63 2

61916581373996975119251441821 66873844
5
+ CF2 CF A
10633363017300311904000000
10395
30055925797598243
1456 2 3186598606475201011
3

2
+
998437671000
3 3
263587545144000

75900812
1995571648453 2 56432 3
2 3
2 +
+
10395
180093375
315 2

339629926756418877268603 1792
70469642338
+
5
3
+ n f CF CF A
767197908896126400000
9
36018675

2677118231310293
1408
1015276 2
(A.14)
2
2 3 +
.
+
2995313013000
9
1925 2
(3)
c3,8
= CF CF2 A
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
W.M. Yao, et al., Particle Data Group, J. Phys. G 33 (2006) 1.

U.K. Yang, et al., CCFR/NuTeV Collaboration, Phys. Rev. Lett. 86 (2001) 2742, hep-ex/0009041.
M. Tzanov, et al., NuTeV Collaboration, Phys. Rev. D 74 (2006) 012008, hep-ex/0509010.
G. Onengut, et al., CHORUS Collaboration, Phys. Lett. B 632 (2006) 65.
C. Adloff, et al., H1 Collaboration, Eur. Phys. J. C 30 (2003) 1, hep-ex/0304003.
S. Chekanov, et al., ZEUS Collaboration, Eur. Phys. J. C 32 (2003) 1, hep-ex/0307043.
A. Aktas, et al., H1 Collaboration, Phys. Lett. B 634 (2006) 173, hep-ex/0512060.
S. Chekanov, et al., ZEUS Collaboration, Phys. Lett. B 637 (2006) 210, hep-ex/0602026.
M.L. Mangano, et al., hep-ph/0105155.
[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]
J.B. Dainton, et al., J. Instrum. (JINST) 1 (2006) P10001, hep-ex/0603016.

G.P. Zeller, et al., NuTeV Collaboration, Phys. Rev. Lett. 88 (2002) 091802, hep-ex/0110059.
A. Aktas, et al., H1 Collaboration, Phys. Lett. B 632 (2006) 35, hep-ex/0507080.
S. Davidson, et al., JHEP 0202 (2002) 037, hep-ph/0112302.
K.S. McFarland, S.O. Moch, hep-ph/0306052.
B.A. Dobrescu, R.K. Ellis, Phys. Rev. D 69 (2004) 114014, hep-ph/0310154.
S. Kretzer, et al., Phys. Rev. Lett. 93 (2004) 041802, hep-ph/0312322.
S. Moch, J.A.M. Vermaseren, A. Vogt, Nucl. Phys. B 688 (2004) 101, hep-ph/0403192.
A. Vogt, S. Moch, J.A.M. Vermaseren, Nucl. Phys. B 691 (2004) 129, hep-ph/0404111.
J.S. Guillen, et al., Nucl. Phys. B 353 (1991) 337.
W.L. van Neerven, E.B. Zijlstra, Phys. Lett. B 272 (1991) 127.
E.B. Zijlstra, W.L. van Neerven, Phys. Lett. B 273 (1991) 476.
E.B. Zijlstra, W.L. van Neerven, Phys. Lett. B 297 (1992) 377.
S. Moch, J.A.M. Vermaseren, Nucl. Phys. B 573 (2000) 853, hep-ph/9912355.
S. Moch, J.A.M. Vermaseren, A. Vogt, Phys. Lett. B 606 (2005) 123, hep-ph/0411112.
J.A.M. Vermaseren, A. Vogt, S. Moch, Nucl. Phys. B 724 (2005) 3, hep-ph/0504242.
A.J. Buras, Rev. Mod. Phys. 52 (1980) 199, and references therein.
A. Vogt, S. Moch, J. Vermaseren, Nucl. Phys. B (Proc. Suppl.) 160 (2006) 44, hep-ph/0608307.
S. Moch, M. Rogal, Nucl. Phys. B 782 (2007) 51, arXiv: 0704.1740 [hep-ph].
S.A. Larin, T. van Ritbergen, J.A.M. Vermaseren, Nucl. Phys. B 427 (1994) 41.
S.A. Larin, P. Nogueira, T. van Ritbergen, J. Vermaseren, Nucl. Phys. B 492 (1997) 338, hep-ph/9605317.
A. Retey, J.A.M. Vermaseren, Nucl. Phys. B 604 (2001) 281, hep-ph/0007294.
S.G. Gorishnii, S.A. Larin, L.R. Surguladze, F.V. Tkachov, Comput. Phys. Commun. 55 (1989) 381.
S.A. Larin, F.V. Tkachev, J.A.M. Vermaseren, NIKHEF-H-91-18.
D.J. Broadhurst, A.L. Kataev, C.J. Maxwell, Phys. Lett. B 590 (2004) 76, hep-ph/0403037.
W.L. van Neerven, A. Vogt, Nucl. Phys. B 603 (2001) 42, hep-ph/0103123.
E.A. Paschos, L. Wolfenstein, Phys. Rev. D 7 (1973) 91.
E. Remiddi, J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725, hep-ph/9905237.
T. Gehrmann, E. Remiddi, Comput. Phys. Commun. 141 (2001) 296, hep-ph/0107173.
D.I. Kazakov, et al., Phys. Rev. Lett. 65 (1990) 1535.
W.L. van Neerven, A. Vogt, Nucl. Phys. B 568 (2000) 263, hep-ph/9907472.
A. Vogt, Comput. Phys. Commun. 170 (2005) 65, hep-ph/0408244.
J.A.M. Vermaseren, math-ph/0010025.
A.L. Kataev, arXiv: 0707.2855 [hep-ph].
Y.L. Dokshitzer, G. Marchesini, B.R. Webber, Nucl. Phys. B 469 (1996) 93, hep-ph/9512336.
S. Moch, J.A.M. Vermaseren, A. Vogt, Nucl. Phys. B 726 (2005) 317, hep-ph/0506288.
K.P.O. Diener, S. Dittmaier, W. Hollik, Phys. Rev. D 72 (2005) 093002, hep-ph/0509084.
S. Catani, D. de Florian, G. Rodrigo, W. Vogelsang, Phys. Rev. Lett. 93 (2004) 152003, hep-ph/0404240.
H.L. Lai, et al., JHEP 0704 (2007) 089, hep-ph/0702268.
R.S. Thorne, A.D. Martin, W.J. Stirling, G. Watt, arXiv: 0706.0456 [hep-ph].
M. Rogal, in: Proceedings of the 2007 Europhysics Conference on High Energy Physics, Manchester, UK.
335
Erratum
Erratum to: Thermal production of gravitinos

[Nucl. Phys. B 606 (2001) 518544]
M. Bolz, A. Brandenburg, W. Buchmller
Deutsches Elektronen-Synchrotron DESY, Hamburg, Germany
Received 21 September 2007
As recently pointed out by Pradler and Steffen [1], Eq. (C.14) has to be replaced by

t
4
.
IBFB = 2 ln(2)T 3 (N + nf ) 64 3 nf eE/T + 1 IBFB
3
(E.1)
The difference between Eqs. (C.14) and (E.1) is a non-vanishing surface term in the BFB case,
which in the partial integration leading from (B.24) to (B.25) was overlooked.
As a consequence, the numerical factor multiplying (N + nf ) changes in Eq. (C.16) from
1.7014 to 1.8782. Correspondingly, also the numerical factor multiplying (N + nf ) in Eq. (44),
and the overall factors in Eqs. (47) and (48) change. The corrected equations read
soft

d 3p
hard
n
(E)
(E)
+
(E)
F
G
G
(2)3

m2g 3 (3)g 2 (N 2 1)T 6
= 1+
32 3 M 2
3m2
CG (T ) =
2

T
ln
)
+
0.5781n
+
0.4992
(N
+
n
f
f ,
m2g

100 GeV 2 mg () 2
TR
,
YG = 1.4 1010
mG
1 TeV
1010 GeV
DOI of original article: 10.1016/S0550-3213(01)00132-8.
E-mail address: buchmuwi@mail.desy.de (W. Buchmller).

doi:10.1016/j.nuclphysb.2007.09.020
(E.2)
(E.3)
M. Bolz et al. / Nuclear Physics B 790 (2008) 336337
G h2 = mG YG (T )nrad (T )h2 c1

mg () 2
TR
100 GeV
= 0.27
.
mG
1 TeV
1010 GeV
337
(E.4)
Compared to Eqs.(47) and (48), the gravitino abundance is

increased by 30%. (Note that also
the prefactor 1/(2 6 ) in Eq. (33) has to be replaced by 1/(4 6 ).)
Acknowledgements
We thank Josef Pradler and Frank Steffen for informing us about the error in our paper.
Reference
[1] J. Pradler, F.D. Steffen, Phys. Rev. D 75 (2007) 023509, hep-ph/0608344.
Erratum
Erratum to: Spectra of neutrinos

from dark matter annihilations
[Nucl. Phys. B 727 (2005) 99138]
Marco Cirelli a, , Nicolao Fornengo b , Teresa Montaruli c,1 ,
Igor Sokalski d , Alessandro Strumia e , Francesco Vissani f
a Physics Department, Yale University, New Haven, CT 06520, USA
b Dipartimento di Fisica Teorica, Universit di Torino and INFN, Sezione di Torino,
via P. Giuria 1, I-10125 Torino, Italy

c University of Wisconsin, Chamberlin Hall, Madison, WI 53706, USA
d INFN, Sezione di Bari, via Amendola 173, I-70126 Bari, Italy
e Dipartimento di Fisica dellUniversit di Pisa and INFN, Italy
f INFN, Laboratori Nazionali del Gran Sasso, Assergi (AQ), Italy
Received 27 September 2007

Available online 5 October 2007
In our original calculations, an erroneous double counting of the prompt neutrino yield in
W -boson decays and a numerical bug in the implementation of the boost for top quark decays
were present. In this Erratum, we present the corrected results and we also update some parameters employed in the computations. The modifications due to the erroneous double counting and
the numerical bug only affect the W + W and t t channels. The update of the parameters leads
to very minor changes. The new figures and tables [Figs. 2, 512, Tables 3 and 4] (replacing the
corresponding figures and tables in the paper) are shown below.
All physics discussions and conclusions are unchanged. These corrected and updated results
are available in electronic form from [1] (Release 3).
DOI of article: 10.1016/j.nuclphysb.2005.08.017.

E-mail address: marco.cirelli@cea.fr (M. Cirelli).

1 On leave of absence from Universit di Bari and INFN, Sezione di Bari, via Amendola 173, I-70126 Bari, Italy.
doi:10.1016/j.nuclphysb.2007.10.001
M. Cirelli et al. / Nuclear Physics B 790 (2008) 338344
339
Fig. 2. Neutrino spectra at production. Upper half: the fluxes of electron and muon neutrinos, for the seven main annihilation channels and for different masses of the parent DM particle (different colors). The solid lines apply to the case
of the Sun, the dotted of the Earth. In all cases, the spectra of antineutrinos are the same as those of neutrinos. Lower
()
half: the same for . (For interpretation of the references to color in this figure legend, the reader is referred to the web
version of this article.)
340
Fig. 5. Neutrino spectra generated by one DM annihilation around the center of Earth. The plots show the spectra of
the three neutrino flavors (the three rows) and assume different DM masses (the three columns). Each plot shows the
+ , cc,
t t, W + W , ZZ. The DM DM channel (not shown) would
open annihilation channels DM DM bb,
produce a line at E = mDM . The dotted lines show the spectra without oscillations while solid lines are the final results
after oscillations. The dashed lines in the upper-left panel have been computed with 13 = 0.1 rad for illustration; all the
other results assume 13 = 0. Neutrino and anti-neutrino spectra are roughly equal: we here show (2 + )/3, in view
of (N ) 2 ( N ). The shaded region is the atmospheric background.
341
Fig. 6. Spectra of through-going + and (summed) generated by DM annihilations around the center of Earth.
The plots assume the DM annihilation rate of Eq. (21), different DM masses and show the main annihilation channels.
For better illustration, some channels have been rescaled by the indicated factor. The shaded region is the atmospheric
background.
Fig. 7. Spectra of fully contained + and (summed) generated by DM annihilations around the center of Earth.
We assume the DM annihilation rate of Eq. (21) and a detector with Mton year exposure. The channel gives a
spectrum peaked at E mDM , due to the monochromatic parent spectrum.
Fig. 8. (Idealized) energy spectra of showers generated by DM annihilations around the center of Earth, in a detector
with Mton year exposure and unit detector efficiency, assuming the DM annihilation rate of Eq. (21).
342
Fig. 9. Neutrino fluxes generated by DM annihilations around the center of Sun. Upper row: e fluxes. Lower row:
the almost equal fluxes of and . We plot the combination (2 + )/3. All spectra are significantly different
from those at production point (not shown here). For the DM DM channel at E = mDM we plotted the survival
probability rather than the energy spectrum, because it is not possible to plot a Dirac function.
Fig. 10. Spectra of through-going + and (summed) generated by DM annihilations around the center of Sun. The
plots assume the DM annihilation rate of Eq. (24), different DM masses and show the main annihilation channels. Some
channels have been rescaled by the indicated factor for better illustration.
343
Fig. 11. Spectra of fully contained + and (summed) generated by DM annihilations around the center of Sun. We
assume the DM annihilation rate of Eq. (24) and a detector with Mton year exposure.
Fig. 12. (Idealized) energy spectra of showers generated by DM annihilations around the center of Sun, in a detector with
Mton year exposure and unit efficiency, assuming the DM annihilation rate of Eq. (24).
Table 3
Ratios of through-going muon rates with over without the effects of the neutrino propagation, for DM annihilations
around the center of the Earth/Sun. E.g., the bottom-right entry means that, for mDM = 1000 GeV, the rate is unaffected
if DM DM W + W annihilations occur in the Earth, and the rate gets reduced to 0.04 of its value if annihilations
occur in the Sun
DM mass
mDM
DM annihilation channels in the Earth/Sun

bb

cc
q q
t t
ZZ
W +W
50 GeV
100 GeV
200 GeV
400 GeV
1000 GeV
1/0.71
1/0.50
1/0.25
1/0.07
1/0.01
0.48/0.68
0.45/0.65
0.55/0.58
0.77/0.50
0.93/0.40
/
/
1.0/0.59
1.0/0.26
1.0/0.09
/
1.0/0.71
1.0/0.39
1.0/0.15
1.0/0.04
/
1.1/0.71
1.0/0.41
1.0/0.16
1.0/0.04
0.50/0.68
0.70/0.62
0.86/0.53
0.95/0.41
0.99/0.28
3.9/3.0
2.0/2.5
1.3/1.7
1.1/0.74
1.0/0.19
0.32/0.61
0.49/0.56
0.75/0.49
0.91/0.40
0.98/0.31
344
Table 4
()
Average percentage energies in units of mDM of produced by DM annihilations around the center of the Earth/Sun,
computed for various annihilation channels and for various DM masses. E.g., the bottom-right entry means that
()
DM DM W + W annihilations with mDM = 1000 GeV produce with average energy equal to 36% mDM =
360 GeV if occurring in the Earth and to 7% mDM if in the Sun
DM mass
mDM
DM annihilation channels in the Earth/Sun

bb

cc
q q
t t
ZZ
W +W
50 GeV
100 GeV
200 GeV
400 GeV
1000 GeV
100/89
100/78
100/57
100/28
100/7.8
5.7/4.8
3.2/3.8
2.8/2.8
2.2/2.0
3.9/1.2
/
/
15/12
15/8.4
15/4.6
/
33/29
33/21
33/14
33/6.1
/
34/31
34/23
35/16
36/7.0
11/11
11/9.0
12/7.2
11/5.4
10/2.7
32/33
25/30
22/25
22/18
24/8.4
12/11
11/8.7
12/6.7
12/4.9
9.9/2.8
Acknowledgements
We thank Joakim Edsjo, Tommy Ohlsson and Mattias Blennow for cross-checking the results
of their novel calculation with ours, from which the two errors corrected in this Erratum were
found.
References
[1] Web pages: http://www.to.infn.it/~fornengo/DMnu.html;
http://www.cern.ch/astrumia/DMnu.html;
http://www.marcocirelli.net/DMnu.html.
Nuclear Physics B 790 [PM] (2008) 345413
Supersymmetric Bethe ansatz and Baxter equations

from discrete Hirota dynamics
Vladimir Kazakov a,1 , Alexander Sorin b , Anton Zabrodin c,d
a Laboratoire de Physique Thorique, de lEcole Normale Suprieure et lUniversit Paris-VI,
24 rue Lhomond, Paris Cedex 75231, France

b Bogoliubov Laboratory of Theoretical Physics, Joint Institute for Nuclear Research,
141980 Dubna, Moscow region, Russia

c Institute of Biochemical Physics, Kosygina str. 4, 119991 Moscow, Russia
d Institute of Theoretical and Experimental Physics, Bol. Cheremushkinskaya str. 25, 117259 Moscow, Russia
Received 12 May 2007; accepted 29 June 2007

Available online 7 August 2007
Abstract
We show that eigenvalues of the family of Baxter Q-operators for supersymmetric integrable spin chains
constructed with the gl(K|M)-invariant R-matrix obey the Hirota bilinear difference equation. The nested
Bethe ansatz for super-spin chains, with any choice of simple root system, is then treated as a discrete
dynamical system for zeros of polynomial solutions to the Hirota equation. Our basic tool is a chain of
Bcklund transformations for the Hirota equation connecting quantum transfer matrices. This approach
also provides a systematic way to derive the complete set of generalized Baxter equations for super-spin
chains.
2007 Published by Elsevier B.V.
1. Introduction
1.1. Motivation and background
Supersymmetric extensions of quantum integrable spin chains were proposed long ago [1,2]
but the proper generalization of the standard methods such as algebraic Bethe ansatz and Baxter
E-mail addresses: kazakov@physique.ens.fr (V. Kazakov), sorin@theor.jinr.ru (A. Sorin), zabrodin@itep.ru
(A. Zabrodin).
1 Membre de lInstitut Universitaire de France.
0550-3213/$ see front matter 2007 Published by Elsevier B.V.
doi:10.1016/j.nuclphysb.2007.06.025
346
V. Kazakov et al. / Nuclear Physics B 790 [PM] (2008) 345413
T Q-relations is still not so well understood, as compared to the case of integrable models with
usual symmetry algebras, and still contains some elements of guesswork.
Bethe ansatz equations for integrable models based on super-algebras are believed to be written according to the general empirical rules [3] applied to graded Lie algebras. Accepting this
as a departure point, one can try to reconstruct other common ingredients of the theory of quantum integrable systems such as Baxter relations and fusion rules. For the super-spin chains based
on the rational or trigonometric R-matrices, the algebraic Bethe ansatz works rather similarly
to the case of usual bosonic spin chains, but the Bethe ansatz equations for a given model do
not have a unique form and depend on the choice of the system of simple roots. The situation
becomes even more complicated when one considers spins in higher representations of the superalgebra, and especially typical ones containing continuous KacDynkin labels. The systematic
description of all possible Bethe ansatz equations and T Q-relations becomes then a cumbersome
task [46] not completely fulfilled in the literature. To our knowledge, a unified approach is still
missing.
In this article we propose a new approach to the supersymmetric spin chains based on the
Hirota-type relations for quantum transfer matrices and Bcklund transformations for them.
Functional relations between commuting quantum transfer matrices are known to be a powerful tool for solving quantum integrable models. They are based on the fusion rules for various
irreducible representations in the auxiliary space of the model. First examples were given in [7,8].
Later, these functional relations were represented in the determinant form [9] and in the form of
the Hirota bilinear difference equation [10,11]. In [1215] it was shown that transfer matrices
in the supersymmetric case are subject to exactly the same functional equations as in the purely
bosonic case. The Hirota form of the functional equations has been proved to be especially useful
and meaningful [16,17]. It is the starting point for our construction.
The Hirota equation [18] is probably the most famous equation in the theory of classical integrable systems on the lattice. It provides a universal integrable discretization of various soliton
equations and, at the same time, serves as a generating equation for their hierarchies. In this
sense, it is a kind of a Master equation for the theory of solitons. It covers a great variety of
integrable problems, classical and quantum.
In our approach, quantization and discretization appear to be closely related in the sense that
solutions to the quantum problems are given in terms of the discrete classical dynamics. What
specifies the problem are the boundary and analytic conditions for the variables entering the
Hirota equation. In applications to quantum spin chains, these variables are parameters of the
representation of the symmetry algebra in the auxiliary space. For representations associated
with rectangular Young diagrams they are height and length of the diagram denoted by a and s,
respectively. In the case of spin chains based on usual (bosonic) algebras gl(K), the boundary
conditions are such that the non-vanishing transfer matrices live in a strip 0 a K in the (a, s)
plane [16,17]. In the case of super-algebras of the type gl(K|M), the strip turns into a domain of
the fat hook type presented in Fig. 1.
In the nested Bethe ansatz scheme for gl(K), one successively lowers the rank of the algebra
gl(K) gl(K 1) (and thus the width of the strip) until the problem gets fully undressed.
This purely quantum technique has a remarkable classical face: it is equivalent to a chain
of Bcklund transformations for the Hirota equation [16,17]. They stem from the discrete zero
curvature representation and associated systems of auxiliary linear equations. This solves the
discrete Hirota dynamics in terms of Bethe equations or the general system of Baxters T Qrelations.
347
Fig. 1. The domain (fat hook) of non-vanishing transfer matrices T (a, s, u) for the supersymmetric spin chain with the
gl(K|M) symmetry.
The aim of this article is to extend this program to the models based on super-algebras
gl(K|M). In this case, there exist two different types of Bcklund transformations. One of them
lowers K while the other one lowers M. The undressing goes until the fat hook is collapsed to
K = M = 0. The result of this procedure does not depend on the order in which we perform these
transformations but the form of the equations does. Different orders lead to different types of
Bethe ansatz equations and Baxters T Q-relations associated with Cartan matrices for different
systems of simple roots. In this way, the abundance of various Bethe equations and T Q-relations
is easily explained and classified. All of them are constructed in our paper. Instead of K + 1
Baxters Q-functions for the bosonic gl(K) algebra we recover (K + 1)(M + 1) Q-functions
(some of them are initially fixed by the physical problem). More than that, we establish a new
equation relating all these Q-functions which is again of the Hirota type. This QQ-relation
opens the most direct and easiest way to construct various systems of Bethe equations. Similar
relations hold for the transfer matrices at each step of the undressing procedure.
Our construction goes through when observables in the Hilbert space of the generalized spin
chain are in arbitrary finite dimensional representations of the symmetry (super-)algebra. We can
also successfully incorporate the case of typical representations carrying the continuous Kac
Dynkin labels, as it is illustrated by examples of super-algebras gl(1|1) and gl(2|1).
Some standard facts and notation related to super-algebras and their representations are listed
in Appendix A. For details see [1922]. Throughout the paper, we use the language of the algebraic Bethe ansatz and the quantum inverse scattering method on the lattice developed in [23]
(see also reviews [1,24] and book [25]). On the other hand, we employ standard methods of
classical theory of solitons [26] and discrete integrable equations [2731].
1.2. A sketch of the results
We consider integrable generalized spin chains with gl(K|M)-invariant R-matrix. The generating function of commuting integrals of motion is the quantum transfer matrix T () (u),
which depends on the spectral parameter u C and the Young diagram . It is obtained as
the (super)trace of the quantum monodromy matrix T () (u) in the auxiliary space carrying the
irreducible representation of the symmetry algebra labeled by :
T () (u) = straux T () (u).
We deal with covariant tensor irreducible representations (irreps) of the super-algebra.
348
The transfer matrices for different are known to be functionally dependent. For rectangular
diagrams = (s a ) with a rows and s columns the functional relation takes the form of the famous
Hirota difference equation:
T (a, s, u + 1)T (a, s, u 1) T (a, s + 1, u)T (a, s 1, u)
= T (a + 1, s, u)T (a 1, s, u),
(1.1)
where the transfer matrix for rectangular diagrams, after some a, s-dependent shift of the spectral
parameter u, is denoted by T (a, s, u). We see that it enters the equation as the -function [29].
Since all the T s commute, the same relation holds for their eigenvalues. We use the normalization in which all non-vanishing T s are polynomials in u of degree N , where N is the number of
sites in the spin chain.
While the functional relation is the same for ordinary and super-algebras, the boundary conditions are different. In the gl(K)-case, the domain of non-vanishing T s in the (s, a) plane is
the half-strip s 0, 0 a K. In the gl(K|M)-case, the domain of non-vanishing T s has the
form of a fat hook. It is shown in Fig. 1. To ensure compatibility with the Hirota equation, the
boundary values at s = 0 and a = 0 should be rather special. In our normalization,
T (0, s, u) = (u s),
T (a, 0, u) = (u + a),
(1.2)

where (u) = i (u i ) is a fixed polynomial of degree N which characterizes the spin chain.
Our goal is to solve the Hirota equation, with the boundary conditions given above, using the
classical methods of the theory of solitons. This program for the case of ordinary Lie algebras
(and for models with elliptic R-matrices) was realized in [16]. In this paper, we extend it to the
case of super-algebras (for models with rational R-matrices).
One of the key features of soliton equations is the existence of (auto) Bcklund transformations (BT), i.e., transformations that send any solution of a soliton equation to another solution
of the same equation. A systematic way to construct such transformations is provided by considering an over-determined system of linear problems (called auxiliary linear problems) whose
compatibility condition is the non-linear equation at hand. We introduce two Bcklund transformations for the Hirota equation. They send any solution with the boundary conditions described
above to a solution of the same Hirota equation with boundary conditions of the same fat hook
type but with different K or M. Specifically, one of them lowers K by 1 and the other one
lowers M by 1 leaving all other boundaries intact. We call these transformations BT1 and BT2.
Applying them successively K + M times, one comes to a collapsed domain which is a union
of two lines, the s-axis and the a-semi-axis shown in Fig. 1, meaning that the original problem
gets undressed to a trivial one. This procedure appears to be equivalent to the nested Bethe
ansatz. Different orders in which we diminish K and M give raise to different dual systems
of nested Bethe ansatz equations. All of them describe the same system. They correspond to
different choices of the basis of simple roots.
Let k, m be indices running from 0 to K and from 0 to M respectively, and let Tk,m (a, s, u)
be the transfer matrices obeying the Hirota equation with the boundary conditions as above but
with K, M replaced by k, m. They are obtained from TK,M (a, s, u) by a chain of BTs. Namely,
Tk1,m s and Tk,m1 s are solutions to the auxiliary linear problems for the Hirota equation for
349
Tk,m s:
Tk,m1
BT2
Tk,m
BT1
Tk1,m
The explicit formulas of these transformations are given below (Eqs. (3.29) and (3.30)). The
functions Tk,m (a, s, u) are polynomials in u for any a, s, k, m but the degree depends only on
k, m. Let Qk,m (u) be the boundary values of the Tk,m (a, s, u), i.e.,
Tk,m (0, s, u) = Qk,m (u s),
Tk,m (a, 0, u) = Qk,m (u + a).
However, they are not fixed for the values of k, m other than k = K, m = M (when QK,M (u) =
(u)) and k = m = 0 (when Q0,0 (u) is put equal to 1) but are to be determined from a solution
to the hierarchy of Hirota equations. In fact these Qs are Baxter polynomial functions whose
roots obey the Bethe equations.
The result of a successive application of the transformations BT1 and BT2 does not depend
on their order. This fact can be reformulated as a discrete zero curvature condition
1
1
U k,m+1
Vk,m = Vk+1,m U k,m
(1.3)
for the shift operators in k and m:

Qk+1,m (u)Qk,m (u + 2)
e2u ,
U k,m (u) =
Qk+1,m (u + 2)Qk,m (u)
Qk,m (u)Qk,m+1 (u + 2)
Vk,m (u) =
e2u .
Qk,m (u + 2)Qk,m+1 (u)
(1.4)
Relation (1.3) is equivalent to the following Hirota equation for the Baxter Q-functions:
Qk,m (u)Qk+1,m+1 (u + 2) Qk+1,m+1 (u)Qk,m (u + 2)
= Qk,m+1 (u)Qk+1,m (u + 2).
(1.5)
This equation represents our principal new result. By analogy with Baxters T Q-relations, we
call Eq. (1.5) the QQ-relation. It provides the most transparent way to derive different systems
of Bethe equations for the generalized spin chain and duality transformations between them.
We also show that a number of more general Hirota equations of the similar type (i.e., acting
in the space spanned by k, m and a particular linear combination of a, s, u) hold for the full set
of functions Tk,m (a, s, u). They lead to a system of algebraic equations for their roots which
generalizes the system of Bethe equations.
The transfer matrices can be expressed through the Q-functions via generalized Baxters T Qrelations. A simple way to represent them is to consider the (non-commutative) generating series
of the transfer matrices for one-row diagrams,
W k,m (u) = Q1
k,m (u)
Tk,m (1, s, u + s + 1)e2su ,
(1.6)
s=0
where the factor in front of the sum is put here for the proper normalization. The operator W
is similar to the wave (or dressing) operator in the Toda lattice theory. We prove the following
350
fundamental operator relations:

1
W k+1,m = U k,m
Wk,m ,
W k,m+1 = Vk,m W k,m ,
(1.7)
which implement the Bcklund transformations on the level of the generating series. These relations allow one to represent W K,M (the quantity of prime interest) as an ordered product of
1
the operators U k,m
and Vk,m along a zigzag path from the point (0, 0) to the point (K, M) on
the (k, m) lattice. This representation provides a concise form of the generalized Baxter relations. Different zigzag paths correspond to different choices of the basis of simple roots, i.e., to
different dual forms of supersymmetric Bethe equations (see Fig. 17).
The normalization (1.2), with the roots i of the polynomial (u) being in general position,
implies that the spins of the inhomogeneous spin chain belong to the vector representation
of the algebra gl(K|M). Higher representations can be obtained by fusing of several copies of
the vector ones. According to the fusion procedure, the corresponding s must be chosen in a
specific string-like way which means that the differences between them are even integers constrained also by some more specific requirements. This amounts to the fact that the Q-functions
become divisible by certain polynomials with explicitly prescribed roots located according to a
similar string-like pattern. If one redefines the Q-functions dividing them by these normalization
factors, then the QQ-relation gets modified and leads, in the same way as before, to the systems
of Bethe equations with non-trivial right-hand sides (sometimes called vacuum parts).
In Section 7, we specify our construction for two popular examples, the gl(1|1) and gl(2|1)
super-algebras. We present the T Q- and QQ-relations for all possible types of Q-functions
Qkm (u) (0 < k K, 0 < m M) as well as Bethe equations for all types of simple roots systems
and representations of the super-algebra (including typical representations with a continuous
component of the KacDynkin label) and compare them with the results existing in the literature.
In our formalism, the construction becomes rather transparent and algorithmic. In this respect,
our method is an interesting alternative to the algebraic Bethe ansatz approach.
2. Fusion relations for transfer matrices and Hirota equation
2.1. Quantum transfer matrices
Let V = CK CM CK|M be the graded linear space of the vector representation of the
super-algebra gl(K|M). The fundamental gl(K|M)-invariant rational R-matrix acts in V V
and has the form [1,2]
R(u) = uI + 2 = uI + 2
K+M
(1)p( ) E E .
(2.1)
, =1
Here I is the identity operator and is the super-permutation, i.e., the operator such that
(x y) = (1)p(x)p(y) y x for any homogeneous vectors x, y V , and E are the generators of the (super-)algebra. In components, they read
I = ,
= (1)p()p() ,
(E ) = .
Note that = I , as in the ordinary case. The notation p(x) is used to denote parity of the
object x (see Appendix A). The R(u) has only even matrix elements. The complex variable u is
351
Fig. 2. The R-matrix.
Fig. 3. The YangBaxter equation.
called the spectral parameter. The R-matrix obeys the graded YangBaxter equation (Fig. 3):

(1)p( ) p( )+p( ) R (u v)R (u)R (v)

(1)p(
)(p( )+p( ))

R (v)R (u)R (u v).
(2.2)
This is the key relation to construct a family of commuting operators, as is outlined below.
The R-matrix is an operator in the tensor product of two linear spaces (not necessarily isomorphic). It is customary to call one of them quantum and the other one auxiliary space. In
Fig. 2 they are associated with the vertical and the horizontal lines respectively. Let us fix a set
of N complex numbers i and consider a chain of N fundamental R-matrices R(u i ) with the
common auxiliary space V . Multiplying them as linear operators in V along the chain, we get an
operator in V N which is called quantum monodromy matrix. In components, it can be written
as
N
,{ }
1 2
TN0,{ii } (u) =
RNN1
N (u N ) R2 2 (u 2 )
1 ,...,N1

N
i1
i=2 p(i )+p(i )
j =1 p(j )
R1011 (u 1 )(1)
(2.3)
It is usually regarded as an operator-valued matrix in the auxiliary space (with indices 0 ,

N ), with the matrix elements being operators in the quantum space V N (see Fig. 4). Setting
0 = N = and summing with the sign factor (1)p( ) , we get the supertrace of the quantum
352
Fig. 4. The monodromy matrix.
monodromy matrix in the auxiliary space,

T (u) = str T (u) =
(1)p( ) T (u)
(2.4)
which is an operator in the quantum space (indices of the quantum space are omitted). It is called
the quantum transfer matrix. Using the graded YangBaxter equation, it can be shown that the
T (u) commute at different u: [T (u), T (u )] = 0. Their diagonalization is the subject of one or
another version of Bethe ansatz.
The (inhomogeneous) integrable spin chain or a vertex model on the square lattice is characterized by the symmetry algebra gl(K|M) and the parameters i (inhomogeneities at the sites
or rapidities). Given such a model, the family of transfer matrices T (u) (2.4) can be included
into a larger family of commuting operators. It is constructed by means of the fusion procedure.
2.2. Fusion procedure
Using the R-matrix (2.1) as a building block, it is possible to generate other gl(K|M)invariant solutions to the (graded) YangBaxter equation. They are operators in tensor products of two arbitrary finite dimensional irreps of the symmetry algebra. We consider covariant
tensor irreps which are labeled (although in a non-unique way, see Appendix A) by Young diagrams . Let V be the corresponding representation space, then the R-matrix acts in V V .
The construction of this R-matrix from the elementary one is referred to as fusion procedure
[7,3236].
Here we consider fusion in the auxiliary space, which allows one to construct, by taking
(super)trace in this space, a large family of transfer matrices commuting with T (u). In words,
the fusion procedure consists in tensor multiplying several copies of the fundamental R-matrix
(2.1) in the auxiliary space and subsequent projection onto an irrep of the gl(K|M) algebra.
As a result, one obtains an R-matrix R () (u) whose quantum space is still V (the space of the
vector representation of gl(K|M)) and whose auxiliary space is the space V of a higher irrep
of this algebra. The fact that the R-matrix obtained in this way obeys the (graded) YangBaxter
equation follows from the observation [7,34] that the projectors onto higher irreps can be realized
as products of the fundamental R-matrices (2.1) taken at special values of the spectral parameter.
These values correspond to the degeneration points of the R-matrices.
In particular, the R-matrix (2.1) has two degeneration points u = 2. Indeed, one can easily
see that

1
d = (K + M)2 (K M) ,
det R(u) = (u + 2)d+ (u 2)d ,
(2.5)
2
353
Fig. 5. The Young diagram decorated by spectral parameters.
R(2) = 2(I ) = 4P ,
(2.6)
where P are projectors onto the symmetric and antisymmetric subspaces in V V . The dimensions of these subspaces are equal to d , and the projectors are complimentary, i.e., P+ P =
P P+ = 0. Therefore, to get the R-matrix with the auxiliary space carrying the = (21 ) irrep
(respectively, the = (12 ) irrep), one should take the tensor product R(u + 2) R(u) (respectively, R(u 2) R(u)) in the auxiliary space and apply the projector P+ (respectively, P ):

1
2
R (1 ) (u) = P R(u 2) R(u) P .
R (2 ) (u) = P+ R(u + 2) R(u) P+ ,
Here, the tensor product notation still implies the usual matrix product in the quantum space.
1
Note that R (2 ) (2) vanishes identically since the projector P+ gets multiplied by the complementary projector P coming from one of the R-matrices in the tensor product. Similarly,
2
R (1 ) (2) = 0.
In a more general case, the procedure is as follows. Let be the Young diagram associated
with a given irrep of the algebra.2 Let n = || be the number of boxes in the diagram and let
n

P : CK|M
V
be the projector onto the space of the irreducible representation. Write in the box with coordinates
(i, j ) (ith line counting from top to bottom and j th column counting from left to right) the
number uij = u 2(i j ), as is shown in Fig. 5. Then R () (u) is constructed as

R () (u) = P
(2.7)
R(uij ) P .
(ij )
Here, the tensor factors are placed from right to left in the lexicographical order, i.e., elements
of the first row from left to right, then elements of the second row from left to right, etc. For
example, for the diagram = (3, 2) the product under the projectors reads R(u) R(u 2)
R(u + 4) R(u + 2) R(u).
Omitting further details of the fusion procedure (see, e.g., [35,36]), we now describe the structure of zeros of the fused R-matrix R () (u) which is essential for what follows. As we have
2 For super-algebras this may be a delicate point since this correspondence is in general not one-to-one.
354
seen, the R-matrix R () (u) is obtained by fusing n = || fundamental R-matrices in the auxiliary space. Then, by the construction, any matrix element of the R () (u) is a polynomial in u
of degree n. However, one can see that at some special values of the spectral parameter the
complimentary projectors get multiplied and the matrix R () (u) vanishes as a whole thing. In
other words, the matrix elements have a number of common zeros. We call them trivial zeros.
A more detailed analysis shows that the number of trivial zeros is n 1 (counting with multiplicities) and thus a scalar polynomial of degree n 1 can be factored out. Namely, this polynomial
is equal to the product of the factors uij = u 2(i j ) over the boxes of the diagram with
(i, j ) except i = j = 1.
We consider here only the representations of the type = (s a ) corresponding to rectangular
Young diagrams with s rows and a columns (rectangular irreps). In this case
a
R (s ) (u) = ra,s (u)R(s ) (u),

a
where matrix elements of R(s ) (u) are polynomials of degree 1 and

ra,s (u) = u1
a
s

(u + 2j 2l),
a, s 1,
(2.8)
j =1 l=1
is the polynomial of degree as 1. For a future reference, we give its representation through the
Barnes function G(z) (a unique entire function such that G(z + 1) = (z)G(z) and G(1) = 1):
ra,s (u) = 2as u1
G( 12 u + s + 1)G( 12 u a + 1)
G( 12 u + 1)G( 12 u + s a + 1)
(2.9)
The quantum monodromy matrix T () with the auxiliary space V is defined by the same
formula as before but with the R-matrix R () . The transfer matrix is
T () (u) = strV T () (u).
(2.10)

The transfer matrices commute for any and u: [T () (u), T ( ) (u )] = 0. This commuting family
of operators extends the family (2.4) which corresponds to the one-box diagram . For the empty
diagram = we formally put T () (u) = 1. For rectangular diagrams = s a we use the special
notation T () (u) := Tsa (u).
2.3. Functional relations for transfer matrices
The transfer matrices are functionally dependent. It appears that all T () (u) can be expressed
through Ts1 (u) or T1a (u) by means of the nice determinant formulas due to Bazhanov and
Reshetikhin [9]. They are the same for all (super)algebras of the type gl(K|M) with K, M 0
[12]:
T () (u) =
=
det
T1j +ij (u 2i + 2)
det
T1 j
1i,j 1
1i,j 1
+ij
(u + 2i 2).
For rectangular diagrams, they read:
(2.11)
355
Fig. 6. Fusion of two fundamental transfer matrices and the corresponding Hirota equation. As usual, each line carries
the vector representation, each crossing corresponds to the insertion of the R-matrix, and the closed lines correspond to
taking traces in the auxiliary space. The insertions of projectors P are interpreted as insertions of the R-matrices with
given values of spectral parameters.
Tsa (u) =
=
det
1
Ts+ij
(u 2i + 2)
det
T1
1i,j a
1i,j s
a+ij
(u + 2i 2).
(2.12)
These formulas should be supplemented by the boundary conditions Ts0 (u) = T0a (u) = 1,
a (u) = 0 for all a, s 0 and n 1. Since all the transfer matrices commute, they
Tsn (u) = Tn
can be diagonalized simultaneously, and the same relations are valid for any of their eigenvalues.
Keeping this in mind, we will often refer to the transfer matrices as scalar functions and call them
T -functions.
Applying the Jacobi identity for determinants to formulas (2.12) (see, e.g., [16,37]), one obtains a closed functional relation between transfer matrices for rectangular diagrams, equivalent
to Hirota difference equation:
a
a
(u 2)Ts1
(u) = Tsa1 (u 2)Tsa+1 (u).
Tsa (u 2)Tsa (u) Ts+1
(2.13)
The bilinear form of the functional relations was discussed in [10,11,16,3740]. Below we illustrate this relation by simple examples.
2.3.1. Simple examples of the functional equation
To figure out the general pattern, it is useful to consider the simplest case of the fusion of two
transfer matrices with fundamental representations in the auxiliary space. The proof is summarized in Fig. 6. Let us represent the first term in Eq. (2.13) as

T11 (u 2)T11 (u) = traux,V V T20 (u 2)T10 (u) .
(2.14)
The index 0 denotes the common quantum space represented in the figure by several vertical
lines, the indices 1, 2 denote the two copies of the auxiliary space V . Inserting I = P+ + P
inside the trace, we get two terms. Using the projector property P2 = P , we immediately see
that the term with P+ is equal to T21 (u 2), by the definition of the latter. The term with P
does not literally coincide with the definition of T12 (u) since the order of the horizontal lines is
reversed. However, plugging P = 14 R(2) and using the YangBaxter equation to move the
vertical lines to the other side of the R-matrix R(2), we come to the equivalent graph with the
required order of the horizontal lines. Finally, we get
T11 (u 2)T11 (u) = T21 (u 2) + T12 (u).
(2.15)
356
Fig. 7. The Hirota-type relation illustrated by rectangular Young diagrams decorated by spectral parameters. To obtain
the first term in the r.h.s., one takes the first row of the second diagram in the l.h.s. and puts it on the top of the first one.
To obtain the second term in the r.h.s., one takes the first column of the first diagram in the l.h.s. and attaches it to the
second one from the left.
Hence we have reproduced the simplest case of Eq. (2.13).

Let us comment on the general case. An illustrative example is given in Fig. 7. The rectangular
Young diagrams are decorated by the values of the spectral parameter
uij = u 2(i j ),
(2.16)
as is explained above. Each box of such decorated diagrams corresponds to a line characterized
by a given value of the spectral parameter. All these lines cross each other and the R-matrices are
associated to the crossing points. The order of the crossings is irrelevant due to the YangBaxter
equation. The reshuffling of spectral parameters in the second and third terms of Eq. (2.13)
(corresponding to the second and the third lines in the figure) is done by the exchange of a row
or a column between the two diagrams of the first line. The decoration of the resulting Young
diagrams follows the rule (2.16). Of course this is not a proof, just an illustration.
2.3.2. Analogy with formulas for characters and identities for symmetric functions
Eqs. (2.11) are spectral parameter dependent versions of the second Weyl formula for characters of (super-)groups (see [20]):
(w1 , . . . , wK ; v1 , . . . , vM ) =
det
i,j =1,...,K+M
Si +ij .
(2.17)
In the theory of symmetric functions such formulas are known as the JacobiTrudi determinant
identities. Here i are the lengths of the rows of the diagram , Sn are super-Schur polynomials
defined as
M

m=1 (1 zvm )
z n Sn ,
(2.18)
=
K
(1
zw
)
k
k=1
n=1
and wk , vm are eigenvalues of the A- and D-parts of the diagonalized element of the super-group
in the matrix realization of the type (A.1).
357
For the rectangular irrep with = s a Eq. (2.17) reads

s a (a, s) =
det
i,j =1,...,a
Ss+ij .
(2.19)
The characters s a satisfy the bilinear relation

2 (a, s) = (a + 1, s)(a 1, s) + (a, s + 1)(a, s 1)
(2.20)
which follows from the Jacobi identity for determinants. It is the spectral parameter independent
version of the Hirota equation.
2.3.3. Normalization and boundary conditions
A simple redefinition of the T -functions by a shift of the spectral parameter,
T (a,s) (u) Tsa (u s + a)
(2.21)
brings Eq. (2.13) to the form

T (a,s) (u + 1)T (a,s) (u 1) T (a,s+1) (u)T (a,s1) (u) = T (a+1,s) (u)T (a1,s) (u)
(2.22)
which appears to be completely symmetric with respect to interchanging of a and s. This is the
famous Hirota bilinear difference equation [18] which is the starting point of our approach to the
Bethe ansatz and generalized Baxter equations in this paper. For convenience, we also give the
determinant representation (2.12) in terms of T (a,s) (u):
T (a,s) (u) =
=
det
T (1,s+ij ) (u + a + 1 i j )
det
T (a+ij,1) (u + s + 1 i j ).
1i,j a
1i,j s
(2.23)
Let us comment on the meaning of Eq. (2.22). On the first glance, there is no much content
in this equation. Its general solution (with the boundary conditions fixed above) is just given
by formulas (2.23) with arbitrary functions T (1,s) (u) or T (a,1) (u). However, in the problem of
interest these functions are by no means arbitrary. They are to be found from certain analytic
conditions. For the finite spin chains with finite dimensional representations at each site these
conditions simply mean that T (a,s) (u) must be a polynomial of degree asN , where N is the
length of the chain, with (as 1)N fixed zeros. These zeros are just the trivial zeros coming
from the fusion procedure. Their location is determined by the scalar factor ra,s (u) (2.8) of the
R-matrix. We thus see that the polynomial T (a,s) (u) for all a, s 1 must be divisible by the
polynomial
N

ra,s (u s + a i ) := (a, s, u).
i=1
This constraint makes the problem non-trivial.

Let us introduce the function
(u) =
N

(u i ),
i=1
(2.24)
358
in terms of which the scalar polynomial factor is written as

(a, s, u) =
ra,s (u s + a i )
i=1
a
s

= 1 (u s + a)
(u s + a + 2j 2l),
a, s 1.
j =1 l=1
The representation through the Barnes function (2.9) allows us to extend this formula to all values
of a and s:
(a, s, u) = 2asN 1 (u s + a)
N

G( 1 (u + s + a) + 1 i )G( 1 (u s a) + 1 i )
2
G( 12 (u s + a) + 1 i )G( 12 (u + s a) + 1 i )
i=1
(2.25)
Extracting it from the T (a,s) (u), we introduce the T -function

T (a, s, u) = 1 (a, s, u)T (a,s) (u)
(2.26)
which is a polynomial in u of degree N for all a, s 0. Note that at a = 0 or s = 0 Eq. (2.25)

yields (0, s, u) = 1/(u s), (a, 0, u) = 1/(u + a), so
T (0, s, u) = (u s),
T (a, 0, u) = (u + a).
(2.27)
It is important to note that the renormalized T -function T (a, s, u) obeys the same Hirota
equation (2.22) as the T -function T (a,s) (u). Indeed, it easy to check that the transformation
T (a,s) (u) f0 (u + s + a)f1 (u + s a)f2 (u s + a)f3 (u s a)T (a,s) (u),
(2.28)
where fi are arbitrary functions, leaves the form of the equation unchanged. Eq. (2.25) shows
that the function (a, s, u) is precisely of this form (the factor 2asN is easily seen to be of this
form, too).
The main difference between the bosonic gl(K) and supersymmetric gl(K|M) cases is in
the boundary conditions for the transfer matrices in the (s, a)-plane. For the algebra gl(K) the
rectangular Young diagrams live in the half-band s 0, 0 a K, while for the super-algebra
the rectangular diagrams live in the domain shown below in Fig. 9. The Hirota equation with
boundary conditions of this type will be our starting point for the analysis of the inhomogeneous
quantum integrable super-spin chains and it will allow us to obtain the full hierarchy of Baxter
relations, the new Hirota equation for the Baxter functions and the nested Bethe ansatz equations
for all possible choices of simple root systems for gl(K|M). This naturally generalizes the known
relations for the spin chains based on the gl(K) algebra.
3. Hierarchy of Hirota equations
Throughout the rest of the paper we deal with rectangular irreps only and use the normalization (2.26), where all the trivial zeros of the transfer matrix are excluded. This normalization
was used in [16,17], for the supersymmetric case see [15].
359
Fig. 8. Hirota equation in the (s, a)-plane.
3.1. Hirota equation and boundary conditions for super-algebra gl(K|M)

As we have seen in the previous section, the functional relations for transfer matrices of
integrable quantum spin chains with spins belonging to representations of the super-algebra
gl(K|M) can be written [12] in the form of the same Hirota equation as in the case of the ordinary Lie algebra gl(K):
T (a, s, u + 1)T (a, s, u 1) T (a, s + 1, u)T (a, s 1, u)
= T (a + 1, s, u)T (a 1, s, u).
(3.1)
It can be schematically drawn in the (s, a)-plane as is shown in Fig. 8. All the non-vanishing
T (a, s, u)s are polynomials in u of one and the same degree N equal to the number of sites in
the spin chain.
To distinguish solutions relevant to Bethe ansatz for a given (super-)algebra, one should specify the boundary conditions in the discrete variables a and s. For the super-algebra gl(K|M) one
can see [12,41] that
T (a, s, u) = 0 if: (i)
(iii)
a < 0 or
a>K
(ii)
s < 0 and a = 0,
and s > M
or
(3.2)
(see Fig. 9, where we use the letters k, m rather than K, M for later references). The latter requirement, that T (a, s, u) vanishes if simultaneously a > K and s > M, comes from the fact that
the Young super-diagrams for gl(K|M) containing a rectangular subdiagram with K + 1 rows
and M + 1 columns are illegal, i.e., the corresponding representations vanish [21]. Note that we
want the Hirota equation to be valid in the whole (s, a) plane, not just in the quadrant a, s 1.
This is why we have to require that T (0, s, u) does not vanish identically on the negative s-axis,
otherwise the Hirota equation would break down at the origin a = s = 0.
The boundary values of T (a, s, u) are rather special. For example, at a = 0 Eq. (3.1) converts
into
T (0, s, u + 1)T (0, s, u 1) = T (0, s + 1, u)T (0, s 1, u)
which is a discrete version of the dAlembert equation with the general solution T (0, s, u) =
f+ (u + s)f (u s) where f are arbitrary functions. In the normalization (2.26) we have
f+ (u) = 1, f (u) = (u) (see (2.27)). Similarly, the boundary function at the half-axis s = 0,
a 0 is normalized to depend on u + a only in which case it has to be equal to f (u + a) =
(u + a). As soon as this is fixed, there is no more freedom left, and the boundary functions at
360
Fig. 9. Boundary conditions for the Hirota equation in the case of the super-algebra gl(k|m). Note that
T (0, s, u) = Qk,m (u s) = 0 on the whole axis < s < +.
the interior boundaries3 are in general products of a function of u + s and a function of u s

on the horizontal line (respectively, of u + a and u a on the vertical line). One more thing to
be taken into account is the identification (up to a sign) of the T -functions on the two interior
boundaries:
T (K, M + n, u) = (1)nM T (K + n, M, u),
n 0.
(3.3)
This equality reflects the fact that the two rectangular Young diagrams of the shapes ((M +
n)K ) and (M K+n ) correspond to the same representation of the algebra gl(K|M) with the Kac
Dynkin label b1 = = bK1 = 0, bK = M + n, bK+1 = = bK+M1 = 0 [21]. Note also
that every point with integer coordinates inside the domain in Fig. 9 corresponds to an atypical
representation while the points on the interior boundaries correspond to typical ones, and in this
respect the coordinate along the boundary can be treated as a continuous number.
Summarizing, we can write:
T (0, s, u) = QK,M (u s),
< s < ,
T (a, 0, u) = QK,M (u + a),
0 a < ,
T (K, s, u) = QK,0 (u + s + K)Q0,M (u s K),

T (a, M, u) = (1)
M(aK)
M s < ,
QK,0 (u + a + M)Q0,M (u a M),
K a < ,
(3.4)
where the polynomial boundary function QK,M (u) = (u) is regarded as a fixed input characterizing the quantum space of the spin chain while the polynomials QK,0 (u) and Q0,M (u) are to be
determined from the solution to the Hirota equation. At K = M = 0 the domain of non-vanishing
T s shrinks to the axis a = 0 and the half-axis s = 0, a 0, and the gauge freedom allows one
to put Q0,0 (u) = 1. It should be noted that the identification (3.3) does not yet imply the coincidence of the Q-functions in the third and the fourth lines of Eq. (3.4) (i.e., on the horizontal
and vertical parts of the interior boundary). In fact the specific form of the boundary conditions
given in Eq. (3.4) (as well as the sign factor (1)M(aK) ) is determined by a consistency with a
more general hierarchy of Hirota equations connecting the T -functions for different values of K
and M. The uniqueness of the boundary conditions (3.4) will be justified in the next subsection.
3 We call the boundaries at a = 0 and s = 0, a 0 exterior and the boundaries inside the right upper quadrant in Fig. 9
interior ones.
361
In the case of the usual Lie algebra gl(K) (M = 0) the boundary conditions (3.4) become the
same as the ones imposed in [17] (in the original paper [16] a gauge equivalent version was used).
The domain of non-vanishing T s in Fig. 9 degenerates so that the vertical strip collapses to a
line. Therefore, the T -functions on the interior boundaries cease to be dynamical variables and
become equal, up to a shift of the argument, to the fixed function QK,0 = (u) characterizing
the spin chain (see [16,17]).
One of the possible setups of the problem is the following: we fix the polynomial QK,M (u)
and then solve the Hirota equation with the aforementioned boundary and analytic conditions.
The result is a finite set of solutions for T (a, s, u) which yield the spectrum of eigenvalues of the
quantum transfer matrix. We proceed by constructing a hierarchy of Hirota equations connecting
neighboring levels of the array in Fig. 9, i.e., equations connecting T -functions for which
K or M differ by 1. The existence of such a hierarchy follows from classical integrability of the
Hirota equation. Decreasing K and M by 1, one can undress step by step the original gl(K|M)
problem to an empty problem formally corresponding to gl(0|0). This procedure appears to be
equivalent to the hierarchical (nested) Bethe ansatz.
3.2. Auxiliary linear problems and Bcklund transformations
Like almost all known nonlinear integrable equations, the Hirota equation (3.1) serves as a
compatibility condition for over-determined linear problems [16,30]. To introduce them, it is
convenient to pass to the new variables
1
1
p = (u s a),
q = (u + s + a),
2
2
1
r = (u s + a).
(3.5)
2
We call them chiral or light-cone variables while the original ones will be referred to as
laboratory variables. Here are the formulas for the inverse transformation,
a = q + r,
s = p r,
u=p+q
(3.6)
and for the transformation of the vector fields:

p = u s ,
q = u + a ,
r = a s .
(3.7)
We set (p, q, r) = T (q + r, p r, p + q) and introduce the following linear problems for

an auxiliary function = (p, q, r):

(p + 1, r + 1)
r
e +
= (p + 1),
(p + 1) (r + 1)

(q + 1, r + 1)
= (q + 1),
e r
(3.8)
(q + 1) (r + 1)
where we indicate explicitly only those variables that are subject to shifts. The compatibility
means that the difference operators

(p + 1, r + 1)
(q + 1, r + 1)
p
r
q
r
e
and e
e +
e
(p + 1) (r + 1)
(q + 1) (r + 1)
commute. This leads to the relation
362
(p + 1) (q + 1, r + 1) + (q + 1) (p + 1, r + 1)
= h(2p, 2q) (r + 1) (p + 1, q + 1),
where h can be an arbitrary function of p and q. In the original variables this equation reads
T (a + 1)T (a 1) + T (s + 1)T (s 1)
= h(u s a, u + s + a)T (u + 1)T (u 1).
From the boundary conditions (3.4) at a = 0 or s = 0 it follows that h = 1 and we obtain the
Hirota equation (3.1).
An advantage of the light-cone variables is their separation in the linear problems: the first
problem does not involve q while the second one does not involve p. However, in contrast to
the laboratory variables a, s, u, they have no immediate physical meaning. Coming back to
the laboratory variables, we set (p, q, r) = (q + r, p r, p + q) and rewrite the linear
problems (3.8) in the form

T (a 1, s + 1, u)T (a, s 1, u + 1)
(a 1, s + 1, u) = (a 1, s, u + 1),
ea s +
T (a, s, u)T (a 1, s, u + 1)

T (a 1, s + 1, u)T (a + 1, s, u + 1)
(a 1, s + 1, u) = (a, s + 1, u + 1).
ea s
T (a, s, u)T (a, s + 1, u + 1)
(3.9)
Because the T -functions can vanish identically at some a, s, we eliminate the denominators by
passing to the new auxiliary function F = T , in terms of which we have
T (a 1, s, u + 1)F (a, s, u) + T (a, s 1, u + 1)F (a 1, s + 1, u)
= T (a, s, u)F (a 1, s, u + 1),
T (a, s + 1, u + 1)F (a, s, u) T (a + 1, s, u + 1)F (a 1, s + 1, u)
= T (a, s, u)F (a, s + 1, u + 1).
(3.10)
Note that the second equation can be obtained from the first one by the transformation
T (a, s, u) (1)as T (s, a, u) (and the same for F ) which leaves the Hirota equation invariant. Note also that the pair of equations can be written in a matrix form as follows:

T (a 1, s, u) T (a, s 1, u)
F (a, s, u 1)
T (a, s + 1, u) T (a + 1, s, u)
F (a 1, s + 1, u 1)

F (a 1, s, u)
= T (a, s, u 1)
.
(3.11)
F (a, s + 1, u)
A remark on the symmetry properties of the linear problems is in order. One can see that while
1 2
2
the Hirota equation (3.1) written for the function (1) 2 (a +s ) T (a, s, u) is form-invariant with
respect to any permutation of the variables a, s, u and changing sign of any variable, the system
of two linear problems (3.10) is not. To make the symmetry explicit, we multiply both sides of
Eq. (3.11) by the matrix inverse to the one in the left-hand side and use the Hirota equation for
T s. In this way we get another pair of linear problems,
T (a + 1, s + 1, u)F (a, s, u) T (a + 1, s, u + 1)F (a, s + 1, u 1)
= T (a, s, u)F (a + 1, s + 1, u),
363
Fig. 10. Boundary conditions for the first Bcklund transformation in the (s, a) plane.
T (a, s, u + 1)F (a, s, u 1) T (a, s 1, u)F (a, s + 1, u)

= T (a + 1, s, u)F (a 1, s, u),
(3.12)
which are equivalent to (and thus compatible with) the pair (3.10) by construction. The set of four
linear problems (3.11), (3.12) possesses the required symmetry (note that the second equation
in (3.12) is symmetric by itself, and its structure resembles the Hirota equation). Furthermore,
the Hirota equation can be derived as a compatibility condition for any two linear problems of
these four, and the other two hold automatically. The four linear problems (3.11), (3.12) can be
combined into a single matrix equation:
0
T (a, s, u 1) T (a, s + 1, u) T (a + 1, s, u)
T (a, s, u 1)
0
T (a 1, s, u)
T (a, s 1, u)
T (a, s + 1, u) T (a 1, s, u)
0
T (a, s, u + 1)
T (a + 1, s, u) T (a, s 1, u)
F (a 1, s, u)
F (a, s + 1, u)
= 0.
F (a, s, u 1)
F (a 1, s + 1, u 1)
T (a, s, u + 1)
(3.13)
The Hirota equation implies that the determinant of the antisymmetric matrix in the left-hand
side vanishes. If this holds, the rank of this matrix equals 2, so there are two linearly independent
solutions to the linear problem (3.13). Their meaning will be clarified below. The symmetric
form of the linear problems was suggested in [42]. For more details on the linear problems for
the Hirota equation and their symmetries see [30,31,42,43] and Appendix B.
There is a remarkable duality between T (a, s, u) and F (a, s, u) [16,30]: one can exchange the
roles of the functions T , F and treat Eq. (3.10) as an over-determined system of linear problems
for the function T with coefficients F . Their compatibility condition is the same Hirota equation
for F :
F (a, s, u + 1)F (a, s, u 1) F (a, s + 1, u)F (a, s 1, u)
= F (a + 1, s, u)F (a 1, s, u).
(3.14)
This can be seen by rewriting (3.10) in yet another equivalent form. Namely, shifting a a + 1
in the first equation and s s 1 in the second, we represent the two equations in the matrix
364
Fig. 11. Two equations for the first Bcklund transformation (BT1) in the (s, a) plane: the variables change within a
brick consisting of two neighboring squares.
form

F (a + 1, s, u) F (a, s + 1, u)
T (a, s, u + 1)
F (a, s 1, u) F (a 1, s, u)
T (a + 1, s 1, u + 1)

T (a + 1, s, u)
= F (a, s, u + 1)
F (a, s 1, u)
(3.15)
which is to be compared with (3.11). It is obvious that they differ by the substitution T F
and changing signs of all variables. Therefore, we get the same Hirota equation for F . We thus
conclude that any solution to the linear problems (3.10), where the T -function obeys the Hirota
equation, provides an auto-Bcklund transformation, i.e., a transformation that sends a solution
of the nonlinear integrable equation to another solution of the same equation. In what follows we
call them simply Bcklund transformations (BT) and distinguish two types of them.
Let us rewrite the linear problems (3.10) changing the order of the terms and shifting the
variables:
T (a + 1, s, u)F (a, s, u + 1) T (a, s, u + 1)F (a + 1, s, u)
= T (a + 1, s 1, u + 1)F (a, s + 1, u),
T (a, s + 1, u + 1)F (a, s, u) T (a, s, u)F (a, s + 1, u + 1)
= T (a + 1, s, u + 1)F (a 1, s + 1, u).
(3.16)
These equations are graphically represented in Fig. 11 in the (s, a) plane. Given polynomials
T (a, s, u) obeying the Hirota equation, we are going to seek for polynomial solutions for F .
It is easy to see that Eqs. (3.16) are not compatible if one imposes the boundary conditions for
F (a, s, u) and T (a, s, u) of the fat hook type with the same K and M. Indeed, applying them in
the corner point of the interior boundary, one sees that the boundary values must vanish identically. However, it is straightforward to verify that Eqs. (3.16) are compatible with the boundary
conditions of the following two types. The boundary conditions for F (a, s, u) can be either
F (a, s, u) = 0 if:
(i)
(iii)
a < 0 or
(ii) s < 0 and a = 0,
a > K 1 and s > M,
or
(3.17)
365
Fig. 12. Particular cases of application of the first Bcklund transformation (BT1) in the form depicted in Fig. 11.
or
F (a, s, u) = 0 if:
(i)
a < 0 or (ii)
(iii)
a>K
and
s < 0 and a = 0,
or
s > M + 1,
(3.18)
which are again of the fat hook type but with the shifts K K 1 or M M + 1.
3.2.1. First Bcklund transformation
The first Bcklund transformation (BT1) T F is given by the linear problems (3.16) with
boundary conditions (3.17). Moreover, the linear problems are also compatible with the more
specific form of the boundary conditions (3.4), where K is replaced by K 1 and M remains the
same. To see this, one should check all cases when one of the three terms in the linear equations
vanishes, which imposes certain constraints on the boundary functions. The first equation at s = 0
(the brick in position 8 in Fig. 12), with T (a, 0, u) as in (3.2), states that F (a, 0, u) depends
only on u + a. Therefore, we can set F (a, 0, u) = QK1,M (u + a), which at this step is just
the notation. Similarly, the second equation at a = 0 (position 5) implies that F (0, s, u) depends
only on u s. At s = 0 it must equal QK1,M (u), therefore, F (0, s, u) = QK1,M (u s).
The consistency of the last two boundary conditions in (3.2) follows from a similar analysis
at the interior boundaries. Indeed, at a = K 1, s > M (position 1) the first equation becomes
QK,0 (u + s + K)Q0,M (u s K)F (K 1, s, u + 1)
= QK,0 (u + s + K)Q0,M (u s K + 2)F (K 1, s + 1, u).
(3.19)
This equation fixes the (u s)-dependent factor in F (K 1, s, u) to be Q0,M (u s (K 1))

while no restrictions on the (u + s)-dependent factor emerge. One is free to call it QK1,0 (u +
s + K 1), so that
F (K 1, s, u) = QK1,0 (u + s + K 1)Q0,M (u s K + 1),
M s < ,
(3.20)
i.e., the boundary condition for F (a, s, u) on the half-line a = K 1, M s < takes the
same form as the 3rd one from Eq. (3.4) for T (a, s, u) on the half-line a = K, M s < (see
Fig. 10). Finally, one can check that the first equation in (3.16) at s = M, a k (position 6) is
consistent with
F (a, M, u) = (1)M(aK+1) QK1,0 (u + s + M)Q0,M (u a M),
K 1 a < ,
(3.21)
366
Fig. 13. Two equations of the second pair of linear problems in the (s, a) plane.
and does not bring any new constraints. The sign factor is not fixed by this argument. To fix it,
one should require consistency with the second Bcklund transformation which we consider in
the next subsection.
3.2.2. Second Bcklund transformation
As is seen from Eq. (3.18), the transformations generated by the linear problems (3.16) are not
able to move the vertical part of the interior boundary from the right to the left. For our purpose,
we need a transformation which would be able to decrease M. Using the duality between T
and F explained in the end of Section 3.2, one can introduce Bcklund transformations of the
required type. The second Bcklund transformation (BT2) T F is obtained from Eq. (3.16)
by exchanging T F and F T :
F (a + 1, s, u)T (a, s, u + 1) F (a, s, u + 1)T (a + 1, s, u)
= F (a + 1, s 1, u + 1)T (a, s + 1, u),
F (a, s + 1, u + 1)T (a, s, u) F (a, s, u)T (a, s + 1, u + 1)
= F (a + 1, s, u + 1)T (a 1, s + 1, u).
(3.22)
These equations are represented graphically in Fig. 13 in the (s, a) plane. As is argued in Section 3.2, their compatibility condition is the Hirota equation (3.1) for T . If it holds, then any
solution F obeys the same Hirota equation and thus provides an auto-Bcklund transformation.
Given a solution T (a, s, u) with the boundary conditions (3.4), Eq. (3.22) are compatible with
the following boundary condition for F (a, s, u):
F (a, s, u) = 0 if:
(i)
a < 0 or (ii)
(iii) a > K
s < 0 and a = 0,
and s > M 1,
or
(3.23)
which directly follow from (3.18) and differ from those for T (3.2) by the shift M M 1.
Moreover, they are also compatible with the more specific form of the boundary conditions (3.4),
where M is replaced by M 1 and K remains the same. In complete analogy with BT1, one
verifies this by applying BT2 on the exterior boundaries (positions 8 and 5 in Fig. 12) and on the
367
Fig. 14. Boundary conditions for the second Bcklund transformation in the (s, a) plane.
interior ones (positions 7 and 2). In particular, on the half-line s = M 1, K a < we have
F (a, M 1, u) = (1)(M1)(aK) QK,0 (u + a + M 1)Q0,M1 (u a M + 1),
K a < ,
(3.24)
the sign being uniquely fixed from the second equation of (3.22) applied in position 2. This
means that the boundary condition for F (a, s, u) on the half-line s = M 1, K a < takes
the same form as the one for T (a, s, u) on the half-line s = M, K a < (see the 4th equation
in (3.4) and Fig. 14).
Our final goal is to undress, using the transformations BT1 and BT2, the original problem
by collapsing the region where the Hirota equation acts: K 0, M 0.
3.3. Hierarchy of Hirota equations and linear problems
We see now that by applying BT1 or BT2 to a solution to the Hirota equation with the boundary conditions (3.4) we can shift the interior boundaries as K K 1 or M M 1 (Figs. 10
and 14, respectively). In this way we come to the problem for F (a, s, u) or F (a, s, u) of the
same kind as the original problem for T (a, s, u). Repeating these steps several times, we arrive
at the hierarchy of the functions Tk,m (a, s, u) such that
TK,M (a, s, u) T (a, s, u),
TK1,M (a, s, u) F (a, s, u),
TK,M1 (a, s, u) F (a, s, u),
etc.,
(3.25)
all of them satisfying the Hirota equation

Tk,m (a, s, u + 1)Tk,m (a, s, u 1) Tk,m (a, s + 1, u)Tk,m (a, s 1, u)
= Tk,m (a + 1, s, u)Tk,m (a 1, s, u),
(3.26)
where k = 0, . . . , K and m = 0, . . . , M. The boundary conditions are as follows

Tk,m (a, s, u) = 0 if:
(i)
a < 0 or (ii)
(iii) a > k
s < 0 and a = 0,
and s > m,
and
Tk,m (0, s, u) = Qk,m (u s),
< s < ,
Tk,m (a, 0, u) = Qk,m (u + a),
0 a < ,
or
(3.27)
368
Tk,m (k, s, u) = Qk,0 (u + s + k)Q0,m (u s k),
m s < ,
Tk,m (a, m, u) = (1)m(ak) Qk,0 (u + a + m)Q0,m (u a m),
ka<
(3.28)
(see Fig. 9). The function QK,M (u) is a fixed polynomial of degree N which depends on the
choice of the quantum (super-)spin chain with the gl(K|M) symmetry. We set Q0,0 (u) = 1 (it
corresponds to an empty chain). The other polynomial functions Qk,m (u) will be determined
from polynomial solutions to the hierarchy of Hirota equations.
The linear problem (3.16) generates a chain of Bcklund transformations BT1:
Tk,m (a + 1, s, u)Tk1,m (a, s, u + 1) Tk,m (a, s, u + 1)Tk1,m (a + 1, s, u)
= Tk,m (a + 1, s 1, u + 1)Tk1,m (a, s + 1, u),
Tk,m (a, s + 1, u + 1)Tk1,m (a, s, u) Tk,m (a, s, u)Tk1,m (a, s + 1, u + 1)
= Tk,m (a + 1, s, u + 1)Tk1,m (a 1, s + 1, u)
(3.29)
(k = 1, . . . , K). The linear problem (3.22) generates a chain of Bcklund transformations BT2:
Tk,m1 (a + 1, s, u)Tk,m (a, s, u + 1) Tk,m1 (a, s, u + 1)Tk,m (a + 1, s, u)
= Tk,m1 (a + 1, s 1, u + 1)Tk,m (a, s + 1, u),
Tk,m1 (a, s + 1, u + 1)Tk,m (a, s, u) Tk,m1 (a, s, u)Tk,m (a, s + 1, u + 1)
= Tk,m1 (a + 1, s, u + 1)Tk,m (a 1, s + 1, u)
(3.30)
(m = 1, . . . , M). Note that (3.30) differs from (3.29) only by the direction of the Bcklund
flow: it is k k 1 in (3.29) and m m + 1 in (3.30). In the 5D linear space with coordinates
a, s, u, k, m, the four equations (3.29), (3.30) act in the hyper-planes
{m = const} {u + s + a = const},
{m = const} {u s a = const},
{k = const} {u + s + a = const},
{k = const} {u s a = const},
respectively. Thus each of them is actually a dynamical equation in three variables rather than
five. It is easy to see that all of them can be transformed to the standard form of the Hirota
equation (B.2) by linear changes of variables.
The two Bcklund transformations can be unified in a matrix equation of the type (3.13). Let
Tk,m (a, s, u) be the antisymmetric matrix from the left-hand side in (3.13) at the level k, m:
Tk,m (a, s, u)
0
Tk,m (a, s, u 1) Tk,m (a, s + 1, u) Tk,m (a + 1, s, u)
0
Tk,m (a 1, s, u)
Tk,m (a, s 1, u)
T (a, s, u 1)
= k,m
,
Tk,m (a, s + 1, u) Tk,m (a 1, s, u)
0
Tk,m (a, s, u + 1)
Tk,m (a + 1, s, u) Tk,m (a, s 1, u) Tk,m (a, s, u + 1)
0
(3.31)
369
then the Bcklund transformations BT1 and (BT2)1 (the transformation inverse to BT2) in the
symmetric form are obtained as the first and the second columns of the matrix equation
Tk1,m (a 1, s, u)
Tk,m+1 (a 1, s, u)
Tk,m+1 (a, s + 1, u)
Tk1,m (a, s + 1, u)
Tk,m (a, s, u)
= 0. (3.32)
Tk,m+1 (a, s, u 1)
Tk1,m (a, s, u 1)
Tk1,m (a 1, s + 1, u 1) Tk,m+1 (a 1, s + 1, u 1)
The first and the second equations in (3.29), (3.30) are obtained in the second and the first lines
of the matrix equation (3.32), respectively. As we have seen above, the rank of the 4 4 matrix
Tk,m is 2, so it has two linearly independent zero eigenvectors. Now we see that they correspond
to the two independent transformations shifting either k or m.
3.4. Bilinear equations for the T -functions with different k and m
In this subsection we derive additional bilinear equations (3.38)(3.43) for the functions
Tk,m (a, s, u) in which both indices k, m undergo shifts by 1. They are also of the Hirota type.
A special case of them (Eq. (3.45)) is particularly important. It provides a bilinear relation for
the Q-functions (the QQ-relation) which will be also derived in Section 4 by other means.
All these additional bilinear relations follow from compatibility of Eqs. (3.29), (3.30) with
shifts in k and in m. To see this more explicitly, we note that these equations admit a remarkable
alternative representation. Namely, it is straightforward to verify that the first equations in (3.29),
(3.30) can be identically rewritten as
Tk1,m (a + 1, s, u)
Tk1,m (a, s + 1, u)

Tk1,m (a, s, u + 1)Tk,m (a, s + 1, u)
Tk,m (a + 1, s, u)
=
eu s
,
Tk1,m (a, s + 1, u)Tk,m (a, s, u + 1)
Tk,m (a, s + 1, u)
Tk,m+1 (a + 1, s, u)
Tk,m+1 (a, s + 1, u)

Tk,m+1 (a, s, u + 1)Tk,m (a, s + 1, u)
u s Tk,m (a + 1, s, u)
e
,
=
Tk,m+1 (a, s + 1, u)Tk,m (a, s, u + 1)
Tk,m (a, s + 1, u)
(3.33)
(3.34)
while the second equations as

Tk+1,m (a, s + 1, u)
Tk+1,m (a + 1, s, u)

Tk+1,m (a, s, u 1)Tk,m (a + 1, s, u)
u a Tk,m (a, s + 1, u)
+e
,
=
Tk+1,m (a + 1, s, u)Tk,m (a, s, u 1)
Tk,m (a + 1, s, u)
Tk,m1 (a, s + 1, u)
Tk,m1 (a + 1, s, u)

Tk,m1 (a, s, u 1)Tk,m (a + 1, s, u)
u a Tk,m (a, s + 1, u)
+e
.
=
Tk,m1 (a + 1, s, u)Tk,m (a, s, u 1)
Tk,m (a + 1, s, u)
(3.35)
(3.36)
In this form they appear as linear problems for the difference operators in the brackets together
with particular solutions. One can notice that they again look like auxiliary linear problems for
the Hirota equation but for a different choice of the variables. For example, Eqs. (3.33), (3.34)
370
should be compared with Eq. (B.1) from Appendix B, where we identify with T and choose
the variables as p1 = k, p2 = m, p3 = 12 (u s) (note that the variables a and 12 (u + s) are kept
constant in the T -functions entering the difference operator in (3.33), (3.34)). We see that the two
equations coincide with the two corresponding linear problems from (B.1) (with 1 = 2 = 1),
with
Tk,m (a + 1, s, u)
k,m (a, s, u) =
Tk,m (a, s + 1, u)
being their common solution. Because they hold for any k, m, the function k1,m+1 (a, s, u) can
be represented in two different ways as follows:

Tk1,m+1 (a, s, u + 1)Tk1,m (a, s + 1, u)
eu s
Tk1,m+1 (a, s + 1, u)Tk1,m (a, s, u + 1)

Tk1,m (a, s, u + 1)Tk,m (a, s + 1, u)
eu s k,m
Tk1,m (a, s + 1, u)Tk,m (a, s, u + 1)

Tk1,m+1 (a, s, u + 1)Tk,m+1 (a, s + 1, u)
=
eu s
Tk1,m+1 (a, s + 1, u)Tk,m+1 (a, s, u + 1)

Tk,m+1 (a, s, u + 1)Tk,m (a, s + 1, u)
eu s k,m .
Tk,m+1 (a, s + 1, u)Tk,m (a, s, u + 1)
Opening the brackets, one finds that the terms multiplied by k,m (a, s, u) and k,m (a, s 2, u +
2) cancel automatically while the terms proportional to k,m (a, s 1, u + 1) give a non-trivial
relation connecting the T -functions with different k and m. It has the form
Tk,m (a, s + 1, u)Tk+1,m+1 (a, s, u + 1) Tk,m (a, s, u + 1)Tk+1,m+1 (a, s + 1, u)
= fk,m (a, u + s)Tk+1,m (a, s, u + 1)Tk,m+1 (a, s + 1, u),
(3.37)
where fk,m (a, u + s) is an arbitrary function of k, m and a, u + s. Comparing it with a similar

equation obtained in the same way from the other pair of linear problems (3.35), (3.36), one can
see that it actually depends on the combination u + s a as well as on k, m (see Appendix C for
details). To fix it, we take s = m, so the first term of the equation vanishes (at a k + 1), and use
boundary conditions (3.28). This fixes the function fk,m to be 1.
Clearly, Eq. (3.37) (with fk,m (a, u + s) = 1) is the Hirota equation of the form (B.2) in the
variables k, m and 12 (u s). The Hirota equation implies the compatibility of the linear problems, i.e., the discrete zero curvature condition for the difference operators in (3.33), (3.34) holds
true. In our situation, it appears to be equivalent to the weak form of this condition, i.e., with
the operators being applied to a particular solution of the linear problems. In other words, the
compatibility of the linear problems (which in general means existence of a continuous family of common solutions) follows, in our case, from the existence of just one common solution
(cf. [44]).
There are other equations of the same type. It is convenient to write them all as the following
chain of equalities:
Tk,m+1 (a, s + 1, u)Tk+1,m (a, s, u + 1)
(3.38)
Tk,m (a 1, s, u)Tk+1,m+1 (a, s, u + 1) Tk,m (a, s, u + 1)Tk+1,m+1 (a 1, s, u)
=
Tk,m+1 (a 1, s, u)Tk+1,m (a, s, u + 1)
(3.39)
1=
371
Tk,m (a 1, s, u)Tk+1,m+1 (a, s 1, u) Tk,m (a, s 1, u)Tk+1,m+1 (a 1, s, u)

.
Tk,m+1 (a 1, s, u)Tk+1,m (a, s 1, u)
(3.40)
These equations have the same structure. In each equation, one of the variables a, s, u enters as
a parameter. More precisely, they act in the hyper-planes
=
{a = const} {u + s = const},
{s = const} {u a = const},
{u = const} {a + s = const},
respectively. The first equality in this chain is already proved. The proof of the other two is
straightforward: one should pass to common denominator, to group together similar terms and
to use Eqs. (3.32). In fact this chain can be continued by three more equations of a similar but
different structure:
Tk,m (a, s 1, u)Tk+1,m+1 (a, s, u + 1) Tk,m (a, s, u + 1)Tk+1,m+1 (a, s 1, u)
Tk,m+1 (a 1, s, u)Tk+1,m (a + 1, s 1, u + 1)
(3.41)
Tk,m (a, s, u + 1)Tk+1,m+1 (a + 1, s, u) Tk,m (a + 1, s, u)Tk+1,m+1 (a, s, u + 1)
=
Tk,m+1 (a, s + 1, u)Tk+1,m (a + 1, s 1, u + 1)
(3.42)
Tk,m (a, s 1, u)Tk+1,m+1 (a + 1, s, u) Tk,m (a + 1, s, u)Tk+1,m+1 (a, s 1, u)
=
.
Tk,m+1 (a, s, u 1)Tk+1,m (a + 1, s 1, u + 1)
(3.43)
They are proved in the same manner. There is no need to prove the first equality separately since
one can just continue the chain of Eqs. (3.38)(3.40) proving that (3.40) is equal to (3.41). Other
forms of these equations and more details can be found in Appendix C. Eqs. (3.41)(3.43) act in
the hyper-planes
1=
{a k + m = const} {u s a = const},
{s + k m = const} {u + s + a = const},
{u k + m = const} {u + s a = const},
respectively. Therefore, Eqs. (3.38)(3.40) and (3.41)(3.43) are actually equations in three variables rather than five. They can be transformed to the standard form of the Hirota equation (B.2)
by linear changes of variables.
We note that Eqs. (3.38)(3.40) can be written in the following concise form:

q Tk+1,m+1
p Tk+1,m
q Tk,m+1
p
= e
e e
(3.44)
e
.
Tkm
Tkm
Tkm
Here Tk,m Tk,m (a, u, s) and (p, q) stands for any one of the pairs (u, s), (u, a) and (a, s).
Restricting the bilinear equations to the boundaries of the fat hook domain, one obtains new
equations which include the Q-functions. For example, setting a = 0 in Eq. (3.37) (or Eq. (3.38))
and using (3.28), we get the QQ-relation mentioned in the Introduction:
Qk,m (u)Qk+1,m+1 (u + 2) Qk,m (u + 2)Qk+1,m+1 (u)
= Qk,m+1 (u)Qk+1,m (u + 2).
(3.45)
372
In the next section, it will be derived by other means. Another new relation, which is of a mixed
(T Q and QQ) type, is the particular case of Eq. (3.43) at a = 0, s = 1:
Tk+1,m+1 (1, 1, u)Qk,m (u) Tk,m (1, 1, u)Qk+1,m+1 (u)
= Qk,m+1 (u 2)Qk+1,m (u + 2).
(3.46)
4. T Q- and QQ-relations
In this section we partially solve the undressing problem for the hierarchy of the T functions Tk,m (a, s, u) and derive the generalized Baxter equations (T Q-relations) which express
T (1, s, u), T (a, 1, u) through the Baxter functions Qk,m (u). This is done by constructing an operator generating series for the T -functions and factorizing it into an ordered product of first
order difference operators, with coefficients being ratios of the Q-functions.4 These operators
obey a discrete zero curvature condition which leads to a bilinear relation for the functions Qk,m
with different values of k, m (the QQ-relation).
4.1. Operator generating series and generalized Baxter relations
We start by introducing the following difference operators of infinite order:
W k,m (u) =

Tk,m (1, s, u + s + 1)
s=0
Qk,m (u)
e2su ,

Tk,m (a, 1, u a + 1)
(1)a e2au
W k,m (u) =
Qk,m (u + 2)
(4.1)
a=0
which represent operator generating series for the transfer matrices corresponding to one-row or
one-column Young diagrams. The denominators are introduced for the proper normalization. Let
us show that the difference operators
Qk+1,m (u)Qk,m (u + 2)
e2u ,
U k,m (u) =
Qk+1,m (u + 2)Qk,m (u)
Qk,m (u)Qk,m+1 (u + 2)
Vk,m (u) =
e2u
Qk,m (u + 2)Qk,m+1 (u)
(4.2)
shift the level indices k, m of the W k,m (u) and W k,m (u). Namely, we are going to prove the
following operator relations:
W k,m (u) = U k,m (u)W k+1,m (u),
W k,m+1 (u) = Vk,m (u)W k,m (u),
(4.3)
W k+1,m (u) = W k,m (u)U k,m (u),

W k,m (u) = W k,m+1 (u)Vk,m (u).
(4.4)
4 For the bosonic case this was done in [9,16,3739]. For the supersymmetric case such equations were conjectured
in [12] (see also [45]).
373
Fig. 15. The discrete zero curvature condition U k,m+1 Vk+1,m = Vk,m U k,m .
To prove the first equation in (4.3), we write

W k1 (u) + e2u W k (u)
= W k1 (u) + W k (u + 2)e2u

Tk1 (1, 0, u + 1) Tk1 (1, s, u + s + 1) Tk (1, s 1, u + s + 2) 2su
=
.
+
+
e
Qk1 (u)
Qk1 (u)
Qk (u + 2)
s=1
(4.5)
(Here and below in the proof we omit the second index m since it is the same everywhere.) To
transform the expression in the square brackets, we use the first equation of the BT1 at a = 0
(position 3 in Fig. 12):
Tk (1, s, u)Qk1 (u s + 1) Tk1 (1, s, u)Qk (u s + 1)
= Tk (1, s 1, u + 1)Qk1 (u s 1).
(4.6)
Shifting it u u + s + 1 and dividing both sides by Qk (u + 2)Qk1 (u), we obtain

Tk (1, s, u + s + 1)Qk1 (u + 2) Tk1 (1, s, u + s + 1) Tk (1, s 1, u + s + 2)
=
.
Qk (u + 2)Qk1 (u)
Qk1 (u)
Qk (u + 2)
Using this, we rewrite the term in the square brackets under the sum in Eq. (4.5) as
Qk (u)Qk1 (u + 2) Tk (1, s, u + s + 1)
Qk (u + 2)Qk1 (u)
Qk (u)
and continue the equality (4.5):
W k1 (u) + e2u W k (u)
Tk1 (1, 0, u + 1) Qk (u)Qk1 (u + 2) Tk (1, s, u + s + 1) 2su

+
e
Qk1 (u)
Qk (u + 2)Qk1 (u)
Qk (u)
s=1
Qk (u)Qk1 (u + 2) Tk (1, s, u + s + 1) 2su

=
.
e
Qk (u + 2)Qk1 (u)
Qk (u)
(4.7)
s=0
In the last step we have noticed that the s = 0 term of the sum multiplied by the ratio of Qs is
just equal to Tk1 (1, 0, u + 1)/Qk1 (u). The sum in the r.h.s. is W k (u), so the first equality in
(4.3) is proved. The proof of the three other equations in (4.3)(4.4) is completely similar.
374
Combining Eqs. (4.3), (4.4), we see that

W k1,m (u)W k1,m (u) = W k,m (u)W k,m (u) = W k,m+1 (u)W k,m+1 (u),
i.e., the operator W k,m (u)W k,m (u) does not depend on k, m. Note that W 0,0 (u) = W 0,0 (u) = 1
as operators, since all the terms in (4.1) are zero except the first one, which is 1 thanks to the
boundary conditions T0,0 (1, 0, u + 1) = Q0,0 (u + 2) = 1, T0,0 (0, 1, u + 1) = Q0,0 (u) = 1.
Therefore, we conclude that the operators W k,m and W k,m are mutually inverse5 :
W k,m (u)W k,m (u) = 1.
(4.8)
In addition, applying Eqs. (4.3), (4.4) many times, we arrive at the following operator relations:
1
1
U 0,m
V0,m1 V0,0 ,
W k,m = U k1,m
1
1
V0,m1
W k,m = V0,0
U 0,m U k1,m ,
(4.9)
where we have skipped u since it is the same everywhere. Taking Eqs. (4.1), (4.9) at k = K,
m = M, we obtain the non-commutative generating functions for the transfer matrices in the
basic representations (s 1 ) or (1a ):

TK,M (1, s, u + s + 1)
s=0
QK,M (u)
1
1
(u) U 0,M
(u)V0,M1 (u) V0,0 (u),
e2su = U K1,M

TK,M (a, 1, u + a + 1) 2au
(1)a
e
QK,M (u + 2a + 2)
a=0
1
1
= V0,0
(u) V0,M1
(u)U 0,M (u) U K1,M (u).
(4.10)
Expanding the right-hand sides in powers of e2u and comparing the coefficients, one obtains
a set of generalized Baxter relations between T s and Qs. In principle, these formulas solve
our original problem: they give solutions to the Hirota equation in terms of the Q-functions
representing the boundary conditions at each level k, m.
4.2. Zero curvature condition and QQ-relation
The Q-functions are polynomials whose roots obey the Bethe equations. Contrary to the case
of bosonic gl(K) algebras, the Bethe equations for super-algebras admit many different forms.
They correspond to all possible undressing paths in the (k, m) plane. Their equivalence can be
established by means of certain duality transformations [14].
Here we suggest an easy transparent argument to derive all these systems of Bethe equations
and the corresponding duality transformations. Namely, we are going to show that the functions
Qk,m (u) obey their own Hirota equation. Given an undressing path, it immediately produces
the chain of Bethe equations. The duality transformation is nothing else than the discrete zero
curvature condition for the operators (4.2) on the k, m lattice.
Eqs. (4.3) imply W k1,m+1 = U k1,m+1 Vk,m W k,m = Vk1,m U k1,m W k,m which gives the discrete zero curvature condition (Fig. 15)
U k,m+1 Vk+1,m = Vk,m U k,m .
(4.11)
5 Using the determinant representation (2.12), it is not difficult to derive this fact directly from their definitions (4.1).
375
Fig. 16. The Hirota equation for the Baxter functions Qk,m (u) in the (k, m) plane.
We remark that it looks a bit non-symmetric because V shifts m by +1 while U shifts k by 1.

Being written through U 1 and V , the zero curvature condition acquires the standard symmetric
form
1
1
U k,m+1
.
Vk,m = Vk+1,m U k,m
(4.12)
As a consequence of it, the following bilinear relation for the Q s is valid (Fig. 16):
Qk,m (u)Qk+1,m+1 (u + 2) Qk+1,m+1 (u)Qk,m (u + 2)
= Qk,m+1 (u)Qk+1,m (u + 2).
(4.13)
It was already derived in Section 3.4 as a particular case of more general T T -relations (3.38),
(3.39) (see (3.45)). This is the Hirota equation in chiral variables (see Appendix B). Strictly
speaking, the zero curvature condition (4.11) implies Eq. (4.13) up to an additional factor in
the r.h.s. depending on k, m which remains unfixed by this argument. However, this factor can
always be eliminated by an appropriate normalization of Qk,m (u)s:
Jk,m
Qk,m (u) = Ak,m
(k,m)
u uj
(4.14)
j =1
i.e., by choosing the coefficients Ak,m . Moreover, the result of Section 3.4 shows that the boundary conditions (3.28) already imply the normalization in which the QQ-relation has the form
(4.13) (see the argument right after Eq. (3.37)).
The zero curvature condition allows us to represent Eqs. (4.10) in a more general form. Given
an arbitrary zigzag path K,M from (K, M) to (0, 0), the r.h.s. of these equations becomes the
ordered product of the shift operators along it (see Fig. 17):

TK,M (1, s, u + s + 1)
s=0
QK,M (u)
e2su =

TK,M (a, 1, u + a + 1) 2au
(1)a
=
e
QK,M (u + 2a + 2)
a=0
V(x,n) (u),
(x,n)K,M
1
(u).
V(x,n)
(4.15)
(x,n)K,M
Here we use the natural notation: x is the vector on the lattice with coordinates (k, m) (k is the
vertical coordinate and m is the horizontal coordinate!), n = (1, 0) or (0, 1) is the unit vector
looking along the next step of the path. In other words, (x, n) is the (oriented) edge of the path
from (K, M) to (0, 0) starting at the point x and looking in the direction n. In the first equation,
the shift operators are ordered from the last edge of the path (ending at the origin) to the first one
while in the second equation the order is opposite. In fact the zero curvature condition implies
376
Fig. 17. The undressing procedure by means of BT1 and BT2 in the (s, a) plane along an arbitrary zigzag path K,M .
This path defines the Cartan matrix which characterizes the basis of generators of the Cartan subalgebra.
that Eq. (4.15) remain true for any path leading from (K, M) to the origin provided the shift
operators are chosen as follows:
1
V(x,n) (u) = U k1,m
(u),
V(x,n) (u) = U k,m (u),

V(x,n) (u) = V 1 (u),
k,m
V(x,n) (u) = Vk,m1 (u),
n = (1, 0),
n = (1, 0),
n = (0, 1),
n = (0, 1).
(4.16)
Some simple examples of Eqs. (4.15) are given in Section 8.

5. Bethe equations
The bilinear QQ-relation (4.13) obtained in the previous section (see also Section 3.4 for an
alternative derivation) gives the easiest and the most transparent way to derive different systems
of Bethe equations and to prove their equivalence. In a similar way, the generalized T T -relations
(3.38)(3.40) can be used to derive a new system of Bethe-like equations for roots of the polynomials Tk,m (a, s, u).
5.1. Bethe equations for roots of Qs
To derive the system of equations for zeros of the polynomials Qk,m (4.14), we put u in
(k,m)
(k+1,m+1)
(k,m)
(k+1,m+1)
the Hirota equation (4.13) successively equal to uj
, uj
2, uj
2, uj
,
(k,m+1)
(k+1,m)
and uj
2, each corresponding to a zero of one of the six Q-functions in the
uj
equation. After proper shifts of k and m such that the arguments of the Q-functions become
(k,m)
(k,m)
and uj
2, we get the relations
uj
(k,m)
(k,m)
(k,m)
(k,m)

Qk,m uj
= Qk,m+1 uj
Qk+1,m uj
+ 2 Qk+1,m+1 uj
+ 2 , (a)

(k,m)
(k,m)
(k,m)
(k,m)
Qk,m uj
2 Qk1,m1 uj
2 , (b)
= Qk,m1 uj
Qk1,m uj

(k,m)
(k,m)
(k,m)
(k,m)
Qk,m uj
2 Qk+1,m+1 uj
2 , (c)
= Qk+1,m uj
Qk,m+1 uj
377
(k,m)
(k,m)
(k,m)
(k,m)

= Qk1,m uj
Qk,m1 uj
Qk,m uj
+ 2 Qk1,m1 uj
+2 ,
(k,m)
(k,m)
(k,m)
(k,m)

Qk,m1 uj
Qk+1,m uj
,
+ 2 = Qk,m1 uj
+ 2 Qk+1,m uj
(k,m)
(k,m)
(k,m)
(k,m)

Qk,m+1 uj
Qk1,m uj
,
2 = Qk,m+1 uj
2 Qk1,m uj
(d)
(e)
(f).
(5.1)
Here 1 k K 1, 1 m M 1 and j runs from 1 to Jk,m . This is the (over)complete set of

Bethe equations for our problem. Their consistency is guaranteed by the Hirota equation (4.13).
To convert them into a more familiar form, let us divide equation (a) by equations (c) and (b)
by (d). Using also equation (e) and (f), it is easy to rewrite the system in the form where each
group of equations contains the Q-functions at three neighboring sites. In this way we obtain the
following sets of equations:
(k,m)
(k,m)
(k,m)
2)Qk+1,m (uj
Qk1,m (uj
)Qk,m (uj
(k,m)
Qk1,m (uj
(k,m)
2)Qk,m (uj
+ 2)
(k,m)
+ 2)Qk+1,m (uj
)
= 1,
(5.2)
(k,m)
)Qk,m (uj
(k,m)
2)Qk,m (uj
Qk,m+1 (uj
Qk,m+1 (uj
(k,m)
(k,m)
(k,m)
(k,m)
+ 2)
Qk+1,m (uj
)Qk,m1 (uj
(k,m)
Qk+1,m (uj
(k,m)
+ 2)Qk,m1 (uj
)
(k,m)
(k,m)
2)Qk,m1 (uj
(k,m)
2)
Qk,m+1 (uj
)Qk1,m (uj
(k,m)
Qk,m+1 (uj
(k,m)
2)Qk1,m (uj
)
+ 2)
(k,m)
+ 2)Qk,m1 (uj
= 1,
(5.3)
= 1,
(5.4)
= 1.
(5.5)
These equations are valid at any point of the (k, m) lattice and do not depend on the choice
of the undressing zigzag path. The figures show the sites of the (k, m) lattice (k and m are
the vertical and horizontal coordinates respectively) which are connected by the corresponding Bethe equation. The point (k, m) is the one between the other two. The edges of the
(k, m) lattice are represented by the arrows which show the directions of the transformations BT1 and BT2. For completeness, we also present here two other equations derived from
(5.1):
(k,m)
)Qk,m (uj
(k,m)
2)Qk,m (uj
Qk1,m (uj
(k,m)
)Qk,m (uj
(k,m)
Qk1,m (uj
(k,m)
2)Qk,m (uj
Qk,m+1 (uj
Qk,m+1 (uj
(k,m)
(k,m)
2)Qk+1,m (uj
(k,m)
(k,m)
+ 2)
(k,m)
+ 2)Qk+1,m (uj
(k,m)
+ 2)
2)Qk,m1 (uj
(k,m)
+ 2)Qk,m1 (uj
)
= 1,
(5.6)
= 1.
(5.7)
It is clear from the figures that these patterns are forbidden for a zigzag path.
378
Let us show how to reduce this 2D array of Bethe equations to a chain. Suppose one fixes a
particular zigzag path K,M from (K, M) to (0, 0). Then, for each vertex of the path (except for
the first and the last ones), one writes the Bethe equations according to the configuration of the
path around the vertex. Let us enumerate vertices of the path by numbers from 0 to K + M so
that the point (k, m) K,M acquires the number k + m. Set

Jk+m
Qk,m (u) Qk+m (u) = Ak+m
(k+m)
u uj
(k, m) K,M .
j =1
Let us also enumerate edges of the path by numbers from 1 to K + M so that the edge joining the
(n 1)th and the nth vertex acquires the number n. (Note that in the course of undressing one
passes these edges in the inverse order.) To the nth edge of the path (oriented according to the
direction of the undressing, i.e., either from the north to the south or from the east to the west)
we assign the sign factor according to the rule:

+1 if the nth edge looks to the south,
pn =
1 if the nth edge looks to the west.
Then the system of the Bethe equations along the path can be written as follows:
(n)
(n)
(n)
Qn+1 (uj + 2pn+1 )Qn (uj 2pn+1 )Qn1 (uj )

(n)
(n)
(n)
Qn+1 (uj )Qn (uj + 2pn )Qn1 (uj 2pn )
= (1)
1+pn pn+1
2
(5.8)
where n runs from 1 to K + M 1. The boundary conditions are Q0 (u) = 1, QK+M (u) =
(u). Any chain of Bethe equations includes K + M 1 equations for the roots of K + M 1
polynomials Qk,m (u) picked along a path as in Fig. 17. All the other [(K + 1)(M + 1) 2]
(K + M 1) = KM Q-functions inside the K M rectangle can be expressed through them by
iterations of the Hirota equation (4.13).
This form of the Bethe equations agrees with the general one suggested in [3]. To see this, let
us redefine the Q-functions by the shift of the spectral parameter:
Qk,m (u) = Q k+m (u k + m),
(5.9)
which is equivalent to

n
p .
Qn (u) = Qn u
(5.10)
=1
In terms of the roots

(n)
(n)
u j = uj
n
(5.11)
=1
n (u) the system of Bethe equations (5.8) acquires a concise form

of the polynomial Q
K+M

b (u (a) Kab )
Q
j
b=1
b (u (a) + Kab )
Q
j
where
= (1)
1+pa pa+1
2
a = 1, . . . , K + M 1,
(5.12)
379
Kab = (pa + pa+1 )a,b pa+1 a+1,b pa a,b+1
(5.13)
is the Cartan matrix for the simple root system corresponding to the chosen undressing path (see,
for example, [14]). So it is natural to think of the KacDynkin diagram for the super-algebras as
a zigzag path on the (k, m) plane, as shown in Fig. 17.
Let us give a remark on the duality transformations. There are (K+M)!
K!M! different ways to choose
the undressing path K,M (Fig. 17), and hence there are as many chains of Bethe equations for
gl(K|M) algebra, all of them describing the same system but for different choices of the simple
roots basis. The transformation from one basis to another can be (and sometimes is) called the duality transformation meaning that the two descriptions of the model are equivalent and in a sense
dual to each other. For particular low rank super-algebras such transformations were discussed
in [46,49,50], and for general super-algebras in [14] (see also [45]). In various solid state applications of supersymmetric integrable models (for example, the tJ model) this transformation
corresponds to the so-called particlehole duality. It is clear that any duality transformation
can be decomposed into a chain of elementary ones. The elementary duality transformation consists in switching two neighboring orthogonal edges of the path (joining at a fermionic node
of the KacDynkin diagram) to another pair of such edges surrounding the same face of the
(k,m)
by Eq. (5.4) at the roots
lattice. It corresponds to replacing Eq. (5.4) at the roots u = uj
u = uj(k+1,m1) or vice versa, which induces also the subsequent change of the Bethe equations
at two neighboring nodes. On the operator level, the elementary duality transformation consists
1
1
in replacing Vk,m
in the products (4.10), according to the zero curvature
Uk,m+1 by U k,m Vk+1,m
condition (4.11).
5.2. Bethe-like equations for roots of T s
Actually, roots of all the polynomial T -functions Tk,m (a, s, u) obey a system of algebraic
equations which generalize the Bethe equations (5.8). They can be derived along the same lines
using, instead of the QQ-relation (3.45) or (4.13), the bilinear T T -relations (3.38)(3.43). Fixing
an undressing zigzag path K,M , we set

Jk+m
Tk,m (a, s, u) Tk+m (a, s, u) = Ak+m (a, s)
(k+m)
u uj

(a, s)
(k, m) K,M .
j =1
(n)
(n)
At a = s = 0 the root uj (0, 0) coincides with uj from the previous subsection. Repeating all
the steps leading to the Bethe equations (5.8), we arrive at the following Bethe-like equations:
Tn+1 (a, s pn+1 , u + pn+1 )Tn (a, s + pn+1 , u pn+1 )Tn1 (a, s, u )
Tn+1 (a, s, u )Tn (a, s pn , u + pn )Tn1 (a, s + pn , u pn )
1+pn pn+1
= (1) 2
,
Tn+1 (a + pn+1 , s, u + pn+1 )Tn (a pn+1 , s, u pn+1 )Tn1 (a, s, u )
Tn+1 (a, s, u )Tn (a + pn , s, u + pn )Tn1 (a pn , s, u pn )
1+pn pn+1
= (1) 2
,
Tn+1 (a pn+1 , s + pn+1 , u )Tn (a + pn+1 , s pn+1 , u )Tn1 (a, s, u )
Tn+1 (a, s, u )Tn (a pn , s + pn , u )Tn1 (a + pn , s pn , u )
= (1)
1+pn pn+1
2
(5.14)
(5.15)
(5.16)
380
Tn+1 (a pn+1 , s + pn+1 , u )Tn (a, s pn+1 , u pn+1 )Tn1 (a + pn , s, u + pn )

Tn+1 (a pn+1 , s, u pn+1 )Tn (a, s + pn , u + pn )Tn1 (a + pn , s pn , u )
1+pn pn+1
= (1) 2
(5.17)
,
Tn+1 (a pn+1 , s + pn+1 , u )Tn (a + pn+1 , s, u pn+1 )Tn1 (a, s pn , u + pn )
Tn+1 (a, s + pn+1 , u pn+1 )Tn (a pn , s, u + pn )Tn1 (a + pn , s pn , u )
1+pn pn+1
= (1) 2
(5.18)
,
Tn+1 (a pn+1 , s, u pn+1 )Tn (a + pn+1 , s + pn+1 , u )Tn1 (a, s pn , u + pn )
Tn+1 (a, s + pn+1 , u pn+1 )Tn (a pn , s pn , u )Tn1 (a + pn , s, u + pn )
= (1)
1+pn pn+1
2
(5.19)
(n)
uj (a, s).
where u
The values of a, s and n are assumed to be such that none of the
T -functions is identically zero. At the boundaries a = 0 or s = 0 Eqs. (5.14), (5.15) coincide
with the Bethe equations (5.8).
6. Algorithm for integration of the Hirota equation
In this section we develop a general algorithm to solve the Hirota equation (3.26) expressing
the functions Tk,m (a, s, u) through the boundary functions Qk,m (u) (3.28). We note that it gives
an operator realization of the combinatorial rules given in [12].
6.1. Shift operators
Our starting point is the alternative representation of the first and second Bcklund transformations given by Eqs. (3.33)(3.36) which we rewrite here in a slightly different form. Eqs. (3.33),
(3.34) read
Tk1,m (a, s, u) = H k ,m (a 1, s, u)Tk,m (a, s, u),
Tk,m+1 (a, s, u) = H k,m+ (a 1, s, u)Tk,m (a, s, u),
(6.1)
(6.2)
and Eqs. (3.35), (3.36) read

Tk+1,m (a, s, u) = H k + ,m (a, s 1, u)Tk,m (a, s, u),
Tk,m1 (a, s, u) = H k,m (a, s 1, u)Tk,m (a, s, u).
(6.3)
(6.4)
Here the difference operators H k ,m (a, s, u) and H k,m (a, s, u) are given by
Tk1,m (a, s, u + 1)
H k ,m (a, s, u) :=
Tk,m (a, s, u + 1)
T
k,m+1 (a, s, u + 1)
H k,m+ (a, s, u) :=
Tk,m (a, s, u + 1)
T
k+1,m (a, s, u 1)
H k + ,m (a, s, u) :=
Tk,m (a, s, u 1)
T
k,m1 (a, s, u 1)
H k,m (a, s, u) :=
Tk,m (a, s, u 1)
Tk1,m (a, s + 1, u) u s
,
e
Tk,m (a, s, u + 1)
Tk,m+1 (a, s + 1, u) u s
,
e
Tk,m (a, s, u + 1)
Tk+1,m (a + 1, s, u) (u +a )
,
+
e
Tk,m (a, s, u 1)
Tk,m1 (a + 1, s, u) (u +a )
.
+
e
Tk,m (a, s, u 1)
(6.5)
(6.6)
(6.7)
(6.8)
381
As we have seen in Section 3.4, these equations are equivalent to Eqs. (3.29) and (3.30) which
have been used to define the Bcklund transformations BT1 and BT2.
The operators introduced above obey, by construction, the weak zero curvature conditions
Tk1,m1 = H k1,m H k ,m Tk,m = H k ,m1 H k,m Tk,m ,
(6.9)
where H k ,m H k ,m (a 1, s, u), H k + ,m H k + ,m (a, s 1, u), H k,m+ H k,m+ (a 1, s, u),

H k,m H k,m (a, s 1, u) and Tk,m Tk,m (a, s, u). As is pointed out in Section 3.4, they imply
the strong, operator form of these conditions:
H k1,m H k ,m = H k ,m1 H k,m .
(6.10)
The operators H k ,m , H k,m generalize the shift operators U k,m , Vk,m introduced in Section 4. We hold the same name for them. Comparing to U k,m , Vk,m , they act to functions of
three variables, not just to functions of the spectral parameter u, and involve non-trivial shifts
in two independent directions. However, the shift operators at a = 0 or s = 0 are effectively
one-dimensional since they do not depend on u + s (or u a):
Qk1,m (u s + 1) Qk1,m (u s 1) u s
H k ,m (0, s, u) =
,
e
Qk,m (u s + 1)
Qk,m (u s + 1)
Qk,m+1 (u s + 1) Qk,m+1 (u s 1) u s
,
e
H k,m+ (0, s, u) =
Qk,m (u s + 1)
Qk,m (u s + 1)
Qk+1,m (u + a 1) Qk+1,m (u + a + 1) (u +a )
,
+
e
H k + ,m (a, 0, u) =
Qk,m (u + a 1)
Qk,m (u + a 1)
Qk,m1 (u + a 1) Qk,m1 (u + a + 1) (u +a )
H k,m (a, 0, u) =
.
+
e
Qk,m (u + a 1)
Qk,m (u + a 1)
(6.11)
They are functionals of Qk,m (u) only. The first (last) two of them, when restricted to the functions
and V ) respectively.
of u s (u + a), are equivalent to (adjoint) operators U k,m and Vk,m (U k,m
k,m
More precisely,
1
H (k+1) ,m (0, s, u)Qk+1,m (u s 1) U k,m (u s 1),
Qk,m (u s 1)
1
H k,m+ (0, s, u)Qk,m (u s 1) Vk,m (u s 1),
Qk,m+1 (u s 1)
(1)a
(u + a 1),
H k + ,m (a, 0, u)(1)a Qk,m (u + a + 1) U k,m
Qk+1,m (u + a + 1)
(1)a
(u + a 1),
H k,(m+1) (a, 0, u)(1)a Qk,m+1 (u + a + 1) Vk,m
Qk,m (u + a + 1)
(6.12)
where it is implied that the operators in the l.h.s. act on functions of u s (u + a).
A simple inspection shows that the shift operators can be written as
Tk1,m (a, s, u + 1) (1)
(a, s + 1, u),
h
H k ,m (a, s, u) =
Tk,m (a, s, u + 1) k1,m
Tk,m+1 (a, s, u + 1) (1)
H k,m+ (a, s, u) =
(a, s + 1, u),
h
Tk,m (a, s, u + 1) k,m+1
(6.13)
(6.14)
382
Tk+1,m (a, s, u 1) (2)

H k + ,m (a, s, u) =
(a + 1, s, u),
h
Tk,m (a, s, u 1) k+1,m
Tk,m1 (a, s, u 1) (2)
H k,m (a, s, u) =
(a + 1, s, u)
h
Tk,m (a, s, u 1) k,m1
(6.15)
(6.16)
where

(1)
h k,m (a, s, u) := Tk,m (a, s, u) 1 eu s
1
,
Tk,m (a, s, u)

1
(u +a )
h (2)
.
k,m (a, s, u) := Tk,m (a, s, u) 1 + e
Tk,m (a, s, u)
(6.17)
(6.18)
(1)
(a, u
From this representation it is obvious that they have non-trivial kernels, Tk,m (a, s, u)fk,m
(2)
a + s) and Tk,m (a, s, u)(1)a fk,m (s, u a + s) respectively, so their common kernel is
(3)
(i)
Tk,m (a, s, u)(1)a fk,m (u a + s), where fk,m are arbitrary functions of their arguments. Modulo these kernels the shift operators H k ,m (a, s, u) and H k,m (a, s, u) can be inverted. We have:
1
H (k+1)
,m (a, s, u)
1

= Tk,m (a, s + 1, u) 1 eu s
= Tk,m (a, s + 1, u)

j =0
1
Tk+1,m (a, s, u + 1)
Tk,m (a, s + 1, u) Tk,m (a, s, u + 1)
1
Tk,m (a, s j + 1, u + j )
Tk+1,m (a, s j, u + j + 1) j (u s )
e
Tk,m (a, s j, u + j + 1)
(6.19)
and
1
H k,(m+1)
(a, s, u)
1

= Tk,m (a + 1, s, u) 1 + e(u +a )
= Tk,m (a + 1, s, u)

j =0
1
Tk,m+1 (a, s, u 1)
Tk,m (a + 1, s, u) Tk,m (a, s, u 1)
(1)j
Tk,m (a j + 1, s, u j )
Tk,m+1 (a j, s, u j 1) j (u +a )
.
e
Tk,m (a j, s, u j 1)
(6.20)
Eqs. (6.1) and (6.4) rewritten in the following equivalent form

1
Tk+1,m (a, s, u) = H (k+1)
,m (a 1, s, u)Tk,m (a, s, u),
(6.21)
1
Tk,m+1 (a, s, u) = H k,(m+1)
(a, s 1, u)Tk,m (a, s, u)
(6.22)
will be useful for integration of the Hirota equation.

As it is clear from the explicit expressions (6.19), (6.20), the inverse shift operators acting
on the function Tk,m (a, s, u) in Eqs. (6.21), (6.22) are represented by sums of fractions whose
numerators and denominators are products of T s containing both positive and negative values of
the arguments a and/or s. The same is also true for products of the shift operators H k ,m (a, s, u)
and H k,m (a, s, u) since the operators es and ea lower values of a and s. The functions
383
Tk,m (a, s, u) are equal to zero at negative integer values of s or a according to the boundary conditions (3.27). Therefore, the numerators and denominators of some ratios could simultaneously
become zero at some values of s or a. We have to define their values in a way consistent with the
hierarchy of Hirota equations. One way to do that is to analytically continue the T -functions to
negative values a = n + and/or s = l + , where n, l N and , R tend to 0. One can
straightforwardly verify that the behavior

Tk,m (a, l + , u) = O l ,
Tk,m (n + , s, u) = O n ,

Tk,m (n + , l + , u) = O n l
(6.23)
is consistent with the Hirota equation and Eqs. (3.29), (3.30). We then notice that both operators
1
H k1
,m (a, s, u) and Hk,m (a, s, u), as well as their products, are non-singular when acting on the
functions Tk,m (a, s, u). Actually, the behavior (6.23) is equivalent to the following prescription
to define the series of the form (6.19) or (6.20) acting on Tk,m (a, s, u): fractions containing
T s at negative values of s and/or a do not give any contribution to the sum. Thus, for finite
1
positive values of s and a, the inverse shift operators H k1
,m (a, s, u) and Hk,m (a, s, u) acting on
Tk,m (a, s, u) contain a finite sum of nonzero terms. We use this prescription in what follows.
Now, we are ready to present a general algorithm of reconstructing the functions Tk,m (a, s, u)
in terms of Qk,m (u) based on Eqs. (6.2), (6.3) and (6.21), (6.22).
6.2. Integration of the Hirota equation
Eqs. (6.3) and (6.21) ((6.2) and (6.22)) are recurrence relations allowing one to express the
functions Tk,m (a, s, u) in terms of the same functions but with smaller values of k, m and/or a, s:
k

1
Tk,m (a, s, u) =
(6.24)
H (a 1, s, u) Tn,m (a, s, u),
p ,m
p=n+1
Tk,m (a, s, u) =
m1

Hk,q + (a 1, s, u) Tk,l (a, s, u),
q=l
Tk,m (a, s, u) =
k1

Tk,m (a, s, u) =
(6.25)

H p+ ,m (a, s 1, u) Tn,m (a, s, u),
p=n
m

(6.26)

1
H k,q
(a, s 1, u) Tk,l (a, s, u).
(6.27)
q=l+1
Here k, l, m, n are positive integer numbers such that

0 l m 1,
0 n k 1,
0 k K,
0 m M,
m
and n=l O n O m O l by definition. Substituting n = a, l = 0 into Eqs. (6.24), (6.25), and
n = 0, l = s into Eqs. (6.26), (6.27), we obtain:
K

1
TK,0 (a, s, u) =
(6.28)
H p ,0 (a 1, s, u) Ta,0 (a, s, u),
p=a+1
384
TK,M (a, s, u) =
M1

HK,q + (a 1, s, u) TK,0 (a, s, u)
(6.29)
q=0
(where 0 < a < K, 0 s < ) and

K1

Hp+ ,M (a, s 1, u) T0,M (a, s, u),

TK,M (a, s, u) =

T0,M (a, s, u) =
p=0
M

(6.30)

1
H 0,q
(a, s
1, u) T0,s (a, s, u)
(6.31)
q=s+1
(where 0 a < , 0 < s < M). This representation allows us to write down the formulas for
the whole set of the nonzero T -functions which do not belong to the boundaries:
M1
K

1
TK,M (a, s, u) =
H K,q + (a 1, s, u)
H (a 1, s, u) Ta,0 (a, s, u)
p ,0
q=0
p=a+1
(6.32)
(where 0 < a < K, 0 s < ) and

K1
M

1
TK,M (a, s, u) =
H p+ ,M (a, s 1, u)
H 0,q
(a, s 1, u) T0,s (a, s, u)
p=0
q=s+1
(6.33)
(where 0 a < , 0 < s < M). Note that the functions Ta,0 (a, s, u) and T0,s (a, s, u) entering
these equations are boundary functions:
Ta,0 (a, s, u) = Qa,0 (u + a + s)
Ta,0 (a, s, u) = 0
(0 s < ),
( s < 0)
(6.34)
and
T0,s (a, s, u) = (1)as Q0,s (u a s)
T0,s (a, s, u) = 0
(0 a < ),
( a < 0),
(6.35)
according to the boundary conditions (3.28).

Let us put a = 1 in Eq. (6.32),
M1
K

1
HK,q + (0, s, u)
Hp ,0 (0, s, u) T1,0 (1, s, u)
TK,M (1, s, u) =
q=0
p=2
(0 s < , 1 < K),
(6.36)
and s = 1 in Eq. (6.33),

K1
M

1
TK,M (a, 1, u) =
H p+ ,M (a, 0, u)
H 0,q
(a, 0, u) T0,1 (a, 1, u)
p=0
(0 a < , 1 < M).
q=2
(6.37)
385
Since the shift operators entering these equations are functionals of the boundary functions
Qk,m (u) only, we immediately obtain explicit expressions for Tk,m (1, s, u) and Tk,m (a, 1, u) in
terms of the boundary functions Qk,m (u).
Let us rewrite the solutions (6.36) and (6.37) in a slightly different form:
M1
K

TK,M (1, s, u) =
(6.38)
H K,q + (0, s, u)
H 1
(0, s, u) T0,0 (1, s, u),
p ,0
q=0
TK,M (a, 1, u) =
K1

H p+ ,M (a, 0, u)
p=0
p=1
M

1
H 0,q
(a, 0, u)
T0,0 (a, 1, u).
(6.39)
q=1
Taking into account the explicit form of T0,0 ,

T0,0 (a, s, u) = 1 at (i)
a = 0 or (ii)
s = 0 and a > 0,
T0,0 (a, s, u) = 0 at (i)
a = 0 and s = 0 or
(ii)
s = 0 and a < 0,
(6.40)
it is easy to see that these solutions can be represented as the generating series
TK,M (1, n, u s + n)en(u s )
n=0
M1

q=0

H K,q + (0, s, u)

H p1
,0 (0, s, u) ,
(6.41)
p=1
TK,M (n, 1, u + a n)en(u +a )
n=0
K1

p=0
M

1
Hp+ ,M (a, 0, u)
H0,q (a, 0, u) .
(6.42)
q=1
Indeed, it is clear that the operator in the r.h.s. of (6.41) is expended in the powers of eu s
as is written in the l.h.s., with some coefficients. To fix them, one applies the both sides to the
function T0,0 (1, s, u) and takes into account that en(u s ) T0,0 (1, s, u) = T0,0 (1, s n, u + n) = 0
unless n = s. The same argument works for the second equality. Using (6.12), one can see that
the operator relations (6.41), (6.42) are identical to (4.10).
Once the functions TK,M (1, s, u) and TK,M (a, 1, u) are constructed, the other T -functions can
be expressed in terms of Qk,m (u) by either iterating Eq. (6.32) with respect to a or Eq. (6.33) with
respect to s. Therefore, setting successively a = 1, 2, . . . (s = 1, 2, . . .) in Eq. (6.32) (Eq. (6.33))
starting with a = 1 (s = 1), one can step by step express TK,M (a, s, u) with different a and s
in terms of Qk,m (u). Eqs. (6.32) and (6.33) solve the problem of the integration of the Hirota
equation for the case of rectangular paths. Using zero curvature conditions (6.10), one can easily
generalize these equations to the case of an arbitrary zigzag path, along the lines of Section 4.2.
7. Higher representations in the quantum space
In the previous sections, it is implied that zeros of the polynomial function (u) are in general position. This corresponds to an inhomogeneous spin chain in the vector representation of
386
gl(K|M) at each site. In principle, this includes all other cases such as spins in higher representations in the quantum space. Indeed, the higher representations can be constructed by fusing
elementary ones according to the fusion procedure outlined in Section 2. There, we have considered fusion in the auxiliary space but for the quantum space the construction is basically the
same. To get a higher representation at a site of the spin chain, one should combine several sites
of the chain carrying the vector representations, with the corresponding s being chosen in a
specific string-like way, and then project onto the higher representation. Before the projection,
the spin chain looks exactly like the ones dealt with in the previous sections. However, zeros of
the function (u) are no longer in general position.
The string-type boundary values impose certain requirements on the location of zeros of the
polynomials T (a, s, u). Indeed, the Hirota equation implies that some of these polynomials must
contain similar string-like factors. From the Hirota equation point of view, the projection onto
a higher representation means selecting a class of polynomial solutions divisible by factors of
this type. In fact, given the boundary values, different schemes of extracting such factors are
possible. They correspond to different types of fusion in the quantum space.
7.1. Symmetric representations in the quantum space
To be more specific, consider the simplest case of symmetric tensor representations (onerow Young diagrams). Fix an integer 1 and consider a combined site i consisting of sites
(labeled by the double index as (i, 1), . . . , (i, )) carrying the vector representation. According
to the fusion procedure, the corresponding parameters i,r form a string: i,r = i 2(r 1),
r = 1, . . . , . Therefore, the boundary values of the T -functions are given by T (0, s, u) = + (u
s), T (a, 0, u) = + (u + a), where

+ (u) = (u)(u + 2) u + 2( 1)
=2
N
N
i

( u
2 + )
i=1
i
( u
2 )
= 0, 1, 2, . . . .
(7.1)

Here (u) = N
i=1 (u i ), as before. The representation through the -function is useful for
the analytic continuation in .
A thorough inspection shows that the Hirota equation is consistent with extracting the following polynomial factor6 from T (a, s, u) for s = 0, 1, . . . , and a 1:
+
(u + s + a)T (a, s, u)
T (a, s, u) = s

N
( us+ai + )
2
T (a, s, u).
= 2(s)N
u+s+ai
(
)
2
i=1
(7.2)
Here T (a, s, u) is a polynomial of degree sN (if s = 0, 1, . . . , ). We extend the definition of

T (a, s, u) to higher values of s by setting T (a, s, u) = T (a, s, u) (s , a 1). Note that the
factor in the right-hand side of (7.2) is a product of functions depending separately on u + s + a
and u s + a. Therefore, if this relation between T and T was valid for all values of s, then
the T s would obey the same Hirota equation in the whole (a, s, u) space (see (2.28)). However,
6 We note that this factor here and in (7.5) below can be obtained directly from the fusion procedure in the quantum
space as the product of trivial zeros of the fused R-matrices.
387
since the definition of T is changed when s , the Hirota equation breaks down in the plane
s = . It is easy to see that it gets modified as follows:

T (a, s, u + 1)T (a, s, u 1) (u + s + a 1) s, T (a, s + 1, u)T (a, s 1, u)

+
(u s) a,1 T (a + 1, s, u)T (a 1, s, u).
= s+(s)(s)
(7.3)
The function (s ) is defined as (s ) = 1 if s and 0 otherwise, so the pre-factor in the

right-hand side equals s+ (u s) if 0 s and + (u s) if s (at a = 1). The boundary
conditions are T (a, 0, u) = T (0, s, u) = 1 (a, s 0). If the point (a, s) belongs to the interior
boundary, then the function T defined by Eq. (7.2) (and by Eq. (7.5) below) may contain some
additional zeros of a similar string-like type.
7.2. Antisymmetric representations in the quantum space
In the case of the antisymmetric fusion (one-column Young diagrams) the parameters i,r also
form a string: i,r = i + 2(r 1), r = 1, . . . , . Comparing to the symmetric fusion, this string
looks to the opposite direction, i.e., the sequence of s increases rather than decreases. The
two types of strings are actually equivalent since they are obtained one from the other by an
overall shift of i s. Our convention here is chosen to be consistent with the general case outlined
below.
The boundary values of the T -functions are given by T (0, s, u) = (u s), T (a, 0, u) =
(u + a), where

(u) = (u)(u 2) u 2( 1)
= 2N
N

i=1
i
( u
2 + 1)
i
( u
2 + 1)
= 0, 1, 2, . . . .
(7.4)
A thorough inspection shows that the Hirota equation is consistent with extracting the following
polynomial factor from T (a, s, u) for a = 0, 1, . . . , and s 1:
T (a, s, u) = a
(u s a)T (a, s, u)
N

( usai + 1)
2
= 2(a)N
T (a, s, u).
us+ai
(

+
1)
2
i=1
(7.5)
Here T (a, s, u) is a polynomial of degree aN (if a = 0, 1, . . . , ). We extend the definition of

T (a, s, u) to higher values of a by setting T (a, s, u) = T (a, s, u) (a , s 1). The modified
Hirota equation for T reads as follows:

(u + a) s,1 T (a, s + 1, u)T (a, s 1, u)
T (a, s, u + 1)T (a, s, u 1) a+(a)(a)

= (u s a + 1) a, T (a + 1, s, u)T (a 1, s, u).
(7.6)
The pre-factor in the second term in the left-hand side equals a (u + a) if 0 a and
(u + a) if a (at s = 1). The boundary conditions are T (a, 0, u) = T (0, s, u) = 1 (a, s 0).
388
Fig. 18. The intersection of Young diagrams and (s a ).
7.3. The general case: Remarks and conjectures

Let us consider the case of a general covariant representation at each site of the chain. We
construct such a site i by fusing || elementary sites with -parameters i + 2(p q), where
(p, q) . We remind the reader that the integer coordinates (p, q) Z2 on a Young diagram
are such that the row index p increases as one goes from top to bottom and the column index q
increases as one goes from left to right, and the top left box of has the coordinates (1, 1).
Given a diagram , we define the polynomial function
(u) =
N

u 2(p q) i =
u 2(p q) ,
i=1 (p,q)
(7.7)
(p,q)
then the boundary values of the T -functions are:

T (0, s, u) = (u s),
T (a, 0, u) = (u + a).
(7.8)
Let (a,
s) be the Young diagram obtained as the intersection of and the rectangular diagram (s a ) (Fig. 18):

(a,
s) s a .
For brevity, we will sometimes denote the cut diagram (a,
s) simply dropping the dependence

on a, s. Let us introduce the polynomial function

(u)
(u) =
u 2(p q) ,
=
\(a,s)
(7.9)
(a,s)
(u)
(p,q)\(a,s)
where the product goes over boxes of the skew diagram \ (a,
s). If is contained in the

(u) = 1.
rectangle (s a ), then we set \(a,s)
Now we are ready to present our first conjecture. We expect that the projection onto the representation in the quantum space means, for the Hirota equation, that we consider the solutions
:
such that the polynomial T (a, s, u) is divisible by the polynomial \(a,s)
(u s + a)T (a, s, u).

T (a, s, u) = \(a,s)
(7.10)
389
For one-row or one-column diagrams this formula yields equations (7.2) and (7.5). Presumably,
this conjecture can be proved by means of the technique developed in [36]. If the point (a, s)
belongs to the interior boundary, then the function T defined by Eq. (7.10) may contain some
additional zeros of a similar string-like type.
We note that the functions T (a, s, u) defined by (7.10) obey the modified Hirota equation:
T (a, s, u + 1)T (a, s, u 1)

1
s

(u + s + a 2j 1) T (a, s + 1, u)T (a, s 1, u)
=
j = s+1

a 1
j = a+1
Here the products like

(u s a + 2j + 1) T (a + 1, s, u)T (a 1, s, u).
n1
s = min(s , a),
j =n
(7.11)
are understood to be 1 and

a = min(a , s)
are respectively the lengths of the sth column and the ath row of the diagram = (a,
s). The
boundary conditions are T (a, 0, u) = T (0, s, u) = 1 (a, s 0). Note that the pre-factors in the
modified Hirota equation are equal to 1 if the rectangle (s a ) is contained in the diagram . In
the opposite case, when the diagram is contained in the rectangle (s a ), the functions T (a, s, u)
coincide with T (a, s, u) and the pre-factors are again equal to 1. Given Eq. (7.10), the derivation
of the modified Hirota equation for T s is straightforward. We present here two simple identities
which appear to be useful:

(u)
(a,s+1)
= (u + 2s)(u + 2s 2) u + 2s 2( s+1 1) ,
(a,s)
(u)

(u)
(a+1,s)
(7.12)
= (u 2a)(u 2a + 2) u 2a + 2( a+1 1) .
(a,s)
(u)
The next challenge is to find out how the polynomial factor extracted from the T -functions
behaves under the Bcklund transformations. Here is our second conjecture. At each step (k, m)
of the chain of the transformations BT1 and BT2 the same relation (7.10) holds,
Tk,m (a, s, u) = k,m \ k,m (a,s) (u s + a)Tk,m (a, s, u),
(7.13)
where the diagram k,m is obtained from = K,M by cutting off K k upper rows and M m
left columns. If K k exceeds the number of rows in the diagram , or M m exceeds the
number of columns in , then we set k,m = . In other words, the transformation BT1 cuts off
the upper row while BT2 cuts off the left column. The coordinates (p, q) on the diagrams k,m
are such that the top left box of any diagram has coordinates (1, 1).
At last, we would like to remark that instead of the function (u) one could use the function

(u) =
(7.14)
u + 2(p q)
(p,q)
and arrive to similar formulas. This can probably be explained by invoking the representation
theory of (super-)Yangians. It suggests7 that the function (u) corresponds to the represen7 We are grateful to V. Tarasov for a discussion on this point.
390
Fig. 19. Reduction of the Young diagram to k,m under K k applications of BT1 and M m applications of BT2.
The full diagram is depicted by thin lines (including the axis), the reduced one, k,m , by the thick lines.
tation associated with the usual Young diagram while the function (u) comes from the
representation associated with the reversed diagram regarded as a skew diagram, see the
first reference in [36]. (The reversed diagram is obtained from by inversion with respect to the
left top corner.) This point needs further clarification.
7.4. Bethe equations with non-trivial vacuum parts
It is known that when spins in the quantum space belong to a higher representation of the
symmetry algebra, Bethe equations (5.2)(5.5) acquire non-trivial right-hand sides (sometimes
called vacuum parts). We are going to show that they actually follow from the equations with
trivial vacuum parts if one partially fixes the roots of the polynomials Qk,m (u) in a special way.
Specifically, we set
Qk,m (u) = k,m (u)Q k,m (u),
(7.15)
which can be regarded as an ansatz suggested by the fusion procedure. Note that it is the particular
case of Eq. (7.13) at a = 0 or s = 0. Accepting this, we are going to substitute it into the QQ This allows us to derive the system of Bethe equations for the
relation to get an equation for Qs.
roots of Qs. The derivation itself does not depend on the validity of the conjectures given above.
To proceed, we need some more notation. Given a Young diagram , let l() = 1 be the
number of its rows or, equivalently, the length of the first row of the transposed diagram . The
short hand notation
lk,m = l(k,m ),

lk,m
= l(k,m )
(7.16)

)
is convenient. In other words, lk,m lk,m
is the minimal rectangle (of height lk,m and length lk,m
containing the diagram k,m . It is obvious that if k,m = , then (see Fig. 19)
lk,m = (k,m )1 = Mm+1 K + k,

= (k,m )1 = Kk+1 M + m.
lk,m
(7.17)
Now we are ready to substitute (7.15) into the Hirota equation for Qs (4.13). We have:
391
is modified in the rectangle 0 k K,

Fig. 20. The domain (shadowed region), where the Hirota equation for Qs
0 m M. The diagram is reflected with respect to its vertical boundary.
k,m (u) = k1,m1 (u)l+ (u)lk,m (u) 1 (u)

k,m
= k1,m (u 2)l+ (u)

k,m
= k,m1 (u + 2)lk,m (u),

provided k,m is not empty (otherwise the right-hand sides are equal to 1), where the string
polynomials l (u) are defined in (7.1) and (7.4). Using these obvious identities, it is straight Q-relation:
forward to obtain the Q

k1,m1 (u + 2)Q k,m (u)
(u + 2lk,m
)Q k1,m1 (u)Q k,m (u + 2) (u 2lk,m + 2)Q
k1,m (u)Q k,m1 (u + 2).

=Q
(7.18)
In this form it is valid if k,m = , otherwise the functions Q obey the standard Hirota equation (4.13). In other words, the Hirota equation gets modified in the region shown in Fig. 20.
The boundary of this region in the (k, m) plane is exactly the boundary of the diagram . The
K,M are fixed to be 1: Q
0,0 (u) = Q K,M (u) = 1.
0,0 and Q
functions Q
The Bethe equations are derived from (7.18) in the same way as in Section 5. They acquire
non-trivial right-hand sides which can be compactly written in terms of the quantities lk,m and
:
lk,m
(k,m)
(k,m)
(k,m)
)Qk,m (uj
2)Q k+1,m (uj
+ 2)
Q k1,m (uj
(k,m)
(k,m)
(k,m)
2)Q k,m (uj
+ 2)Q k+1,m (uj
)
Q k1,m (uj
(k,m)
(k,m)
(k,m)
)Qk,m (uj
2)Q k,m1 (uj
+ 2)
Q k,m+1 (uj
(k,m)
=
=
(uj
(k,m)

+ 2lk+1,m
)
(k,m)
2lk,m+1 )
(uj
(uj
(k,m)
(k,m)
(k,m)
(k,m)
2)Q k,m (uj
+ 2)Q k,m1 (uj
)
(uj
Q k,m+1 (uj

)Q k,m1 (u(k,m)
+ 2) (u(k,m)
+ 2lk+1,m
)
Q k+1,m (u(k,m)
j
j
j
=
,
(k,m)
(k,m)
(k,m)
Q k+1,m (uj
+ 2)Q k,m1 (uj
)
(uj
2lk,m )
(k,m)
(k,m)
(k,m)
)Qk1,m (uj
2) (uj
2lk,m+1 )
Q k,m+1 (uj
(k,m)
(k,m)
2)Q k1,m (uj
)
Q k,m+1 (uj
(k,m)
(uj
)
+ 2lk,m
)
+ 2lk,m
2lk,m )
(7.19)
(7.20)
(7.21)
(7.22)
392
These equations are listed here in the same order as Eqs. (5.2)(5.5). For empty diagrams, lk,m

and lk,m
are put equal to 0. Using formulas (7.17), we can represent the right-hand sides in a
more explicit but less compact form:
k1,m (u(k,m) )Q
k,m (u(k,m) 2)Q
k+1,m (u(k,m) + 2)
Q
j
j
j
k1,m (u(k,m) 2)Q
k,m (u(k,m) + 2)Q k+1,m (u(k,m) )
Q
j
j
j
=
(u(k,m)
+ 2m 2M + 2Kk+1 )
j
(k,m)
(uj
+ 2m 2M + 2Kk )
(7.23)
k,m+1 (u(k,m) )Q
k,m (u(k,m) 2)Q
k,m1 (u(k,m) + 2)
Q
j
j
j
k,m+1 (u(k,m) 2)Q
k,m (u(k,m) + 2)Q k,m1 (u(k,m) )
Q
j
j
j
=
(u(k,m)
2k + 2K 2Mm )
j
(k,m)
(uj
2k + 2K 2Mm+1 )
(k,m)
(k,m)
)Qk,m1 (uj
+ 2)
Q k+1,m (uj
(k,m)
k,m1 (u(k,m) )
Q k+1,m (uj
+ 2)Q
j
(k,m)
(k,m)
k1,m (u
)Q
2)
Q k,m+1 (uj
j
(k,m)
(k,m)
k1,m (u
2)Q
)
Q k,m+1 (uj
j
(7.24)
,
(k,m)
=
=
(uj
+ 2m 2M + 2Kk )
(k,m)
(uj
2k + 2K 2Mm+1 )
(k,m)
2k + 2K 2Mm )
(uj
.
(k,m)
(uj
+ 2m 2M + 2Kk+1 )
(7.25)
(7.26)
These equations are the building blocks to make up the chain of Bethe equations for any undressing path.
For example, the chain of the Bethe equations for the simplest path (K, M) (0, M)
(0, 0) is as follows. Moving down from (K, M) to (0, M), we have the equations
k+1,M (u(k,M) + 2)
k1,M (u(k,M) )Q k,M (u(k,M) 2)Q
Q
j
j
j
k+1,M (u(k,M) )
k1,M (u(k,M) 2)Q k,M (u(k,M) + 2)Q
Q
j
j
j
=
(u(k,M)
+ 2Kk+1 )
j
(k,M)
(uj
+ 2Kk )
(7.27)
where k = 1, 2, . . . , K 1. They agree with the chain of Bethe equations presented in [47,48]
for the bosonic case. At the corner, the equation is
(0,M)
(0,M)
)Q0,M1 (uj
+ 2)
Q 1,M (uj
(0,M)
(0,M)
+ 2)Q 0,M1 (uj
)
Q 1,M (uj
(0,M)
(uj
(0,M)
(uj
+ 2K )
+ 2K 21 )
(7.28)
Finally, moving to the left from (0, M) to (0, 0), we have the equations
0,m1 (u(0,m) + 2)
0,m+1 (u(0,m) )Q 0,m (u(0,m) 2)Q
Q
j
j
j
0,m1 (u(0,m) )
0,m+1 (u(0,m) 2)Q 0,m (u(0,m) + 2)Q
Q
j
j
j
(0,m)
(uj
(0,m)
(uj
+ 2K 2Mm )
+ 2K 2Mm+1 )
(7.29)
393
Fig. 21. Boundary conditions for the transfer matrix T (a, s, u) T1,1 (a, s, u) for the gl(1|1) super-algebra.
where m = 1, 2, . . . , M 1. One can see that the differences of arguments of the -functions
in the numerator and the denominator are equal to the doubled KacDynkin labels (A.2) corresponding to the diagram .
One can also write down the chain of equations for an arbitrary path, similarly to Eq. (5.8).
However, the general form of the right-hand sides is not very illuminating.
The last remark concerns the form of the ansatz (7.15). Instead of (7.15) one could use a
similar ansatz

Qk,m (u) = k,m u + 2(K k) 2(M m) Q k,m (u)
(7.30)
with the function (u) defined in (7.14). This leads to a similar system of Bethe equations with
non-trivial vacuum parts. There is a one-to-one correspondence between solutions of the both
systems.
8. Examples
8.1. Baxter and Bethe equations for gl(1|1)
Let us consider in detail the gl(1|1) case. In this case all elements of transfer matrix
T1,1 (a, s, u) T (a, s, u) lay on the boundaries of the fat hook domain, as we see in Fig. 21.
We read off from this figure:
T (0, 0, u) = Q1,1 (u) (u) =
(u j ),
j =1
T (a, 0, u) = Q1,1 (u + a),
a 0,
T (0, s, u) = Q1,1 (u s),
< s < ,
T (a, 1, u) = (1)
a1
Q1,0 (u + a + 1)Q0,1 (u a 1),
T (1, s, u) = Q1,0 (u + s + 1)Q0,1 (u s 1),
a 1,
s 1.
(8.1)
Using the Hirota equation (4.13) we get the only non-trivial QQ-relation:
Q0,0 (u)Q1,1 (u + 2) Q1,1 (u)Q0,0 (u + 2) = Q0,1 (u)Q1,0 (u + 2),
which gives, together with Q0,0 (u) = 1,
Q1,0 (u)Q0,1 (u 2) = (u) (u 2).
(8.2)
394
This relation allows us to exclude, for example, Q0,1 or Q1,0 from Eq. (8.1) and to obtain, in
particular, the Baxter T Q-relations:

Q0,1 (u s 1)
(u + s + 1) (u + s 1) ,
T (1, s, u) =
Q0,1 (u + s 1)

Q1,0 (u + s + 1)
T (1, s, u) =
(8.3)
(u s + 1) (u s 1) .
Q1,0 (u s + 1)
From the second relation, using regularity of the left-hand side, we obtain the following Bethe
equation:
1=
(uj(1,0) )
(1,0)
(uj
2)
(8.4)
It is a free fermion equation. We can also obtain it from the undressing procedure Q1,1
Q1,0 Q0,0 = 1, putting u = u(1,0) + s 1 in Eq. (8.2). For another undressing procedure,
Q1,1 Q0,1 Q0,0 = 1, or by canceling poles at u = u(0,1) s + 1 in the first equation in
(8.3), we get the Bethe equation
(0,1)
1=
(uj
(0,1)
(uj
+ 2)
(8.5)
Note that the Bethe equations (8.4), (8.5) do not depend on the representation parameter s
whereas the eigenvalues T (1, s, u) do depend on it. In principle, nothing prevents this parameter
to be continued analytically to the complex plane. This is precisely the continuous label of typical
(long) irreducible representations [21] of the super-algebra in auxiliary space.
Let us consider higher rectangular irreps in the physical space. To cover representations with
an arbitrary spin (one-row diagrams with boxes), we take a special form of the function (u),
as in Eq. (7.1):

(u) = (u)(u + 2) (u + 2) u + 2( 1) ,
(8.6)
describing a string of length . (Comparing to the previous section, wehave changed the notation slightly denoting the polynomial function with roots i by (u) = j (u j ).) According
to (7.15), we have:
Q0,1 (u) = Q 0,1 (u),

1,0 (u).
Q1,0 (u) = (u)(u + 2) u + 2( 2) Q
From the first equation in (8.3) we obtain the transfer matrix T1,1 (1, s, u) T (1, s, u):
T (1, s, u + 1) =

0,1 (u s)
1
Q
(u + s + 2) (u + s)
(u + s + 2j ).
0,1 (u + s)
Q
(8.7)
j =1
The last factor represents trivial zeros of T (a, s, u). It has the same origin as the one in Eq. (7.2).
However, one should note that because T (1, s, u) lives on the interior boundary, this factor contains more zeros than the one in Eq. (7.2) (see the remark in the end of Section 7.1). The Bethe
equation (8.5) becomes:
1=
(uj(0,1) )
(0,1)
(uj
+ 2)
(8.8)
395
It is easy to generalize these results to the case of an arbitrary physical spin l at each site
l = 1, 2, . . . , N of the chain. In this case we take
(u) =
N

(u l )(u l + 2) (u l + 2l 2).
l=1
This yields the transfer matrix

N

N

0,1 (u s)
Q
T (1, s, u + 1) =
(u l + s + 2l ) (u l + s)
0,1 (u + s)
Q
l=1
N
l 2

l=1
(u l + s + 2j ).
(8.9)
l=1 j =0
The last factor represents the trivial zeros which can be absorbed by normalization. Introducing
a new renormalized function T with the trivial zeroes removed,

T (1, s, u + 1) = T (1, s, u + 1)
N
l 1

(u l + s + 2j )
(8.10)
l=1 j =1
we obtain, redefining the parameters l = l + l ,

N

N

0,1 (u s)
Q

T (1, s, u + 1) =
(u l + s + l ) (u l + s l )
0,1 (u + s)
Q
l=1
(8.11)
l=1
which is essentially the same transfer matrix eigenvalue as for the rational limit of Eqs. (A.9),
(A.10) from [51], with the definition of the gl(1|1) S-matrix (3.1), (3.2) from [52]. The Bethe
equations are
1=
(0,1)
l + s + l
(0,1)
l + s l
uj
l(=j )=1
uj
(8.12)
For the complete comparison one should exchange in these papers the parameters s, s
r of the auxiliary and quantum spaces and take the rational limit. Some simple shifts in the
arguments and definitions of Bethe roots are also necessary. Note that unlike [51], where they
are the soliton charges, both representation labels and s can be now considered as continuous
parameters of the typical representation of gl(1|1). This corresponds to the limit of large charges
and large period of these soliton charges.
8.2. Baxter and Bethe equations for gl(2|1)
In the gl(2|1) case,8 the transfer matrices T2,1 (a, s, u) T (a, s, u) are on the boundaries
of the fat hook except T (1, s, u) in the middle row, as we see in Fig. 22. There are six Qfunctions Qk,m (u) (k = 0, 1, 2; m = 0, 1), two of them being fixed by the boundary conditions.
8 The construction of the Baxter Q-operators and T Q-relations for the models based on gl(2|1) (and U (gl(2|1)))
q
were recently discussed in [53,54]. We thank Z. Tsuboi for bringing these works to our attention.
396
Fig. 22. Boundary conditions for T (a, s, u) T2,1 (a, s, u) for the gl(2|1) super-algebra.
The T -functions are expressed through them in the following way:

T (0, 0, u) = Q2,1 (u) (u) =
(u j ),
j =1
T (a, 0, u) = Q2,1 (u + a),
a 0,
T (0, s, u) = Q2,1 (u s),
< s < ,
T (a, 1, u) = Q2,0 (u + a + 1)Q0,1 (u a 1)(1)a2 ,

T (2, s, u) = Q2,0 (u + s + 2)Q0,1 (u s 2),
a 2,
s 1.
(8.13)

8.2.1. Baxter equations for the KacDynkin diagram 1
(Here and below the subscript 1 means the fundamental representation in the quantum space.)
The operator generating series (4.15) reads

T (a, 1, u + a + 1)
a=0
Q2,1 (u + 2a + 2)
e2au
1
= U 1,0 U 2,0 V2,0

Q1,0 (u)
Q2,0 (u) Q1,0 (u + 2)
=
e2u
e2u
Q1,0 (u + 2)
Q2,0 (u + 2) Q1,0 (u)
1

Q2,0 (u) Q2,1 (u + 2)
2u
.
e
Q2,0 (u + 2) Q2,1 (u)
(8.14)
In particular,
T (2, 1, u + 3) Q2,0 (u + 6) (u + 4) Q1,0 (u + 4) Q2,0 (u + 6) (u + 2)
=
(u + 4)
Q2,0 (u + 4) (u + 6) Q1,0 (u + 2) Q2,0 (u + 4) (u + 6)
Q1,0 (u) Q2,0 (u + 6) (u + 2) Q2,0 (u + 6) (u)
+
. (8.15)
Q1,0 (u + 2) Q2,0 (u + 2) (u + 6) Q2,0 (u + 2) (u + 6)
From this equation and equation
T (2, s, u) =
Q2,0 (u + s + 2)
T (2, 1, u s + 1)
Q2,0 (u s + 4)
which follows from Eq. (8.13), we obtain
397
T (2, s, u)
(u s + 2)
Q2,0 (u + s + 2) (u s + 2) Q1,0 (u s + 2) Q2,0 (u + s + 2) (u s)
=
Q2,0 (u s + 2) (u s + 4)
Q1,0 (u s) Q2,0 (u s + 2) (u s + 4)
Q1,0 (u s 2) Q2,0 (u + s + 2) (u s)
Q2,0 (u + s + 2) (u s 2)
+
.
Q1,0 (u s)
Q2,0 (u s) (u s + 4)
Q2,0 (u s) (u s + 4)
(8.16)
Note that the representation index s can be treated as a continuous parameter. It corresponds to
the continuous label of typical representations of the gl(2|1) super-algebra.
8.2.2. Spins in higher irreps in the quantum space
As an illustrative example to Section 5, as well as for the purpose of comparison with the
result of [6] (Eq. (E.1) obtained from the specific S-matrix by the algebraic Bethe ansatz), we
consider here spins in the quantum space in the irrep (r 2 ) (Young diagrams with two rows of
length r). At the end, the number r will be treated as a continuous label.
Let us introduce new definitions, following Eq. (7.7) of the previous section:

(u) 2,1 (u) = (u 2) 2 (u) 2 (u + 2) 2 (u + 2r 4) (u + 2r 2),
(8.17)
2

2,0 (u) = (u 2) (u) 2 (u + 2) 2 (u + 2r 6) (u + 2r 4),
(8.18)
1,0 (u) = (u)(u + 2) (u + 2r 4).

Here (u) = j (u j ). Then, according to Eq. (7.15),
Q2,0 (u) = 2,0 (u)Q 2,0 (u),
Q1,0 (u) = 1,0 (u)Q 1,0 (u).
(8.19)
(8.20)
(8.21)
In this notation, Eq. (8.16) takes the form

T (2, s, u)
1,0 (u s + 2) Q
2,0 (u + s + 2) 1,0 (u s + 2) (u s)
Q 2,0 (u + s + 2) Q
=
1,0 (u s) Q
2,0 (u s + 2) 1,0 (u s) (u s + 2)
Q 2,0 (u s + 2)
Q
Q 1,0 (u s 2) Q 2,0 (u + s + 2) 1,0 (u s 2) 2,0 (u s + 2) (u s)
2,0 (u s)
1,0 (u s)
2,0 (u s) (u s + 2)
Q 1,0 (u s)
Q
Q 2,0 (u + s + 2) 2,0 (u s + 2) (u s 2)
,
+
2,0 (u s) (u s + 2)
Q 2,0 (u s)
where we redefined the transfer matrix extracting trivial zeros:
2 (u s + 2) 2,0 (u + s + 2)
T (2, s, u) = T (2, s, u)
(u s + 4) 2,0 (u s + 2)

= T (2, s, u) (u s)(u s + 2)2,0 (u + s + 2) .
(8.22)
For the reason already discussed in the case of gl(1|1), T is not equal to T of Eq. (7.10).
Calculating the ratios of -functions in Eq. (8.22), we find for the following result for the
transfer matrix in physical irrep (r 2 ) and the auxiliary irrep (s 2 ):
398
Tr (2, s, u) =
2,0 (u + s + 2) Q 1,0 (u s + 2) Q 2,0 (u + s + 2) (u s 2)

Q
2,0 (u s + 2)
Q
Q 1,0 (u s) Q 2,0 (u s + 2) (u s + 2r)
1,0 (u s 2) Q 2,0 (u + s + 2) (u s 2)
Q
Q 1,0 (u s)
Q 2,0 (u s) (u s + 2r)
2,0 (u + s + 2) (u s 4)
Q
(u s 2)
.
Q 2,0 (u s) (u s + 2r 2) (u s + 2r)
(8.23)
k,m (u + k m) and setting (u)
= (u + r 1)
Using the notation of Eq. (5.9) Q k+m (u) = Q
we bring Eq. (8.23) to the form
Tr (2, s, u) =
2 (u + s) Q 1 (u s + 1) Q 2 (u + s) (u
s r 1)
Q
2 (u s) Q 1 (u s 1) Q 2 (u s) (u
s + r + 1)
Q
1 (u s 3) Q
2 (u + s) (u
s r 1)
Q
1 (u s 1) Q
2 (u s 2) (u
s + r + 1)
Q
2 (u + s)
Q
(u
s r 3) (u
s r 1)
.
(u
s
+
r
1)
(u
s + r + 1)
Q2 (u s r 1)
(8.24)
This transfer matrix eigenvalue coincides, after shifting r r 1 and making an easy generalization to the inhomogeneous chain (i.e., to spins in arbitrary irreps at each site of the chain),
with Eq. (E.1) taken from [6] (see Appendix E, where their result is rewritten in our notation).
Note also that the spin label r, as well as s, can be treated as continuous parameters here. Hence
our method is general enough to describe the transfer matrices in all possible typical and atypical
irreps in quantum and auxiliary spaces.
(2,0)
(1,0)
+ s 2 and u = uj
+ s in Eq. (8.23), we write the
Canceling the poles at u = uj
following Bethe equations:
(2,0)
(uj
+ 2r 2)
(2,0)
(uj
1 =
4)
1,0 (u(2,0) )
Q
j
1,0 (u(2,0) 2)
Q
j
(1,0)
Q 1,0 (uj + 2)
2,0 (u(1,0) )
Q
j
(1,0)
(1,0)
Q 1,0 (uj 2) Q 2,0 (uj + 2)
(8.25)
or, in the notation of Eq. (8.24),

(2)
(u
j + r + 1)
(2)
(u
j r 1)
1 =
1 (u(2) + 1)
Q
j
1 (u(2) 1)
Q
j
(1)
2 (u(1) 1)
Q 1 (uj + 2) Q
j
(1)
Q 1 (u(1)
j 2) Q2 (uj + 1)
(8.26)
As an example, let us also consider two other equations for different choices of the Kac
Dynkin diagram.
399

8.2.3. Bethe equations for the KacDynkin diagram 1
Let us express T (1, s, u) through Qs using Eq. (4.10):

T (a, 1, u + a + 1)
a=0
Q2,1 (u + 2a + 2)
e2au
1
= V0,0
U1,1 U 2,1

1

Q0,1 (u + 2)
Q1,1 (u) Q0,1 (u + 2)
2u
2u
=
e
e
Q0,1 (u)
Q1,1 (u + 2) Q0,1 (u)

Q2,1 (u) Q1,1 (u + 2)
e2u .
Q2,1 (u + 2) Q1,1 (u)
(8.27)
In particular,
Q1,1 (u)
Q0,1 (u) (u + 2) Q1,1 (u + 4)
T (1, 1, u + 2)
=
(u + 4)
Q1,1 (u + 2) Q0,1 (u + 2) (u + 4) Q1,1 (u + 2)
Q0,1 (u) (u + 2)
.
+
Q0,1 (u + 2) (u + 4)
(8.28)
Canceling the poles in Eq. (8.28), it is straightforward to write the following Bethe equations for
gl(2|1) super-algebra:
(1,1)
(uj
(1,1)
(uj
1=
+ 4)
+ 2)
(1,1)
Q0,1 (uj
(1,1)
Q0,1 (uj
(0,1)
+ 4)
(0,1)
+ 2)
Q1,1 (uj
Q1,1 (uj
+ 2)
(1,1)
Q1,1 (uj
+ 4)
(1,1)
Q1,1 (uj )
(8.29)

8.2.4. Bethe equations for the KacDynkin diagram 1
Using Eq. (4.16) or Eq. (4.15), we write the operator generating series:

T (a, 1, u + a + 1)
a=0
Q2,1 (u + 2a + 2)
e2au
1
U2,1
= U 1,0 V1,0

1
Q1,0 (u)
Q1,0 (u) Q1,1 (u + 2)
2u
2u
=
e
e
Q1,0 (u + 2)
Q1,0 (u + 2) Q1,1 (u)

Q2,1 (u) Q1,1 (u + 2)
e2u .
Q2,1 (u + 2) Q1,1 (u)
(8.30)
In particular,
Q1,0 (u + 4) (u + 2) Q1,0 (u + 4) Q1,1 (u) (u + 2)
T (1, 1, u + 2)
=
+
(u + 3)
Q1,0 (u + 2) (u + 4) Q1,0 (u + 2) Q1,1 (u + 2) (u + 4)
Q1,1 (u)
.
Q1,1 (u + 2)
(8.31)
400
Canceling the poles in Eq. (8.28), we write the following Bethe equations:
(1,0)
1=
Q1,1 (uj
(1,0)
Q1,1 (uj
2)
(1,1)
(uj
(1,1)
(uj
,
(1,1)
+ 2)
Q1,0 (uj
(1,1)
Q1,0 (uj
+ 2)
(8.32)
8.3. Examples of the integration algorithm

To illustrate the general algorithm of Section 6, we apply it to gl(2), gl(2|1) and gl(2|2)
(super-)algebras.
8.3.1. gl(2) algebra
In this case Eq. (6.36) reads
T2,0 (1, s, u) = H 21
,0 (0, s, u)T1,0 (1, s, u),
0 s < .
(8.33)
Substituting explicit expressions for H 21

,0 (0, s, u) (6.19) and T1,0 (1, s, u) (6.34), we immediately obtain the explicit formula for non-vanishing functions which do not belong to the boundaries:
T2,0 (1, s, u + s 1) = Q1,0 (u + 2s)Q1,0 (u 2)
s
Q2,0 (u + 2j )
.
Q1,0 (u + 2j 2)Q1,0 (u + 2j )
j =0
(8.34)
8.3.2. gl(2|1) super-algebra

In this case Eq. (6.36) reads
T2,1 (1, s, u) = H 2,0+ (0, s, u)H 21
,0 (0, s, u)T1,0 (1, s, u)
H 2,0+ (0, s, u)T2,0 (1, s, u),
0 s < .
(8.35)
Substituting explicit expressions for H 2,0+ (0, s, u) (6.11) and T2,0 (1, s, u) (8.34), we obtain,
after straightforward calculations, the following result for non-vanishing functions which do not
belong to the boundaries:
T2,1 (1, s, u + s 1)

s1

Q2,0 (u + 2j + 2)
Q2,1 (u)
= Q1,0 (u + 2s)
+ Q1,1 (u 2)
(s 1) .
Q1,0 (u)
Q1,0 (u + 2j )Q1,0 (u + 2j + 2)
j =0
(8.36)
The step function (s) = 1 at s 0 and 0 otherwise. In the calculation, we have used Eq. (3.45)
to substitute
Q1,0 (u 2)Q2,1 (u) Q1,0 (u)Q2,1 (u 2)
= Q1,1 (u 2).
Q2,0 (u)
401
8.3.3. gl(2|2) super-algebra

In this case Eqs. (6.36) and (6.37) read
T2,2 (1, s, u) = H 2,1+ (0, s, u)H 2,0+ (0, s, u)H 21
,0 (0, s, u)T1,0 (1, s, u)
H 2,1+ (0, s, u)H 2,0+ (0, s, u)T2,0 (1, s, u)
H 2,1+ (0, s, u)T2,1 (1, s, u),
0s<
(8.37)
and
1
T2,2 (a, 1, u) = H 1+ ,2 (a, 0, u)H 0+ ,2 (a, 0, u)H 0,2
(a, 0, u)T0,1 (a, 1, u),
0 a < .
(8.38)
Substituting explicit expressions for the shift operators, we finally obtain:

T2,2 (1, s, u + s 1)

= Q1,0 (u + 2s)
Q2,2 (u) Q2,2 (u)Q1,1 (u 2)Q2,0 (u + 2)

+
(s 1)
Q1,0 (u)
Q2,1 (u)Q1,0 (u)Q1,0 (u + 2)
Q2,2 (u 2)Q2,1 (u + 2)
(s 1)
Q2,1 (u)Q1,0 (u + 2)
+ Q1,2 (u 2)
s2

j =0

Q2,0 (u + 2j + 4)
(s 2)
Q1,0 (u + 2j + 2)Q1,0 (u + 2j + 4)
(8.39)
and
T2,2 (a, 1, u a + 1)

= Q0,1 (u 2a)
+
Q2,2 (u) Q2,2 (u)Q1,1 (u + 2)Q0,2 (u 2)

+
(a 1)
Q0,1 (u)
Q1,2 (u)Q0,1 (u)Q0,1 (u 2)
Q2,2 (u + 2)Q1,2 (u 2)
(a 1)
Q1,2 (u)Q0,1 (u 2)
+ Q2,1 (u + 2)
a2

j =0

(1)j Q0,2 (u 2j 4)
(a 2) .
Q0,1 (u 2j 2)Q0,1 (u 2j 4)
(8.40)
In the calculation, we have used Eq. (3.45) to substitute

Q1,1 (u 2)Q2,2 (u) Q1,1 (u)Q2,2 (u 2) = Q1,2 (u 2)Q2,1 (u)
and
Q0,1 (u)Q1,2 (u + 2) Q0,1 (u + 2)Q1,2 (u)
= Q1,1 (u + 2).
Q0,2 (u)
The functions T2,2 (1, s, u) and T2,2 (a, 1, u) form a complete set of non-vanishing T -functions
which do not lie on the boundaries of the fat hook domain for the case of the gl(2|2) superalgebra.
402
9. Discussion
In this paper we have dealt with finite dimensional representations of gl(K|M) and the periodic spin chains based on the rational R-matrices. We believe that the method is powerful
enough to incorporate various generalizations, such as extension to twisted and open spin chains,
to infinite-dimensional representations of non-compact version of the symmetry group as well as
to models with more general R-matrices, including various exotic ones, as the Hubbard R-matrix
or the recently constructed AdS/CFT S-matrix. We expect that the Hirota relation should be the
same for all these problems, with only difference being in the boundary and analyticity conditions.
For example, the extension to spin chains with twisted boundary conditions can be accomplished by replacingthe polynomial Q-functions by Bloch polynomials, i.e., functions of the
form Q(u) = Aeu j (u uj ), where is related to the twisted boundary condition. This leads
to appropriate simple modifications in the Bethe equations.
The extension to supersymmetric spin chains with trigonometric R-matrices is also straightforward. The Hirota equation and the boundary conditions remain the same. The only change is
again in the analytical properties of the T -functions: in the trigonometric case they are trigonometric polynomials, i.e., finite products of sin((u uj )) or sinh((u uj )). A hypothetical
generalization to the elliptic case is much more interesting. As far as we know, supersymmetric
quantum spin chains with elliptic R-matrices were never discussed in the literature. On the other
hand, the bosonic case suggests [16] that the functional relations between commuting integrals
of motion are given by the same Hirota equation. The solutions are sought in the form of elliptic
polynomials, i.e., finite products of Jacobi theta-functions ((u uj )| ). From this point of
view, it would be interesting to analyze elliptic polynomial solutions to the Hirota equation with
boundary conditions of the fat hook type. They might solve (as yet hypothetical) supersymmetric
quantum integrable models with elliptic R-matrices.
It is also important to elaborate, within the framework of Hirota equations, a more direct
approach to spin chains in typical representations of super-algebras (the ones having a continuous
label). One would like to understand them not only in the sense of analytic continuation with
respect to the representation label (as we demonstrated in this paper by simple examples) but
also to find their place on other levels of the construction. A characterization of solutions to
the Hirota equation that are responsible for spin chains in typical representations would be of
particular importance.
Acknowledgements
We would like to thank I. Cherednik, S. Khoroshkin, I. Kostov, I. Krichever, P. Kulish, M. Nazarov, K. Sakai, D. Serban, V. Tarasov, V. Tolstoy, Z. Tsuboi and P. Vieira for discussions at
different stages of this work. The work of V.K. has been partially supported by European Union
under the RTN contracts MRTN-CT-2004-512194 and by the ANR program INT-AdS/CFTANR36ADSCSTZ. A.S. would like to thank the LPTENS for the hospitality during his visit. The
work of A.S. was partially supported by the RFBR Grant No. 06-01-00627-a, RFBR-DFG Grant
No. 06-02-04012-a, DFG Grant 436 RUS 113/669-3, the Program for Supporting Leading Scientific Schools (Grant No. NSh-5332.2006.2), and the HeisenbergLandau Program. A.Z. thanks
LPTENS, where most of the work was done, for the hospitality during his visit. The work of A.Z.
was partially supported by RFBR grant 06-02-17383, by grant RFBR-06-01-92054-CEa , by the
403
grant for support of scientific schools Nsh-8004.2006.2, by the ANR project GIMP No. ANR05-BLAN-0029-01, by INTAS 03-51-6346 and by grant NWO 047.017.015.
Appendix A. Super-algebras and their representations
For completeness, we list here some basic objects and notation related to Lie super-algebras
gl(K|M) and their representations. The main references are [1922].
We use the notation p() for the Z2 -grading of the index : p() = 0 if is bosonic and
p() = 1 if is fermionic. The same notation is used to denote the grading of any objects
(vectors, operators, . . . ) with definite parity. Objects with definite parity are called homogeneous.
A matrix A with matrix elements Aij (we imply that they are usual numbers, i.e., p(Aij ) = 0)
is said to be even (odd) if p(i) + p(j ) is even (odd) for all non-vanishing elements of A. The
R-matrix (2.1) is an even matrix.
Let X, Y be two graded spaces,
, fj be the corresponding (homogeneous) basis
and let ei
vectors. For any two vectors x = i xi ei , y = i yi fi we have

(xi ei ) (yj fj ) =
(1)p(i)p(j ) xi yj (ei fj ),
x y =
i,j
i,j
so the components of the vector x y in the basis ei fj are (1)p(i)p(j ) xi yj . The action of the operator A B in X Y is defined on homogeneous objects as A B(x y) =
(1)p(x)p(y) A(x) B(y). Matrix elements of the tensor product of even matrices Aij , B are:
p() p(i)+p(j ) i
Aj B .
(A B)i
j = (1)
This rule explains the origin of the sign factors in the graded YangBaxter equation.
The Lie super-algebra gl(K|M) can be most transparently defined through its matrix realization: it is the set of block matrices

A B
g=
(A.1)
C D
such that A, B, C, D are respectively K K, K M, M K and M M matrices. The even
subalgebra gl(K|M)0 has B = C = 0, the odd subalgebra gl(K|M)1 has A = D = 0. For homogeneous elements, the bracket is given by

[g, g ] = gg (1)p(g)p(g ) g g.
Note that gl(K|M)0 = gl(K)gl(M). For elements g realized as above the supertrace is defined
by str g = tr A tr D.
A basis for gl(K|M) consists of matrices Eij with entry 1 at position (i, j ) and 0 otherwise.
A Cartan subalgebra of gl(K|M) is spanned by the elements Eii , i = 1, . . . , K + M. The set of
generators of gl(K|M) consists of the Eii and the elements Ei,i+1 and Ei+1,i , i = 1, . . . , K +
M 1. The space dual to the Cartan subalgebra is spanned by the linear forms i : g Aii
(i = 1, . . . , K) and i : g Dii (i = 1, . . . , M), where g is given by (A.1). Let us choose the
basis in the space dual to the Cartan subalgebra to be 1 , . . . , K , 1 , . . . , M . On this space there
is a bilinear form induced by the supertrace in the super-algebra:
(i |j ) = ij ,
(i |j ) = (i |j ) = 0,
(i |j ) = ij .
Even (bosonic) roots are i j and i j (i = j ), odd (fermionic) roots are (i j ). There
are several choices of simple root systems. The distinguished simple root system has the form
404
i = i i+1 for i = 1, . . . , K 1, K = K 1 , K+j = j j +1 for j = 1, . . . , M 1. All

other simple root systems are obtained from this one by reflections with respect to odd roots
with (|) = 0.
Elements of the space dual to the Cartan subalgebra are called the weights. The weight is an
expression of the form
=
K
i i +
i=1
M
j j .
j =1
Let be a Young diagram formed by a sequence of non-negative non-increasing integers:

= (1 , 2 , . . .), 1 2 0. To such a diagram one assigns the weight i = i
(i = 1, . . . , K), j = j (j = 1, . . . , M), where j = max(j K, 0) and j is the height of the
j th column of the diagram . It is implied that K+1 M.
A KacDynkin label is a sequence b1 , b2 , . . . , bK , . . . , bK+M1 , where all bj except bK are
non-negative integers while bK may be any real number. There is a one-to-one correspondence
between finite-dimensional irreducible representations (irreps) of the super-algebra gl(K|M) and
the KacDynkin labels [19]. We consider covariant tensor irreps of the super-algebra gl(K|M).
One can assign the following KacDynkin label to any highest weight associated with a Young
diagram as above:
bi = i i+1 ,
bK = K + 1 ,
i = 1, . . . , K 1,
bj +K = j j +1 ,
j = 1, . . . , M 1.
(A.2)
Therefore, one associates a tensor irrep of gl(K|M) to any Young diagram. (The diagrams containing a rectangular subdiagram with K + 1 rows and M + 1 columns correspond to vanishing
representations, so such diagrams are illegal, similarly to diagrams containing K + 1 rows for
gl(K).) However, for super-algebras this correspondence is not one-to-one. Different Young diagrams may correspond to equivalent irreps (i.e., to the same KacDynkin label). In particular,
this is the case for rectangular diagrams9 when M + n columns of K boxes are replaced by M
columns of K + n boxes [21].
There is a large class of finite-dimensional irreps of gl(K|M) which cannot be associated
with a Young diagram. Given an irreducible tensor representation with the highest weight ,
there is
a one-parametric family of finite-dimensional irreps with the highest weight (c) =
+c K
i=1 i , where c is a real parameter. This yields the KacDynkin label with a non-integer
bK = K + 1 + c.
One distinguishes typical and atypical irreps [19] (in physical terminology, long and short
irreps, respectively). An irrep with the highest weight is atypical iff there exists at least one
pair (i, j ), i = 1, . . . , K, j = 1, . . . , M, such that
( + , i j ) = 0,
(A.3)
where
=
1
1
(K M 2i + 1)i +
(K + M 2j + 1)j .
2
2
K
i=1
j =1
9 For brevity, we call irreps corresponding to the Young diagrams of rectangular shape rectangular irreps.
405
For typical irreps, there is no such pair (i, j ) that (A.3) holds. All irreps with bK M are typical.
All rectangular irreps corresponding to diagrams having a rows and s columns with a < K or
s < M are atypical.
Appendix B. Auxiliary linear problems for the Hirota equation
Let = (p1 , p2 , p3 ) be a function of three variables. For brevity, we denote
(p1 + 1, p2 , p3 ) := 1 ,
(p1 , p2 + 1, p3 ) := 2 ,
(p1 + 1, p2 + 1, p3 ) := 12 ,
etc.
Let be any cyclic permutation of 123. Consider the following system of three linear equations for a function = (p1 , p2 , p3 ):

= 123, 231, 312,

= e ,
e +
(B.1)

where are parameters and /p . Their compatibility implies the Hirota equation for :
1 1 23 + 2 2 13 + 3 3 12 = 0.
(B.2)
To see this, consider the first and the second linear equations. Their compatibility means that
the difference operators

12
23
1
2
3
2
e
e 3
and e
e + 1
1 2
2 3
commute. A simple calculation shows that their commutativity is equivalent to the condition that
the function
1 1 23 + 3 3 12
:= h(p1 , p3 )
2 13
does not depend on p2 . Compatibility with the third linear problem implies that the function
h must be a constant equal to 3 , whence the Hirota equation follows. In the case when the
function h can be fixed in some other way (for example, from the boundary conditions), just
two linear problems are enough to represent the Hirota equation as a discrete zero curvature
condition. Moreover, the specific form of the coefficient functions in the difference operators
(B.1) implies that the compatibility follows from the existence of just one non-trivial common
solution (cf. [44]).
In terms of the function = the linear problems acquire the form
+ = 0,
= 123, 231, 312.
(B.3)
From the first and the second ones we have

2 =
3 2 + 1 23
,
3
1 =
3 1 2 13
.
3
Plugging this into the Hirota equation, we obtain another linear problem compatible with the
previous ones:
1 23 1 + 2 13 2 + 3 12 3 = 0.
(B.4)
406
The four linear problems can be combined into a single matrix equation as follows [43]:

1
2
1 23
0
3

0
3
1
2 13 2
= 0.
2
1
0
3 12 3
1 23
2 13
3 12
(B.5)
The determinant of the antisymmetric matrix in the left-hand side is equal to (1 1 23 +2 2 13 +

3 3 12 )2 . It vanishes if obeys the Hirota equation, and the rank of the matrix is 2 in this case,
so only two of the four equations are linearly independent.
One may regard system (B.3) as linear equations for with coefficients . Shifting the variables p p 1, p p 1, and then passing to the new variables p1,2,3 p1,2,3 , one
sees that the form of this system remains the same. Since the Hirota equation is invariant under
the simultaneous change of the signs of all variables, the compatibility condition is the same
Hirota equation for :
1 23 1 + 2 13 2 + 3 12 3 = 0.
(B.6)
Let us pass to the laboratory variables x1 , x2 , x3 according to the formulas

1
p1 = (1 x1 + 2 x2 + 3 x3 ),
2
1
p3 = (1 x1 + 2 x2 3 x3 ),
2
1
p2 = (1 x1 2 x2 + 3 x3 ),
2
where = 1 is a fixed set of signs (clearly, there are 23 = 8 possible choices). The inverse
transformation reads
x1 = 1 (p2 + p3 ),
x2 = 2 (p1 + p3 ),
x3 = 3 (p1 + p2 ).
We introduce the functions T (x1 , x2 , x3 ) = (p1 , p2 , p3 ) and F (x1 , x2 , x3 ) = (p1 , p2 , p3 ),

where the variables x and p are connected by the formulas given above. In the laboratory
variables, the system of four linear problems takes the form
F23
0
T12
T13
1 T1123
T
0
T23
2 T1223
12
F13
(B.7)
= 0,
T13
T23
0
3 T1233 F12
1 T1123
2 T1223
3 T1233
where we denote T1 T (x1 +1 , x2 , x3 ), T12 T (x1 +1 , x2 +2 , x3 ), T1123 T (x1 +21 , x2 +

2 , x3 + 3 ), etc. (and similarly for F ).
The compatibility of these linear problems implies the Hirota equation
1 T1123 T23 + 2 T1223 T13 + 3 T1233 T12 = 0.
Shifting the variables (x x ), we get the equation
1 T (x1 + 1 , x2 , x3 )T (x1 1 , x2 , x3 ) + 2 T (x1 , x2 + 2 , x3 )T (x1 , x2 2 , x3 )
+ 3 T (x1 , x2 , x3 + 3 )T (x1 , x2 , x3 3 ) = 0.
Note that it is the same for any choice of the s. Linear equations (B.7) provide inequivalent
Bcklund transformations for it. However, only four of them (corresponding, say, to the choices
407
1 = 2 = 3 = 1, 1 = 2 = 3 = 1, 1 = 2 = 3 = 1, 1 = 2 = 3 = 1) are actually
different because the simultaneous change of signs of all s means passing to the conjugate
system of linear problems, where the roles of T and F are interchanged. Eq. (3.13) in the main
text corresponds to the choice x1 = a, x2 = s, x3 = u, 1 = 2 = 3 = 1, 1 = 2 = 3 = 1.
Appendix C. On bilinear equations (3.38)(3.43) and their compatibility with (3.29), (3.30)
Here we complete the proof of Eq. (3.38) and present some more details on compatibility of
bilinear equations (3.29), (3.30) and (3.38)(3.43).
As we have seen, the pair of Eqs. (3.33), (3.34) implies the relation
= fk,m (a, u + s)Tk+1,m (a, s, u + 1)Tk,m+1 (a, s + 1, u),
(C.1)
where fk,m (a, u + s) is an arbitrary function of k, m and a, u + s. In the same way, the pair of
Eqs. (3.35), (3.36) implies the relation
= gk,m (s, u a)Tk,m+1 (a 1, s, u)Tk+1,m (a, s, u + 1),
(C.2)
where gk,m (s, u a) is an arbitrary function of k, m and s, u a. On the other hand, we have
the identity
(C.3)
Tk,m+1 (a, s + 1, u)Tk+1,m (a, s, u + 1)
,
=
Tk,m+1 (a 1, s, u)Tk+1,m (a, s, u + 1)
(C.4)
which is straightforwardly proved by passing to the common denominator, grouping together
similar terms and using Eqs. (3.29), (3.30). Its left-hand side is fk,m (a, u + s), and the right-hand
side is gk,m (s, u a). Therefore, we conclude that
fk,m (a, u + s) = gk,m (s, u a).
This equality is possible only if fk,m (a, u + s) actually depends on the difference of its arguments, i.e., on u + s a. This fact has been used in Section 3.4 to prove that fk,m (a, u + s) = 1.
The system of bilinear equations (3.38)(3.43) can be represented in the form
Tk+1,m+1 (a, s, u)
Tk,m+1 (a, s + 1, u)Tk+1,m (a, s, u + 1)

,
Tk,m (a, s, u)
Tk,m (a, s + 1, u)Tk,m (a, s, u + 1)
Tk+1,m+1 (a, s, u) Tk,m+1 (a, s, u 1)Tk+1,m (a + 1, s, u)

=
,
e a eu
Tk,m (a, s, u)
Tk,m (a, s, u 1)Tk,m (a + 1, s, u)
Tk+1,m+1 (a, s, u) Tk,m+1 (a, s + 1, u)Tk+1,m (a + 1, s, u)

=
,
e a e s
Tk,m (a, s, u)
Tk,m (a, s + 1, u)Tk,m (a + 1, s, u)
Tk+1,m+1 (a, s, u) Tk,m+1 (a 1, s + 1, u 1)Tk+1,m (a + 1, s, u)

=
,
e s eu
Tk,m (a, s, u)
Tk,m (a, s, u 1)Tk,m (a, s + 1, u)
Tk+1,m+1 (a, s, u) Tk,m+1 (a, s + 1, u)Tk+1,m (a + 1, s 1, u + 1)

=
,
e a e u
Tk,m (a, s, u)
Tk,m (a, s, u + 1)Tk,m (a + 1, s, u)
e u e s
(C.5)
(C.6)
(C.7)
(C.8)
(C.9)
408
es ea
Tk+1,m+1 (a, s, u)
Tk,m (a, s, u)
Tk,m+1 (a 1, s + 1, u 1)Tk+1,m (a, s, u + 1)

. (C.10)
Tk,m (a 1, s, u)Tk,m (a, s + 1, u)
These equations have a similar structure: different commuting difference operators in the lefthand sides act on the same fraction of the T -functions. It is a simple exercise to verify that these
equations are self-consistent if Eqs. (3.29), (3.30) are satisfied. The self-consistency conditions
obviously follow from the commutativity of the difference operators in the left-hand sides. Now,
let us demonstrate that all the equations from (3.29), (3.30) are encoded in the system (C.5)
(C.10). To this end, we subtract the sum of Eqs. (C.5) and (C.9) (Eqs. (C.7) and (C.8)) from
Eq. (C.7) (Eq. (C.6)) and find that the resulting equation reproduces the first equation from (3.30)
(respectively, the second equation from (3.29)). The remaining second (first) equation from BT2
(3.30) (BT1 (3.29)) results from the consistency condition of Eqs. (C.6) and (C.8) (respectively,
Eqs. (C.5) and (C.9)).
The connections between the different bilinear relations for the transfer matrices Tk,m (a, s, u)
become more transparent if one represents them in a matrix form. Namely, two pairs of
Eqs. (3.38), (3.42) and (3.39), (3.41) can be rewritten as

Tk,m (a, s + 1, u) Tk+1,m+1 (a, s + 1, u)
Tk+1,m+1 (a, s, u + 1)
Tk,m (a + 1, s, u) Tk+1,m+1 (a + 1, s, u)
Tk,m (a, s, u + 1)

Tk+1,m (a, s, u + 1)
= Tk,m+1 (a, s + 1, u)
(C.11)
Tk+1,m (a + 1, s 1, u + 1)
and

Tk,m (a, s + 1, u) Tk+1,m+1 (a, s + 1, u)
Tk+1,m+1 (a, s, u 1)
Tk,m (a + 1, s, u) Tk+1,m+1 (a + 1, s, u)
Tk,m (a, s, u 1)

Tk,m+1 (a 1, s + 1, u 1)
= Tk+1,m (a + 1, s, u)
,
Tk,m+1 (a, s, u 1)
(C.12)
respectively. Then, multiplying both sides of Eqs. (C.11) and (C.12) by the matrices inverse
to the ones in their left-hand sides (for the calculation of the determinant Eq. (3.40) is used),
the resulting equations reproduce respectively the first and the second equations from the linear
systems BT1 and BT2 (3.29), (3.30). Similarly, representing the two equations of the last line of
the matrix equation (3.32) in the matrix form

Tk+1,m+1 (a 1, s, u) Tk+1,m+1 (a, s, u 1)
cTk+1,m (a + 1, s, u)
Tk,m (a 1, s, u)
= Tk+1,m (a, s 1, u)
Tk,m (a, s, u 1)

Tk+1,m+1 (a, s + 1, u)
Tk,m (a, s + 1, u)
Tk+1,m (a, s, u + 1)
,
(C.13)
multiplying the both sides by the matrix inverse to the one in the l.h.s., and using Eq. (3.42) to
calculate the determinant of this matrix, one arrives at Eqs. (3.41), (3.43).
Appendix D. An alternative derivation of the Bethe equations
Here we give a direct derivation of the Bethe equations from the pair of linear problems
(Bcklund transformation) (3.29). It does not use the Hirota equation for the Q-functions.
We know that we can use the k (respectively, m) flow of the BT1 (respectively, BT2) to
decrease rank K (respectively, M) of the original problem to K 1 (respectively, M 1). For
the step by step passing from (K, M) to (0, 0) we can choose any zigzag path of the type drown
409
in Fig. 17. The interior boundaries of the domain of non-vanishing T s move towards the exterior
ones until the domain collapses to the horizontal and vertical lines, so that the original problem
gets undressed. The final solution can be formulated in terms of Bethe equations. Let us derive
them for the simplest path with just one turn.
First, using the transformation BT1 at each step, we move the horizontal interior boundary
a = K, s M to the half-line a = 0, s M, as shown in Fig. 10. Consider the second equation
from Eq. (3.29) at s = 0 (Fig. 12, position 4):
Tk,M (a, 1, u + 1)Qk1,M (u + a) Qk,M (u + a)Tk1,M (a, 1, u + 1)
= Qk,M (u + a + 2)Tk1,M (a 1, 1, u),
(D.1)
where k = 1, . . . , K, or, denoting Pka (u) Tk,M (a, 1, u a + 1), Qk,M (u) Qk (u):
a1
a
(u)Qk (u) = Pk1
(u 2)Qk (u + 2),
Pka (u)Qk1 (u) Pk1
k = 1, . . . , K.
(D.2)
We know that Pka (u) and Qk (u) should be polynomials in u. We set:

Qk (u) =
Jk

(k)
u uj
(D.3)
j =1
Taking Eq. (D.2) at zeroes of Qk1 (u), Qk (u) and Qk (u + 2), we obtain the equations
(k1) (k1)
(k1)

a1 (k1)
a
Pk1
uj
Qk uj
= Pk1
uj
2 Qk uj
+2 ,
(k)
(k)
(k)

a1 (k)
Pka uj Qk1 uj = Pk1
uj 2 Qk uj + 2 ,
(k)
(k)
(k)

(k)

a
Pka uj 2 Qk1 uj 2 = Pk1
uj 2 Qk uj 2 .
(D.4)
Dividing the second equation (with the shift a a + 1) by the third one and excluding the ratio
of P s with the help of the first equation, we arrive at the following standard system of Bethe
equations:
(t)
(t)
Qt+1 (u(t)
j + 2)Qt (uj 2)Qt1 (uj )
(t)
(t)
Qt+1 (u(t)
j )Qt (uj + 2)Qt1 (uj 2)
= 1
(j = 1, . . . , Jt , t = 1, . . . , K 1). (D.5)
Then, using the transformation BT2 Eq. (3.30) at each step, we move (as shown in Fig. 14)
the vertical interior boundary s = M, a 0 to the half-line s = 0, a 0. Consider the second
equation of (3.30) at s = 0 (Fig. 12, position 4). In a similar way, we get the Bethe equations for
the roots u(mM)
of Q0,m (u) := QmM (u):
j
(t)
(t)
(t)
Qt+1 (uj )Qt (uj 2)Qt1 (uj + 2)

(t)
(t)
(t)
Qt+1 (uj 2)Qt (uj + 2)Qt1 (uj )
= 1
(j = 1, . . . , Jt , t = 1, . . . , M + 1).
(D.6)
(0)
uj
To get the missing equation for zeros

of the function Q0 (u) = Q0,M (u), we use the second
equation of BT1 at k = 1, m = M, and the second equation of BT2 at k = 0, m = M:
P1a (u)Q0 (u) P0a (u)Q1 (u) = P0a1 (u 2)Q1 (u + 2),
R1a (u)Q0 (u) P0a (u)Q1 (u) = P0a1 (u 2)Q1 (u + 2),
(D.7)
410
where P1a (u) = T1,M (a, 1, u a + 1), R1a (u) = T0,M1 (a, 1, u a + 1). At the fermionic roots
J
(0)
of Q0 (u) = j 0=1 (u uj ) the ratio of these two equations gives the missing system of Bethe
equation:
(0)
Q1 (u(0)
j )Q1 (uj + 2)
(0)
Q1 (u(0)
j + 2)Q1 (uj )
= 1,
j = 1, . . . , J0 .
(D.8)
The boundary conditions for the system of nested Bethe equations are: QK (u) = QK,M (u) =
(u) (a given polynomial function), QM (u) = Q0,0 (u) = 1.
Note that our numbering of the Qs here differs slightly from the one in Section 4.2: the
Qs here are numbered from M to K while the same Qs there have the numbers from 0 to
K + M. Taking this into account, it is easy to see that the chain of Bethe equations obtained here
coincides with (5.8), where the signs are chosen as p1 = p2 = = pM = 1, pM+1 = pM+2 =
= pM+K = 1.
This is only one of possible sets of Bethe equations for the same model. We could also apply
the elementary moves BT1 and BT2 in any other order. At each change of the direction we obtain
an equation of the fermionic type (D.8). The different sets of equations lead to the same solution
for the T -functions T (a, s, u) TK,M (a, s, u) at the highest level of the hierarchy. The different
systems of Bethe equations are known to be related by certain duality transformations [14,45].
As we have seen in the main text, they admit a natural explanation in terms of a discrete zero
curvature condition on the (k, m) lattice.
Appendix E. Comparison of the results for gl(2|1) algebra with the results of [6]
(2)
(1)
Let us compare Eq. (6.1) of [6], with the notation = 12 u, k = 12 uk , k = 12 uk , b =

s + 12 ,
Q1 (u) =
Ju

u u(1)
j ,
Q2 (u) =
j =1
Jv

u u(2)
j ,
(u) =
j =1
(u rj ),
j =1
with our Eq. (8.23). The Baxter T Q-relation from [6] for the (b, 1/2) irrep in the auxiliary space
and (bi , 1/2) irreps at the sites of the chain now reads as follows:
Ts|{rl } (u) =
Q2 (u + s) Q1 (u s + 1) Q2 (u + s) + (u s)
Q2 (u s) Q1 (u s 1) Q2 (u s) (u s)
Q1 (u s 3) Q2 (u + s) + (u s)
Q1 (u s 1) Q2 (u s 2) (u s)
Q2 (u + s) + (u s) + (u s 2)
+
.
Q2 (u s 2) (u s) (u s 2)
(E.1)
We see that it coincides with our Eq. (8.23) up to the redefinition of -functions.
Let us give also the Bethe equations ensuring the polynomiality of Ts|{rl } (u):
(2)
1. Canceling poles at u = uj + s,
(2)
(uj )
(2)
+ (uj )
(2)
Q1 (uj + 1)
(2)
Q1 (uj 1)
(E.2)
411
we arrive at the Bethe equations for the fermionic node.

(1)
2. Canceling the poles at u = uj + s + 1,
1 =
(1)
Q1 (u(1)
j + 2) Q2 (uj 1)
(1)
Q1 (u(1)
j 2) Q2 (uj + 1)
(E.3)
we get the Bethe equations for the bosonic node. (We note a mistake in the denominator of
the r.h.s. of Eq. (6.3) of [6].)
(2)
3. The condition of canceling poles at u = uj + b + 3 is the same as the first one.
References
[1] P. Kulish, E. Sklyanin, On solutions of the YangBaxter equation, Zap. Nauchn. Sem. LOMI 95 (1980) 129160,
J. Sov. Math. 19 (1982) 1956 (in English).
[2] P. Kulish, Integrable graded magnetics, Zap. Nauchn. Sem. LOMI 145 (1985) 140163.
[3] E. Ogievetsky, P. Wiegmann, Factorized S-matrix and the Bethe ansatz for simple Lie groups, Phys. Lett. B 168
(1986) 360366;
N. Reshetikhin, P. Wiegmann, Towards the classification of completely integrable quantum field theories (the Bethe
ansatz associated with Dynkin diagrams and their automorphisms), Phys. Lett. B 189 (1987) 125131.
[4] Z. Maassarani, Uq osp(2, 2) lattice models, J. Phys. A: Math. Gen. 28 (1995) 13051324.
[5] P.B. Ramos, M.J. Martins, One-parameter family of an integrable spl(2|1) vertex model: Algebraic Bethe ansatz
and ground state structure, Nucl. Phys. B 474 (1996) 678714.
[6] M.P. Pfannmuller, H. Frahm, Algebraic Bethe ansatz for gl(2, 1) invariant 36-vertex models, Nucl. Phys. B 479
(1996) 575, cond-mat/9604082.
[7] P. Kulish, N. Reshetikhin, On GL(3)-invariant solutions of the YangBaxter equation and associated quantum systems, Zap. Nauchn. Sem. LOMI 120 (1982) 92121 (in Russian), J. Sov. Math. 34 (1986) 19481971 (in English).
[8] N. Reshetikhin, The functional equation method in the theory of exactly soluble quantum systems, Sov. Phys.
JETP 57 (1983) 691696.
[9] V. Bazhanov, N. Reshetikhin, Restricted solid-on-solid models connected with simply laced algebras and conformal
field theory, J. Phys. A: Math. Gen. 23 (1990) 14771492.
[10] A. Klmper, P. Pearce, Conformal weights of RSOS lattice models and their fusion hierarchies, Physica A 183
(1992) 304350.
[11] A. Kuniba, T. Nakanishi, Rogers dilogarithm in integrable systems, hep-th/9210025, preprint HUTP-92/A046.
[12] Z. Tsuboi, Analytic Bethe ansatz and functional equations for Lie super-algebra sl(r + 1|s + 1), J. Phys. A: Math.
Gen. 30 (1997) 79757991.
[13] Z. Tsuboi, Analytic Bethe ansatz related to a one-parameter family of finite-dimensional representations of the Lie
super-algebra sl(r + 1|s + 1), J. Phys. A: Math. Gen. 31 (1998) 54855498.
[14] Z. Tsuboi, Analytic Bethe ansatz and functional equations associated with any simple root systems of the Lie superalgebra sl(r + 1|s + 1), Physica A 252 (1998) 565585.
+ 1|s + 1)) PerkSchultz
[15] Z. Tsuboi, Nonlinear integral equations and high temperature expansion for the Uq (sl(r
model, Nucl. Phys. B 737 (2006) 261290, cond-mat/0510458.
[16] I. Krichever, O. Lipan, P. Wiegmann, A. Zabrodin, Quantum integrable systems and elliptic solutions of classical
discrete nonlinear equations, Commun. Math. Phys. 188 (1997) 267304, hep-th/9604080.
[17] A. Zabrodin, Discrete Hirotas equation in quantum integrable models, Int. J. Mod. Phys. B 11 (1997) 31253158,
hep-th/9610039;
A. Zabrodin, Hirota equation and Bethe ansatz, Teor. Mat. Fys. 116 (1998) 54100, Theor. Math. Phys. 116 (1998)
782819 (in English).
[18] R. Hirota, Discrete analogue of a generalized Toda equation, J. Phys. Soc. Jpn. 50 (1981) 37853791.
[19] V. Kac, Lie super-algebras, Adv. Math. 26 (1977) 896;
V. Kac, Lecture Notes in Mathematics, vol. 676, Springer-Verlag, New York, 1978, pp. 597626.
[20] A. Baha Balantekin, I. Bars, Dimension and character formulas for Lie supergroups, J. Math. Phys. 22 (1981) 1149.
[21] I. Bars, B. Morel, H. Ruegg, KacDynkin diagrams and supertableaux, J. Math. Phys. 24 (1983) 22532262.
[22] M. Gould, R. Zhang, Classification of all star irreps of gl(m|n), J. Math. Phys. 31 (1990) 25522559.
412
[23] L. Faddeev, L. Takhtadjan, Quantum inverse scattering method and the XY Z Heisenberg model, Usp. Mat.
Nauk 34 (5) (1979) 1363.
[24] L. Faddeev, The Bethe ansatz, SFB 288 Preprint No. 70 (1993);
L. Faddeev, Algebraic aspects of Bethe-ansatz, Int. J. Mod. Phys. A 10 (1995) 18451878, hep-th/9404013.
[25] N. Bogoliubov, A. Izergin, V. Korepin, Quantum Inverse Scattering Method and Correlation Functions, Cambridge
Univ. Press, Cambridge, 1993.
[26] V. Zakharov, S. Manakov, S. Novikov, L. Pitaevsky, Theory of Solitons, Nauka, Moscow, 1980.
[27] T. Miwa, On Hirotas difference equations, Proc. Jpn. Acad. Ser. A 58 (1982) 912.
[28] E. Date, M. Jimbo, T. Miwa, Method for generating discrete soliton equations I, J. Phys. Soc. Jpn. 51 (1982) 4116
4127.
[29] M. Jimbo, T. Miwa, Solitons and infinite dimensional Lie algebras, Publ. RIMS 19 (1983) 9431001.
[30] S. Saito, N. Saitoh, Linearization of bilinear difference equations, Phys. Lett. A 120 (1987) 322326;
S. Saito, N. Saitoh, Gauge and dual symmetries and linearization of Hirotas bilinear equations, J. Math. Phys. 28
(1987) 10521055.
[31] A. Zabrodin, Hirotas difference equations, Teor. Mat. Fys. 113 (1997) 179230, Theor. Math. Phys. 113 (1997)
13471392, solv-int/9704001 (in English).
[32] P. Kulish, N. Reshetikhin, E. Sklyanin, YangBaxter equation and representation theory: I, Lett. Math. Phys. 5
(1981) 393403;
P. Kulish, E. Sklyanin, Quantum Spectral Transform Method. Recent Developments, Lecture Notes in Physics,
vol. 151, Springer, Berlin, 1982, pp. 61119.
[33] I. Cherednik, On the properties of factorized S matrices in elliptic functions, Yad. Phys. 36 (1982) 549557, Sov. J.
Nucl. Phys. 36 (1982) 320324 (in English);
I. Cherednik, On quantum deformations of irreducible finite-dimensional representations of glN , Sov. Math.
Dokl. 33 (1986) 507510.
[34] I. Cherednik, On special basis of irreducible representations of degenerated affine Hecke algebras, Funk. Anal.
Prilozh. 20 (1) (1986) 8788 (in Russian).
[35] M. Nazarov, Yangians and Capelli identities, Amer. Math. Soc. Transl. 181 (1998) 139163, q-alg/9601027.
[36] M. Nazarov, V. Tarasov, On irreduciiblity of tensor products of Yangian modules, Int. Math. Res. Notices (1998)
125150, q-alg/9712004;
M. Nazarov, V. Tarasov, On irreduciiblity of tensor products of Yangian modules associated with skew Young
diagrams, Duke Math. J. 112 (2002) 343378, q-alg/0012039.
[37] A. Kuniba, T. Nakanishi, J. Suzuki, Functional relations in solvable lattice models I: Functional relations and representation theory, Int. J. Mod. Phys. A 9 (1994) 52155312, hep-th/9309137.
[38] A. Kuniba, J. Suzuki, Analytic Bethe ansatz for fundamental representations of Yangians, Commun. Math. Phys. 173
(1995) 225264, hep-th/9406180.
(1)
[39] A. Kuniba, Y. Ohta, J. Suzuki, Quantum JacobiTrudi and Giambelli formulae for Uq (Br ) from analytic Bethe
ansatz, J. Phys. A: Math. Gen. 28 (1997) 62116226, hep-th/9506167.
(1)
[40] Y. Zhou, P. Pearce, Solutions of functional equations of restricted An1 fused lattice models, Nucl. Phys. B 446
(1995) 485510, hep-th/9502067.
[41] T. Deguchi, P. Martin, An algebraic approach to vertex models and transfer matrix spectra, Int. J. Mod. Phys.
A 7 (Suppl. 1A) (1992) 165196;
P. Martin, V. Rittenberg, A template for quantum spin chain spectra, Int. J. Mod. Phys. A 7 (Suppl. 1B) (1992)
707730.
[42] N. Shinzawa, S. Saito, A symmetric generalization of linear Bcklund transformation associated with the Hirota
bilinear difference equation, J. Phys. A: Math. Gen. 31 (1998) 45334540, solv-int/9801002.
[43] N. Shinzawa, Symmetric linear Bcklund transformation for discrete BKP and DKP equations, J. Phys. A: Math.
Gen. 33 (1998) 39573970, solv-int/9907016;
N. Shinzawa, R. Hirota, The Bcklund transformation equations for the ultradiscrete KP equation, nlin.SI/0212014.
[44] I. Krichever, Characterizing Jacobians via trisecants of the Kummer variety, math.AG/0605625.
[45] N. Beisert, V.A. Kazakov, K. Sakai, K. Zarembo, Complete spectrum of long operators in N = 4 SYM at one loop,
JHEP 0507 (2005) 030, hep-th/0503200.
[46] P.A. Bares, I.M.P. Karmelo, J. Ferrer, P. Horsch, Charge-spin recombination in the one-dimensional supersymmetric
tJ model, Phys. Rev. B 46 (1992) 1462414654.
[47] P. Kulish, N. Reshetikhin, Diagonalization of GL(N ) invariant transfer matrices and quantum N -wave system (Lee
model), J. Phys. A: Math. Gen. 16 (1983) L591L596.
413
[48] E. Mukhin, V. Tarasov, A. Varchenko, Bethe eigenvectors of higher transfer matrices, J. Stat. Mech. Theor. Exp. 8
(2006) P08002, math.QA/0605015.
[49] F.H.L. Essler, V.E. Korepin, Higher conservation laws and algebraic Bethe anstze for the supersymmetric tJ
model, Phys. Rev. B 46 (1992) 91479162.
[50] F.H.L. Essler, V.E. Korepin, K. Schoutens, New exactly solvable model of strongly correlated electrons motivated
by high Tc superconductivity, Phys. Rev. Lett. 68 (1992) 29602963, cond-mat/9209002;
F.H.L. Essler, V.E. Korepin, K. Schoutens, Exact solution of an electronic model of superconductivity in (1 + 1)dimensions, cond-mat/9211001.
[51] P. Fendley, K.A. Intriligator, Scattering and thermodynamics in integrable N = 2 theories, Nucl. Phys. B 380 (1992)
265292, hep-th/9202011.
[52] P. Fendley, K.A. Intriligator, Scattering and thermodynamics of fractionally charged supersymmetric solitons, Nucl.
Phys. B 372 (1992) 533558, hep-th/9111014.
[53] A. Belitsky, S. Derkachov, G. Korchemsky, A. Manashov, Baxter Q-operator for graded SL(2|1) spin chain, J. Stat.
Mech. 0701 (2007) P005, hep-th/0610332;
A. Belitsky, Baxter equation for long-range SL(2|1) magnet, hep-th/0703058.
[54] V. Bazhanov, Z. Tsuboi, work in progress;
Z. Tsuboi, Talk at the Melbourne Meeting From Statistical Mechanics to Conformal and Quantum Field Theory.
A class of partially solvable two-dimensional quantum

models with periodic potentials
M.V. Ioffe a, , J. Mateos Guilarte b , P.A. Valinevich a
a Department of Theoretical Physics, Sankt-Petersburg State University, 198504 Sankt-Petersburg, Russia
b Departamento de Fisica Fundamental and IUFFyM, Facultad de Sciencias, Universidad de Salamanca,
37008 Salamanca, Spain

Received 9 June 2007; received in revised form 28 June 2007; accepted 16 July 2007
Available online 21 July 2007
Abstract
The supersymmetric approach is used to analyse a class of two-dimensional quantum systems with periodic potentials. In particular, the method of SUSY-separation of variables allowed us to find a part of the
energy spectra and the corresponding wave functions (partial solvability) for several models. These models
are not amenable to conventional separation of variables, and they can be considered as two-dimensional
generalizations of Lam, associated Lam, and trigonometric Razavy potentials. All these models have the
symmetry operators of fourth order in momenta, and one of them (the Lam potential) obeys the property
of self-isospectrality.
PACS: 03.65.-w; 03.65.Fd; 11.30.Pb
Keywords: Supersymmetry; Partial solvability; 2-dim periodic potentials
1. Introduction
The supersymmetric (SUSY) approach has provided a powerful impulse for new developments in analytical studies in quantum mechanics. To date, most new results have been obtained
for one-dimensional quantum systems (see the monographs and reviews [1]). Over the last two
decades, indubitable progress has been achieved also for higher-dimensional systems within the
E-mail addresses: m.ioffe@pobox.spbu.ru (M.V. Ioffe), guilarte@usal.es (J.M. Guilarte), pavel@PV7784.spb.edu

(P.A. Valinevich).
doi:10.1016/j.nuclphysb.2007.07.010
M.V. Ioffe et al. / Nuclear Physics B 790 [PM] (2008) 414431
415
framework of SUSY quantum mechanics [25]. The most essential results have been found
for two-dimensional quantum mechanics. In particular, the new approach for the construction
of completely integrable systems with symmetry operators of fourth order in momenta was
proposed [3]. Two new methodsSUSY-separation of variables and two-dimensional shapeinvariance [4,5]have allowed the problem of solvability to be tackled for two-dimensional
quantum systems, beyond the standard separation of variables.
The method of SUSY-separation of variables has been applied successfully to investigate the
spectra and wave functions of some models on the whole plane x = (x1 , x2 ), which are not
amenable to conventional separation of variables: the Morse potential [46], the PschlTeller
potential [7,8], and some others [6,9].
The starting point of this approach is the analysis of solutions of the intertwining relations
between a pair of two-dimensional Hamiltonians of the Schrdinger form:
H (
x )Q = Q H (
x );
H =
(2)
+ V (
x );
x ) = H (
x )Q+ ;
Q+ H (
H = (2) + V (
x ).
(1)
(2)
In general, the intertwining operators

are the operators of second order in momenta, i.e. in
derivatives. In [7] (see also [8]), the particular class of intertwining operators Q was considered:
operators with twisted reducibility

Q = (Q+ ) = l + l (
x ) (3 )lk +k + k (
x) ,
,
xl
(3)
where (
x ), (
x ) are two different functions (superpotentials), 3 is the Pauli matrix, and summation over k, l = 1, 2 is implied. In this class of models, both Hamiltonians H and H are
quasifactorized [7]

2
H = l + l (
x ) +l + l (
x ) = (2) + l (
x ) l l (
x );

2
(2)
H = l + l (
(4)
x ) +l + l (
x ) = + l (
x ) l l (
x ),
and hence their energy spectra are non-negative.
It was shown in [7,8] in a general form that Eqs. (1) and (3) lead to the following representation
for superpotentials , in terms of four functions 1,2, :
= 1 (x1 ) + 2 (x2 ) + + (x+ ) + (x ),
= 1 (x1 ) + 2 (x2 ) + (x+ ) (x ),
where x = (x1 x2 )/ 2. These functions 1,2, satisfy the equation

1 (x1 ) + (x+ ) + (x ) + 2 (x2 ) + (x+ ) (x ) = 0.
By substitutions, this becomes a purely functional equation with no derivatives:

1 (x1 ) + (x+ ) + (x ) = 2 (x2 ) + (x+ ) (x ) .
(5)
The solution of Eq. (5) is necessary to build the solutions of intertwining relations (1) for the
potentials V , V and for the supercharges Q . In particular,

2

V (
x ) = 12 (x1 ) 1 (x1 ) + 22 (x2 ) 2 (x2 ) + +
(x+ ) +
(x+ )
2

+
(x )
(x ) ,
416

Q = 12 22 2 + (x+ ) + (x ) 1 2 + (x+ ) (x ) 2

12 (x1 ) 1 (x1 ) + 22 (x2 ) 2 (x2 ) + 2+ (x+ ) (x ).
(6)
Thus, in order to find the systems with intertwining (1) by supercharges of the form (3), it is necessary to solve (5). This equation seems to be rather complicated, but it appeared to be solvable in
a general form. The two-dimensional generalization of the PschlTeller potential, investigated
in [7,8], was based just on a particular solution of Eq. (5).
In the present paper we focus our attention on the 1,2, , solutions which are periodic in the
variables x1 and x2 . These solutions will be further used to build a class of two-dimensional
potentials V (
x ) of the form (6)not amenable to standard separation of variableswhich are
periodic along x1 , x2 with the same periods. By means of the method of supersymmetric separation of variables [4,5] we shall derive the partial solvability of these periodic models: several
energy eigenvalues and corresponding eigenfunctions will be found. Until now the supersymmetric approach has been used for the analysis of periodic potentials in one-dimensional case
only (see for example [1012]). To the best of our knowledge, this is1 the first attempt to study
analytically two-dimensional periodic potentials, not amenable to separation of variables, within
the framework of SUSY quantum mechanics.
It is appropriate to make some remarks concerning the possible spectra of two-dimensional
periodic systems. The general statements are well known from textbooks (mainly on solid state
theory) [14]: the spectra of d-dimensional (d 2) models with periodic potentials in general
have a band structure similar to that of the one-dimensional case. However, in contrast to d = 1,
no strict results on the (anti)periodicity properties of the band edge wave functions are known
[15], and therefore analysis of the spectra of two-dimensional systems is much more complicated.
In some sense, the situation resembles the non-periodic case: no analogues of the oscillation theorem are known for d 2 quantum systems. This is a reason why the analysis of multidimensional
excited bound states is much more difficult than in the one-dimensional situation.
To imagine the variety of possible structures of the band spectrum, one can consider (contrary
to the rest of this paper) the simplest two-dimensional periodic systems which allow separation
of variables (see, for example, [16] and references therein). After separation of variables, both
one-dimensional problems have the band structure of energies 1 (k1 ), 2 (k2 ). Let us assume that
they have a finite number (one or two) of band gaps (as in Sections 4.2 and 4.3 below). Then,
the positions of the two-dimensional band edges depend crucially on the parameters of onedimensional bands 1 , 2 . In particular, these bands may be overlapped, so that the band gaps (or
at least some part of them) of the two-dimensional spectrum, E = 1 + 2 , may even disappear.
It is natural to consider two limiting cases: of almost free particles (the band gaps are vanishing)
and of tight binding particles (the band gaps are very wide).
The structure of this paper is as follows. Section 2 is devoted to the solution of the functional
equation (5): the explicit expressions for all four functions 1,2, will be found. In Section 3
the symmetry properties of the general solution of Eq. (5) are analysed. In Section 4, which
is the main part of the paper, a new class of two-dimensional systems, partially solvable by
means of SUSY-separation of variables, is constructed. The potentials of these systems are periodic on the plane, and they can be considered as two-dimensional generalizations (not amenable
to separation of variables) of fairly well-known [1012,1719] one-dimensional potentials: the
Lam potential, the associated Lam potential and the trigonometric Razavy potential. Using
the known positions of band edges and corresponding wave functions for these one-dimensional
1 See some remarks in [13, Section 5] and [8].
417
models with a finite number of gaps, some energy eigenvalues and their eigenfunctions for twodimensional periodic generalizations will be obtained explicitly (partial solvability). Section 5
includes a discussion of certain specific properties of the models of the previous section and
some limiting cases. In particular, all two-dimensional models constructed are integrable: the
symmetry operators of fourth order in momenta commute with the Hamiltonians. In addition, the
two-dimensional Lam model obeys the property of self-isospectrality, like its one-dimensional
prototype [10].
2. The general solution of the functional equation
In this section we analyse the functional equation (5), which plays an important role in our
approach. This analysis was started in Appendix of the paper [7], where the necessary conditions
for the existence of its solutions were derived. First, we will remind briefly the main steps of
this derivation. In the formulas below we shall imply that the functions 1,2 , depend on the
corresponding arguments: 1,2 = 1,2 (x1,2 ) and = (x ), unless otherwise stated.
Acting by (12 22 ) on both sides of (5), we obtain:

1
1
1
1
2
(7)
1 1 2 2 ( + ) =

1 ( + ).
1
2
2 2
1
It is now convenient to introduce a new unknown function, :

1/2
+ 1 1 2
(x1 , x2 ).
Substitution of this definition into (7) gives:

1

1 + 2 2 = 0,
1
2
x
x
)
d 2
and therefore depends only on ( 1 1 (
( )
1

1/2
+ = 1 1 2

x1
1 ( )
d
1 ( )
x2
2 ()
d),
2 ()
and the general solution of (7) is:

2 ()
d
.
2 ()
(8)

2 ()
d
.
2 ()
(9)
From the initial equation (5) we also have that:

1/2
+ + = 2 1 2

x1
1 ( )
d
1 ( )
x2
From Eqs. (8), (9) we could already obtain the solutions for , but we have to check that
these solutions will indeed depend on proper arguments. Thus, the constraints for the function
and functions 1,2 are obtained from the equations:
= 0.
The result is:

1
1
1 2
1
+
=
,
2 2 2
1 1
1
2

2

1 1 2 2
12
2

1 + 2
.
=
21
22
2
1
(10)
(11)
418
Here we disregard the trivial solution 0, for which + = = 0, 1,2 are arbitrary, but the
potentials (6) are amenable to separation of variables.
Otherwise one can exclude , dividing Eq. (10) by Eq. (11):
1 22 2 12
= 22 2 21 2 + 1 1 2 2 .

(12)
1
2
There is no separation of variables in (12), but it will appear after applying the operator 1 2 , so
that:
(1 /1 ) (2 /2 )
=
2a = const.
(12 )
(22 )
, integrating again and taking into account (12), one has that
Integrating, multiplying by 1,2
1,2
must satisfy the equation:
4
2
( )21,2 = a1,2
+ b1,2
+ c,
(13)
where a, b, c are arbitrary real constants. All solutions of this equation can be expressed in terms
of elliptic functions, and they are described for different ranges of parameters, for example,
in [20].
The subsequent discussion will depend crucially on the sign of the discriminant D = b2 4ac.
In the present paper we restrict ourselves to the case of D > 0, i.e. when the quadratic polynomial
4 + b 2 + c has two different real roots, which will be denoted as r and r (let r > r ).
a1,2
1
2
1
2
1,2
For positive values of the discriminant, three types of solutions of (13) exist, depending on the
dn(x|k)
relative position of the roots r1,2 . These solutions are proportional either to cn(x|k)
sn(x|k) or to sn(x|k)
1
or to sn(x|k)
, where sn, cn, dn are the well-known elliptic Jacobi functions [17].
Equations (13) are the necessary conditions for the functions 1,2 to satisfy Eq. (5). But are
these conditions also sufficient? To answer this question, we must solve Eqs. (10), (11) for function , with arbitrary elliptic functions 1,2 , satisfying (13). From Eq. (13), the argument of the
function can be written in the form:
2

x1
x2
(1 r1 )(22 r2 )
1
1 ( )
2 ()

.
(14)
d
d
=
ln
1 ( )
2 ()
2a(r1 r2 ) (12 r2 )(22 r1 )
The module sign is not important here, since
(12 r1 )(22 r2 )
(12 r2 )(22 r1 )
0
(15)
because of (13). In terms of this new variable , Eq. (10) takes the form
1
2 +
()
,
=
1
()
4( 2 )
where the notation sign(1 (x1 )2 (x2 )) was used.
The solution of this differential equation is2 :
(16)
() = 0
4
1
0 = const.
2 This solution coincides with the solution in the simpler case of a constant (not depending on ) value of .
(17)
419
Formally substituting (17) into the l.h.s. of Eq. (16), we obtain:

1
()
2 + 4
.
=
1
()
4( 2 )
Let us prove that the extra term in the nominator of the r.h.s. actually vanishes; i.e. that
0. The variable may be considered as a function of 12 and 22 , and hence
22
2
+
.
= 1
12
22
In turn, due to Eq. (5), one can rewrite derivatives over i2 in terms of derivatives over i :

di
a(2i2 r1 r2 )

sign i
=
=
.

2
2
i
i
di i
2 a(i2 r1 )(i2 r2 )
Gathering everything together and using
=
i
= 2(i ), we find that:

(r1 r2 )
(r1 r2 )
sign 1 1 1 +
sign 2 2 2 ,
2
2
2(21 r1 r2 )
2(22 r1 r2 )
which is zero for all values of x1,2 since at the points where (i ) = 0, the denominator has no
singularity (it tends to (r1 r2 ) for i 0).
Now, according to Eqs. (8), (9) solution (17) gives in terms of functions 1,2 :
= 0
2 1
1
2|1 2 | 2
4
1
(18)
The important fact is that without any new constraints on 1,2 (x1,2 ) (besides Eq. (13)) the functions depend on the proper arguments x :
For any pair of functions 1,2 satisfying (13) there exists a pair of functions given by (18),
such that they are solutions of (5).
Closing this section we shall derive an alternative expression for , to be used later (see
Section 4). Let us define the function
21 2 + 2a12 22 + b12 + b22 + 2c
.
(2 1 )2
One can check by straightforward calculations, that
G
2 + 1 2
G
.
=2 2 1 2
G
2 12
Then, taking into account

=
1 2 + 1 2
22 12
Eq. (19) can be rewritten in compact form as:

G
2 = 0.
(19)
420
2 = Const G, where the constant can be calculated by considering this expression in

Hence,
some specific point: Const = 20 /(4(b2 4ac)), with arbitrary constant 0 . We therefore obtain
the desired alternative expression for :
2
=
b2
20
2 + 2a12 22 + b12 + b22 + 2c
1 2
.
4ac
(2 1 )2
(20)
By a completely analogous calculation, or directly from Eq. (5), the expression for + can be
derived:
2
+
=
b2
20
2 + 2a12 22 + b12 + b22 + 2c
1 2
.
4ac
(2 + 1 )2
(21)
Formulas (20) and (21) are very convenient to prove that satisfy the same Eq. (13), but
with different coefficients. Direct calculations show that:
2
4
2
2
2
4
2

= 2 b2 4ac
b
+ 0 a
+ b
+ c.

8
0
(22)
Let us note that although the discriminant D = b2 4ac was chosen to be positive, the analogous
discriminant D for Eq. (22), which is equal to D = 4ac, can be non-positive as well.
3. Symmetries of the functional equation
As pointed out in a previous paper [7, Section 3.2], the functional equation (5)
has two discrete
symmetries: S1 , S2 . After the change of variables y1,2 = x , y (y1 y2 )/ 2 = x1,2 , Eq. (5)
becomes:

1 (y+ ) + (y1 ) + (y2 ) = 2 (y ) + (y1 ) (y2 ) ,
which, by rearrangement of terms, can be brought to:

+ (y1 ) 1 (y+ ) + 2 (y ) = (y2 ) 1 (y+ ) 2 (y ) .
The last form coincides with (5) up to interchanged 1,2 and . Hence, if (1 , 2 , + , ) is a
solution of (5), then the set (+ , , 1 , 2 ) is also a solution (this discrete symmetry was called
S1 ). By writing the four functions in brackets, we mean that the first of them should be put in the
place of 1 in (5), the second one in the place of 2 , and the last two in the place of + and ,
correspondingly.
This symmetry can be observed explicitly from the general solutions (20), (21). Namely, if
we calculate 1,2 from with the use of analogues of (20), (21):
2
1,2
=
+ 2a
2 2 + b
2 + b
2 + 2c
+

+
2+
( + )2
b 2 4a c
20
(23)
then we must arrive at the same functions 1,2 . Indeed, calculating the r.h.s. of (23), one obtains
2
1 = 0 12 /(8c), and by proper choice of the arbitrary constant 0 (which can be imaginary)
one has 1 = 1 . Thus, 2 (y2 ) = 1 (y1 )(+ (y+ ) + (y ))/( (y ) + (y+ )), and by comparing it with (5) we see that 2 = 2 too.
There is another symmetry S2 . If (1 , 2 , + , ) is a solution of (5), then (1 , 2 , 1+ , 1 )
is also a solution of (5). This was shown in [7] directly from the functional equation (5), and now
421
we shall derive it from the general solution (18). It can be rewritten in the form:
1
1
1
0 2 + 2 2 1 2
=
.
2 |b| 12
2
1
If one performs the change (1 , 2 ) (1 , 2 ) then , and
1
1
1
0 2 2 2 1 2 Const

= ,
2 |b| 12 +
2
1
which completes the proof.

It should be noted that these symmetries may play an important role. In particular, the first
one, S1 , was applied in [7] to combine the shape-invariance with SUSY-separation of variables
in the 2D PschlTeller model.
4. SUSY-separation of variables for some models with periodic potentials
4.1. The algorithm of SUSY-separation
In the general expressions (6) for the potential V (
x ) one sees the very special dependence on
coordinates (x1 , x2 ): there are two terms that do not mix x1 and x2 , and two mixing terms that
depend on the light-cone variables x :
V (
x ) = V1 (x1 ) + V2 (x2 ) + V+ (x+ ) + V (x );
2

V1,2, = 1,2,
1,2,
,
(24)
all terms being represented in supersymmetric form. Just the last two terms of V (
x ) show that
separation of variables in the Schrdinger equation is not possible.
The method of SUSY-separation of variables (see details in [4,5,7]) allows us to partially
solve some two-dimensional models, i.e. to find a part of their spectra and corresponding wave
x ) is
functions. Due to the intertwining relations (1), the subspace spanned by zero modes n (
closed under the action of the Hamiltonian H . Therefore, the wave functions (
x ) can be built as
x ) of the supercharge Q+ . In turn, the zero modes n (
x)
linear combinations of zero modes n (
can be found through the specific similarity (gauge) transformation of the supercharge (3) to
the separated form:

q+ = e(x ) Q+ e(x ) = 12 22 12 1 + 22 2 ,
(25)
where the function is given by:

x+
(
x)
x
+ ( ) d +

() d .
(26)
This transformation allows us to reduce the two-dimensional quantum problem for n (

x ) to a
pair of one-dimensional problems:
n (
x ) = e(x ) n (
x );
n (
x ) n (x1 )n (x2 ),
where n and n are the eigenfunctions of one-dimensional Schrdinger equations:

n + 12 1 n = n n ,

n + 22 2 n = n n
(27)
(28)
422
(n are the constants of separation). Let us note that the one-dimensional potentials in (27), (28)
coincide exactly with the first two terms in the two-dimensional potential (24). Moreover, they
take the form typical of potentials within the framework of one-dimensional SUSY quantum
mechanics [1]. In this sense two-dimensional potentials (24) can be thought of as generalizations
of one-dimensional models (27), (28).
Our goal in this section is to build a new class of partially solvable models within the framework of the approach described above. In practice, this means that we have to present a new class
of functions 1,2 (x1,2 ), which obey the following special requirements:
they must obey the functional equation (5), i.e. Eq. (13).
the one-dimensional models (27) and (28) with superpotentials 1,2 must be exactly solvable
in order to provide for the normalizable zero modes n .
In such a way we hope to find by means of (18) (or by means of (20), (21)), and hence to
build the two-dimensional potential (6). The model with this potential will be partially solvable
by means of SUSY-separation of variables.
At the last stage, one has to build the gauge-transformed Hamiltonian:
h(
x ) e(x ) H (
x )e(x ) = 12 22 + 2(+ + )1 + 2(+ )2
+ 12 1 + 22 2 ,
(29)
such that
and, by direct calculations, find the matrix C,
h
= C ;
(0 , 1 , . . . , N ).
This can be done by the action of the operator h on zero modes n :

hn = 2n + 2(+ + )1 + 2(+ )2 n
(30)
after expressing the r.h.s. as a linear combination of k .

According to the prescriptions of SUSY-separation of variables (see more details in [4,5,7]),
the eigenvalues Ek of the matrix C give part of the spectrum of the Hamiltonian H . The corresponding wave functions k (
x ) are obtained as:
x ),
x ) = B (
k (
where the matrix B is a solution of the matrix equation
B C = B
Its diagonal elements coincide with Ek .
with the diagonal matrix .
Up to now, this program was performed successfully for the following cases.
(1) A two-dimensional generalization of the Morse potential [4,5] with
1 (x) = 2 (x) exp (x).
(2) A two-dimensional generalization of the PschlTeller potential [7,8] with
1 (x) = 2 (x) sinh1 (x).
(31)
423
In both models, the functions , which were originally guessed directly from Eq. (5), can
now be obtained by means of Eqs. (18), (20), (21).
Below we shall consider new specific choices of functions 1,2 that will satisfy all the aforementioned conditions. The common feature of these new models is that the functions 1,2 are
periodic and therefore lead to partially solvable two-dimensional potentials, V (
x ), periodic on
the plane (x1 , x2 ).
4.2. The two-dimensional Lam potential
From the variety of elliptic functions 1,2 , which give solutions of Eq. (13), we have to choose
the subclass leading to exactly solvable one-dimensional Schrdinger equations (27) and (28)
with periodic potentials. Let us start from the particular 2K-periodic solutions3
1 (x) = 2 (x) = k 2
sn(x|k) cn(x|k)
,
dn(x|k)
(32)
where the standard notations [17] for Jacobi elliptic functions sn(x|k),
cn(x|k), dn(x|k) with
/2
modulus k (0, 1] and the complete elliptic integral K(k) 0 d/ 1 k 2 sin2 were used.
Later on, for simplicity we skip the argument k as long as no confusion appears.
Thus, the one-dimensional potentials V1,2 of (24), which coincide with potentials in (27)
and (28), have the form of the Lam equation:
V1,2 (x) = 2k 2 sn2 (x) k 2 .
(33)
We notice that (33) is only the simplest l = 1 case of the general Lam potential V = l(l +
1)k 2 sn2 (x) with arbitrary integer l. For higher values, l > 1, the superpotentials 1,2 do not
satisfy the basic equation (13), and for this reason are not considered here. Since the elliptic
function sn(x) is 4K-periodic, and 2K-antiperiodic sn(x + 2K) = sn(x), the potential (33) is
periodic with the period 2K.
The spectrum and the wave functions for the Lam potential (33) are known exactly [17]: it has
one bound band with energies (0, 1 k 2 ) and the continuous band with energies (1, ),
with the band gap between them. The band edge eigenfunctions are
0 = 0;
1 = 1 k 2 ;
2 = 1;
0 (x) = dn(x);
1 (x) = cn(x);
2 (x) = sn(x),
(34)
and the wave functions obey the well-known [17] (anti)periodicity property of band edge eigenfunctions for one-dimensional periodic potentials: 0,1 (x + 2K) = 0,1 (x), 2 (x + 2K) =
2 (x).
According to formulas (20), (21), the choice of (32) for 1,2 (x1,2 ) leads to very compact
explicit expressions for (x ):
cn( 2x)
+ (x) = (x) = B
(35)
,
sn( 2x)
where the new constant B = 0 /(4 1 k 2 ).

3 One can check straightforwardly that this expression satisfies Eq. (13). It can be transformed to the third type of
solutions of (13), mentioned in Section 2, by the Landen transformation [17] accompanied by a suitable shift of argument.
424
The two-dimensional potential V (

x ) can be obtained from (6):

V (
x ) = 2k 2 sn2 (x1 ) + sn2 (x2 ) 1
B(B + 2 dn(x1 + x2 )) B(B 2 dn(x1 x2 ))

+
2B 2 .
+
sn2 (x1 + x2 )
sn2 (x1 x2 )
(36)
This can be considered as a two-dimensional generalization (not amenable to separation of variables) of the Lam potential (33). The potential (36) is periodic under the shifts of x1,2 with
the periods 2K. The coefficients of both terms in (36), singular at (x1 x2 ) 0, are such that
no fall to the center occurs for arbitrary value of parameter B. The plot of the two-dimensional
potential (36) is represented in Fig. 1(a) for a specific choice of parameter values.
To start the procedure of SUSY-separation of variables, it is easy to calculate the function (
x)
by (26):

B
(1 dn(x1 x2 ))(1 + dn(x1 + x2 ))
(
x ) = ln
(37)
.
(1 + dn(x1 x2 ))(1 dn(x1 + x2 ))
2 2
The functions n for the one-dimensional energies n at band edges are:
x ) = dn(x1 ) dn(x2 );
0 (
1 (
x ) = cn(x1 ) cn(x2 );
2 (
x ) = sn(x1 ) sn(x2 ).
We must act through the operator h (see (29), (30)) on all n in order to find the matrix C and
its eigenvalues. For this model, the action of the gauge-transformed Hamiltonian h on n can
be written as:
hn = 2n n

2 2B
2
sn(x2 ) cn(x2 ) dn(x1 )1 sn(x1 ) cn(x1 ) dn(x2 )2 n .
sn (x1 ) sn2 (x2 )
x ) = 0 (
x ) = e(x ) 0 , is the lowest eigenfunction
One can calculate that h0 = 0, i.e. 0 (
of H with energy E0 = 0. Further calculations show that:

1
1
2(1 k 2 ) 2 2B(1 k 2 )
=
.
h(
x)
2
2 2B
2
2
Diagonalizing the matrix on the r.h.s. and following the algorithm of Section 4.1, one obtains the
eigenfunctions n (
x ) for (36):
0 = e(x ) dn(x1 ) dn(x2 );

E1 2(1 k 2 )
(
x)
1 = e
cn(x1 ) cn(x2 ) +
sn(x1 ) sn(x2 ) ;
2 2B

E2 2(1 k 2 )
2 = e(x ) cn(x1 ) cn(x2 ) +
sn(x1 ) sn(x2 ) ,
2 2B

and their energy eigenvalues: E0 = 0, E1,2 = 2 k 2 k 4 + 8B 2 (1 k 2 ). At first sight, this
result may be in conflict with the obvious non-negativeness of the spectrum of the Hamiltonian
H due to its quasi-factorizability (4): the energy E1 above seems to be negative for some values
of B. The resolution of this paradox is quite simple, since for the wave functions of interest each
possible singularity must be normalizable. The wave functions above possess the power singularities at (x1 x2 ) 0 via the multiplier e according to expression (37). These singularities
(a)
(b)
(c)
(d)
425
Fig. 1. Panel (a): The two-dimensional Lam potential (36) for the parameters k 2 = 0.30 and B = 0.05. The period of
potential along x1 , x2 is 2K(k) = 3.42. We used a cut-off of the potential at 2.0. Panels (b), (c), (d): The squares
of three known wave functions 02 , 12 , 22 (see Eqs. (38)) for two-dimensional Lam potential with the same values
of parameters and suitable cut-offs. The energy eigenvalues are E0 = 0, E1 = 1.38, E2 = 2.02.
are normalizable for B 2 < 1/2, and just for only these values of B is the eigenvalue E1 positive.
The squared wave functions 0 (
x )2 , 1 (
x )2 , 2 (
x )2 are given in Figs. 1(b)(d) for the same
parameter values as in Fig. 1(a).
The periodicity properties of elliptic Jacobi functions allow us to check that all three Bloch
eigenfunctions [14] have vanishing quasi-momenta, i.e. they are 2K-periodic:

n (x1 + 2Km1 , x2 + 2Km2 ) = exp ik1 (n ) 2Km1 + ik2 (n ) 2Km2 n (x1 , x2 ),
n ) = (k1 (n ),
where m1 , m2 are arbitrary integer numbers, and the quasi-momentum vector k(
Although we cannot find the whole structure of the spectrum of the model, the zero
k2 (n )) = 0.
values of the quasi-momenta hint that perhaps the states obtained correspond to the band edges
(in analogy with the one-dimensional case). In any case, we analytically find three eigenfunctions
of the generalized two-dimensional Lam Hamiltonian (36) for B 2 < 1/2, and one of them,
0 (
x ), is certainly the ground state; i.e., the lower edge of the bound band.
4.3. The two-dimensional associated Lam potential
Since the functional equation (5) is homogeneous, one can consider the more general form
of (32) and (35), multiplying the functions 1,2 by an arbitrary parameter l and keeping
426
unchanged:
sn(x) cn(x)
;
dn(x)
cn( 2x)
+ (x) = (x) = B
.
sn( 2x)
1 (x) = 2 (x) = lk 2
(38)
(39)
Thus, the one-dimensional potentials V1,2 are:

V1,2 (x1,2 ) = l(l + 1)k 2 sn2 (x1,2 ) + l(l 1)k 2
cn2 (x1,2 )
dn2 (x1,2 )
l2k2.
(40)
These potentials coincide with the so-called associated Lam potential [17], with well-known
band structure [11,12] for different (not only integer) values of the parameter l. Here we consider only4 the simplest case l = 2, where the spectrum has two bound bands and the continuous
band with the following energy values of the band edges and analytical expressions for the corresponding wave functions with necessary (anti)periodic conditions under xi xi + 2K [11]:
0 = 0;
0 = 0 = dn2 (x);
1 = 5 3k 2 2 4 3k 2 ;
1 = 1 =

cn(x) 2 2
3k sn (x) + 1 ;
dn(x)
2 = 5 2k 2 2 4 5k 2 + k 4 ;
2 = 2 =

sn(x) 2 2
3k sn (x) + 1 ;
dn(x)
3 = 5 2k 2 + 2 4 5k 2 + k 4 ;
3 = 3 =

sn(x) 2 2
3k sn (x) + 2 ;
dn(x)
4 = 5 3k 2 + 2 4 3k 2 ;
4 = 4 =

cn(x) 2 2
3k sn (x) + 2 ,
dn(x)
where 1,2 = 2 4 3k 2 , 1,2 = 2 k 2 4 5k 2 + k 4 . The zero modes of q+ are again

n = n (x1 )n (x2 ).
The two-dimensional potential is:
2

cn (x1 ) cn2 (x2 )
V (
x ) = 6k 2 sn2 (x1 ) + sn2 (x2 ) + 2k 2
+
8k 2
dn2 (x1 ) dn2 (x2 )
B(B + 2 dn(x1 + x2 )) B(B 2 dn(x1 x2 ))

+
2B 2 ,
+
(41)
sn2 (x1 + x2 )
sn2 (x1 x2 )
which is the two-dimensional generalization of the associated Lam potential (40). Due to properties of elliptic Jacobi functions, it is again periodic under xi xi + 2K mi with an arbitrary
integer mi .
In order to find a part of spectrum of the system (41), we must study the action of the operator h on the functions n found above. Again, the matrix C is block-diagonal since h0 = 0, and
therefore the lowest energy eigenvalue (the lower edge of the first bound band) vanishes. Other
4 This approach seems to be applicable to the higher values of l > 2 as well, although further calculations will be much
more complicated.
zero modes n are mixed with each other by the action of h:

1
1
2 2Ba11 2 2Ba12
0
2E1
2E2
0
22Bb12 2
2 22Bb11
h(
x) =
,
3
2 2Bb21
2 2Bb22
3
0
2E3
4
0
2 2Ba21 2 2Ba22
2E4
4
427
(42)
where the coefficients aij , bij have a rather involved form:

i (i 3j )
Mi Li 3j
;
aij = (1)j
;
j (2 1 )
j (2 1 )

Li = 1 k 2 i2 2k 2 (3 + i ).
Mi = 1 k 2 i2 + 6k 4 (3 + i );
bij = (1)j
Diagonalizing the matrix on the r.h.s. of (42), one obtains the energy eigenvalues En , n =
1, 2, 3, 4, and eigenfunctions n (
x ) for (41) analogously to the previous subsection. In this case
n ) for n = 1, 2, 3, 4 are zero.
also the quasi-momenta k(
4.4. The two-dimensional trigonometric Razavy potential
Although the two models studied in the previous subsections lead to very different potentials
the two-dimensional Lam and associated Lam potentialsthe forms of the initial solutions
1,2 are very similar to each other, and the functions simply coincide. The difference can be
described by the parameter :
1,2 (x) = k 2
sn(x|k) cn(x|k)
,
dn(x|k)
(43)
where = 1 for the Lam system, and = 2 for the associated Lam.
In this subsection, we consider the limiting case when
2
; k 0,
(44)
k2
being a new arbitrary finite parameter, and being the same as in (35), but with k 0.
Taking into account that in this limit
sn(x|k) sin x;
cn(x|k) cos x;
dn(x|k) 1,
we obtain:
1,2 (x) = sin(2x);
+ (x) = (x) = B cot( 2x),
(45)
and the one-dimensional potentials V1,2 (x) takes the form:

2
(1 cos 4x) 2 cos 2x.
(46)
2
These potentials coincide with the periodic potentials used by M. Razavy [19] for the description
of torsional oscillations of certain molecules. One must choose = 2; n = 0 in Eq. (4) of [19]
in order to identify the trigonometric Razavy potential with our Eq. (46). The trigonometric
Razavy potential admits [21] partial solvability: a few of its band edge levels (eigenvalues and
wave functions) were found analytically for different values of the integer parameter n. Actually,
V1,2 (x) =
428
in our case of n = 0 only the lowest band edge eigenfunction with 0 = 0 was found:

0 (x) = 0 (x) = exp cos 2x .

2
(47)
Our choice for 1,2, leads to the following two-dimensional periodic potential (24), which
can be considered as a generalization of the one-dimensional trigonometric Razavy potential:
2
(cos 4x1 + cos 4x2 ) 2(cos 2x1 + cos 2x2 )
2
B(B + 2 ) B(B 2 )
+
+
+ 2 2B 2 .
sin2 ( 2x+ ) sin2 ( 2x )
V (
x) =
(48)
For this potential, the method of SUSY-separation of variables provides the analytical expression
for the lowest (E0 = 0) band edge wave function:

0 (
x ) = exp (
x ) 0 (x1 )0 (x2 )
B

sin(x1 x2 ) 2

=
(49)
exp
+
cos
2x
)
,
(cos
2x
1
2
sin(x1 + x2 )
2
where (
x ) was calculated directly from (26), and 0 , 0 from (47).
5. Discussions and conclusions
In Section 3 two different symmetries (S1 , S2 ) of the functional equation (5) were presented.
Here we apply them to the Lam and associated Lam potentials discussed above.
First, S1 symmetry applied to any solution (1 , 2 , + , ) leads to the potential
2
2

V (S1 ) = +
(x1 ) +
(x1 ) +
(x2 )
(x2 )

+ 12 (x+ ) 1 (x+ ) + 22 (x ) 2 (x ) ,
(50)
which differs from (6) only in the change of variables x1,2 = x ,. Since the Laplace operator has
(S )
the same form in x1,2 coordinates, new wave functions n 1 of (50) are obtained from the old
(S1 )
ones n of (6) simply as n (x1 , x2 ) = n (x+ , x ).
In contrast to S1 , the effect of S2 symmetry could in principle be more promising for the
building of new models. Performing the S2 -transformation of the associated Lam potential,
generated by (38)(39) with l = 2, we obtain a new potential:
cn2 (x1 )
cn2 (x2 )
x ) = 6k 2 sn2 (x1 ) + 2k 2 sn2 (x2 ) + 2k 2 2
+ 6k 2 2
8k 2
V (S2 ) (
dn (x1 )
dn (x2 )
B(B 2 dn(x1 + x2 )) B(B + 2 dn(x1 x2 ))

+
2B 2 .
+
cn2 (x1 + x2 )
cn2 (x1 x2 )
It is connected to (41) as:

B
(S2 )
V
(x1 , x2 ; B) = V x1 , x2 + K;
,
1 k2
(51)
and the wave functions of V (S2 ) can be obtained easily from those of V by the same change of
coordinates and parameters. The analogous conclusion is also suitable for the Lam potential,
the case l = 1.
429
In the main textSection 4we in fact considered only one of the superpartner Hamiltonians H, H from the basic intertwining relations (1). It is well known [1] that in one-dimensional
SUSY quantum mechanics these superpartners are usually almost isospectral, i.e. they have the
same spectra but up to normalizable zero modes of the supercharges Q . The absence of such
zero modes in the case of non-periodic potentials means the spontaneous breaking of SUSY. The
situation is different [10] for some one-dimensional periodic potentials, where the two partner
potentials V , V are related by a discrete symmetry, but SUSY is not broken. These potentials
were called self-isospectral, and the particular examples are given by the one-dimensional
Lam (33) and associated Lam (40) potentials. Anothernon-self-isospectralclass of onedimensional periodic models also exists [11,12], where (quite the contrary) the SUSY intertwining relations provide a variety of new solvable periodic potentials.
In the non-periodic models on the whole plane [35] both second-order supercharges Q
may have zero modes. It is interesting to consider the two-dimensional periodic supersymmetric
models from the point of view of their self-isospectrality. For example, the superpartner of the
generalized Lam potential (36) has the form:

V (
x ) = 2k 2 sn2 (x1 ) + sn2 (x2 ) 1
B(B 2 dn(x1 + x2 )) B(B + 2 dn(x1 x2 ))

+
2B 2 ,
+
sn2 (x1 + x2 )
sn2 (x1 x2 )
i.e., it differs from V (
x ) only by the signs in front of the functions dn(x1 x2 ) in the nominators.
It is easy to check that the reflection (x1 , x2 ) (x1 , x2 ) merely turns V into V and vice versa:
V (x1 , x2 ) = V (x1 , x2 ).
Therefore, the spectra of the Hamiltonians H and H coincide, and the two-dimensional generalized Lam potential (36) obeys the property of self-isospectrality. The analogous proof also
works for the two-dimensional associated Lam potential (41).
It is significant that while in a certain sense the self-isospectrality for one-dimensional models
(with first-order supercharges) renders supersymmetry useless, this is not the case in the twodimensional situation (with second order supercharges). Indeed, since all the two-dimensional
models considered above satisfy the supersymmetric intertwining relations (1) with second-order
supercharges Q , the corresponding Hamiltonians H (irrespective of properties of the superpartners H ) commute with the operators
R = Q Q+ ,
where Q are given by (3). This means that all these systems are completely integrableR plays
the role of the symmetry operators. It is clear that these symmetry operators R annihilate the wave
functions constructed in Section 4, since they were built simply as linear combinations of zero
modes of Q+ . However, the action of the operators R on other (as yet unknown) wave functions
of H may be very nontrivial.
In this paper the one-dimensional systems with a finite number of bands were represented by
the band edge wave functions only. The wave functions inside the bands for the Lam potential,
however, are also known. Two linearly independent wave functions with energy are given by:
(x) =
H (x ) xZ()
;
e
(x)
dn2 (),
(52)
430
where the Jacobi theta-functions, H , , and the Jacobi zeta-function, Z, are defined in the theory
of elliptic functions (see [17]). One could use these wave functions, instead of band edge functions above, to construct the additional eigenvalues of the two-dimensional periodic models. This
task seems to be much more difficult technically and will be considered elsewhere. Here we wish
to illustrate how the wave functions (52) coincide with the band edge wave functions (34) in the
limits 0, (1 k 2 ), 1. Indeed, these limits correspond to the following values
of : K + iK ,

K, 0, where K is the associated complete elliptic integral K (k) = K(k ) = K( 1 k 2 ). One
then has to substitute these limiting values into H , , Z in (52). As a result, just Eqs. (34) are
derived.
Let us also mention the limit k 1 of the two-dimensional Lam and associated Lam systems (36) and (41). In this limit, sn(x|k = 1) = tanh x; cn(x|k = 1) = sech x; dn(x|k = 1) =
sech x. Therefore, for k 1 Eqs. (38) and (39) lead to
1,2 (x) = l tanh x;
+ (x) = (x) =
B
,
sinh( 2x)
where l = 1 for the Lam and l = 2 for the associated Lam systems. Substitution of these
expressions into two-dimensional potentials (36), (41) gives exactly the potential of the twodimensional generalization of PschlTeller model. This was studied in detail in [7,8], where its
partial solvability and complete integrability were demonstrated.
We should notice in conclusion that even in one-dimensional quantum mechanics very limited number of exactly solvable periodic problems is known [1012]. There is no need to stress
the interest of finding analytically solvable higher dimensional models like those described in
this paper. Such models would be very desirable both as a basis for perturbation theory and for
further study of general properties. The importance of the investigation of this kind of model
is increasing owing to the development of modern physical technologies, which have led to
the manufacture of new materials: a variety of superlattices, films, quantum two-dimensional
dots etc. A study of these materials and the corresponding devices should be based on twodimensional (and three-dimensional) Schrdinger equations with periodic potentials.
Acknowledgements
The work was partially supported by the Spanish MEC grant FIS2006-09417 (J.M.G.), by
project VA013C05 of the Junta de Castilla y Len (J.M.G. and M.V.I.) and by the Russian grants
RFFI 06-01-00186-a (M.V.I. and P.A.V.) and RNP 2.1.1.1112 (M.V.I.). P.A.V. is indebted to the
International Centre of Fundamental Physics in Moscow and the non-profit foundation Dynasty
for financial support. M.V.I. is grateful to the University of Salamanca for kind hospitality and to
Dr. A. Ganguly for useful elucidations on elliptic functions.
References
[1] F. Cooper, A. Khare, U. Sukhatme, Phys. Rep. 251 (1995) 268;
G. Junker, Supersymmetric Methods in Quantum and Statistical Physics, Springer, Berlin, 1996;
B.K. Bagchi, Supersymmetry in Quantum and Classical Mechanics, Chapman and Hall/CRC, Boca Raton, 2001.
[2] A.A. Andrianov, N.V. Borisov, M.V. Ioffe, Phys. Lett. 105A (1984) 19;
A.A. Andrianov, N.V. Borisov, M.V. Ioffe, M.I. Eides, Phys. Lett. 109A (1985) 143.
[3] A.A. Andrianov, M.V. Ioffe, D.N. Nishnianidze, Phys. Lett. A 201 (1995) 103;
A.A. Andrianov, M.V. Ioffe, D.N. Nishnianidze, Theor. Math. Phys. 104 (1995) 1129.
[4] M.V. Ioffe, J. Phys. A: Math. Gen. A 37 (2004) 10363.
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
431
F. Cannata, M.V. Ioffe, D.N. Nishnianidze, J. Phys. A: Math. Gen. A 35 (2002) 1389.
F. Cannata, M.V. Ioffe, D.N. Nishnianidze, Theor. Math. Phys. 148 (2006) 960.
M.V. Ioffe, P.A. Valinevich, J. Phys. A: Math. Gen. A 38 (2005) 2497.
M.V. Ioffe, J. Mateos Guilarte, P.A. Valinevich, Ann. Phys. 321 (2006) 2552.
M.V. Ioffe, J. Negro, L.M. Nieto, D.N. Nishnianidze, J. Phys. A: Math. Gen. A 39 (2006) 9297.
G. Dunne, J. Feinberg, Phys. Rev. D 57 (1998) 1271.
A. Khare, U. Sukhatme, J. Math. Phys. 40 (1999) 5473;
U. Sukhatme, A. Khare, Comment on: Self-isospectral periodic potentials and supersymmetric quantum mechanics, quant-ph/9902072;
A. Khare, U. Sukhatme, J. Math. Phys. 42 (2001) 5652;
A. Khare, U. Sukhatme, Periodic potentials and supersymmetry, quant-ph/0402206.
D.J. Fernndez C., J. Negro, L.M. Nieto, Phys. Lett. A 275 (2000) 338;
G. Dunne, M. Shifman, Ann. Phys. 299 (2002) 143;
D.J. Fernndez C., B. Mielnik, O. Rosas-Ortiz, B.F. Samsonov, Phys. Lett. A 294 (2002) 168;
D.J. Fernndez C., B. Mielnik, O. Rosas-Ortiz, B.F. Samsonov, J. Phys. A: Math. Gen. A 35 (2002) 4279;
B.F. Samsonov, M.L. Glasser, J. Negro, L.M. Nieto, J. Phys. A: Math. Gen. A 36 (2003) 10053;
D.J. Fernndez C., A. Ganguly, Exactly solvable associated Lam potentials and supersymmetric transformations,
quant-ph/0608180;
F. Correa, L.M. Nieto, M. Plyushchay, Phys. Lett. B 644 (2007) 94.
A.A. Andrianov, M.V. Ioffe, D.N. Nishnianidze, J. Phys. A: Math. Gen. A 32 (1999) 4641.
C. Kittel, Quantum Theory of Solids, John Wiley and Sons, New York, 1963.
M.S.P. Eastham, The Spectral Theory of Periodic Differential Equations, Scottish Academic Press, Edinburgh, 1973.
V.G. Baryakhtar, E.D. Belokolos, O.D. Dmitriiev, Three-dimensional exactly solvable model of electron movement
in the periodic separable Lam potential, cond-mat/0611496.
E. Whittaker, G. Watson, A Course of Modern Analysis, Cambridge Univ. Press, 1980;
P.F. Byrd, M.D. Friedman, Handbook of Elliptic Integrals for Engineers and Scientists, Springer-Verlag, Berlin,
1971;
H. Bateman, E. Erdelyi, Higher Transcendental Functions, vol. 3, McGrawHill, New York, 1955;
N.I. Akhiezer, Elements of the Theory of Elliptic Functions, Amer. Math. Soc., Providence, 1990;
F.M. Arscott, Periodic Differential Equations, Pergamon Press, London, 1963;
W. Magnus, S. Winkler, Hills Equation, Wiley, New York, 1966.
H. Li, D. Kusnezov, F. Iachello, J. Phys. A: Math. Gen. A 33 (2000) 6413.
M. Razavy, Phys. Lett. 82A (1981) 7.
A.M. Perelomov, Integrable Systems of Classical Mechanics and Lie Algebras, Birkhuser, Boston, MA, 1990
(Appendix B).
A.V. Turbiner, Commun. Math. Phys. 118 (1988) 467;
M.A. Shifman, Int. J. Mod. Phys. A 4 (1989) 2897;
V.V. Ulyanov, O.B. Zaslavskii, Phys. Rep. 216 (1992) 179;
F. Finkel, A. Gonzlez-Lpez, M.A. Rodrguez, J. Phys. A: Math. Gen. A 32 (1999) 6821.
Counting BPS operators in N = 4 SYM

F.A. Dolan
Institiid Ard-linn Bhaile tha Cliath (Dublin Institute for Advanced Studies), 10 Burlington road, Dublin 4, Ireland
Received 27 April 2007; accepted 27 July 2007
Abstract
The free field partition function for a generic U (N ) gauge theory, where the fundamental fields transform
in the adjoint representation, is analysed in terms of symmetric polynomial techniques. It is shown by these
means how this is related to the cycle polynomial for the symmetric group and how the large N result may
be easily recovered. Higher order corrections for finite N are also discussed in terms of symmetric group
characters. For finite N , the partition function involving a single bosonic fundamental field is recovered and
explicit counting of multi-trace quarter BPS operators in free N = 4 super-YangMills discussed, including
a general result for large N . The partition function for quarter BPS operators in the chiral ring of N = 4
super-YangMills is analysed in terms of plane partitions. Asymptotic counting of BPS primary operators
with differing R-symmetry charges is discussed in both free N = 4 super-YangMills and in the chiral
ring. Also, general and explicit expressions are derived for SU(2) gauge theory partition functions, when
the fundamental fields transform in the adjoint, for free field theory.
Keywords: Characters; Partition functions; Gauge theory; N = 4 super-YangMills
1. Introduction
For the past while there has been intense interest in finite N partition functions for Yang
Mills theories, especially in super-symmetric ones, particularly with regard to the construction
of BPS states and the counting thereof [115]. Much attention has been devoted to this issue
for N = 4 super-YangMills theory, it being by now the archetypal example of a conformal
field theory for which we have a dual description in terms of string theory, by means of the
AdS/CFT conjecture [16]. An example is in the assiduous efforts that have been made to exE-mail address: fdolan@stp.dias.ie.
doi:10.1016/j.nuclphysb.2007.07.026
F.A. Dolan / Nuclear Physics B 790 [PM] (2008) 432464
433
plain the entropy of certain BPS black holes in AdS5 S5 [1720] in terms of microscopic
counting of dual operators in N = 4 super-YangMills, with gauge group SU(N ), by means of
partition functions [1]. While this problem remains unsolved to date, essentially due to the difficulty of defining what is meant by these dual operators for finite N , there are other equally
interesting sectors of (super-)YangMills theories where more progress with counting has been
made, examples being in free N = 4 super-YangMills and in chiral ring sectors involving BPS
operators.
For N = 4 super-YangMills, the half BPS sector consists of multi-trace operators involving
a single bosonic operator, Z. Similarly, the quarter BPS sector consists of multi-trace operators
involving two bosonic operators Z, Y while the eighth consists of ones involving three bosonic
(All these operators are here assumed to
operators Z, Y , X and two fermionic operators , .
belong to the Lie algebra of U (N ).)
For the chiral ring, the commuting/anti-commuting of these operators is at the heart of why
we can write very concise and elegant generating functions for the finite N multi-trace partition
functions [1]. Analysis of these partition functions, in terms of the counting of operators, has
become a sophisticated industry where such approaches as the so-called Plethystic Program
have provided substantial results [3,9].
This paper is devoted largely to the issue of partition functions for free field theory and particularly to the counting of gauge invariant multi-trace operators for the case of two bosonic
fundamental fields. This is of relevance to the quarter BPS sector of N = 4 super-YangMills in
the free field limit when the operators Z, Y do not commute, in contrast to the chiral ring when
they do.
The partition function for a free, massless quarkgluon gas was computed long ago [21].
This involved taking particle statistics into account using coherent state techniques and then
imposing the gauge singlet condition by integrating over the relevant gauge group. With some
modifications to the expression thus derived we may write the multi-trace partition function for
some generic bosonic/fermionic fundamental fields in terms of an integral over the gauge group,
involving the single particle partition function [22,23]. This is the starting point here.
For U (N ) gauge theories we may easily write down the integral, though, even for this case,
its evaluation is far from simple. One approach (which we adapt here for the SU(2) case) is to
rewrite the expression in terms of an N -fold contour integral, whereby it may in principle be
evaluated by summing the contributions from poles inside origin centred unit discs, in each of
the N complex planessimilar techniques have been used in [10]. Due to the number of poles
this becomes unfeasible for higher values of N . Another approach is to use the fact that the
complex integral that interests us provides for an inner product for symmetric polynomialssee
Macdonald [24, pp. 363372], for a related discussion. Taking this point more seriously reveals
an alternative route to evaluating the free field theory partition function which exposes not only
the large N case in an almost trivial way, but also how and where this differs from the finite N
case.
This treatment also reveals an alternative interpretation of the free field partition function at
finite N it is related to a gauge group average of the cycle polynomial for the symmetric permutation group (after a certain identification of letters with gauge group valued variables). This
point is not dwelt on further here though makes the connection between the partition function
and Polya enumeration explicit. For single trace operators at large N this connection has already
434
been made [22,25] for N = 4 super-YangMills, whereby the partition function for single trace
operators is related to the cycle polynomial for the cyclic permutation group.1
Another issue is how to use the expression for the free field partition function to give explicit counting of gauge invariant multi-trace operators in a YangMills theory, with gauge group
U (N ). The case for one bosonic fundamental field has been widely discussed and, for finite N ,
the operators are counted by partition numbers of non-negative integers into at most N parts
(which is here denoted by pN (n)no closed formula for these numbers for arbitrary N , n exist,
though they have a nice generating function). Here, this result is re-derived, from a symmetric
function perspective, by employing the well-known CauchyLittlewood formula.
To proceed further with counting, for the quarter BPS sector of N = 4 super-YangMills, for
instance, character methods prove to be both natural and indispensable. Characters in relation to
conformal field theories prove to be very convenient for encoding the allowable representations
[15,2933] and for studying related partition functions, [3336]. For N = 4 super-YangMills,
it was shown in [33] that, if we are to distinguish among primary operators with differing
conformal dimensions, spins and R-symmetry charges, such counting is most easily achieved
using reductions of the full N = 4 superconformal characters, in certain limits that isolate
corresponding sectors of short/semi-short operators. (One such limit corresponds to the index
constructed in [1].) This point may be easily illustrated for quarter BPS primary operators in
N = 4 super-YangMills, the case of two bosonic fundamental fields here. (See [37] for an explicit construction and counting of quarter BPS operators.) For N = 4 super-YangMills, the
counting of quarter BPS primary operators is complicated by the fact that, if we are to keep track
of differing R-symmetry representations, any partition function restricted to this sector must be
expanded in terms of U (2) characters (or two-variable Schur polynomials). Denoting some partition function restricted to this sector by Z(t, u), where t, u are letters corresponding to the fields
Z, Y , then by expanding
Z(t, u) =
n

n=0 m=0
N(n,m) s(n,m) (t, u),
s(n,m) (t, u) =
t n+1 um t m un+1
,
t u
(1.1)
in terms of two-variable Schur polynomials s(n,m) (t, u), we obtain the numbers N(n,m) of gauge
invariant quarter BPS operators belonging to the [m, n m, m] SU(4)R R-symmetry representation and having conformal dimension n + m (so that they are superconformal primary highest
weight states in the corresponding quarter BPS supermultiplets). (The case m = 0 counts gauge
invariant half BPS primary operators.)
Here, the free field partition function is thus expanded in terms of Schur polynomials, depending on the same variables as the one particle partition function, the two boson case being
a specialisation. This is quite naturally achieved using the CauchyLittlewood formula (and, if
we include fermions, another formula due to Littlewood). Generally, we may obtain a result that
relates the counting numbers to a sum over Kronecker coefficients. These arise naturally in the
theory of the symmetric permutation group, though remain somewhat mysterious from a combinatorial perspective.
Specialising to the two boson case, a recursive procedure is employed here for the counting of
multi-trace quarter BPS operators in free field theory at finite N . This issue was given considerable discussion for the large N case in [33]here the results of [33] are generalised in terms of
1 For more recent applications of symmetric polynomials and Polya enumeration to super-symmetric quantum mechanical models, analysed in terms of Fock space methods, see [2628].
435
a generating function that may be employed to count quarter BPS operators for any R-symmetry
charges, at large N . Asymptotic counting is also addressed in the latter case for the numbers
N(n,m) in (1.1) for large n and fixed m.
To complete the discussion, counting of quarter BPS operators in the chiral ring of N = 4
super-YangMills is investigated in terms of expanding over U (2) characters as in (1.1). An
explicit formula is given for the corresponding finite N partition function, with a short combinatorial interpretation given in terms of plane partitions, and specialised to large N . For the
latter case, the exponential behaviour of the numbers N(n,m) in (1.1) is found for n and m both
comparably large. This behaviour is consistent with a special case addressed in [3,9]. By way
of completion, a similar discussion for an arbitrary number of bosonic fundamental fields in the
chiral ring is included.
Two appendices are included; Appendix A establishes some notation used for partitions and
gives some standard results for the symmetric group and symmetric polynomials, Appendix B
gives some tables of numbers of quarter BPS operators in free N = 4 super-YangMills, with
gauge group U (N ), for which explicit formulae are given in the main text. Footnotes contain
further details and points of clarification.
2. Free field partition functions
We start from the single particle partition function which is here denoted by f (t) for some
variables2 t = (t1 , t2 , . . .). The general form of f (t) is

f (t) =
(2.1)
a i ti ,
i
where each ti is a letter corresponding to a fundamental field and ai are signs, being +1 for a
bosonic field, or 1 for a fermionic field.
For compact gauge Lie group G, the multi-trace partition function is then given by [21] (see
[22,23] for refinements)

1
n
n
ZG (t) = dG (g) exp
(2.2)
,
f t R g
n
G
n=1
Hurwitz) measure dG (g) for g G (so that

involving the Haar (or G-invariant, or
d
(g)
F
(g)
=
d
(g)
F
(gh)
=
G
G
G
G
G dG (g) F (hg) for all h G and G dG (g) = 1)
and where R (g) is the character for the R representation of G, assuming that the fundamental
fields transform in identical gauge group representations R.
For G = U (N ), so that for any matrix U U (N ) we may write U = V V , where V is a
unitary matrix and = diag(ei1 , . . . , eiN ), 0 i 2 , and for some F (U ) = F (), independent of V , then we may write

dU (N ) (U ) F (U ) =
U (N )
1
(2)N N !
2
N
0 j =1
dj
e k eil
2 F (),
(2.3)
1k<lN
2 In what follows roman letters are used to denote a collection of variables and, for x = (x , . . . , x ), y = (y , . . . , y )
i
j
1
1
for example, the shorthand x is used to mean (x1 , . . . , xi ) and z = xy to mean z = (z11 , . . . , zij ) where zrs = xr ys .
The latter convenient notation has been used by Macdonald [24].
436
which is, of course, related to the Weyl parametrisation of U (N ). Thus, for such F (U ), the
left-hand side of (2.3) simplifies to an integral over the N torus. Of course, as R (U ) generally
depends on linear combinations of tr(U j )k tr(U l )m for various non-negative integers j , k, l, m
then any function of R (U ) is an example of such an F (U ).
We are interested in the case where R = Adj is the adjoint representation so that for U (N ) we
have that Adj (U ) = tr U tr U (while for SU(N ) then Adj (U ) = tr U tr, U 1). For U (N ) we
then find that, using (2.2) with (2.3),

2
N
N

1
1
ik
il
2
n
in(j k )
ZU (N ) (t) =
e e
dj
exp
e
f t
.
(2)N N !
n
j
=1
n=1
j,k=1
1k<lN
0
(2.4)
We may write (2.4) as an N -fold contour integral by first making the variable change zj = eij
so that the integrals in (2.4) are around unit circles in each zj complex plane and then we obtain

N
1
1
1
dzi
1
n
exp
,
(z) z
f t pn (z)pn z
ZU (N ) (t) =
(2.5)
(2i)N N!
zi
n
i=1
n=1

n
where (z) = 1i<j N (zi zj ) is the Vandermonde determinant and pn (z) = N
i=1 zi is a
power symmetric polynomialsee Appendix A for a brief discussion of symmetric polynomials.
This integral may then in principle be evaluated by deforming the contours so as to extract the
residues at poles within the discs |zj | < 1, 1 j N .
A crucial observation is that, for some N variable symmetric polynomials g(z), h(z), then

N

dzi
1
g, hN = h, gN =
(2.6)
(z) z1 g(z)h z1 ,
N
(2i) N !
zi
i=1
acts as an inner productthis is easy to see in terms of Schur polynomials which provide an
orthonormal basis for symmetric polynomials. The reader may now wish to peruse Appendix A
where notation regarding partitions and a short discussion of symmetric polynomials is included.
2.1. The general and large N cases for U (N )
For application of inner products to (2.5) we have that, in terms of power symmetric polynomials p (z) for partitions ,

1

a
a
1
n
1
exp
=
f t pn (z)pn z
f tn n pn (z)an pn z1 n
n
nan an !
n=1 an =0
n=1
1

f (t)p (z)p z1 ,
z
(2.7)
with the definitions of

z =
nan an !,
n=1
f (t) =

a
f tn n ,
n=1
(2.8)

being in terms of the frequency representation of (1a1 , 2a2 , . . .) with n1 nan = ||, the
weight of the partition (note that the frequency representation of is simply a convenient
re-ordering of the parts of ).
437
In (2.7) the numbers z have a standard combinatorial

interpretationfor a given
permutation

Sm with a1 1-cycles, a2 2-cycles, etc., so that n1 nan = m = ||, then z = n1 nan an !
is the size of the centraliser Z = { ; Sm , 1 = } of Sm . (This may be easily seen
as under conjugation of by then can permute the cycles of length n among themselves
in an ! ways and/or render a cyclic rotation on each of the individual cycles in nan ways.) More
details of the symmetric group are to be found in Appendix A.
We may immediately observe that (2.7) represents a sum over cycle polynomials of the symmetric group Sm . This is given by, for letters u1 , u2 , . . . , um ,3
Cm (u) =
a1 +2a2 ++mam ,m
a1 ,...,am 0
m

n=1
1
nan a
n!
un an =
1
u ,
z
(2.9)
m
where m means that is any partition of msee Appendix A for notationand

u =
un an ,
(2.10)
n=1
in terms of the frequency representation of above. Identifying un = f (tn )pn (z)pn (z1 ) then
we may rewrite
ZU (N ) (t) =

m=0
1
(2i)N N!

N

dzi
(z) z1 Cm (u),
zi
(2.11)
i=1
the sum of the U (N ) group averages of each of the cycle polynomials Cm (u). (Physically, the
interpretation is that the cycle index for the symmetric permutation group accounts for particle
statistics while integration over the gauge group imposes the gauge singlet condition. For purposes of clarity, the U (N ) case has been focused upon here, though from the form of (2.2) it is
easy to see how this generalises for other gauge groups whereby the letters un = f (tn )R (g n )
for the fundamental fields transforming in identical gauge group representations, R.)
Directly from (2.7), in terms of the inner product (2.6), then
ZU (N ) (t) =
1
1
2
,
f (t)p , p N =
f (t)
z
z
(2.12)
||
()N
where on the right-hand side of (2.12) we have used an expression for the inner product of two
power symmetric polynomials expressed in terms of the characters of the symmetric group, given
in Appendix A. (Here () means the number of non-zero parts of the partition .)
3 This formula is easy to see from the definition of the cycle polynomial for a subgroup G of S . This is given by
m
1
1 j1 (g)
u1
um jm (g) =
|Kg |u1 j1 (g) um jm (g) ,
|G|
|G|
gG
Kg
where ji (g) denotes the number of i cycles in the unique decomposition of g into disjoint cycles and Kg denotes the
conjugacy classes of G with class representatives g. The size of the conjugacy class Kg is given by |Kg | = |G|/|Zg |
where Zg is the centraliser of g G. For the present case then, G = Sm , |Sm | = m!, and |Z | = z , where gives the
cycle structure of Sm , and thus, for the corresponding conjugacy class K , |K | = m!/z .
438
Using a result of Appendix A (essentially orthogonality relations for symmetric group characters), (2.12) may be rewritten as

1
2
.
f (t) +
f (t)
ZU (N ) (t) =
(2.13)
z
||N
||>N
||
()N
In the large N limit, ZU (N ) (t) simplifies considerably as only the first term in (2.13) need be
considered. Using the frequency representation of then
ZU () (t) =
f (t) =

a
f tn n =
n=1 an =0
n=1
1
,
1 f (tn )
(2.14)
a result which has been obtained using Polya counting methods for single trace operators and
saddle point approximations [22,23].
Higher order corrections
in ||, the weight of the partition , to (2.13) may be obtained by suc
cessive evaluation of ||,()N ( )2 . One method is to employ the MurnaghanNakayama
rule, used to compute using skew hooks and Young diagrams. (A readable account of the
MurnaghanNakayama rule may be found in [38], though of course it is explained in many standard textbooks that discuss the symmetric group.)
For the case of || = N + 1 then we may observe that,
2
2 2

2

(2.15)
=
= z , = 1N+1 ,
N+1
()N
N+1
()N+1
since the partition = (1N+1 ) is the only one excluded among those partitions of N + 1 with
() N . By applying the MurnaghanNakayama rule we may determine, for = (1L ),

2
1 for || = L,
=
(2.16)
0 otherwise,
since in this case is just a sign. (This may be easily seen as there is only one possible way to
remove successive skew hooks, which in this case are just column Young diagrams of length i ,
from the (1L ) column Young diagram to leave one of normal shape, in this case, another column
Young diagram.) Thus, using (2.15) with (2.16) in (2.13), we obtain

1
1
2
.
f (t)
f (t) +
f (t)
ZU (N ) (t) =
(2.17)
z
z
||N+1
||=N+1
||>N +1
||
()N
By a similar line of argument we may do the same for the case of || = N + 2. We have that,
for 1 = (1N+2 ) and 2 = (2, 1N+1 ),
2
2 2 2
2 2
(2.18)
=
1 2 = z 1 2 .
N+2
()N
N+2
()N+2
We may determine, for = (2, 1L ),

2
2
= (a1 1) for || = n1 nan = L + 2,
0
otherwise.
(2.19)
439
Using (2.16), (2.19) with (2.18) in (2.17) then we obtain

ZU (N ) (t) =
||N +2
f (t)
N +1||N+2
||>N +2
1
f (t)
z
||=N+2
1
f (t)
z
2
1
,
f (t)
z
(2.20)
||
()N
where, in the frequency representation of ,

1
(a1 1)2
=
=

z
z

1
1
1
nan an !.
+
a1 ! (a1 1)! (a1 2)!
(2.21)
n=2
We may proceed in this manner to compute explicit higher order corrections though this
becomes cumbersome save for the first few cases as shown. (For || > N + 2 the corrections
will
contributions
from (2.16) and (2.19) as well as extra ones coming from

always involve
2
2
(
)
||,()||
||,()N ( ) .)
2.2. The one boson case for U (N )
For the case of one bosonic fundamental field (applicable to half BPS operators for N = 4
super-YangMills), we have f (t) = t in (2.5), so that we may write
ZU (N ) (t) =
1
(2i)N N !

N
N

dzi
1
(z) z1
.
zi
1 tzj zk1
i=1
j,k=1
(2.22)
To evaluate this integral we may use the CauchyLittlewood formula,

M
L

i=1 j =1
1
=
1 xi yj
s (x1 , . . . , xL )s (y1 , . . . , yM ),
(2.23)
()min{L,M}
where the sum on the right-hand side is over all partitions such that the corresponding Young
diagrams have no more than min{L, M} rows, () min{L, M}. With xi = tzi , yi = zi1 ,
i = 1, . . . , N , in (2.23), so that s (tz1 , . . . , tzN ) = t || s (z), and employing also (2.6) and the
orthonormality of Schur polynomials, we may easily obtain,

t || s , s N =
t || .
ZU (N ) (t) =
(2.24)
()N
()N
By changing summation variables so that i i+1 = ai , i = 1, . . . , N 1, N = aN then we

may write
ZU (N ) (t) =

a1 ,...,aN =0
t a1 +2a2 ++N aN = PN (t),
(2.25)
440
where4
PN (t) =
N

i=1
1
.
1 ti
(2.26)
Of course this is nothing other than the generating function for the number pN (n) of partitions
of n into no more than N parts since, by definition,

(2.27)
||,n = pN (n),
()N
so that by the above
pN (n)t n = PN (t).
(2.28)
n=0
This makes explicit the connection between ZU (N ) (t) and the partition numbers pN (n).
2.3. The SU(2) gauge group case

Here we first consider f (t) = kj =1 tj in (2.1) so that the variables 0 ti < 1 represent k bosons in the single particle partition function. For such fields transforming in the adjoint representation of SU(2) then (2.2) simplifies significantly. For any U SU(2) we may
write U = V V , where V is unitary and = diag(ei , ei ), for 0 < 2 , so that for
F (U ) = F ( ), then in usual Weyl parametrisation,
1
dSU(2) (U ) F (U ) =
2
d sin2 F ( )
0
SU(2)
1
=
4
d (1 cos )F

d (1 cos )F

,
2
4
0
1
=
2
2
(2.29)
where
F ( ) = F ( + ) is assumed in writing the last line. In the present case, F (U ) = F ( ) =

n
n
2i + e2i + 1 = 2 cos 2 + 1,
n1 f (t )Adj (U )/n, where Adj (U ) = tr(U ) tr(U ) 1 = e
so that
1
ZSU(2) (t) =
2
2
d (1 cos )
0
k

j =1
1
.
(1 tj )(1 tj ei )(1 tj ei )
(2.30)
4 1/P (t) = (1 t n ) is commonly called the Euler function, denoted by (t). P (t) = p(n)t n acts as
n=1
n=0
a generating function for the number of unordered partitions of n, p(n). Note that pN (n) = p(n) for n N , i.e., the
number of partitions of n into no more than N parts is the same as the total number of partitions of n so long as n N .
441
Making the variable change z = ei , and using that F ( ) = F ( ) is even, then

1
ZSU(2) (t) =
2i

1
dz
,
(1 z)
z
(1 tj )(1 tj z)(1 tj z1 )
k
(2.31)
j =1
where the integral is around the unit circle |z| = 1. The residues in (2.31) may be easily computed
since all the relevant (simple) poles in the disc |z| < 1 occur at the points z = tj . Thus
ZSU(2) (t) =
k
k

tik1
1 ti2 j =1
i=1
j =i
1
.
(ti tj )(1 ti tj )(1 tj )
(2.32)
This partition function has an interesting interpretation from a group theory perspective. We
may write,

1+t
=
n (z)t n ,
1
(1 tz)(1 tz )
(2.33)
n=0
where
1
j (z) =
zj + 2 zj 2
1
2
z z
12
1
j Z,
2
(2.34)
is an SU(2) character, corresponding to the spin j irreducible representation, Rj . Now the integral in (2.31) acts as an SU(2) inner product,

dz
1
j , k =
(2.35)
(1 z)j (z)k z1 = j k ,
2i
z
for j, k N. Thus, from (2.31) with (2.33) and (2.35),
k

1 tj2 ZSU(2) (t) =
j =1
n1 nk , 1t1 n1 tk nk ,
(2.36)
n1 ,...,nk =0
acts as a generating function for the number of singlets

in the decomposition of the SU(2) representation Rn1 Rnk .5 By using that n (z) = nj=n zj we may use the Cauchy residue
theorem to compute explicitly that
n1 nk , 1 =
2n1

j1 =0
2nk
(j1 ++jk ,n1 ++nk j1 ++jk ,n1 ++nk +1 ).
(2.37)
jk =0
5 Generating functions for products of Lie algebra representations have been considered elsewhere, in [39] for instance. A generating function for the number of singlets in n products of the fundamental times n products of the
anti-fundamental representations for SU(N ) was found by Gessel [40] in terms of Toeplitz determinants involving Bessel
functions. See also [41] for a nice physics oriented discussion of similar issues.
The special case of R1/2 R1/2
1 2n , of singlet representations.
(2n products of the fundamental) for SU(2), contains a Catalan number, n+1
n
442
If we modify the one particle partition function to include k bosons and k fermions and hence

consider (2.1) in the form f (t, t) = kj =1 tj jk=1 tj then we may similarly as above evaluate

(1 tj )(1 tj z)(1 tj z1 )
dz
1
,
(1 z)
ZSU(2) (t, t) =
(2.38)
2i
z
(1 tj )(1 tj z)(1 tj z1 )
1j k
1jk
where the contour is around the unit disc |z| = 1. So long as k > k then (2.38) receives contributions from only those simple poles at z = tj so that for this case we obtain
k
tikk1
ZSU(2) (t, t)
k>k =
1 ti2
i=1

1jk
1j k,j =i

.
(2.39)
For k k then (2.38) also receives contributions from poles at z = 0. For instance
k
ZSU(2) (t, t) k=k =

i=1
1
ti (1 ti2 )

1j,jk

1j,jk
j =i

tj (1 tj )
,
tj (1 tj )
(2.40)
where the last term on the right-hand side of (2.40) comes from the simple pole at z = 0.
These formulae should be useful for computing the multi-trace partition functions, for fundamental fields transforming in an SU(2) gauge group, in other sectors of N = 4 super-YangMills.
For instance, after a suitable identification of the variables tj , tj with variables in single particle
partition functions for semi-short sectors of N = 4 super-YangMills, described in detail in [33],
then (2.40) should allow for an explicit expression for corresponding multi-trace partition functions. They may also be useful for computing the N = 4 superconformal index of [1] for SU(2)
gauge group, or at least for restrictions of it such as described in [33] or [42].
3. Counting operators in free N = 4 super-YangMills
In this section, the counting of half and quarter BPS operators for free N = 4 super-Yang
Mills, when the fundamental fields transform in the adjoint representation of U (N ), is discussed
in some detail.
3.1. Counting operators directly
We may, of course, proceed to count multi-trace half and quarter BPS primary operators directly, in terms of the fundamental fields, Z, for half BPS operators and Z, Y , for quarter BPS
operators.
(Z, Y ) forms a U (2) doublet, where U (2) has generators given by a subset of the SU(4)R
generators, Hi , Ei , 1 i 3, where Hi are the Cartan sub-algebra generators and Ei
are ladder operators satisfying (in the ChevalleySerre basis) [Hi , Ej ] = Kij Ej , with
[Kij ] being the usual SU(4) Cartan matrix. The U (2) generators consist of the SU(2) generators H2 , E2 , where explicitly [(H1 , H2 , H3 ), E2 ] = (1, 2, 1)E2 , along with the
443
generator H1 + H2 + H3 , whose eigenvalues give the conformal dimensions in this case,

[H1 + H2 + H3 , (Z, Y )] = (Z, Y ) [33]. Explicitly, we have that [E2+ , Z] = 0, [E2 , Z] = Y ,
[(H1 , H2 , H3 ), Z] = (0, 1, 0)Z, [(H1 , H2 , H3 ), Y ] = (1, 1, 1)Y so that an operator involving
nZs and mY s transforms in the [m, n m, m] SU(4)R R-symmetry representation.
For k-trace half BPS primary operators transforming in the [0, n, 0] SU(4)R R-symmetry
representation, with conformal dimension n, then in terms of the fundamental field Z a basis is
provided by,

tr Z n1 tr Z nk ,
k
ni = n.
(3.1)
i=1
We have that, due to trace identities for finite N , tr(Z n ) for n > N is expressible in terms of a sum
over multi-trace operators of the form (3.1), for k > 1, and thus, a minimal basis for multi-trace
half BPS primary operators consists of (3.1) for all 1 k n and with every ni N , ordered so
that n1 n2 nk 0, i.e., so that (n1 , . . . , nk ) is a partition of n where each part ni N .
With this restriction, the number of multi-trace half BPS primary operators for a given n is
N(n) = pN (n),
(3.2)
since the number pN (n), in (2.27), of partitions of n into N parts is the same as the number
of partitions of n in which each part is N see [43] for a simple proof employing generating
functions.
For quarter BPS operators belonging to the [m, nm, m] SU(4)R R-symmetry representation,
a basis for k-trace operators is

n1j m1j
nkj mkj
tr
(3.3)
Z Y
Z Y
nij = n,
mij = m,
tr
,
j
i,j
i,j
where there is a choice of ordering in each trace. (Note that the m = 0 case corresponds to the
half BPS case already considered.) Using the basis provided by (3.3) for all allowable k, then
to avoid over-counting of multi-trace quarter BPS operators, the cyclicity of each trace and also
trace identities for finite N must be accounted for. Assuming that this is done, let M(n,m) denote
the number of elements in this minimal basis for multi-trace operators of the form (3.3). Then, to
obtain the number N(n,m) of multi-trace quarter BPS primary operators in the SU(4)R representation [m, n m, m], the number of U (2) descendants, in the SU(4)R representation [m, n m, m],
of multi-trace quarter BPS primary operators, in SU(4)R representations [j, n + m 2j, j ],
0 j m 1, must be subtracted from M(n,m) . (These descendants arise due to the relation
[E2 , Z] = Y . Acting with (E2 )mj on the highest weight state in the SU(4)R representation
[j, n + m 2j, j ] we obtain a descendant in the SU(4)R representation [m, n m, m].) The number of such U (2) descendants coincides with N(n+mj,j ) , the number of corresponding primary
operators. In this way, we obtain that
M(n,m) = N(n,m) + N(n+1,m1) + + N(n+m1,1) + N(n+m) ,
(3.4)
so that N(n,m) = M(n,m) M(n+1,m1) may be obtained recursively for each m.

We may illustrate by counting all multi-trace quarter BPS primary operators in the [1, n
1, 1] R-symmetry representation. In this case a basis for (k + 1)-trace operators is provided by

tr Z n1 tr Z nk tr Z j Y ,
k

i=1
ni = n j.
(3.5)
444
Cyclicity of traces implies that we may arrange Y as shown, to avoid over-counting. U (N ) trace
identities imply, similarly as for the half BPS case, that a minimal basis for multi-trace operators
requires j < N and each ni N in (3.5) for every 1 k n j , so that (n1 , . . . , nk ) forms
a partition of n j , with every part N . Thus, by a similar argument as for the half BPS

case, M(n,1) = N1
j =0 pN (n j ). Finally, to ensure that only primary operators are counted
then we must subtract off contributions from descendants of half BPS primary operators in the
[0, n + 1, 0]SU(4)R representation, of which there are pN (n + 1). Using (3.4) with (3.2) we then
conclude that
N(n,1) =
N1
pN (n j ) pN (n + 1),
(3.6)
j =0
gives the number of multi-trace quarter BPS primary operators in the [1, n 1, 1] R-symmetry
representation.
Counting in this fashion becomes more difficult for greater m and now a procedure is described employing symmetric polynomials to find a generating function for the numbers of
multi-trace quarter BPS primary operators in the [m, n m, m] SU(4)R representation, for
m = 0, 1, 2 at finite N and for any n, m at large N . This generating function is subsequently
used to provide asymptotic counting for fixed m, large n in the large N limit.
3.2. Counting operators via expansion of partition functions in Schur polynomials
For k bosonic fundamental fields, we may take f (t) =
written as
ZU (N ) (t) =
1
(2i)N N!

N
dzi
i=1
zi
j =1 tj
N
k

(z) z1
j =1 r,s=1
in (2.1) so that (2.5) may be
1
1 tj zr zs1
(3.7)
Often it is the case that such partition functions should be expanded in terms of s (t), the k
variable Schur polynomial labelled by partitions . An example is provided by (1.1) for counting
multi-trace quarter BPS operators. We may use the CauchyLittlewood formula (2.23) to expand
in this way, to obtain

N s (t),
ZU (N ) (t) =
(3.8)
()min{k,N 2 }
where
N =
1
(2i)N N!

N

dzi
(z) z1 s zz1 ,
zi
(3.9)
i=1
has components zi zj1 , 1 i, j N .

where
From Macdonald [24] we have that
s (xy) =
s (x)s (y),
zz1
,||
(3.10)
445
in terms of Kronecker coefficients,

1
1
=
( ) ( ) ( ) =
,
||!
z
S||
(3.11)
||
being a sum over irreducible S|| characters evaluated at S|| , related to a sum over irreducible S|| characters evaluated on the conjugacy classes labelled by the partitions on the
right of (3.11). Using (2.6) along with the orthonormality property of Schur polynomials we find
that
N =
(3.12)
.
||
()N
The situation becomes much more involved if we include also k fermionic fields, so that (2.1)

may be written in the form f (t, t) = kj =1 tj jk=1 tj , and attempt to expand ZU (N ) (t, t) in
terms of products of Schur polynomials s (t)s (t). Such expansion is required for counting, for
instance, for the free field partition function in the eighth BPS sector of N = 4 super-Yang
Mills. In this case the partition function is expanded, analogous to (1.1), in terms of SU(2|3)
characters, which may be expressed in terms of a linear combination of products of two-variable
and three-variable Schur polynomials. (See [33] for a discussion of counting for the eighth BPS
sector along these lines.) Including fermions, (3.7) becomes modified by
ZU (N ) (t, t) =
1
(2i)N N !

N
dzi
i=1
zi

(z) z1
N

1 tj zr zs1
1j k r,s=1
1jk
1 tj zr zs1
(3.13)
To achieve the expansion, we may use the CauchyLittlewood formula (2.23) along with another
formula of Littlewood,
M
L

(1 + xi yj ) =
i=1 j =1
s (x1 , . . . , xL )s (y1 , . . . , yM ),
(3.14)
()L,( )M
where is the partition conjugate to (where the rows and columns of the Young diagram
corresponding to are interchanged) and where the sum is restricted to those whereby the
M. We
corresponding Young diagrams have at most L rows, () L, and M columns, ()
may thus write

ZU (N ) (t, t) =
(3.15)
N, s (t)s (t),
2
)N
()min{k,N 2 } ()k,(
where
(1)||
N, =
(2i)N N!

N

dzi
(z) z1 s zz1 s zz1 .
zi
(3.16)
i=1
Obviously these numbers are considerably more involved than those in (3.9). We may of course
use (3.10) again to interpret (3.16) in terms of Kronecker coefficients.
446
3.3. Counting quarter BPS operators by symmetric polynomial methods

The two bosonic fundamental field case is now focused upon.6 In particular, the numbers
N(n,m) in (1.1) are evaluated using results of the last sub-section.
We may proceed to evaluate N recursively. The simplest case is for N = N(n) , whereby
introducing a formal variable t then it is clear, by (2.23) with (2.22), (2.25) and (2.26), that

N
N

1
dzi
1
1
(3.17)
N(n) t n =
(z)
z
= PN (t),
N
(2i) N !
zi
1 tzj zk1
n=0
i=1
j,k=1
so that, by (2.28), N(n) is given by (3.2).
More generally to evaluate N(n,m) from (3.9) we may use, for y = zz1 ,
s(m) (y)s(n) (y) = s(n,m) (y) + s(n+1,m1) (y) + + s(n+m1,1) (y) + s(n+m) (y),

s(m) zz1 =
s (z)s z1 ,
(3.18)
m
()N
where the expression in the first line of (3.18) may be easily seen using Young tableaux multiplication rules while (2.23) determines the expression in the second line. From (3.9) with (3.18),
we may find a useful generating function, in terms of a formal variable t , for the numbers in (3.4)
as follows,
(m)
FN (t) =
M(n,m) t n
n=0
1
=
(2i)N N!
1

1
dzi
(z) z s(m) zz
s(n) zz1 t n
zi
i=1
n=0

N
N
1
1
dzi
1
1
z
=
(z)
z
s
(z)s
(2i)N N!
zi
1 tzj zk1
m
i=1
j,k=1
=

m
()N
()N
||
t s s , s s N ,
(3.19)
so that we may write

1
1
dt 1 (m)
(m1)
N(n,m) =
F (t) n+1 FN
(t) ,
2i
t tn N
t
(3.20)
6 The two boson case leads to an interesting generalisation of an identity in [44] involving LittlewoodRichardson
, the coefficients that appear in the decomposition s (x)s (x) = c s (x). With f (t) in (2.1) given
coefficients c

by f (t) = t1 + t2 , and expanding appropriately the corresponding integrand in (3.7) using (2.23); then using (2.6), the
= 0 if || = || + ||)
orthonormality of Schur polynomials and the result (2.14), we obtain (note that c
ZU () (t1 , t2 ) =

,
t1 || t2 || s s , s s =

,,
||+||
which reduces to Theorem 4.1 of [44] if we take t1 = t2 = t .
2
t1 || t2 || c
=
n1
1
1 t1 n t2 n
which allows for recursive determination of N(n,m) .

Applying this to the case of N = N(n,1) we have, from (3.18)

s(1) zz1 = s(1) (z)s(1) z1 ,
447
(3.21)
so that, from (3.19),

FN(1) (t) =

(N(n,1) + N(n+1) )t n =
t || s(1) s , s(1) s N .
(3.22)
()N
n=0
Using (again, this may be easily seen from Young tableaux multiplication rules)
s(1) (z)s (z) =
N
(3.23)
s+er (z),
r=1
for {er ; 1 r N, er es = rs } being usual orthonormal vectors, we find that

FN(1) (t) =
N

(N(n,1) + N(n+1) )t n =
t ||
s+er , s+es N .
()N
n=0
(3.24)
r,s=1
Now for any partition , s+er , s+es N vanishes unless er = es for any r, s and r1 r > 0
for r = 2, . . . , N , due to
s(1 ,...,r1 ,r +1,...,N ) (z) = 0 for r1 = r , r > 1.
(3.25)
Changing summation variables to those in (2.25) then we have, with the definition (2.26),
FN(1) (t) =

(N(n,1) + N(n+1) )t n
n=0
t a1 ++N aN +
a1 ,...,aN 0
N1
ti
i=0
N1
t a1 ++N aN
r=1 a1 ,...,aN 0
ar 1
t a1 ++N aN =
a1 ,...,aN 0
1
PN1 (t).
1t
(3.26)
Thus, using (2.28), (3.2) with (3.26),7

N(n,1) =
n
pN1 (j ) N(n+1) =
j =0
n
pN1 (j ) pN (n + 1).
(3.27)
j =0
7 This formula agrees with (3.6) due to N 1 p (n j ) = n

j =0 pN 1 (j ) which follows because the corresponding
j =0 N
generating functions match,

1
N

n=0 j =0

pN (n j )t n = 1 + t + + t N 1 PN (t) =
n

1
pN 1 (j )t n .
PN 1 (t) =
1t
n=0 j =0
448
For the case of N = N(n,2) we have that, from (3.18),

s(2) zz1 = s(2) (z)s(2) z1 + s(1,1) (z)s(1,1) z1 ,
so that, from (3.19), we have

(2)
t || s(2) s , s(2) s N + s(1,1) s , s(1,1) s N .
FN (t) =
(3.28)
(3.29)
()N
Using
s(2) (z)s (z) =
N

r=1
s(1,1) (z)s (z) =
s+2er (z) +
s+er +es (z),
1r<sN
s+er +es (z),
(3.30)
1r<sN
along with (3.25) and

s(1 ,...,r1 ,r +2,...,N ) (z) = s(1 ,...,r +1,r1 +1,...,N ) (z),
(3.31)
for the cases where r1 = r , we may obtain, with the definition (2.26),
1 t N+1
1
PN1 (t) +
PN2 (t),
(3.32)
2
(1 t)(1 t )
(1 t)(1 t 2 )
||
where
||the first contribution comes from t s(2) s , s(2) s N while the second comes from
t s(1,1) s , s(1,1) s N . Since the partition number pk (n) = 0 for n = 1, 2, . . . we may
write, using (2.28),8
(2)
FN (t) =

1
Pk (t) =
pk (n i 2j )t n .
(1 t)(1 t 2 )
(3.33)
n,i,j =0
Thus, from (3.32) with (3.27),

N(n,2) =
n+1
pN1 (j )
j =0
i,j =0 (pN2 (n i 2j ) + pN1 (n i 2j ))
if n N,

+
i,j =0 (pN2 (n i 2j ) + pN1 (n i 2j )
pN1 (N + 1 n i 2j ))
if n N + 1.
(3.34)
8 This is a special case of the following: for any f (n), n Z, that satisfies f (n) = 0, n = 1, 2, . . . , then we may (at
least formally) write
Pk (t)

n=0
f (n)t n =

n,i1 ,...,ik =0
f (n i1 2i2 kik )t n .
449
Tables of the numbers (3.2), (3.27) and (3.34) are given in Appendix B for some few cases of
n, N . Notice from these tables that the numbers N(n,m) below the diagonal line N n + m for
a given n are the same for all N . This is a general feature that derives from values of N(n,m) for
N n + m, which numbers may be obtained from a corresponding generating function that is
now constructed.
Using these techniques, we may provide a consistency check of (3.17), (3.26), (3.32) along
with a general result for N(n,m) for high enough values of N , N m + n. This employs the
orthogonality property of power symmetric polynomials p (z) (in the large N limit) along with
1
p (z)
s(n) (z) =
z
n
i 2
i n

1
1
1
=
i1 +2i2 ++nin ,n p1 (z)i1
p2 (z)
pn (z) . (3.35)
i1 !i2 ! in !
2
n
i1 ,...,in =0
Using the trivial identity p (xy) = p (x)p (y) then from (2.6), (3.19) with (3.35) we have
(m)
F
(t) =

n=0 n m

n=0 n m
1
p p , p p t n
z z
1
p , p t n
z z

z n
=
t ,
z z
(3.36)
n=0 n m
where for (1a1 , 2a2 , . . .) being the frequency representation of and (1b1 , 2b2 , . . .) being that of
then has frequency representation (1a1 +a2 , 2a2 +b2 , . . .) so that || = n + m. This agrees with
(m)
FN (t) in a series expansion up to O(t Nm ) (since the last equation in (3.36) is also valid for
finite N so long as || = n + m N , by a result of Appendix A). Now, since
(aj + bj )!
z
=
,
z z
aj !bj !
(3.37)
j =1
we obtain from (3.36) that,

(m)
(t) =
F
n=0 a1 ,...,an =0 b1 ,...,bm =0

m
b1 ++mbm ,m
b1 ,...,bm =0
a1 ++nan ,n b1 ++mbm ,m

b1 ,...,bm =0
j =1

(aj + bj )! j aj 1
t
aj !bj !
1 tj
j =1 aj =0
b1 ++mbm ,m
m

(aj + bj )! n
t
aj !bj !
1
(1 t j )bj +1
j =1
j >m

j >m
1
.
1 tj
For the first few cases we have that, with PN (t) as defined in (2.26),
for m = 0,
P1 (t)
(m)
P
(t)
for m = 1,
F (t) = 1t
2
P (t) for m = 2,
(1t)(1t 2 )
(3.38)
(3.39)
450
whose series expansion agrees with (3.17), (3.26), (3.32) up to O(t Nm ) for, respectively, m =
0, 1, 2. We may use (3.20) with (3.36) to determine N(n,m) exactly for N n + m.
3.4. Asymptotic counting of quarter BPS operators at large N
Asymptotic counting for the one boson case in the large N limit, for which, with PN (t) as
defined in (2.26), with p(n) being the total number of (unordered) partitions of n,
ZU () (t) = P (t) =
p(n)t n ,
(3.40)
n=1
is the multi-trace partition function, entails finding an asymptotic value for the partition number
p(n) for large n. This may be achieved by performing a saddle point approximation of p(n) =
1
n1 . The function P (t) has a large singularity at t = 1, but in addition has
2i dt P (t)t
singularities at all other roots of unitysee [45] on the validity of ignoring these contributions
asymptotically. This method was used by Hardy and Ramanujan to find their celebrated formula,
here given in a less detailed form as,

2
1
n ,
p(n) exp
(3.41)
3
4n 3
which was improved by Rademacher to give p(n) exactly. Their method relied crucially on the
modular properties of P (t).
Focusing now on the two bosonic fundamental field case for which, in the large N limit,
(3.20) with (3.38) gives exact counting, at issue is first finding asymptotic values for the numbers
Q(n, m, b) = Q(n, m, b1 , . . . , bm ), with constraint equation m
j =1 j bj = m, defined by
1

1
=1+
Q(n, m, b)t n .
j )bj +1
1 tj
(1
t
j >m
j =1
n=1
m

(3.42)
Having found these we may then attempt to find the dominant contribution to (3.20) with (3.38)
for large N . In order to give asymptotic values for Q(n, m, b) we may follow [46] and apply a
formula due to Meinardus which gives a general result for the generating function
1 tn
an
=1+
n=1
r(n)t n .
(3.43)
n=1
A detailed version of Meinardus theorem may be found in [46] but for purposes of brevity we
may note that it implies that, as n ,
1/(+1)

r(n) C n exp A ( + 1) ( + 1)n
(3.44)
( + 1)/ ,
s
where (s) = j =1 j is the Riemann zeta function and the constants C, , , A are determined
by the auxiliary Dirichlet series,
D(s) =

aj
j =1
js
(3.45)
which must converge for Re(s) > , a positive real number, and possess an analytic continuation
in the region Re(s) c, 1 < c < 0, such that, in this region, D(s) is analytic except at a simple
451
pole at s = where it has residue A. In terms of , A then

(12D(0))/2(+1)
1
C=
A ( + 1) ( + 1)
exp D (0),
2(1 + )

1
= D(0) 1 /( + 1).
2
(3.46)
Applying Meinardus theorem to the case of (3.42), clearly we have

D(s) =
m

bj
j =1
js
+ (s),
so that, assuming m =
residue A = 1. Using
(3.47)
j =1 j bj
1
bj ,
D(0) = +
2
m
j =1
is fixed, D(s) has a simple pole at s = = 1 where it has
m
1 1
exp D (0) =
,
2 j =1 j bj
then, from (3.44) with (3.46), we may easily determine that, as n ,

mj=1 bj

m
6n
1
2
1
exp
n .
Q(n, m, b)
3
j bj
4n 3
(3.48)
(3.49)
j =1
This reduces to (3.41) when bj = 0, 1 j m, whereby Q(n, 0, . . . , 0) = p(n). Using (3.20)

for (3.38) with (3.42) and (3.49) then, as n ,

N(n,m) =
Q(n, m, b)
Q(n + 1, m 1, b)
b1 ,...,bm 0

m
j =1 j bj =m
1

4n 3
b1 ,...,bm1 0
m1
j =1 j bj =m1
m

6n
2
exp
n ,
(3.50)
since Q(n, m, m, 0, . . . , 0), for b1 = m, bj = 0, j > 1, dominates over all other terms in (3.50).
This gives asymptotic values for the numbers in (1.1), for counting quarter BPS operators,
transforming in [m, n m, m] SU(4)R representations, in the large N limit of free N = 4 superYangMills, as previously described.
4. Counting operators in the chiral ring of N = 4 super-YangMills
For the purposes of counting operators in the chiral ring of N = 4 super-YangMills, we
denote corresponding multi-trace partition functions by CU (N ) (t).
The generating function for CU (N ) (t) for the case of one bosonic fundamental field has been
written in the form [1,3,9]
C(, t) =

n=0

1
=
N CU (N ) (t),
1 t n
N=0
(4.1)
452
so that acts as a chemical potential for the rank of the gauge group U (N ). The equivalence
CU (N ) (t) = ZU (N ) (t) = PN (t), with ZU (N ) (t) as in (2.25), is actually a special case of the qBinomial theorem. Writing, see [43] for notation,

(a; q)k = (1 a)(1 aq) 1 aq k1 ,
(4.2)
then the q-Binomial theorem is, for |x|, |q| < 1,

(a; q)k
k=0
(q; q)k
xk =
(ax; q)
.
(x; q)
(4.3)
(Identifying = x and q = t and setting a = 0 in (4.3), so that 1/(; t) = C(, t) above and
1/(t; t)N = PN (t) in (2.26), then CU (N ) (t) = ZU (N ) (t) = PN (t) straightforwardly. This special
case of the q-Binomial theorem is due to Euler.)
For the two boson case, so that the single particle partition function is given by f (t, u) =
t + u for some t , u, then the generating function for the finite N chiral ring partition function
CU (N ) (t, u) is given by [1,3,9]
C(, t, u) =

n,m=0

1
=
N CU (N ) (t, u).
1 t n um
(4.4)
N=0
This function is more difficult to analyse in terms of counting though has been investigated by
Stanley [47] in relation to partitionsthere it has been dubbed the double Eulerian generating
function. Through use of the CauchyLittlewood formula, then we may expand CU (N ) (t, u) in
terms of partitions of N as,

h (t)h (u),
CU (N ) (t, u) =
(4.5)
N
where

h (t) = s 1, t, t 2 , . . . ,
(4.6)
so that using an identity for Schur polynomials to be found in [24,47] then

N
1i<j N (1 t i j +j i )(1 ui j +j i )
(tu) i=1 (i1)i .
CU (N ) (t, u) =
N
i=1 (t; t)i +N i (u; u)i +Ni
N
(4.7)
Eq. (4.5) with (4.6) has a natural interpretation in terms

of plane partitions in that, for being
all column-strict plane partitions of shape , || = i,j ij ,9

||
h (t) = s 1, t, t 2 , . . . =
t .
(4.8)
9 See [47] for a detailed description of plane partitions. Briefly, a column-strict plane partition of shape is an array =
(ij ) of non-negative integers with finitely many non-zero entries, that is arranged in a Young tableaux with shape see
Appendix Asuch that the numbers ij are weakly decreasing along each row, ij
i j +1 0, and strictly decreasing
down each column, ij > i+1 j 0. The sum of the parts of is given by | | = i,j ij . (Note that in contrast to the
definition in [47], here we are allowing ij = 0, for some i, j , to be a part of the plane partition with shape .)
453
Obviously, (4.5) with (4.8) generalise for other chiral ring sectors. (For a different connection between the double Eulerian generating function and major indices of permutations see
[47, p. 385].) As an illustration of (4.5) with (4.8), we may consider the case N = 2 whereby
= (2, 0), (1, 1) gives the two possible partitions of 2. For = (2, 0) (corresponding to a Young
diagram with a single row of two boxes) 11 12 0 gives all column-strict plane partitions
of shape (2, 0), while for = (1, 1) (corresponding to a Young diagram with a single column of
two boxes) then 11 > 21 0 gives all column-strict plane partitions of shape (1, 1). Thus,
h(2,0) (t) =
t 11 +12 =
1
,
(1 t)(1 t 2 )
t 11 +21 =
t
,
(1 t)(1 t 2 )
11 ,12 0
11 12
h(1,1) (t) =
11 ,21 0
11 >21
(4.9)
so that, from (4.5) for N = 2,

CU (2) (t, u) = h(2,0) (t)h(2,0) (u) + h(1,1) (t)h(1,1) (u)
1 + ut
,
=
(1 t)(1 t 2 )(1 u)(1 u2 )
(4.10)
which is the correct result as may be verified by extracting the 2 coefficient in an expansion
of (4.4) up to O( 2 ).
In the large N limit,
CU () (t, u) =

n1 ,n2 0
n1 +n2 >0
1
,
1 t n1 un2
(4.11)
upon which attention is shortly focused.

For the numbers N(n,m) N (n,m) counting quarter BPS primary operators for the chiral ring
of N = 4 super-YangMills, belonging to [m, n m, m] SU(4)R R-symmetry representations,
as in (1.1), we have10

2
1
dt du CU (N ) (t, u)s(n,m) t 1 , u1 t 1 u1 .
N (n,m) =
(4.12)
2
8
(n,m) ,
These may be more conveniently evaluated in terms of the numbers in (3.4) M(n,m) M
counting all chiral ring quarter BPS operators in the [m, n m, m] SU(4)R representation, given
by

(n,m) = 1
dt du CU (N ) (t, u)t n1 um1 ,
M
(4.13)
(2i)2
(n,m) M
(n+1,m1) . Defining P (n) to be the number of column-strict
so that N (n,m) = M

plane partitions of shape so that || = i,j ij = n, then, from (4.5) with (4.8) and (4.13),
10 This formula employs the orthonormality relation of Schur polynomials described here and has appeared in a similar
context in [33, Appendix B].
454
(n,m) =
M
P (n)P (m). Thus,

P (n)P (m) P (n + 1)P (m 1) ,
N (n,m) =
N
(4.14)
N
counts chiral ring quarter BPS primary operators in SU(4)R representations [m, n m, m] for
any n, m at finite N .
4.1. Asymptotic counting for chiral ring BPS operators at large N
For asymptotic counting of operators in the chiral ring of N = 4 super-YangMills at large N ,
a relatively crude method is employed here which nevertheless captures the exponential behaviour of counting numbers of interest. This method is based on saddle point approximations of
functions near a dominant singularitysee [45] for a useful summary. (Often for physical applications in thermodynamics, e.g., for entropy formulae, we are interested only in the exponential
behaviour of such numbers anyhow.)
To illustrate, we consider the one boson case in the large N limit again. We first find a convenient approximating function as follows,

1
n
P (t) =
= exp
ln 1 t
1 tn
n=1
n=1

2
s
exp ds ln 1 t
(4.15)
= exp
,
6 ln t
0
which has an easier singularity structure. (The approximation in the second step may be justified by the EulerMaclaurin formula for approximating sums by integrals.) Using (4.15) then for
large enough n,

1
1
2
n1
dt P (t)t
dt eg(t) , g(t) =
p(n) =
(4.16)
n ln t.
2i
2i
6 ln t
We may approximate the latter
integral for large n by noting that the dominant contribution is at
6n

/
1 for which
the saddle point t = e

2
2 3 2/3n
g (t ) =
6n e
= ,
n,
g (t ) = 0,
g(t ) =
(4.17)
3
so that, for t = t t ,
p(n) e
e
2n/3
2n/3
Thus,

ln p(n)
1
2i
1
2
2
n,
3
dt e 2 t
1
2
ds e 2 s =
2
1
2
2n/3
(4.18)
(4.19)
455
which captures the correct behaviour of ln p(n) for large n, according to (3.41).
We may proceed analogously for the quarter BPS chiral ring multi-trace partition function at
large N , (4.11), which we approximate by

1
n1 n2
CU () (t, u) =
=
exp
ln
1
t
u
1 t n1 un2
n1 ,n2 0
n1 +n2 >0
n1 ,n2 0
n1 +n2 >0

(3)
v w
exp
= exp
dv dw ln 1 t u
.
ln t ln u
(4.20)
0 0
In this case we have, from (4.13),

(n,m) 1
dt du eg(t,u) ,
M
(4.21)
(2i)2
where
(3)
g(t, u) =
(4.22)
n ln t m ln u,
ln t ln u
for n, m large. The dominant contribution to the integral, for n, m both large and of the same
order, occurs about the point (t , u ) (1, 1) where
2 )1/3
t = e( (3)mn
u = e( (3)nm
for which

g(t , u ) = 3 3 (3)nm,
2 )1/3
(4.23)
=
= 0,
g(t, u)
g(t, u)
t
u
(t ,u )
(t ,u )
1

2
2 1/3
g(t,
u)
= 2 (3)1 n5 m1 3 e2( (3)mn ) = ,
2
t
(t ,u )
1

2 1/3
g(t, u)
= 2 (3)1 m5 n1 3 e2( (3)nm ) = ,

2
u
(t ,u )
1

2 1/3
2 1/3
= (3)1 n2 m2 3 e( (3)mn ) +( (3)nm ) = .
g(t, u)
tu
(t ,u )
(4.24)
So long as m, n are both large and of the same order, the saddle point approximation is justified
and we obtain

1
2
2
(n,m) e3 3 (3)nm 1
dv dw e 2 (v +w +2 vw)
M
4 2
= h(, , )e
3 3 (3)nm
(4.25)
where
h(, , ) =
1
1
2 1/3
2 1/3
(3)m2 n2 3 e( (3)mn ) ( (3)nm ) . (4.26)
2 3
We thus have that for n, m both comparably large,

(n,m) 3 3 (3)nm,
ln M
(4.27)
456
so that

(n,m) M
(n+1,m1) ) ln M
(n,m) 3 3 (3)nm.
ln N (n,m) = ln(M
(4.28)
It is difficult to check the consistency of this result given the dearth of literature on these types
of multi-variable generating functions and their asymptotic behaviour, however, we may consider
the simpler function, also considered in [9],
CU () (t, t) =

n=1

1
=
E(r)t r ,
(1 t n )n+1
(4.29)
r=0
where, in terms of the counting numbers N (n,m) , from (1.1),

CU () (t, t) =
n

(n m + 1)N (n,m) t
n=0 m=0
n+m
E(r) =
[ 2 r]
(r 2m + 1)N (rm,m) .
m=0
(4.30)
Hence, E(r) counts quarter BPS primary operators in the chiral ring of N = 4 super-YangMills,
transforming in [m, n m, m] SU(4)R representations, with the same conformal dimensions
r = n + m. Extracting the dominant contribution to ln E(r) from (4.30), which occurs at the
maximum value of m, mM = [ 12 r], and using (4.28), we obtain

33
ln E(r) ln N (rmM ,mM )
(4.31)
2 (3)r 2 .
2
By considering (4.29) directly, we may employ Meinardus theorem (described in the third
section) to find the behaviour of ln E(r) as r . Note, however that Meinardus theorem may
not be applied directly to (4.29) since the corresponding auxiliary Dirichlet series (3.45), with
aj = j + 1, has two simple poles. To overcome this difficulty we split (4.29) into a product of two
functions, both separately amenable to application of Meinardus theorem. One is the reciprocal
of the Euler function, P (t) in (4.15). The other, the MacMahon function, is given by
M(t) =

n=1

1
=
q(r)t r ,
n
n
(1 t )
(4.32)
r=0
and has been considered in a similar context as here in [3].11 Writing

CU () (t, t) = P (t)M(t),
with P (t) as in (4.15), so that, using (4.29),

E(r) =
p(r1 )q(r2 ),
(4.33)
(4.34)
r1 ,r2 0
r1 +r2 =r
we may find the asymptotic behaviour of ln E(r), as r , by extracting the dominant contribution from (4.34) using the asymptotic behaviour of p(r), q(r). The auxiliary Dirichlet series
11 The relation of M(t) in (4.32) to plane partitions is given a description in [46]. Briefly, q(r) gives the number

of ordinary plane partitions , so that ij i+1 j > 0, ij i j +1 > 0, with | | = i,j ij = r. p(r) < q(r) as
ordinary partitions are a special case of plane partitions. In fact, the formula for ln q(r) found here is a special case of
a more exact asymptotic formula first found by Wright [48] for the number of plane partitions q(r) of the number r.
457
for M(t) in (4.32) is, from (3.45) with aj = j ,

D(s) = (s 1),
(4.35)
which has a simple pole at s = = 2, at which the residue is A = 1. Thus, from (3.44),

3 3
ln q(r)
(4.36)
2 (3)r 2 .
2
This is consistent with (4.31) as the dominant contribution to ln E(r) comes from the r1 = 0 term
in (4.34) (since p(r) q(r) as r ) so that ln E(r) ln q(r).
It has not escaped attention that the method used here, to capture the exponential behaviour
(n,m) , may be easily extended to chiral ring sectors
of asymptotic values for the numbers M
other than the quarter BPS one. Suppose, for simplicity, that Zj , 1 j k 1, are commuting
bosonic fundamental
fields, in the U (N ) Lie algebra, so that the single particle partition function
is given by f (t) = k1
j =1 tj , in terms of the corresponding letters tj . Let M(m1 ,...,mk1 ) denote the
number of independent operators involving products of m1 Z1 s, m2 Z2 s, etc., in corresponding
multi-trace operators. The multi-trace partition function, in the large N limit, is given by,

1
CU () (t) =
(4.37)
nk1 ,
n1
1 t1 tk1
n ,...,n 0
1
k1
n1 ++nk1 >0
which may be crudely approximated by, similarly as before,12

k1

vk1
v1
CU () (t) exp
dvj ln 1 t1 tk1
0
j =1

(1)k+1 (k)
= exp
.
ln t1 ln tk1
(4.38)
Thus, without going into as much detail, for the analogue of (4.27) we have (assuming mj are all
comparably large)

(m ,...,m ) g(t , . . . , t ) = k k (k)m1 mk1 ,
ln M
(4.39)
1
k1
1
k1
) is the value of
where g(t1 , . . . , tk1
g(t1 , . . . , tk1 ) =
(1)k+1 (k)
m1 ln t1 mk1 ln tk1 ,
ln t1 ln tk1
) (1, . . . , 1), where

at the saddle point (t1 , . . . , tk1

(ln t1 , . . . , ln tk1
) = k (k)m1 mk1 (1/m1 , . . . , 1/mk1 ),
(4.40)
(4.41)
12 For Li (x) =
j n
n
j 1 x /j being the usual polylogarithm, with Lin (1) = (n), n > 1, Lin (0) = 0, then with the
convention Li1 (x) = ln(1 x), the following integral
1
dx
Lin (zx) = Lin+1 (z),
x
may be useful for showing this, after a suitable change of variables.
458
so that
g(t1 , . . . , tk1 )
tj
(t ,...,t
= 0,
j = 1, . . . , k 1.
(4.42)
k1 )
Eq. (4.39) is consistent with a result implied by Meinardus theorem. The function,13
CU () (t, . . . , t)
1 tn
nk2 /(k2)!
n=1
c(k, r)t r ,
(4.43)
r=0
has auxiliary Dirichlet series, from (3.45) with aj = j k2 /(k 2)!,

D(s) =
1
(s + 2 k),
(k 2)!
(4.44)
which has a simple pole at s = = k 1 at which the residue is A = 1/(k 2)!, so that,
from (3.44),

k k
(k 1) (k)r k1 .
ln c(k, r)
(4.45)
k1
Eq. (4.45) is precisely the result that may be obtained from (4.39) if we maximise the product m1 mk1 , subject to the constraint k1
j =1 mj = r, for which the solution is mj = mM =
r/(k 1) (relaxing the constraint that mj be non-negative integers, which is irrelevant asymp (mM ,...,mM ) ln c(k, r).
totically), so that ln M
This is applicable to counting multi-trace operators in the eighth BPS chiral ring sector for
N = 4 super-YangMills with fundamental fields Z, Y , X involving m1 Zs, m2 Y s, m3 Xs.
Expanding the corresponding partition function (4.37), with k = 4, in terms of Schur polynomials s(m1 ,m2 ,m3 ) (t), m1 m2 m3 0, similar to (1.1), the expansion coefficients N (m1 ,m2 ,m3 )
count spinless multi-trace primary operators transforming in the [m2 + m3 , m1 m2 , m2 m3 ]
SU(4)R R-symmetry representation, with conformal dimensions m1 + m2 + m3 [33]. Just as
in (4.28), asymptotically ln N(m1 ,m2 ,m3 ) ln M(m1 ,m2 ,m3 ) . This counting, however, ignores con which it may be important to include in order to give
tributions of the fermionic fields , ,
correct counting of eighth BPS chiral ring operators.
5. Conclusions
There are some obvious questions not answered by this work. The first is whether or not the
approach in the second section using symmetric polynomials can give insight into thermodynamics at finite N , such as for the Hagedorn transition, for example. While it gives the large N
expression (2.14) in an elementary way, its wider applicability or usefulness to such questions
is unclear. The approach is undoubtedly useful for finding exact expressions for counting numbers (as in (3.2), (3.27) and (3.34) for quarter BPS operators) and (3.9), (3.16) may be useful
13 This may be easily seen from (4.37), as the number of solutions to k1 m = n, where m are non-negative
j
j =1 j

k2 /(k 2)!. More properly,
which,
to
leading
order
in
large
n,
behaves
like
n
integers, is the binomial number n+k2
k2
we should split the product CU () (t, . . . , t) into pieces separately amenable to Meinardus theorem, as for the prior case
for k = 3, however, just as for that case, the numbers c(k, r) dominate, and so other contributions are ignored here.
459
for analysing counting for more complicated sectors of N = 4 super-YangMills, with gauge
group U (N ).
The second question is how the arguments employing symmetric polynomial techniques here
may be extended to other gauge groups, the most pertinent being perhaps SU(N ). Arguments
here employing (2.6) and the orthonormality property of Schur polynomials should remain
largely unaffected for SU(N ). Exact values for counting numbers obtained here should require
some modification for SU(N ), though asymptotic values may be unchanged.
The third question concerns asymptotic values for counting numbers and how these may be
improved. The asymptotic counting formulae given in such papers as [3,9] for chiral ring sectors are special cases of formulae such as those of Hardy and Ramanujan, Meinardus, etc., all of
which derive from single variable generating functions. It is hoped that the expressions (3.50),
(4.28), (4.39), given here for asymptotic counting of BPS operators, that distinguishes between
differing R-symmetry charges, represents a serious attempt at going beyond consideration of
single variable generating functions.14 Improving upon these formulae will require more sophisticated techniques, perhaps along the lines used to find those of Hardy and Ramanujan or
Meinardus and employing any modular properties of the multi-variable functions involved. This
issue may also be important for microscopic counting for black holes, as the BPS solutions found
thus far, for N = 4 superconformal symmetry, depend on special values of R-symmetry charges
[1720]see [50] for a related detailed discussion.
Thus far, the elegant results for finite N partition functions for chiral ring sectors have been
interpreted from a largely geometric perspectiveit may be interesting to investigate more how
such results are related to the theory of random matrices and/or symmetric polynomials.
Acknowledgements
I warmly thank Yang-Hui He, Paul Heslop, Hugh Osborn, Christian Romelsberger and Christian Saemann for useful comments and discussions. This work is supported by an IRCSET (Irish
Research Council for Science, Engineering and Technology) Post-doctoral Fellowship.
Appendix A. Partitions, symmetric group characters, symmetric polynomials and inner
products
A generic partition is any finite or infinite sequence = (1 , 2 , . . .) of non-negative integers in decreasing order 1 2 0 containing only finitely many non-zero terms. Often
it is convenient to omit zero entries. The non-zero entries are called the parts of the number of
which
we denote by (). The sum of the parts of is called the weight of which we denote by
|| = i i . If || = L then is a partition of L and we write L. For convenience we sometimes write in its frequency representation which is a reordering of the entries in , indicating
the number of times each successive
non-negative integer occurs, (1a1 , 2a2 , . . .) so that exactly
an of the parts of equal n and || = n1 nan .
14 After submission of the first version of this paper to the electronic archive, I received an e-mail from Hai Lin pointing
out an interesting comparison between (4.28) here and (2.14)of [49], obtained in quite a different context. The two
formulae are essentially the same given the numerical value 3 3 (3) = 3.189 . . ., correct to three decimal places.
460
In terms of standard Young diagrams, corresponds to a Young diagram of shape , with 1

boxes in the first row, 2 boxes in the second row, etc.; the number of parts () is simply the
number of rows and the weight || is the total number of boxes.
For the symmetric group, SN , the irreducible representations are labelled by partitions
N see [38] for a useful summaryso that, for X ( ), SN , being a corresponding matrix representation, then the character of SN in the representation X is ( ) = tr(X ( )).
The characters are class functions so that they take a constant value on conjugacy classes and,
recalling that for SN the conjugacy classes K are labelled by partitions N , corresponding to
the cycle structure of a class representative, then ( ) = for all K . With z as defined
1/2
in (2.8), a crucial property of SN characters is the orthogonality of the matrix [z ] . This

gives rise to the orthogonality relations, for , N (see also Chapter IV of [51] for a related
discussion)
1
1
(A.1)
( ) ( ) =
= ,
N!
z
SN
and
N
= z .
(A.2)
N
A convenient basis for N variable symmetric polynomials are Schur polynomials s (z) =
s (z1 , . . . , zN ) labelled by = (1 , . . . , N ). They may be expressed in a number of ways
[24,47]. For convenience we write them as
s (z) = a+ (z)/a (z),
(A.3)
where , the Weyl vector, is given by = (N 1, N 2, . . . , 1, 0) and

a+ (z) =
sgn( )z (1) 1 +N1 z (j ) j +Nj z (N) N = det zi j +Nj ,
(A.4)
SN
with

a+ (z) = det zi Nj =
(zi zj ) = (x),
(A.5)
1i<j N
being the Vandermonde determinant. Schur polynomials s (z) have

a standard interpretation as
corresponding to the characters of irreducible U (N ) (or, for i zi = 1, SU(N )) Lie algebra
representations. Here, gives the shape of the Young tableaux for the corresponding U (N ) Lie
algebra representation.
For = (1 , . . . , N ) and = (1 , . . . , N ) where i , i Z then, from the definition
of (2.6) along with (A.3),

s , s N =
(A.6)
sgn( ) =
sgn( ) ,
SN
SN
where, for any = (1 , . . . , N ), = ( + ) is the shifted Weyl reflection of by ,

with the action of SN on being given by (1 , . . . , N ) = ( (1) , . . . , (N) ). (Eq. (A.6) is
a reflection of s (x) = sgn( )s (x) for any partition and SN note that this property is
useful for showing (3.25), (3.31). has a standard interpretation in terms of U (N ) Lie algebra
461
representationsfor the Verma module with dominant integral highest weight having orthonormal basis labels , 1 2 N 0, then , for = idSN , are the orthonormal basis
labels for the highest weights of all invariant sub-modules. This fact may be exploited to derive
the Weyl character formula (A.3) for the irreducible U (N ) Lie algebra representation with dominant integral highest weight having orthonormal basis labels , or, alternatively, Young tableaux
of shape .)
When , are partitions so that 1 N 0 and 1 N 0 then (A.6) reduces
to a well-defined inner product,
s , s N = ,
(A.7)
so that in this case the Schur polynomials are orthonormal. Note that in order that s (x) be nonzero for some arbitrary partition then () N , so that (A.7) is zero for () > N or () > N .
Another basis for symmetric polynomials are the power symmetric polynomials, p (z), for
= (1 , . . . , L ) L, which are defined by
p (z) = p1 (z)p2 (z) pL (z),
pn (z) =
N
zin .
(A.8)
i=1
Note that there is no longer the restriction that () N as for Schur polynomials.
Symmetric group characters may be used to relate the two bases for symmetric polynomials
[24,47] so that, with the definition of z in (2.8),
s (z) =
1
p (z)
z
(A.9)
N
(a theorem of Frobenius) and, for L,

s (z).
p (z) =
(A.10)
L
()N
(N )
(Eq. (A.9) with = 1 for all N is useful for obtaining (3.35).)

Regarding the inner product (2.6), then using (A.10) along with (A.7), we then have that, for
L, M,

.
p , p N = LM
(A.11)
L
()N
Orthogonality of symmetric group characters implies, from (A.2), that for ||, || N
then (A.11) simplifies to, with the definition of z in (2.8),
p , p N = z .
(A.12)
Appendix B. Tables
This appendix gives some tables (Tables B.1B.3) of numbers of quarter BPS operators in
free N = 4 super-YangMills, with gauge group U (N ).
462
Table B.1
Numbers of multi-trace half BPS primary operators, with conformal dimension n and belonging to [0, n, 0] R-symmetry
representations, for free N = 4 SYM with U (N ) gauge group. (For every N there is one [0, 0, 0] and [0, 1, 0]
representationthese are omitted above.)
N(n)
Nn
10
11
1
2
3
4
5
6
1
2
2
2
2
2
1
2
3
3
3
3
1
3
4
5
5
5
1
3
5
6
7
7
1
4
7
9
10
11
1
4
8
11
13
14
1
5
10
15
18
20
1
5
12
18
23
26
1
6
14
23
30
35
1
6
16
27
37
44
Table B.2
Numbers of multi-trace quarter BPS primary operators, with conformal dimension n + 1 and belonging to [1, n 1, 1]
R-symmetry representations, for free N = 4 SYM with U (N ) gauge group. (n = 0, 1 cases are all zero.)
N(n,1)
Nn
10
11
2
3
4
5
6
7
1
1
1
1
1
1
1
2
2
2
2
2
2
4
5
5
5
5
2
5
7
8
8
8
3
8
12
14
15
15
3
10
16
20
22
23
4
13
23
30
34
36
4
16
30
41
48
52
5
20
40
57
69
76
5
23
49
74
92
104
Table B.3
Numbers of multi-trace quarter BPS primary operators, with conformal dimension n + 2 and belonging to [2, n 2, 2]
R-symmetry representations, for free N = 4 SYM with U (N ) gauge group. (n = 0, 1 cases are all zero.)
N(n,2)
Nn
10
11
3
4
5
6
7
8
3
3
3
3
3
3
5
6
6
6
6
6
10
14
15
15
15
15
14
21
25
26
26
26
21
36
44
48
49
49
27
50
66
74
78
79
36
73
101
118
126
130
44
96
142
171
188
196
55
130
200
251
281
298
65
163
267
346
398
428
References
[1] J. Kinney, J.M. Maldacena, S. Minwalla, S. Raju, An index for 4-dimensional super-conformal theories, hepth/0510251.
[2] D. Martelli, J. Sparks, S.T. Yau, SasakiEinstein manifolds and volume minimisation, hep-th/0603021.
[3] S. Benvenuti, B. Feng, A. Hanany, Y.H. He, Counting BPS operators in gauge theories: Quivers, syzygies and
plethystics, hep-th/0608050.
[4] D. Martelli, J. Sparks, Dual giant gravitons in SasakiEinstein backgrounds, Nucl. Phys. B 759 (2006) 292, hepth/0608060.
[5] A. Basu, G. Mandal, Dual giant gravitons in AdSm Y n (SasakiEinstein), hep-th/0608093.
[6] A. Butti, D. Forcella, A. Zaffaroni, Counting BPS baryonic operators in CFTs with SasakiEinstein duals, hepth/0611229.
463
[7] A. Hanany, C. Romelsberger, Counting BPS operators in the chiral ring of N = 2 supersymmetric gauge theories
or N = 2 braine surgery, hep-th/0611346.
[8] L. Grant, K. Narayan, Mesonic chiral rings in CalabiYau cones from field theory, hep-th/0701189.
[9] B. Feng, A. Hanany, Y.H. He, Counting gauge invariants: The plethystic program, hep-th/0701063.
[10] D. Forcella, A. Hanany, A. Zaffaroni, Baryonic generating functions, hep-th/0701236.
[11] S. Lee, S. Lee, J. Park, Toric AdS4 /CFT3 duals and M-theory crystals, hep-th/0702120.
[12] G. Mandal, N.V. Suryanarayana, Counting 1/8 BPS dual giants, hep-th/0606088.
[13] A. Sinha, N.V. Suryanarayana, Two-charge small black hole entropy: String-loops and multi-strings, JHEP 0610
(2006) 034, hep-th/0606218.
[14] N.V. Suryanarayana, Half-BPS giants, free fermions and microstates of superstars, JHEP 0601 (2006) 082, hepth/0411145.
[15] A. Barabanschikov, L. Grant, L.L. Huang, S. Raju, The spectrum of YangMills on a sphere, hep-th/0501063.
(1998) 231, hep-th/9711200.
[17] J.B. Gutowski, H.S. Reall, Supersymmetric AdS(5) black holes, JHEP 0402 (2004) 006, hep-th/0401042.
[18] J.B. Gutowski, H.S. Reall, General supersymmetric AdS(5) black holes, JHEP 0404 (2004) 048, hep-th/0401129.
[19] Z.W. Chong, M. Cvetic, H. Lu, C.N. Pope, General non-extremal rotating black holes in minimal five-dimensional
gauged supergravity, Phys. Rev. Lett. 95 (2005) 161301, hep-th/0506029.
[20] H.K. Kunduri, J. Lucietti, H.S. Reall, Supersymmetric multi-charge AdS(5) black holes, JHEP 0604 (2006) 036,
hep-th/0601156.
[21] B.-S. Skagerstam, On the large Nc limit of the SU(Nc ) colour quarkgluon partition function, Z. Phys. C 24 (1984)
97.
[22] B. Sundborg, The Hagedorn transition, deconfinement and N = 4 SYM theory, Nucl. Phys. B 573 (2000) 349,
hep-th/9908001.
[23] O. Aharony, J. Marsano, S. Minwalla, K. Papadodimas, M. Van Raamsdonk, The Hagedorn/deconfinement phase
transition in weakly coupled large n gauge theories, Adv. Theor. Math. Phys. 8 (2004) 603, hep-th/0310285.
[24] I.G. Macdonald, Symmetric Functions and Hall Polynomials, second ed., Clarendon Press, Oxford, 1995.
[25] A.M. Polyakov, Gauge fields and spacetime, Int. J. Mod. Phys. A 17S1 (2002) 119, hep-th/0110196.
[26] E. Onofri, G. Veneziano, J. Wosiek, Supersymmetry and combinatorics, math-ph/0603082.
[27] R. De Pietri, S. Mori, E. Onofri, The planar spectrum in U (N )-invariant quantum mechanics by Fock space methods:
I. The bosonic case, JHEP 0701 (2007) 018, hep-th/0610045.
[28] M. Bonini, G.M. Cicuta, E. Onofri, Fock space methods and large N , J. Phys. A 40 (2007) F229, hep-th/0701076.
[29] V.K. Dobrev, E. Sezgin, in: J.D. Hennig, W. Lcke, J. Tolar (Eds.), Spectrum and Character Formulae of SO(3, 2)
Unitary Representations, in: Lecture Notes in Physics, vol. 379, Springer, Berlin, 1990.
[30] V.K. Dobrev, Positive energy representations of non-compact quantum algebras, in: H.D. Doebner, et al. (Eds.), Proceedings of the Workshop on Generalized Symmetries in Physics, Clausthal, July 1993, World Scientific, Singapore,
1994.
[31] F.A. Dolan, Character formulae and partition functions in higher dimensional conformal field theory, J. Math.
Phys. 47 (2006) 062303, hep-th/0508031.
[32] V.K. Dobrev, Characters of the positive energy UIRs of D = 4 conformal supersymmetry, hep-th/0406154.
[33] M. Bianchi, F.A. Dolan, P.J. Heslop, H. Osborn, N = 4 superconformal characters and partition functions, Nucl.
Phys. B 767 (2007) 163, hep-th/0609179.
[34] G.W. Gibbons, M.J. Perry, C.N. Pope, Partition functions, the Bekenstein bound and temperature inversion in antide Sitter space and its conformal boundary, Phys. Rev. D 74 (2006) 084009, hep-th/0606186.
[35] J.L. Cardy, Operator content and modular properties of higher dimensional conformal field theories, Nucl. Phys.
B 366 (1991) 403.
[36] D. Kutasov, F. Larsen, Partition sums and entropy bounds in weakly coupled CFT, JHEP 0101 (2001) 001, hepth/0009244.
[37] E. DHoker, P. Heslop, P. Howe, A.V. Ryzhov, Systematics of quarter BPS operators in N = 4 SYM, JHEP 0304
(2003) 038, hep-th/0301104.
[38] B.E. Sagan, The Symmetric Group, Representations, Combinatorial Algorithms, and Symmetric Functions,
Wadsworth and Brooks/Cole Mathematics Series, California, 1991.
[39] L. Begin, C. Cummins, P. Mathieu, Generating functions for tensor products, hep-th/9811113.
[40] I.M. Gessel, Symmetric functions and P-recursiveness, J. Combin. Theory Ser. A 53 (1990) 257.
[41] J. Carlsson, B.H.J. McKellar, SU(N ) glueball masses in 2 + 1 dimensions, Phys. Rev. D 68 (2003) 074502, heplat/0303016.
464
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
Y. Nakayama, Finite N index and angular momentum bound from gravity, hep-th/0701208.
G.E. Andrews, R. Askey, R. Roy, Special Functions, Cambridge Univ. Press, Cambridge, 1999.
J.F. Willenbring, Stable Hilbert series of S(g)K for classical groups, math/0510649.
A.M. Odlyzko, Asymptotic enumeration methods, in: R.L. Graham, M. Grtschel, L. Lovsz (Eds.), Handbook of
Combinatorics, vol. 2, North-Holland, Amsterdam, 1995.
G.E. Andrews, in: G.C. Rota (Ed.), The Theory of Partitions, in: Encyclopedia of Mathematics and Its Applications,
vol. 2, AddisonWesley, Reading, MA, 1976.
R.P. Stanley, Enumerative Combinatorics, vol. 2, Cambridge Univ. Press, Cambridge, 1999.
E.M. Wright, Asymptotic partition formulae, I: Plane partitions, Quart. J. Math. Oxford Ser. 2 (1931) 177189.
H. Lin, J.M. Maldacena, Fivebranes from gauge theory, Phys. Rev. D 74 (2006) 084014, hep-th/0509235.
H.K. Kunduri, J. Lucietti, H.S. Reall, Do supersymmetric anti-de Sitter black rings exist?, JHEP 0702 (2007) 026,
hep-th/0611351.
D.E. Littlewood, The Theory of Group Characters and Matrix Representations of Groups, Clarendon Press, Oxford,
1940.
The generalized non-linear Schrdinger model

on the interval
Anastasia Doikou , Davide Fioravanti, Francesco Ravanini
University of Bologna, Physics Department, INFN Section, Via Irnerio 46, Bologna 40126, Italy
Abstract
The generalized (1 + 1)-D non-linear Schrdinger (NLS) theory with particular integrable boundary
conditions is considered. More precisely, two distinct types of boundary conditions, known as soliton preserving (SP) and soliton non-preserving (SNP), are implemented into the classical glN NLS model. Based
on this choice of boundaries the relevant conserved quantities are computed and the corresponding equations of motion are derived. A suitable quantum lattice version of the boundary generalized NLS model is
also investigated. The first non-trivial local integral of motion is explicitly computed, and the spectrum and
Bethe ansatz equations are derived for the soliton non-preserving boundary conditions.
1. Introduction
After various investigations (cf. below for detailed references) it is now well established that
any integrable system (with finite or infinite degrees of freedom) based on the higher rank algebras glN or Uq (glN ) may be endowed with two distinct types of integrable1 boundary conditions.
These boundary conditions are known as soliton preserving (SP), traditionally recognized in the
framework of integrable quantum spin chains (finite number of degrees of freedom) [18], and
soliton non-preserving (SNP) originally introduced in the context of classical integrable field
theories (infinite number of degrees of freedom) [9], and further investigated in [1012]. SNP
E-mail addresses: doikou@bo.infn.it (A. Doikou), fioravanti@bo.infn.it (D. Fioravanti), ravanini@bo.infn.it

(F. Ravanini).
1 They are defined as those boundary conditions that preserve the integrability of the system.
doi:10.1016/j.nuclphysb.2007.08.007
466
A. Doikou et al. / Nuclear Physics B 790 [PM] (2008) 465492
boundary conditions have been also introduced and studied for integrable quantum lattice systems [6,1316], and their quantum integrability was first shown in [13]. From the physical point
of view the SP boundary conditions oblige a particle-like excitation to reflect to itself: no multiplet changing occurs. The SNP boundary conditions, on the other hand, force an excitation to
reflect to its conjugate, namely to an excitation carrying the conjugate representation. From
the algebraic perspective the two types of boundary conditions are associated with two distinct
algebras, i.e. the reflection algebra [1,2] and the twisted Yangian respectively [17,18]. The study
of the underlying algebraic structures defined by the YangBaxter and reflection equations is in
general of great consequence, both at classical and quantum level, not only for integrable systems
per se, but also for other relevant problems in theoretical physics. For instance, in the context of
the AdS/CFT correspondence [19] it is known that from the string theory side the relevant classical integrable model is a sigma model (see e.g. [20,21]), that is a field theory with infinite degrees
of freedom. From the quantum gauge theory side on the other hand the loop contributions are
apparently given by integrable quantum 1D lattice models with a finite number of degrees of freedom [22,23]. As a consequence, a crucial point would be the formulation of a discrete-quantum
counterpart of the aforementioned classical sigma model. In this respect, the knowledge of the
discrete-quantum Lax operator would facilitate the derivation of the relevant Hamiltonian, and of
the other charges in involution, as well as of the exact nested Bethe ansatz equations. In fact, up
to date only the asymptotic forms of the would-be-exact Bethe ansatz equations are known (see
e.g. [23]). Thus a rigorous derivation of these equations would undoubtedly be of great physical
significance, as proven when corrections to the asymptotic regime are available [24,25].
In the frame of classical continuum theories the SNP boundary conditions have been primarily investigated up to now [912]. Therefore it is of great importance to further analyze the other
set of boundary conditions, i.e. the SP ones within this context. In the present study we examine
both SP and SNP boundary conditions for the classical generalized NLS model, and by using
Hamiltonian methods [26] we derive the relevant integrals of motion, and also specify the corresponding classical equations of motion. Note that in [27] the quantum glN NLS model on the half
line was studied, based on the reflection algebra (i.e. SP boundaries), primarily from the point of
view of the underlying symmetry algebras. It should be stressed that although in most classical
continuum theories the SNP boundary conditions have been analyzed (see e.g. [9,11,12]) this is
the first time that such boundary conditions are implemented within the generalized NLS model.
Here we consider the classical NLS model as a simple example, however our main motivation
is to search for all possible boundary conditions in other classical theories such as affine Toda
field theories, principal chiral models and others. From the viewpoint of quantum lattice models
on the other hand the extensively analyzed boundary conditions are the SP ones, thus we focus
here mostly on the SNP case for a lattice version of NLS. In particular we consider a suitable
lattice version of the NLS model [28,29] with SNP boundary conditions, and we derive the exact
spectrum and the corresponding Bethe ansatz equations.
The outline of this article is as follows: in Section 2 we present the generic algebraic setting
for classical models on the full line and on the interval. More precisely we introduce the classical
YangBaxter equation and the underlying algebra for the system on the full line. In the case of
a system on the interval we distinguish two types of boundary conditions based on the classical
versions of reflection algebra (SP) and twisted Yangian (SNP). Next the NLS model on the full
line is reviewed and an explicit derivation of the local integrals of motion by solving the auxiliary
linear problem [26] is presented. In Section 3 being guided by the same logic and adopting
Sklyanins formulation [2] we derive the integrals of motion of the glN NLS system on the
interval with SP and SNP boundary conditions. Moreover, the corresponding classical equations
467
of motion are obtained for various boundary conditions. In addition the usual NLS model with a
reflecting impurity is investigated in the same spirit. In Section 4 a suitable lattice version of the
NLS model is investigated. After a brief review on the model with periodic boundary conditions
we deal with the model with open boundaries. First the spectrum and Bethe ansatz equations
are derived for the usual lattice NLS. Finally, the SNP boundary conditions are considered for
the generalized NLS system. The first non-trivial local integral of motion is explicitly specified
for particular choice of boundary conditions, and the spectrum and Bethe ansatz equations are
deduced for the simplest boundaries.
2. The general setting
The line of attack which will be adopted for the study of the glN NLS model with integrable
boundaries is based on the solution of the so-called auxiliary linear problem [26]. It is therefore
necessary to recall at least the basics regarding this formulation. Define the transition matrix T ,
being a solution of the following set of equations
T
= U(x, t, )T ,
x
T
= V(x, t, )T ,
t
T (x, x, t, ) = I
(2.1)
(2.2)
(2.3)
with U, V being in general n n matrices with entries functions of complex valued fields, their
derivatives, and the spectral parameter . The solution of (2.1) for x > y may be written as:
x

U(x , t, ) dx .
T (x, y, ) = P exp
(2.4)
The fact that T is a solution of Eq. (2.1) will be extensively used subsequently for obtaining the
relevant integrals of motion. Compatibility conditions of the two differential equation (2.1), (2.2)
lead to the zero curvature condition
V + [U, V] = 0,
U
(2.5)
giving rise to the corresponding classical equations of motion of the system under consideration.
There exists an alternative description, known as the r matrix approach (Hamiltonian formulation). In this picture the underlying classical algebra is manifest in analogy to the quantum
case as will become quite transparent later. Let us first recall this method for a general classical
integrable system on the full line. The existence of the classical r-matrix, satisfying the classical
YangBaxter equation

r12 (1 2 ), r13 (1 ) + r23 (2 ) + r13 (1 ), r23 (2 ) = 0,
(2.6)
guarantees the integrability of the classical system. Indeed, consider the operator T (x, y, ) satisfying

T1 (x, y, t, 1 ), T2 (x, y, t, 2 ) = r12 (1 2 ), T1 (x, y, t, 1 )T2 (x, y, t, 2 ) .
(2.7)
Making use of the latter equation one may readily show for a system in full line:

ln tr T (x, y, 1 ) , ln tr T (x, y, 2 ) = 0
(2.8)
468
i.e. the system is integrable, and the charges in involutionlocal integrals of motionmay be
attained by the expansion of the object ln tr{T (x, y, )}, based essentially on the fact that T also
satisfies (2.1).
Our main aim here is to consider the glN NLS model on the interval. For this purpose we
follow the line of action described in [26], but using Sklyanins formulation for the system on
the interval or on the half line (see also [30] for the sine Gordon on the half line). We briefly
describe this process below for any classical integrable system on the interval. In this case one
has to derive a modified transition matrix T , based on Sklyanins formulation and satisfying
the following Poisson bracket algebras R, T, i.e. classical versions of the reflection algebra and
twisted Yangian respectively. It will be convenient for our purposes here to introduce some useful
notation. Let
() = r12 () for SP,

r12
T () = T 1 () for SP,
V = antid(1, 1, . . . , 1) or
t1
r12
() = r12 () = V1 r12
()V1 for SNP,
t
T () = V T ()V for SNP,
V = antid(i, i, . . . , i)
for n even only.
(2.9)
In general V can be any matrix such that V 2 = I, for instance V = I is also a possible choice (see
e.g. [9]). Then the defining relations describing the classical reflection algebra and the twisted
Yangian respectively, may be written in the following compact form2 :

T1 (1 ), T2 (2 ) = r12 (1 2 ), T1 (1 )T2 (2 )
(1 + 2 )T2 (2 ) T2 (2 )r12
(1 + 2 )T1 (1 ).
+ T1 (1 )r12
(2.10)
To construct the generating function of the integrals of motion one also needs c-number representations of the algebra R(T) satisfying (2.10) for SP and SNP respectively, and also

K1 (1 ), K2 (2 ) = 0.
(2.11)
The modified transition matrices, compatible with the corresponding algebras R, T are given by
the following expressions [2]:
T (x, y, t, ) = T (x, y, t, )K ()T (x, y, t, )
(2.12)
and the generating function of the involutive quantities is defined as

t (x, y, t, ) = tr K + ()T (x, y, t, ) .
(2.13)
Due to (2.10) it can be shown that (3.10)

t (x, y, t, 1 ), t (x, y, t, 2 ) = 0, 1 , 2 C.
(2.14)
Technical details on the proof of classical integrability are provided in Appendix A.

By expanding ln t () in powers of 1 one recovers the local integrals of motion of the considered system, and this is achieved in the subsequent sections. Among the local integrals of
motion there exist naturally the Hamiltonian, which also provides information regarding the corresponding equations of motion. This is in fact the formulation we are going to assume for the
NLS system on the half line, although an alternative strategy would be to derive the modified
2 Note that the classical versions of the reflection equation and the twisted Yangian are provided in general by more
involved expressions for generic r matrices. In the present study we focus on r matrices satisfying r12 () = r21 ()
(r12 () = r21 ()), and in this case (2.10), are valid.
469
Lax pair, compatible with the boundary conditions chosen, and hence the associated equations of
motion (see e.g. [9]). Nonetheless, the rigorous derivation of the modified Lax pair is essentially
based on the existence of local integrals of motion [26], therefore the viewpoint adopted here is
arguably the most natural.
3. Classical local integrals of motion
The main objective in this section is to solve the auxiliary linear problem for the generalized
NLS model on the interval. Before however we proceed to the study of the model on the interval
we shall briefly review the system on the full line. In any case, these results will be relevant for
the boundary case as well.
3.1. The generalized NLS on the full line
We shall hereafter focus on the glN NLS model. Consider the classical r matrix
r() =
where P =
Eij Ej i
(3.1)
i,j =1
P is the permutation operator, and (Eij )kl = ik j l . The Lax pair for the generalized NLS model
is given by the following expressions [26]:
U = U0 + U1 ,
V = V0 + V1 + 2 V2
where (see also [31])
N1

1
U1 =
Eii EN N ,
2i
U0 =
i=1
V0 = i
N1

(3.2)

N1
( i EiN + i EN i ),
i=1

N1
( i EiN i EN i ),
i j Eij |i |2 EN N i
i,j =1
i=1
V1 = U0 ,
V2 = U1
(3.3)
and i , j

i (x), j (y) = i (x), j (y) = 0,
satisfy3 :

i (x), j (y) = ij (x y).
(3.5)
From the zero curvature condition (2.5) the classical equations of motion for the generalized
NLS model are entailed i.e.

i (x, t)
2 i (x, t)
j (x, t)2 i (x, t), i, j {1, . . . , N 1}.
+ 2
i
(3.6)
=
2
t
x
j
3 The Poisson structure for the generalized NLS model is defined as:
{A, B} = i
i L
dx

A
A
B
B
.
i (x) i (x) i (x) i (x)
(3.4)
470
It is clear that for N = 2 the equations of motion of the usual NLS model are recovered.
As already mentioned to obtain the local integrals of motion of NLS one has to expand T
(2.4) in powers of 1 [26]. Let us consider the following ansatz for T as ||

1
T (x, y, ) = I + W (x, ) exp Z(x, y, ) I + W (y, )
(3.7)

where W is off diagonal matrix i.e. W = i =j Wij Eij , and Z is purely diagonal Z =
N
i=1 Zii Eii . Also
(n)

Zii
,
n
Zii () =
Wij =
n=1
W (n)

ij
n=1
(3.8)
.
(n)
(n)
Inserting the latter expressions (3.8) in (2.1) one may identify the coefficients Wij and Zii (see
Appendix B for a detailed analysis). Notice that as i the only non-negligible contribution
(n)
from Z (n) comes from the ZN N term, and is given by:

N1
(n)
+
dx i (x)WiN (x).
L
(n)
ZN N (L, L) = iLn,1
(3.9)
i=1 L
(n)
It is thus sufficient to determine the coefficients WiN in order to extract the relevant local integrals of motion (see also [31]). Indeed solving (2.1) one may easily obtain:
(1)
(2)
WiN
(x) = i i (x),
WiN
(x) = i (x),

3
(3)
k (x)2 i (x), . . . .
WiN (x) = i i (x) i 2
(3.10)
k
From the latter formulae (3.10) and taking into account (3.7), (3.9) the local integrals of motion
of NLS may be readily extracted from ln tr T (L, L, ), i.e.
L
I1 = i
dx
L
dx
L
N1

i (x)i (x) i (x) i (x) ,
i=1
I3 = i
i (x) i (x),
i=1
I2 =
2
N1

dx

2
2

k (x) + i (x)i (x) ,

i (x)
N1

....
(3.11)
i=1
The corresponding familiar quantities for the generalized NLS are given by:
I1
,
i
and apparently
N =
P =
I2
,
i
H=
{H, P} = {H, N } = {N , P} = 0.
I3
,
i
(3.12)
(3.13)
Again in the special case where N = 2 the well-known local integrals of motion for the usual
NLS model on the full line are recovered.
471
3.2. The generalized NLS on the interval

After the review on the NLS on the full line we can come to our main concern, which is
the evaluation of the integrals of motion after implementing integrable boundary conditions. We
shall investigate subsequently both SP and SNP boundary conditions.
3.2.1. SP boundary conditions
Let us first consider the NLS model with SP boundary conditions. For this purpose c-number
solutions of the classical reflection equation are needed. A general non-diagonal K matrix satisfying the classical reflection equation is given by (see also [6])
K() = D + A + i 1 ,
D = E11 c
N1

Eii + EN N ,
A = 2k(E1N + EN1 ),
c = 4k 2 + 1.
(3.14)
i=2
Apparently in the case where k = 0 a diagonal solution is recovered

N1

K() =
Eii + EN N + i 1 .
(3.15)
i=1
The more general diagonal K matrix is given by (see also e.g. [3,5])
l

K() =
Eii +
i=1
Eii + i 1 .
(3.16)
i=l+1
The solution (3.15) may be seen as a special case of (3.16) for l = N 1.

Henceforth we set x = 0, y = L, and we focus on the case with diagonal boundaries provided by (3.15), (3.16), and K = I. Also K + () = K(, + ), K () = K(, ). The quantity
under expansion is

ln tr K + ()T (0, L, )K ()T (0, L, )

1
= ln tr 1 + W (0) K + () 1 + W (0) eZ(0,L)

1

1 + W (L) K () 1 + W (L) eZ(0,L) ,
(3.17)
where the objects with hat are simply the same as before but now . The technical details
of the relevant computations are presented in Appendix C.
(I) Let us first consider the simple boundary conditions described by (3.15). Gathering all
the information provided by Eqs. (C.2), and by explicit computations concerning the i
expansion (see Appendix C, case (b) for fore details) we conclude that the integrals of motion
for the NLS on the interval are given by:
0
I1 = 2i
L
dx
N1

i=1
i (x) i (x),
472
0
I3 = 2i
dx

2 N1
2

j (x) + i (x)i (x)

i (x)
j =1
i=1
+ 2i +
N1

N1

i (0) i (0) 2i
i=1
N1

i (L) i (L),
...,
(3.18)
i=1
the quantity I2 as expected is trivial, as in the case of sine-Gordon model on the half line. Recall
that in the whole line the quantity I2 corresponds essentially to the momentum, which is not a
conserved quantity any more. To obtain the number of particles and the Hamiltonian we simply
have to divide by 2i
I1
I3
(3.19)
,
H=
and {H, N } = 0.
2i
2i
It is clear that different choices of boundary conditions lead to distinct boundary contributions to
the integrals of motion. A more detailed description of complicated diagonal and non-diagonal
boundaries is presented in Appendix C. Notice also that for N = 2 the boundary Hamiltonian
presented in [2] is recovered. Of course we could have considered Schwartz boundary conditions
at x = L i.e. (L), (L)

= 0 and trivial right boundary K = I (that is the system is
considered on the half line), then the boundary terms appearing at x = L in the expressions of
the integrals of motion would disappear.
As already mentioned the equations of motion will be derived based on the existence of a
boundary Hamiltonian rather than on the existence of a modified Lax pair. In general, among the
integrals of motion there exists a Hamiltonian (3.18) such that the relations below
N =

i (x, t)
= H(0, L), i (x, t) ,
t
L x 0

i (x, t)
= H(0, L), i (x, t) ,
t
(3.20)
give rise to the classical equations of motion. Indeed considering the Hamiltonian H (3.18),
(3.19), we end up with the following set of equations
N1

i (x, t)
2 i (x, t)
j (x, t)2 i (x, t),
+
2
=
2
t
x
j =1

i (x, t)
i (x, t)
+
=
= 0,
i (x, t)
i (x, t)
x
x
x=0
x=L
i {1, . . . , N 1}.
(3.21)
In general the boundary Hamiltonian for the generalized NLS model may be expressed as

0
N1

2 N1
2

j (x) + (x) (x) + B,
H = dx
(3.22)
i (x)
i
i=1
j =1
where B is the boundary potential. One may write the equations of motion for a generic boundary
potential B. It is clear that the bulk part remains intact as in (3.21), and what is only modified is
the boundary conditions at x = 0, x = L depending naturally on B, i.e.

i (x, t)
B
B
i (x, t)
=
= 0.
+
+
(3.23)
x
x
i x=0
i x=L
473
Two more examples of diagonal boundaries are presented below:

(II) Consider the boundary conditions described by (3.16). The corresponding contributions to
the integrals of motion due to the presence of non-trivial boundaries are computed in Appendix C,
case (b). In this case the boundary potential (see (C.6)) is given by
N1

B=
i (0) i (0) + i (0) i (0)
i=l + +1
N1

i (0) i (0)
i=1
l

i (L) i (L) + i (L) i (L) +
i (L) i (L),
i=l +1
(3.24)
i=1
and consequently the boundary conditions for the equations of motion at x = 0, x = L now
read as:

j (x)
j + (x)
+
=
= 0, j 1, . . . , l ,
j + (x)
j (x)
x
x
x=0
x=L

j l + 1, . . . , N 1 .
j + (0) = j (L) = 0,
(3.25)
The previous case (I) may be seen as a special case of the more general diagonal boundary
conditions by setting l = N 1. Ultimately, one would like to investigate the SP boundary
conditions in the context of affine Toda field theories, something that has not been achieved up
to date. In this case, it is naturally anticipated that the corresponding equations of motion should
explicitly depend on the parameters , l , contrary to the case analyzed in [9], where no extra
free parameters associated to the boundaries occur. It is also worth stressing that in the context
of integrable spin chains the integers l appear explicitly in the corresponding Hamiltonian as
well as in the associated symmetry of the model. More precisely, it was shown in [5] that the
open spin chain with diagonal boundary conditions associated to integers l = l is gll glNl
invariant (or Uq (gll ) Uq (glNl ) invariant in the trigonometric case). The symmetry breaking
for the quantum glN NLS model due to presence of non-trivial integrable boundaries is also
discussed in [27].
(III) Finally we consider the case where K = I (Appendix C, case (c)). The boundary potential in this case is
B=
N1

N1

i (0) i (0) + i (0) i (0) +
i (L) i (L) + i (L) i (L)
i=1
i=1
(3.26)
and apparently we end up with simple Dirichlet boundary conditions

i (0) = i (L) = 0,
i {1, . . . , N 1}.
(3.27)
Note that the N = 2 case in particular was investigated classically on the half line in [32], whereas
the NLS equation on the interval was studied in [33].
3.2.2. SNP boundary conditions
Recall that in this case the object under consideration, compatible with the underlying algebra,
that is the classical version of the twisted Yangian, is

ln tr K + T (0, L, )K V T t (0, L, )V
(3.28)
474
and we choose here for simplicity K = I. Note however that a generic solution of the classical
twisted Yangian is given by any matrix K = K t (see [14]). We shall choose in what follows
V = antid(1, . . . , 1). By expanding (3.28) in powers of 1 , along the lines described in Appendix C, explicit expressions for the integrals of motion are entailed (see (C.15), (C.16)). It is worth
pointing out, bearing in mind expressions (C.15), that non-local contributions to the integrals of
motion arise. This is quite an intriguing fact and it definitely merits further investigation, which
however will be undertaken in a forthcoming work. Nevertheless, based on the formulas (C.15),
(C.16) we may explicitly express the first non-trivial conserved quantity, which is somehow a
modified number of particles, i.e.
Nm =
0
N1

dx i (x) i (x) +
i=1 L
0
dx 1 (x) 1 (x).
(3.29)
Notice that the SNP boundary modify dramatically the number of particles (see (3.11), (3.12)).
Indeed, the variation due to the integrable boundary conditions is not limited to the addition of
certain boundary terms to the bulk quantity, as is customary, but it gives rise to an alteration
of the bulk expression itself. This is a very interesting and definitely non-conventional aspect
that has not been encountered before, especially in the context of continuum integrable theories.
Note finally that in the special case N = 2 the modified number of particles reduces to the
usual number of particles, which is a conserved quantity for the sl2 NLS model with diagonal
boundary conditions (see (3.18)).
3.3. The NLS model with reflecting impurity
A physically relevant example will be discussed in what follows. More precisely we shall
restrict our attention to the usual NLS model, and within the framework described in the previous
section we shall examine the problem of reflecting impurities attached to the ends of the system.
According to [2] one may consider a more general solution of the reflection equation. Consider
the classical Lax operator satisfying

L1 (1 ), L2 (2 ) = r12 (1 2 ), L1 (1 )L2 (2 ) ,
(3.30)
recall r() is given in (3.1). For example consider the L operator associated to the classical Lie
algebra sl2 :

+ S3 S1 iS2
L() =
(3.31)
S1 + iS2 S3
where apparently Sa obey

{Sa , Sb } = i
abc Sc
(3.32)
i=1
being the usual antisymmetric tensor. One may easily express the later matrix in terms of
canonical variables (x, X), i.e.

+ xX x2 X + 2x
L() =
(3.33)
.
X
xX +
475
Degenerate cases of the matrix above are for instance the Toda chain and the DST model (see
e.g. [34] and references therein) with Lax operators given by

L
Toda
() =
+X
ex
ex
0

,
DST
() =

+ xX x
.
X
1
(3.34)
Consider the following generating function of the integrals of motion

ln tr K+ ()T (L, 0, )K ()T (L, 0, ) ,
(3.35)
K is a generic dynamical type solution of the classical reflection equation [2], i.e.
K () = L( )K ()L1 ( ),
(3.36)
L can be any solution of (3.30), K are any c-number solutions of classical reflection equation.
Note that the Poisson brackets for K in the classical reflection equation are considered with
respect to the canonical variables (x, X). Here we shall deal with a simple example, that is K =
I and = 0 and L given by (3.33) (for simplicity set = 0) then it is clear that
K () = + 2
x X
X
(x )2 X
x X

.
(3.37)
Finally the boundary contribution to the Hamiltonian is given by (see also Appendix B)
1
1
I3(b) = h31 h1 h2 + h3 + h 31 h 1 h 2 + h 3 + 2i(0) (0)
3
3
2i(L) (L),
(3.38)

h1 = 2Z+ ,
h2 = 2(0)(0)
2i x+ Z+ (0) + (x+ )1 Z+ (0)
h0 = 1,
,

+ 2 (x+ )1 Z+ (0) + x+ Z+ (0) ,

h3 = 2i(0) (0)
(3.39)
+ 4Z+ (0)(0)
in particular h n = ()n hn :
where Z = x X . Analogous expressions to (3.39) are given for h,
+
+
0 L, (x , X ) (x , X ). Based on the latter expressions the boundary part of the HamilI
(b)
3
tonian may be deduced. Indeed, bearing in mind that the boundary potential is given by B = 2i
and taking into account (3.38), (3.39) we conclude
B = b(x+ , Z+ , 0) b(x , Z , L)
(3.40)
where we define

1
2
x1 Z (x) + xZ (x)
b(x, Z, x) = xZ2 (x) + x1 Z2 (x)
i

4 3
(x)(x)
+ (x) (x)
Z
3i
(3.41)
and as expected the boundary contribution of the Hamiltonian is solely expressed in terms of the
canonical variables x , X as well as the boundary values of the fields and their derivatives (see
also [35] for a similar treatment of the classical sine-Gordon model).
476
4. A quantum lattice version of NLS

4.1. Review on periodic lattice NLS
Let us first present the general algebraic framework associated to the discrete quantum version
of the NLS model, introduced and studied for the periodic case in [28,29]. In the quantum level
the key object as is well known is the L operator satisfying:
R12 (1 2 )L1n (1 )L2n (2 ) = L2n (2 )L1n (1 )R12 (1 2 )
(4.1)
where the R-matrix associated to the glN Yangian is

R() = iP,
(4.2)
and obeys of course the YangBaxter equation [36,37], i.e.

R12 (1 2 )R13 (1 )R23 (2 ) = R23 (2 )R13 (1 )R12 (1 2 ).
(4.3)
We shall focus in this and the subsequent section in the simplest sl2 Yangian. In this case a
simple solution of Eq. (4.1) is given by (on a detailed description of the underlying algebra see
e.g. [28,29])

1 i
+ 2 n n i n
L0n () =
(4.4)
,
i n
1
where the n , n satisfy canonical commutation relations
1
(4.5)
nm .
In fact this solution may be thought of as the quantum analogous of the NLS. Indeed the classical
x +
limit of the L operator (4.4) gives U (3.3) (for further details see [28]). Set n = xnn dx (x)
then as 0

2
i
(x)
L() = 1 + U(x, ) + O , where U(x, ) =

(4.6)
(x)
0
[n , m ] =
is equivalent to U (3.3) of NLS up to a gauge transformation i.e.
note that (x) = (x),

and U
1 + hx h1 ,
U = hUh
h = eix 2 .
(4.7)
It is more convenient for our purposes here to use L with a rescaled spectral parameter matrix.
i
Let us multiply (4.4) by i
and also set = then the rescaled L matrix may be written as:

1 + Nn i n
L0n () =
(4.8)
i n
Nn = 1 + 2 n n . For the special value = 0 the L operator reduces to a projector

1 0
L( = 0) =
.
0 0
(4.9)
Due to the fact that the algebra (4.1) is equipped with a coproduct one may build tensorial
representations and construct a spin chain like system with periodic boundary conditions, by
477
introducing the quantities

T0 () = L0L . . . L01 ,
and
t () = tr0 T0 ()
(4.10)
with T () being essentially the quantum analogous of (2.4) and apparently by virtue of (4.1)

t (), t ( ) = 0.
(4.11)
By expanding ln t ( ) around = 0 we find the corresponding involutive quantities exactly as in
the classical case [28]. It is easy to see that due to (4.9) t (0) = 1, then one finds (for more details
we refer the reader to [28])
ln t ( ) = C1 + 2 C2 + 3 C3 + ,
[Cn , Cm ] = 0
(4.12)
with
C1 =
L
1
Nn ,
n=1
C2 =
pn =
n=1
C3 =
L

n+1 n
n=1
hn =
n=1

1
2
N
,
22 n
1

n+1 n1 (Nn + Nn+1 )n+1 n + 32 N3n .
(4.13)
n=1
From the latter objects one may derive lattice analogous of the classical quantities (3.12),

N = (C1 L)|0 dx (x)(x),

L
1

dx (x x ),
P = C2 +
2
22 0

L
H = C3 +
(4.14)
dx x x + ()2 .
32 0
Notice that the expressions above are symmetric to , so one can set = and obtain
the familiar expressions for the NLS system (3.12). Note also that the existence of an obvious pseudo-vacuum allows the implementation of Bethe ansatz techniques [28,38], however our
aim here is to extend such computations in the case of the integrable open spin chain, which is
discussed in the subsequent sections.
4.2. Open lattice NLS
We come now to the case with open boundary conditions, which is our main concern. The
underlying algebra in this case is defined by the reflection equation [1]
R12 (1 2 )K1 (1 )R21 (1 + 2 )K2 (2 )
= K2 (2 )R21 (1 + 2 )K1 (1 )R12 (1 2 ).
(4.15)
The tensorial type solutions of the reflection equation as well known is given by [2]
T0 () = T0 ()K0 (, , c )T01 ()
(4.16)
478
K (, , k ) is c-number solution of the reflection equation with the most general form given
by [39]
K () = z i + 2k ( + + ).
(4.17)
L1 () = L()
is given by:
Note that the explicit expression of

1
i n
L 0n () =
.
i n i + Nn
(4.18)
The corresponding generating function of the conserved quantities of the open system is

t () = tr0 K0+ (, + , k + )T0 ()
(4.19)
K + (, + , k + ) = K( i, + , k + )t , K is again a c number solution of the reflection equation.
And due to (4.15) it is clear that integrability of the quantum system is ensured i.e.

t (), t ( ) = 0, , C.
(4.20)
In the remaining of this section we specify the spectrum of the lattice NLS model with diagonal boundary conditions by means of the Bethe ansatz technique [38]. Focusing on diagonal
boundaries should be sufficient given that in [6] the spectral equivalence between systems with
diagonal and non-diagonal boundaries was shown by means of appropriate gauge transformations, but only for spin chains associated to the fundamental representation of sl2 . Presumably
there exist suitable gauge transformations for the system under consideration, such that the spectral equivalence is guaranteed. We shall further comment on this point on a separate publication.
When both boundaries are diagonal there exists an obvious reference state for the transfer state
=
N

n
with n n = 0.
(4.21)
n=1
Based on this observation one may in a straightforward manner derive the spectrum and the
corresponding Bethe ansatz equations. We provide directly the results avoiding the technical
details (for a detailed description we refer the reader to [2]). The spectrum of the transfer matrix
is given by
() = g()b1 ()
M

( j + i)( + j )
( j )( + j i)
j =1
+ h()b2 ()
M

( j i)( + j 2i)
,
( j )( + j i)
(4.22)
j =1
where we define
h() = (i + 1 + )L ,
g() = (i + 1)L ,
i
b1 () =
( + i )( + i + ),
i2
( + i + i)( i + i + ).
b2 () =
i2
(4.23)
479
The corresponding Bethe ansatz equations arising as analyticity conditions on the spectrum are:
g(i +
h(i +
i
2 ) b1 (i
i
2 ) b2 (i
+
+
i
2)
i
2)
M

i j i i + j i
.
i j + i i + j + i
(4.24)
j =1
Notice that although we deal with an open spin chain and one would expect a leading order of 2L,
we see a leading order of L exactly as in the periodic case. The same phenomenon occurs in the
boundary lattice Liouville model [40], and is presumably associated to the degenerate nature of
the L matrix (see also similar comments in [34]). It should be emphasized that the Bethe ansatz
equations (4.24) are of particular significance given that their thermodynamic analysis yields for
instance consequential information regarding the corresponding bulk as well as boundary exact
S matrices of the model.
5. The generalized lattice NLS
We shall deal in what follows with the lattice quantum version of the glN NLS model. Recall
that the glN Yangian R matrix, solution of the YangBaxter equation (4.3), is given in (4.2). The
relevant L operator in this case is given by (see also [28])

N1
N
N

(j 1)

i (j ) (j )
L() = +
(5.1)

Ejj +
E1j + (j 1) Ej 1 .
E11 +
j =1
j =2
j =2
Notice that here we set implicitly i = 1. It will be also useful for the following to define
N
2
we choose here V = antid(1, . . . , 1), which gives rise to the following explicit form:

N1
N1

i
(j
)
(j
)

Ejj
++
L()
=
EN N +
L()
= V1 Lt1 ( + i)V1 ,
j =1
(j 1) EjN + (j 1) EN j
(5.2)
j =1
(5.3)
j =2
where j = N j + 1. Recall that in general we could have chosen any V such that V 2 = I.
We shall focus hereafter in the case of SNP boundary conditions, given that they are not so
widely known compared to the SP ones, especially in the context of integrable lattice models.
The main objective in this section is to derive the exact spectrum and the corresponding Bethe
ansatz equations. Note that in the SNP case the underlying algebra is defined by the following
relation (twisted Yangian, see e.g. [17])
R12 (1 2 )K1 (1 )R 21 (1 + 2 )K2 (2 )
= K2 (2 )R 12 (1 + 2 )K1 (1 )R21 (1 2 )
(5.4)
and in analogy to the classical case we define

t1
R()
= V1 R12
( + i)V1 .
(5.5)
480
The generating function of the quantum integrals of motion in this case is defined as:

t () = tr K + ()T ()K ()T () , with
T () = L0L () . . . L01 (), T () = L 01 () . . . L 0L ()
(5.6)
K are c-number solutions of the twisted Yangian (5.4). In fact, it was shown in [14] that any
matrix K = K t is a solution of the twisted Yangian. In Appendix D an explicit computation of
local integrals of motion for particular boundary conditions is presented. Based on these findings
we present the explicit form of the boundary momentum in the case where K = V , i.e.

L
L1
N1
N1
(j ) (j )
(j ) (j )
i 2
(j ) (j )
Pd =
(5.7)
L L + 1 1
.
Nn 2
n n+1 +
2
n=1
n=1 j =1
j =1
I, which is the case will be examined subsequently,

We could have of course considered
but for the sake of simplicity we considered the aforementioned boundary conditions, which
from a technical point of view are much easier to deal with. By comparing the bulk momentum
given in Appendix D (D.4) with (5.7) we conclude that the periodic terms in (D.4) are replaced
essentially by the last two boundary terms in (D.7), whereas the bulk part remains intact. Following the logic of (4.14), it is expected that the expression (5.7) should be a regularization of the
continuum generalized NLS model momentum with particular SNP boundary conditions, exactly
as it happens for the periodic NLS model (indeed compare the bulk continuum expression (3.11)
with the discrete periodic analogous (D.6)). Comments on higher conserved charges may be also
found in Appendix D.
To deduce the spectrum and Bethe ansatz equations of the generalized NLS model with SNP
boundary conditions we shall restrict our attention to another simple case i.e. K = I. Again the
spectral equivalence between systems with diagonal and non-diagonal boundaries is discussed in
[14,16], for spin chains associated to the fundamental representation of glN . The first step toward
the diagonalization of the transfer matrix (5.6) is the derivation of a reference state. Indeed, in
this case there exists an obvious reference state, that is
=
L

n :
n(i) n = 0,
n {1, . . . , L}, i {1, . . . , N 1}.
(5.8)
n=1
and consequently
The corresponding eigenvalue may be easily derived using the fact that L, L
T , T satisfy
T1 (1 )R 12 (1 + 2 )T2 (2 ) = T2 (2 )R 12 (1 + 2 )T1 (1 ).
(5.9)
Taking into account the latter relation we conclude that the actions of the transfer matrix on the
pseudo vacuum provides the following eigenvalue:
() = a ()g1 () +
(0)
N1

gn () + b L ()gN ()
n=2
where we define
i
i
b()
=
+ ,
a() = ,
i2 ( 1)
N +1
gn () =
, 1n<
,
2
i2
(5.10)
481
g N+1 = 1, N odd,
2
N +1
(5.11)
< l N.
2
The functions gn are essentially boundary contributions to the spectrum.
To determine the general eigenvalue form we shall adopt the analytical Bethe ansatz formulation [41]. The basic assumption within this framework is that the structure of any eigenvalue is
similar to the pseudo-vacuum eigenvalue i.e.
gl = gl( + i),
() = a L ()g1 ()A1 () +
N1

gn ()An () + b L ()gN ()AN ().
(5.12)
n=2
The so-called dressing functions An may be explicitly determined by imposing certain physical
and algebraic requirements, such as analyticity, crossing, etc. We do not give all the details of the
formulation (for a more detailed description of the process we refer the reader to [4,6,14]), but
the explicit expressions for the dressing functions are given by:
A1 () =
Al+1 =
(1)
M

+ j +
j =1
+ j
(l)
M

(1)
(1)
(l)
+ j
(l+1)
M
(1)
i
2
i
2
(l)
i j
il
2
+ (l)
j
j =1
(1)
j +
i
2
i
2
(l)
j
il
2
+ (l+1)
(l+1)
j =1 + j
il
2
il
2
il
2
i
2
i
2
i
il
2
(l+1)
j
(l+1)
j
il
2
il
2
i
2
i
2
1l<
N 1
,
2
N 1
<lN
2
Al () = Al( + i),
(5.13)
and in particular for N = 2n + 1

An+1 () =
(n)
M

(n)
+ j
in
2 i
(n)
+ j in
2
j =1
(n+1)
M
+ j
j =1
+ j
(n+1)
(n+1)
(n)
in
2
in
2
in
2 i
(n)
j in
2
i
2
i
2
(n+1)
(n+1)
j
j
in
2
in
2
i
2
i
2
(5.14)
Finally Bethe ansatz equations follow as analyticity requirements upon the spectrum, and they
are written explicitly as:
(i) N = 2n + 1:

L
i
(1)
a i +
l1 + (1 l1 )
2
=
(l)
M

(l)
e 2 i
(l) (l)
j e 2 i
j =1
l = 1, . . . , n 1,
(l)
+ j
(l+ )
M
j =1
(l)
(l)
(l+ )
(l+ )
e1 i + j
e1 i j
482

(n)
M
(n)
(n)
(n)
(n)
(n) (n)
(n)
(n)
(n)
e 1 i =
e2 i j e2 i + j e1 i j e1 i + j
2
j =1
(n1)
M
(n)
(n)
(n1)
(n1)
e1 i + j
e1 i j
(5.15)
j =1
with = 1, M (N+1) = 0 and define en () =
in
2
+ in
2
(ii) N = 2n: In this case the Bethe ansatz equations for l = 1, . . . , n 1 are the same as in the
previous case. What is only modified is the last set of equations, which takes the form:
e 1
(n)
i =
(n)
M

(n)
(n) (n)
(n)
e 2 i j e 2 i + j
j =1
(n1)
M
j =1
(n)
(n1) 2 (n)
(n1)
2
i j
e1 i + j
.
e1
(5.16)
Such type of boundary conditions were first introduced in [13] for the sl3 spin chain, whereas
generalizations investigated in [6,14] from the physical (Bethe ansatz) as well as the algebraic
point of view. The associated symmetries were studied in detail in [1416], while in [15] both
SP and SNP boundary conditions were examined in parallel. The interesting observation is that
the RHS of the equations above in the case where N = 2n + 1 coincide with the ones associated
to the osp(1|2n) algebra. In any case the Bethe ansatz equations (5.15) are somehow folded
compared to the usual glN ones. This is expected given that folding occurs at the algebraic level
as well (Dynkin diagrams), and only the subalgebra invariant under charge conjugation survives
after the implementation of these rather unconventional boundary conditions (see also relevant
comments in [14]). The case of SP boundaries can be also treated along the same lines, and the
entailed spectrum and Bethe ansatz equations will have the usual glN structure (the expressions
are omitted here for brevity). More precisely the LHS of the Bethe ansatz equations will have
exactly the same form as the usual glN BAE for an open chain (see e.g. [3,5,6]), while the RHS
will depend explicitly on the actions of the diagonal entries of the L, L1 on the pseudo-vacuum.
6. Discussion
To summarize, SP and SNP boundary conditions were studied for the classical generalized
NLS model, and the boundary integrals of motions as well as the relevant classical equations of
motion were explicitly derived. This was the first time, to our knowledge, that SNP boundaries
were implemented in the context of the generalized NLS model. Nevertheless, there are still
several open questions especially regarding the locality of some of the integrals of motion for
particular choices of left/right boundaries, which however will be left for future investigations. In
the same spirit the usual (sl2 ) NLS model with reflecting impurities was also analyzed. Moreover,
a suitable lattice version of the generalized NLS model was considered and the SNP boundary
conditions were implemented, given that in general they are much less studied in this context. For
this choice of boundary conditions we were able to specify the first non-trivial local integral of
motion i.e. the boundary momentum. We also derived the spectrum and Bethe ansatz equations
for the simplest left/right boundaries.
483
Although SP boundary conditions are somehow the obvious ones in the framework of lattice
integrable models, they have not been really considered up to now in classical continuum integrable theories. Therefore, it will be our next goal to impose the SP boundary conditions to other
well known classical systems such as (massless) affine Toda field theories, principal chiral models, etc. In addition, the investigation of the generalized (m)KdV hierarchies [42] with integrable
boundaries is another very interesting direction to pursue together with their quantization into an
appropriate lattice version (see e.g. [43] and references therein). Once this point is clarified, the
study of the underlying dynamical symmetries constrained by integrable boundary conditions
could be discussed in full generality along the lines described in [44], and this would definitively
shed new light on the character of the different integrable boundary conditions. More precisely,
it would be of great consequence to examine how the so-called hidden symmetries constructed
in [44] are modified in the presence of non-trivial integrable boundaries, and in particular in the
case of (quantum) twisted Yangians. Finally, an interesting direction to pursue is the explicit
derivation of the modified Lax pairs by means of the boundary integrals of motion. We hope to
report on all these issues in forthcoming publications.
Acknowledgements
This work was supported by national and local (Bologna section) INFN through grant TO12,
Italian Ministry of University and Research through a PRIN grant, NATO Collaborative Linkage
Grant PST.CLG.980424, and the European Network EUCLID; Integrable models and applications: from strings to condensed matter, contract number HPRN-CT-2002-00325.
Appendix A
In this appendix the classical integrability for models with both types of boundary conditions,
SP and SNP, is reviewed. The first step in order to prove the classical integrability is to show that
the quantities introduced in (2.12) are indeed representations of the algebras defined by (2.10).
To achieve this we shall need in addition to (2.7) the following set of algebraic relations emerging
essentially from (2.7), i.e.

T1 (1 ), T2 (2 ) = r12 (1 2 )T1 (1 )T2 (2 ) T1 (1 )T2 (2 )r12 (1 2 ),

T1 (1 ), T2 (2 ) = T1 (1 )r12
(1 + 2 )T2 (2 ) T2 (2 )r12
(1 + 2 )T1 (1 ),

T1 (1 ), T2 (2 ) = T1 (1 )r12 (1 + 2 )T2 (2 ) T2 (2 )r12 (1 + 2 )T1 (1 ).
(A.1)
Our aim now is to show that (2.10) are satisfied by (2.9):

T1 (1 ), T2 (2 )

= T1 (1 )K1 (1 )T1 (1 ), T2 (2 )K2 (2 )T2 (2 ) =

= T1 (1 )T2 (2 ) K1 (1 )K2 (2 )r12 (1 2 ) + r12 (1 2 )K1 (1 )K2 (2 )

+ K1 (1 )r12
(1 + 2 )K2 (2 ) K2 (2 )r12
(2 )K1 (1 ) T1 (1 )T2 (2 )
(1 + 2 )T2 (2 ) T2 (2 )r12
(1 + 2 )T1 (1 )
+ T1 (1 )r12
+ r12 (1 2 )T1 (1 )T2 (2 ) T1 (1 )T2 (2 )r12 (1 2 )
(A.2)
484
and making use of (2.11), (A.3) and (A.1) we end up to (2.10). Recall also that c-number solutions of the above equations satisfy the following

r12 (1 2 ), K1 (1 )K2 (2 )
= K2 (2 )r12
(1 + 2 )K1 (1 ) K1 (1 )r12
(1 + 2 )K2 (2 ),
(A.3)
which is equivalent to (2.11).

We may now show exploiting (2.10) and (A.3) the classical integrability (2.14). Indeed consider the following object

+
K1 (1 )T1 (1 ), K2+ (2 )T2 (2 )
(A.4)
then taking the trace in both spaces 1 and 2, and considering the defining relations (2.10), (2.11)
we end up with

t (1 ), t (2 ) = tr12 K1+ (1 )K2+ (2 )r12 (1 2 )T1 (1 )T2 (2 )
K1+ (1 )K2+ (2 )T1 (1 )T2 (2 )r12 (1 2 )
(1 + 2 )T2 (2 )
+ K1+ (1 )K2+ (2 )T1 (1 )r12

+
+
K1 (1 )K2 (2 )T2 (2 )r12 (1 + 2 )T1 (1 ) .
(A.5)
Finally moving appropriately the factors of the products within the trace and using (A.3) it is
straightforward to show

t (1 ), t (2 ) = 0, 1 , 2 C
(A.6)
and this concludes our proof. Similar arguments hold also in the case of dynamical boundaries
discussed in Section 3.3.
Appendix B
We present here some technical details on the derivation of the conserved quantities for the
generalized NLS model on the full line. The first step is to insert the ansatz (3.7) in Eq. (2.1).
Then we separate the diagonal and off diagonal part and obtain the following expressions:
Z = U1 + (U0 W )(D) ,
W + W Z = U0 + (U0 W )(O) + U1 W,
(B.1)
where the superscripts

the diagonal and off diagonal part of the product U0 W .
(D), (O) denote
Recall that W = i =j Wij Eij , Z = i Zii Eii then it is straightforward to obtain:
(U0 W )(D) =
(O)
(U0 W )

N1
( i WN i Eii + i WiN EN N ),
i=1
( i WNj Eij + i Wij ENj ).
i =j,i =N,j =N
Substituting the latter expressions (B.2) in (B.1), we obtain
(B.2)
485
N1

Z(L, L, ) = iL
Eii EN N
i=1

N1
+
L
dx ( i WN i Eii + i WiN EN N ).
(B.3)
i=1 L
And recalling that the leading contribution in the expansion of (ln tr T )T is given in (3.7)as
i is coming from the ZN N term we conclude:

N1
ZN N (L, L, ) = iL +
dx i (x)WiN (x).
L
(B.4)
i=1 L
Due to (B.4) it is obvious that in this case it is sufficient to derive the coefficients WiN only. In
any case one can show that the coefficients Wij satisfy the following equations:

2
Wij Eij i
(WN i EN i WiN EiN ) +
EiN
i WN2 i EN i + i WiN
i =j
i =N
=
( i EiN + i EN i ) +
i =N
i =N
( i WNj Eij + i Wij ENj )
i =j,i =N,j =N
( j WNj Wij Eij + i WiN Wj N Ej N ).
(B.5)
i =j,i =N,j =N
(n)

Wij
(n)
Finally setting Wij =
n=1 n and using (B.5) we find expressions for WiN reported in (3.10).
In the case with integrable boundary conditions, we shall need in addition to (3.10) the following
objects:

(1)
(2)
(1)
(1) (1)
WN i = iWN i +
WNj Wj i ,
WN i = i i ,
i =j,i =N,j =N
(1)
(1) (1)
Wj i = iWj N WN i ,
(3)
(2)
(1) (1) (1)
WN i = iWN i + WiN WN i WN i
(1)
(2)
WNj Wj i ,
i =j,i =N,j =N
(2)
Wij
(1) (2)
= iWiN WNj
(1) (1) (1)

iWj N WNj Wij .
(B.6)
Appendix C
In what follows we evaluate the boundary terms contributing to the Hamiltonian for right
and left boundary described by the more general, diagonal and non-diagonal, solutions of the
reflection equation (SP boundary conditions). Moreover, for the SNP boundary conditions we
identify the corresponding integrals of motions, and we explicitly evaluate the first non-trivial
charge.
C.1. SP boundary conditions
We shall expand the generic object (3.17) keeping of course only diagonal contributions.
More precisely, as in the bulk case due to the fact that the leading order is eiL as i
486
the only non-negligible part is coming from the EN N terms, hence we shall only consider such
contributions:

1
hn
,
1 + W (0, ) K + () 1 + W (0, ) N N =
n
n=0
1 + W (L, )

hn
K () 1 + W (L, ) N N =
,
n
n=0

L, )
= iL +
Z(0, L, ) Z(0,
NN
1 ()n
(n)
ZN
N (0, L)
n=1
(C.1)
Again considering only the contribution of the term eiL as i we end up with the following expression

ln tr K + ()T (0, L, )K ()T (0, L, )
(n)

hn + h n

Z
(0,
L)
h
h
n
m
= iL +
(C.2)
1 ()n N N n
+ ln
+
.
n
n+m
n=1
n=0
n,m=0
Recall from (C.2) that the boundary contribution lies basically in the logarithmic function, hence
one has to expand the log i.e.

hn + h n

hn h m
fn
+
=
ln
(C.3)
n
n+m
n
n=0
n,m=0
n=0
where fn , provide essentially the boundary contribution to the integrals of motion (plus possible
total derivatives from the bulk part) for the left right boundary respectively. The interesting observation is that the boundary contribution decouples nicely to terms associated to left and right
boundary separately, i.e. no mixing occurs
1
1
f2 = h21 + h2 h 21 + h 2 ,
2
2
1
1
f3 = h31 h1 h2 + h3 + h 31 h 1 h 2 + h 3 , . . . .
3
3
f1 = h1 + h 1 ,
(C.4)
(a) We first consider generic non-diagonal boundary conditions described by (3.14). One can
explicitly evaluate hn for the generic case:

h1 = 2ik + 1 (0) 1 (0) + i + ,
h0 = 1,
N1

i (0) i (0),
h2 = 2k + 1 (0) 1 (0) 4k +2
i=2
N1
N1

h3 = 2ik + 1 (0) 1 (0) + 2i +
j (0) j (0) 2i
j (0) j (0)
j =1

j (0)2 1 (0) 2 1 (0) + 2ik + 32 1 2
+ 2ik +
1
3
2
j =1

N1

4ik +2

j (0) j (0) + j (0) j (0) ,
487
(C.5)
....
j 2
In fact it is clear from the expansion of the left and right boundary contribution that h n =
(1)n hn : 0 L, + , k + k .
Given that
there is an overall derivative from the bulk part of Z3 , giving rise to a boundary
term (2i i i i ), we conclude that the boundary contribution to the conserved quantity I3
(i.e. the Hamiltonian) is given by
I3 = f3 + 2i
N1

i (0)i (0) 2i
N1

i=1
i (L)i (L) = 2iB,
(C.6)
i=1
where B is the boundary potential, and recall that f3 is given by (C.4). For diagonal boundary,
terms proportional to k apparently disappear, and the Hamiltonian reduces to (3.18). For a
purely antidiagonal boundary only terms proportional to k survive, all the other terms disappear.
(b) The more general diagonal boundary conditions are described by (3.16). We associate
the integers l and the free parameters to the right/left boundaries. In this case from the
expansions (C.1) and taking into account (3.10), (B.6) we find:
h1 = i + ,
h0 = 1,
N1

h2 = 2
i (0) i (0),
i=l + +1
+
h3 = 2i
N1

i (0) i (0) + 2i
i (0) i (0) + 2i
N1

i=l + +1
i=1
i (0) i (0),
. . . (C.7)
i=1
, + .
similar expressions of course hold for h n , i.e. h n = (1)n hn : 0 L, l +
lN1
Taking into account (C.4), (C.6) and derivative contribution from the bulk (2i i=1 i i ) we
conclude that the boundary contribution to the Hamiltonian is given by:

I3 = 2i
N1

i (0) i (0) + i (0) i (0)
i=l + +1
N1

i (L) i (L) + i (L) i (L)
i=l +1
+ 2i +
i (0) i (0)
i=1
i (L) i (L) .
(C.8)
i=1
Note that when l = N 1 one recovers the diagonal limit of the more general case (a).
(c) Finally the case where K = I may be treated in the same spirit. In this case we find that
h0 = 1,
...
h1 = 0,
h2 = 2
N1

i=1
i (0) i (0),
h3 = 2i
N1

i=1
i (0) i (0),
(C.9)
488
and the corresponding boundary contribution to the Hamiltonian is given by

N1

N1

I3 = 2i
i (0) i (0) + i (0) i (0)
i (L) i (L) + i (L) i (L) .
i=1
i=1
(C.10)
C.2. SNP boundary conditions
Recall that in this case the object under consideration is given in (3.28), also we consider for
simplicity K = I and we choose V = antid(1, . . . , 1). Before we proceed with the evaluations
of the integrals of motion let us first introduce some useful notation
W ij () = Wji (),

Z ii () = Zii (),
and
(n)

1
f
1 + W ()
= 1 + F () where F () =
,
n
(C.11)
n=1
recall j = N j + 1, and also one may easily compute that

2

f(2) = W (1) W (2) , . . . .
f(1) = W (1) ,
(C.12)
We shall need the following contributions in order to evaluate the corresponding integrals of
motion:

1 + W (0, ) 1 + W (0, ) = 1 + H (),
1

1
1 + W (L, )
= 1 + H ().
1 + W (L, )
(C.13)
Also bearing in mind that the leading contributions in the considered expansion, as i , are
coming from ZN N and Z ii for i {2, . . . , N } we may write:
ZN N (0, L, ) + Z ii (0, L, )
= iL +
Z (n) (0, L) + (1)n Z (n) (0, L)

NN
ii
n=1
(C.14)
Gathering all the information given above we end up with the following expression

ln tr T (0, L, )V T t (0, L, )V
(n)
(n)

ZN N (0, L) + (1)n Z11 (0 L)
= iL +
n
n=1
+ ln 1 + HN N () + H N N () +
eZii (0,L,)ZNN (0,L,) HiN ()H N i () .
i=2
(C.15)
Finally taking into account the information provided above (C.11)(C.15) we may express the
first non-trivial integrals of motion as:
(1)
(1)
I1 = ZN
N (0, L) Z11 (0, L)
= i
0
N1

dx i (x) i (x) i
i=1 L
0
dx 1 (x) 1 (x).
489
(C.16)
(n)
in (C.15) non-local terms seem to arise in the
Notice that in general due to the presence of Z11
higher integrals of motion, which is quite an unusual issue and shall be addressed elsewhere.
Nevertheless, a straightforward computation of the higher charges, based on the explicit expression (C.15), may prove the locality or not of the higher integrals of motions. Moreover, the
(n)
presence of Zii alters the structure of the bulk part of the integrals as well. The latter integral of
I1
motion (C.16) gives rise to a modified number of particles, Nm = i
.
Appendix D
In this appendix we shall evaluate the first integrals of motion of the quantum discrete glN
NLS model with SNP boundary conditions. This model may be also regarded as a higher rank
algebraic extension of the sl2 DST model (see e.g. [34]), holding a special place between the
glN quantum spin chainsextensions of the Heisenberg modeland the glN generalization
of the Toda chain. To explicitly specify the local integrals of motions of the model with open
boundary conditions we shall, as usual, consider the asymptotic expansion of the generating
function T (). We shall focus here on the simple case where both left and right boundaries are
given by K () = antid(1, . . . , 1), and effectively we shall expand
t () = tr T ()T () where T () = T t ()
(D.1)
and recall T () is given by (5.6). Indeed after expanding in powers of 1 we obtain

1 (2)
T + O 3 ,
2
3
1 (2)
T
+
O
,
2
1
T () E11 + T (1) +
1
T () E11 + T (1) +
(D.2)
where the quantities T (1,2) , T (1,2) are defined below
L

N
N

(j 1)
(j 1)
(1)
Nn E11 +
1
E1j +
L
Ej 1 ,
T = i
j =2
n=1
T (1) = i
(2)
n E11 +
N
Nn Nm E11 +
L1

Nn
N

j =2
N

j =2
n=1
(j 1)

(j 1)
L
E1j
(j 1)
E1j ,
j =2
N
L1

(j 1) (j 1)
n+1 E11
n=1 j =2
n>m
(j 1)
1
Ej 1
j =2
n=1
j =2
Ej 1 +
N

j =2
(j 1) (j 1)
1
Ejj
j =2
(j 1)
L

n=2
Nn E1j +
N

j =2
(j 1)
L1 Ej 1
490
T (2) = 2
nN
m E11 +
N
n
N
N

j =2
n=1
(j 1)
(j 1) (j 1)
n+1 E11
n=1 j =2
n<m
L1

N
L1

(j 1)
E1j +
(j 1) (j 1)
1
Ejj
j =2
(j 1)
j =2
n Ej 1 +
N
(j 1)
L1 E1j
j =2
n=2
Ej 1 ,
j =2
where Nn =
N1

(j ) (j )
n n , N
n = Nn + .
(D.3)
j =1
In the expressions above all the lower indices denote the site of the spin chain, while the upper
indices denote the component of the (N 1)-dimensional vector fields.
may easily obtain first the bulk integrals of motion by considering the expansion tr T () =
We
In
(1,2) we obtain
.
n n Indeed for the bulk case after simply taking the trace of T

L
L1
N1

N1
(j ) (j )
(j ) (j )
Nn ,
I2 = 2
Nn Nm +
n n+1 +
L 1 . (D.4)
I1 = i
n=1 j =1
n<m
n=1
j =1
The quantities identified as the number of particles and the momentum in the NLS model are
given by the following expressions

1
1 1 2
Nd = I1 ,
(D.5)
Pd =
I I2
i
i 2 1
and more precisely
Nd =
Nn ,
n=1
i
Pd =
2
N2n
L1
N1

n=1 j =1
n=1
(j ) (j )
n n+1
N1

(j ) (j )
1 L
(D.6)
j =1
We come now to the open NLS model, and we consider the expansion of (D.1). The first charge
of the open model is zero, that is the number of particles is not a conserved quantity anymore.
The second charge is given by

L
L1
N1

N1
(j ) (j )
(j ) (j )
(j ) (j )
2
2
I2 =
(D.7)
L L + 1 1
Nn 2
n n+1 +
n=1
n=1 j =1
j =1
I2
2i ,
which is obviously modified due to the presence

and corresponds to the momentum Pd =
of the open boundaries. The third charge again is trivial, involving only boundary terms. We do
not compute any higher conserved charges, but we may rather safely conjecture that the only
non-trivial conserved charges are the even ones.
References
[1] I.V. Cherednik, Theor. Math. Phys. 61 (1984) 977.
[2] E.K. Sklyanin, J. Phys. A 21 (1988) 2375.
[3] H.J. de Vega, A. Gonzalez-Ruiz, Nucl. Phys. B 417 (1994) 553;
H.J. de Vega, A. Gonzalez-Ruiz, Phys. Lett. B 332 (1994) 123.
491
[4] L. Mezincescu, R.I. Nepomechie, Nucl. Phys. B 372 (1992) 597;

S. Artz, L. Mezincescu, R.I. Nepomechie, J. Phys. A 28 (1995) 5131.
[5] A. Doikou, R.I. Nepomechie, Nucl. Phys. B 521 (1998) 547;
A. Doikou, R.I. Nepomechie, Nucl. Phys. B 530 (1998) 641.
[6] D. Arnaudon, J. Avan, N. Cramp, A. Doikou, L. Frappat, E. Ragoucy, J. Stat. Mech. 0408 (2004) P005;
D. Arnaudon, N. Cramp, A. Doikou, L. Frappat, E. Ragoucy, J. Stat. Mech. 0502 (2005) P007.
[7] A. Doikou, Nucl. Phys. B 725 (2005) 493;
A. Doikou, SIGMA 3 (2007) 009.
[8] W. Galleas, M.J. Martins, Phys. Lett. A 335 (2005) 167;
R. Malara, A. Lima-Santos, J. Stat. Mech. 0609 (2006) P013;
W.-L. Yang, Y.-Z. Zhang, JHEP 0412 (2004) 019.
[9] P. Bowcock, E. Corrigan, P.E. Dorey, R.H. Rietdijk, Nucl. Phys. B 445 (1995) 469;
P. Bowcock, E. Corrigan, R.H. Rietdijk, Nucl. Phys. B 465 (1996) 350.
[10] G.M. Gandenberger, hep-th/9911178.
[11] G. Delius, N. Mackay, Commun. Math. Phys. 233 (2003) 173.
[12] G.W. Delius, N.J. MacKay, B.J. Short, Phys. Lett. B 522 (2001) 335;
G.W. Delius, N.J. MacKay, B.J. Short, Phys. Lett. B 524 (2002) 401, Erratum;
N.J. MacKay, J. Phys. A 35 (2002) 7865.
[13] A. Doikou, J. Phys. A 33 (2000) 8797.
[14] D. Arnaudon, N. Cramp, A. Doikou, L. Frappat, E. Ragoucy, Int. J. Mod. Phys. A 21 (2006) 1537.
[15] A. Doikou, J. Math. Phys. 46 (2005) 053504.
[16] N. Cramp, A. Doikou, J. Math. Phys. 48 (2007).
[17] G.I. Olshanski, Twisted Yangians and infinite-dimensional classical Lie algebras, in: P.P. Kulish (Ed.), Quantum
Groups, in: Lecture Notes in Mathematics, vol. 1510, Springer-Verlag, 1992, p. 103;
A. Molev, M. Nazarov, G.I. Olshanski, Russ. Math. Surv. 51 (1996) 206.
[18] A.I. Molev, E. Ragoucy, P. Sorba, Rev. Math. Phys. 15 (2003) 789;
A.I. Molev, in: M. Hazewinkel (Ed.), Handbook of Algebra, vol. 3, Elsevier, 2003, p. 907.
[19] J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231;
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253;
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105.
[20] I. Bena, J. Polchinski, R. Roiban, Phys. Rev. D 69 (2004) 046002.
[21] V.A. Kazakov, A. Marshakov, J.A. Minahan, K. Zarembo, JHEP 0405 (2004) 024.
[22] J.A. Minahan, K. Zarembo, JHEP 0303 (2003) 013.
[23] N. Beisert, V. Dippel, M. Staudacher, JHEP 0407 (2004) 075.
[24] A. Rej, D. Serban, M. Staudacher, JHEP 0603 (2006) 018.
[25] G. Feverati, D. Fioravanti, P. Grinza, M. Rossi, J. Stat. Mech. 0702 (2007) P001.
[26] L.D. Faddeev, L.A. Takhtakajan, Hamiltonian Methods in the Theory of Solitons, Springer-Verlag, 1987.
[27] M. Mintchev, E. Ragoucy, P. Sorba, J. Phys. A 34 (2001) 8345.
[28] A. Kundu, O. Ragnisco, J. Phys. A 27 (1994) 6335.
[29] A. Kundu, B. Basumallick, Mod. Phys. Lett. 7 (1992) 61.
[30] A. MacIntyre, J. Phys. A 28 (1995) 1089.
[31] A.P. Fordy, P.P. Kulish, Commun. Math. Phys. 89 (1983) 427.
[32] V.O. Tarasov, Inv. Probl. 7 (1991) 435;
R.F. Bikbaev, V.O. Tarasov, J. Phys. A 24 (1991) 2507.
[33] A.S. Fokas, A.R. Its, nlin.SI/0412008.
[34] E.K. Sklyanin, solv-int/9908002;
E.K. Sklyanin, nlin.SI/0009009;
V.B. Kuznetsov, M. Salermo, E.K. Sklyanin, J. Phys. A 33 (2000) 171.
[35] P. Baseilhac, G. Delius, J. Phys. A 34 (2001) 8259.
[36] R.J. Baxter, Ann. Phys. 70 (1972) 193;
R.J. Baxter, J. Stat. Phys. 8 (1973) 25;
R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, 1982.
[37] V.E. Korepin, Theor. Math. Phys. 76 (1980) 165;
V.E. Korepin, G. Izergin, N.M. Bogoliubov, Quantum Inverse Scattering Method, Correlation Functions and Algebraic Bethe Ansatz, Cambridge Univ. Press, 1993.
[38] L.D. Faddeev, L.A. Takhtajan, J. Sov. Math. 24 (1984) 241;
L.D. Faddeev, L.A. Takhtajan, Phys. Lett. A 85 (1981) 375.
492
[39] H.J. de Vega, A. Gonzalez-Ruiz, J. Phys. A 26 (1993) L519.

[40] A. Doikou, J. Stat. Mech. (2006) P09010.
[41] V.I. Vichirko, N.Yu. Reshetikhin, Theor. Math. Phys. 56 (1983) 805;
N.Yu. Reshetikhin, Lett. Math. Phys. 7 (1983) 205.
[42] V.G. Drinfeld, V.V. Sokolov, J. Sov. Math. 30 (1984) 1975;
L.A. Dickey, Soliton Equations and Hamiltonian Systems, World Scientific, 1991.
[43] D. Fioravanti, M. Rossi, J. Phys. A 35 (2002) 3647.
[44] D. Fioravanti, M. Stanishkov, Phys. Lett. B 447 (1999) 277;
D. Fioravanti, M. Stanishkov, Nucl. Phys. B 577 (2000) 500.
Hierarchy of N = 8 mechanics models

Evgeny Ivanov a , Olaf Lechtenfeld b, , Anton Sutulin a
a Bogoliubov Laboratory of Theoretical Physics, JINR, 141980 Dubna, Russia
b Institut fr Theoretische Physik, Leibniz Universitt Hannover, Appelstrae 2, 30167 Hannover, Germany

Abstract
Using the N = 4 superspace approach in one dimension (time), we construct general N = 8 supersymmetric mechanics actions for the multiplets (b, 8, 8 b) classified in hep-th/0406015, with the main focus
on the previously unexplored cases of (8, 8, 0), (7, 8, 1) and (6, 8, 2), as well as on (5, 8, 3) for completeness. N = 8 supersymmetry of the action amounts to a harmonicity condition for the Lagrangian with
respect to its superfield arguments. We derive the generic off-shell component action for the root multiplet (8, 8, 0), prove that the actions for all other multiplets follow from it through automorphic dualities
and argue that this hierarchical structure is universal. The bosonic target geometry in all cases is conformally flat, with a unique scalar potential (except for the root multiplet). We show that the N = 4 superfield
constraints respect the full R-symmetry and find the explicit realization of its quotient over the manifest
R-symmetry on superfields and component fields. Several R-symmetric N = 4 superfield Lagrangians with
N = 8 supersymmetry are either newly found or reproduced by a simple universal method.
PACS: 11.30.Pb; 11.15.-q; 11.10.Kk; 03.65.-w
Keywords: Supersymmetry; Mechanics; Superfield
1. Introduction
Particle models with extended supersymmetry supply non-trivial examples of supersymmetric
mechanical systems and of supersymmetric quantum mechanics (see, e.g., [13]). In many cases
E-mail addresses: eivanov@theor.jinr.ru (E. Ivanov), lechtenf@itp.uni-hannover.de (O. Lechtenfeld),

sutulin@theor.jinr.ru (A. Sutulin).
doi:10.1016/j.nuclphysb.2007.08.014
494
E. Ivanov et al. / Nuclear Physics B 790 [PM] (2008) 493523
they are related by dimensional reduction to some higher-dimensional supersymmetric field theory, e.g., N = 4 super-YangMills theory (see, e.g., [4]), and as such provide a simplified setting
for studying salient features of these theories. Other models of this kind represent supersymmetric extensions of certain intrinsically one-dimensional systems, like conformal mechanics
[2,510], or CalogeroMoser integrable models [1114]. From the mathematical point of view,
extended d = 1 supersymmetry, as compared with its higher-dimensional counterpart, exhibits
rather unusual features, such as the so-called automorphic duality between multiplets with the
same number of fermions but with different distributions of the bosons to the physical and the
auxiliary sector [1517]. Furthermore, the d = 1 systems with extended supersymmetry display
special types of bosonic target geometries which are not encountered in higher-dimensional supersymmetric field theories [18,19].
Until now, most exhaustively studied have been the mechanics models with 2 N 4 supersymmetries (see, e.g., the review [1] and references therein). Clearly, the higher is N , the more
restrictions on the dynamics and target geometry of the corresponding mechanics models are expected. From this point of view, models with N = 8 supersymmetry are of obvious interest. The
corresponding supersymmetry algebra can be obtained by direct dimensional reduction from the
four-dimensional N = 2 Poincar superalgebra. Therefore, some of N = 8 mechanics models are
dimensional reductions of N = 2 super-YangMills theory or models of self-interacting N = 2
hypermultiplets in four dimensions. The d = 1 reduction of the gauge multiplet can yield either
one of the off-shell N = 8 multiplets (5, 8, 3) and (2, 8, 6) (in the classification of Refs. [10,20]1 ),
depending on whether one performs the reduction on the level of the gauge potential superfield
[2124] or the superfield strength [2527]. The d = 1 reduction of the generic hypermultiplet
model (which exists off shell only in harmonic superspace [28,29]) produces a general onedimensional sigma model with hyper-Khler target geometry and an infinite number of auxiliary
fields. Finally, the d = 1 reduction of the general off-shell sigma model for twisted N = (4, 4) supermultiplets in two dimensions leads to the most general model of the N = 8 multiplet (4, 8, 4)
[30,31].
It should be pointed out, however, that by dimensional reduction one does not exhaust
the full set of off-shell d = 1 supermultiplets and corresponding supersymmetric mechanics
systems. For instance, in the case of N = 4 and N = 8 supersymmetry there exist off-shell
multiplets with field contents (4, 4, 0) and (8, 8, 0), respectively (so-called root [32,33] or
extreme multiplets). These cannot be recovered by dimensional reduction because the suitable d > 1 off-shell multiplets always include auxiliary fields. Therefore, in order to compile
a complete list of d = 1 supermultiplets and relevant supersymmetric mechanics models, we
better not refer to dimensional reduction but proceed solely from the d = 1 supersymmetry
algebra under consideration. This strategy was pursued in the papers [9,10,20] co-authored
by two of the present authors (E.I. and O.L.). Using superfield techniques, we derived the
full set {(b, 4, 4 b) | b = 0, . . . , 4} of off-shell N = 4 multiplets with four physical fermions [9] and the full set {(b, 8, 8 b) | b = 0, . . . , 8} of linear off-shell N = 8 multiplets with
eight physical fermions [20]. These multiplets can be defined either by manifestly N = 8 supersymmetric constraints in N = 8, d = 1 superspace or by equivalent sets of constraints in
N = 4, d = 1 superspace, with one N = 4 supersymmetry being manifest and a second one
hidden.
1 The symbol (b, f, a) denotes the off-shell d = 1 multiplet with f physical fermionic, b physical bosonic and a auxiliary bosonic fields, with f = b + a.
495
In [20], by exploiting the N = 4 superfield formalism, the free actions for all mentioned N = 8
multiplets were constructed, and particular examples of non-trivial N = 8 invariant actions were
presented. Later on, some of these multiplets (as well as their nonlinear cousins) and associated
N = 8 mechanics were studied in more detail in [2527,30] and [3440]. The general sigmamodel type action for the multiplet (5, 8, 3), both in N = 4 superfield formulation and in N = 8
harmonic superfield formulation, had been constructed earlier in [2124]. Superconformally invariant actions for the multiplets (3, 8, 5) and (5, 8, 3), in N = 4 superfield formulation, were
given in [10]. Manifestly N = 8 supersymmetric general actions in terms of constrained N = 8
superfields were given for the multiplet (4, 8, 4) (in bi-harmonic N = 8 superspace) [31] and
(2, 8, 6) (in conventional N = 8, d = 1 superspace) [25,26].
As for the remaining N = 8 multiplets, namely (0, 8, 8), (1, 8, 7), (6, 8, 2), (7, 8, 1) and
(8, 8, 0), no detailed analysis of the corresponding models or construction of generic invariant
actions have been carried out yet. One reason is that no convenient N = 8 superfield formalism is known for describing these multiplets. Although their defining off-shell constraints can be
written in conventional N = 8 superspace [20], these constraints cannot be cast (at least not in a
simple way) in the form of Grassmann analyticity conditions and then solved in terms of analytic
superfields living on N = 8 superspaces of smaller Grassmann dimension. On the other hand,
by dimensional reasoning, the sigma-model type actions of these multiplets cannot be written
as integrals of the relevant constrained N = 8 superfields over the full N = 8 superspace. Typically, one may overcome this difficulty by solving the constraints through negative-dimension
prepotentials; the appearance of new non-geometric gauge invariances in such a prepotential
formulation renders it impractical however.
Another, much more feasible way around is to employ the off-shell N = 4 superfield approach. In this paper, we use it to find the general N = 8 supersymmetric actions for the so far unexplored multiplets (8, 8, 0), (7, 8, 1) and (6, 8, 2). For completeness, we also study the (5, 8, 3)
case within the same setting, for the N = 4 superfield splitting (5, 8, 3) = (4, 4, 0) (1, 4, 3)
which was not fully discussed in the literature before. We derive general conditions for the existence of the hidden N = 4 supersymmetry in the superfield actions of the constituent N = 4
superfields (and hence for the full N = 8 supersymmetry). We show that in all cases it is the
harmonicity of the Lagrangian with respect to its superfield arguments which guarantees this
second N = 4 supersymmetry. For all cases considered we derive the general bosonic action
and, in addition, the full general component action for the root multiplet (8, 8, 0). We argue
that the generic sigma-model type off-shell component action of each N = 8 model (b, 8, 8 b)
can be obtained from this (8, 8, 0) parent action via automorphic duality and demonstrate this
property for b 5. This hierarchical structure of the N = 8 mechanics models generalizes the
N = 4 hierarchy established in [33].
The target bosonic geometry is always conformally flat, with the conformal factor obeying the
appropriate b-dimensional harmonicity condition (with b 5). We also present explicit examples
of actions enjoying invariance under the corresponding R-symmetry. It is shown that the N = 4
superfield constraints respect the full SO(b) SO(8 b) symmetry, which becomes the maximal
R-symmetry group SO(8) of the N = 8, d = 1 Poincar superalgebra in the extreme case of
the multiplet (8, 8, 0). We display the realization of the hidden parts of these R-symmetries
both on the N = 4 superfields and on the component fields and give the explicit form of the
corresponding invariant N = 4 actions and their bosonic cores. In all cases, except for (8, 8, 0),
an N = 8 potential term exists and is written down.
496
2. The (8, 8, 0) mechanics

2.1. N = 8 invariant action in N = 4 superspace
We will use the N = 4 superspace parametrized by z = (t, ia ), with the spinor covariant
derivatives D ia = ia + i ia t satisfying the anticommutation relation

D ia , D kb = 2i ik ab t .
(2.1)
The multiplet (8, 8, 0) is presented by two different off-shell N = 4 supermultiplets (4, 4, 0)

described by the quartet superfields i and q aA , and A being doublet indices of two extra
SU(2) groups of the PauliGrsey type commuting with the N = 4 superalgebra generators (and
with the covariant derivatives as well) [20]. These superfields are subjected to the constraints
D a(i k) = 0,
D i(a q b)A = 0.
(2.2)
The hidden N = 4 supersymmetry extending the explicit one to N = 8, d = 1 supersymmetry is

realized by the following transformations mixing i and q aA :
1
i = A D ia qaA ,
2
1
q aA = A D ia i ,
2
(2.3)
where A is the corresponding Grassmann parameter.

The most general sigma-model type action of these two multiplets in N = 4 superspace is
given by

gen
S(8) = dt d 4 L(, q),
(2.4)
where the Grassmann integration measure is defined as
1 ai k b
(2.5)
D Da Di Dkb .
4!
To extract the most general N = 8 invariant subclass of the actions (2.4), we should impose
on (2.4) the condition that it is invariant under (2.3) up to a total derivative in the integrand
d 4 =
L = A Dai Gi aA (, q).
(2.6)
Gi aA
is a function of the involved superfields, arbitrary for the moment. The t -derivative
Here
on its own cannot appear in (2.6), since it does not show up in the transformation laws (2.3).
Taking into account that the constraints (2.2) imply
1
1
D ia q bA = ba D ic qcA ba iA ,
2
2
1
1
D ia k = ki D j a j ki a ,
2
2
(2.7)
and equating the coefficient before the independent spinors iA and a in both sides of the condition (2.6), we find the conditions of the hidden N = 4 (and hence full N = 8) supersymmetry
of the action (2.4)
(a)
Gi aA
L
=
,
i
qaA
(b)
L
Gi aA
= BA
.
q aB
i
(2.8)
497
As the integrability condition of this system it follows that L should satisfy the dimension-8
harmonicity equation
(8) L = ((q) + () )L = 0,
(2.9)
where
(q) =
2
,
q aA qaA
() =
2
.
i i
(2.10)
One more corollary of (2.8) is

(q) Gi aA = () Gi aA = 2
2L
i qaA
(8) Gi aA = 0.
(2.11)
It is instructive to write (2.8) in a vector form, by the correspondence (aA) n,

(i) :

(a) [ G]
G n = n L,
n sd = 0,

n G n = L,
(b) [m G n]
(2.12)
sd = 0,
where |sd means self-dual part of the corresponding R 4 field strength. Thus, the 4 4 matrix
G n should satisfy two independent Euclidean self-duality conditions with respect to each vector
index, and also its two divergences should be related to L as in (2.12).
It is worth pointing out that Eq. (2.9) is the only integrability condition for L which follows
from the general conditions of the invariance of the N = 4 superfield action (2.4) under the extra
N = 4 supersymmetry (2.3). For any L satisfying (2.9), Eqs. (2.8) and their consequence (2.11)
define the corresponding function Gi aA and so ensure the N = 8 supersymmetry of the relevant
action. Although the precise form of this function has no practical value, it can be explicitly found
for some particular solutions of (2.9) (see Section 2.2). The fact that (2.9) is the necessary and
sufficient condition of the N = 4 superfield action (2.4) to possess N = 8 supersymmetry can be
also confirmed by considering the full off-shell component form of this action (see Section 2.4).
This action is parametrized by two arbitrary functions (q) L| and L|2 and it is invariant under
the component form of the hidden N = 4 supersymmetry (2.3) if and only if
(q) L| = () L|
(8) L = 0.
(2.13)
Let us also note that the action (2.4) is defined up to adding to L some terms (q, ) which
simultaneously obey two dimension-4 Laplace equations
(q) = () = 0.
(2.14)
Such terms do not contribute to the action, but can be used, e.g., to simplify the Lagrangian in
one or another case.
2.2. Bosonic target metric and examples
Using the definition (2.5), Eqs. (2.7) and the following corollaries of the latter
D ia kA = 4iki t q aA ,
D ia b = 4iba t i
(2.15)
2 Hereafter, the vertical bar means the restriction to the first component in the expansion of the corresponding superfield expression.
498
(they can be derived using the basic anticommutation relation (2.1)), it is easy to find the purely
bosonic part of the general Lagrangian (2.4)

1
gen
dt ((q) L|)(t q)2 (() L|)(t )2 .
S(4+4)bos =
(2.16)
2
Note that the minus sign before the second term does not mean negative energy since the free
Lagrangian is just q 2 2 . The N = 8 supersymmetric case corresponds to the choice

1
gen
S(8)bos =
(2.17)
dt G(8) (t q)2 + (t )2 , G(8) (q, ) = (q) L| = () L|.
2
So the bosonic metric in this case is necessarily conformally flat (in agreement with the ansatz
proposed in [19]) and it satisfies the dimension-8 harmonicity condition
((q) + () )G(8) = 0,
(2.18)
since the Lagrangian L does. In Section 2.4 the full off-shell component structure of the superfield Lagrangian in (2.4) is established. The Lagrangian (2.16) follows from it in the limit when
all fermions are put equal to zero.
The general N = 8 supersymmetry conditions (2.8) are essentially simplified for the particular
cases when the N = 8 supersymmetric Lagrangian in addition possesses invariance with respect
to one or both SO(4) symmetries realized on two independent vector indices. For instance, requiring SO(4) acting on the index (aA) to be preserved restricts Gi aA to the form
Gi aA = q aA Gi (, x),
x q aA qaA ,
(2.19)
and L to L(, x). In this particular case Eqs. (2.8) take the following form
(a)
Gi
= 2 Lx ,
i
(b) 2Gi + xGi =
L
.
i
(2.20)
From the second condition follows that Gi obeys an even stronger condition than self-duality:
it should be pure gauge,
Gi =
F
.
i
(2.21)
Then the set (2.20) is reduced to the following two equations for the function F :
() F = 4Lx ,
2F + xFx = A(x) + L,
(2.22)
A(x) being an integration constant (bearing no i -dependence). From these equations, for any L
satisfying the dimension-8 harmonicity condition (2.9) one can restore the appropriate function F
(up to an unessential freedom of adding solutions of the corresponding homogeneous equation)
and further G by Eq. (2.21), thus explicitly proving N = 8 supersymmetry of the relevant N = 4
superfield action.
The N = 8 supersymmetry conditions are most simple in the case of full SO(4) SO(4)
invariance, in which case the Lagrangian can depend only on the two invariants x = q aA qaA and
y = i i (this case was briefly considered in [20]). The matrix Gi aA in this case is reduced
to
Gi aA = i q aA G(x, y)
(2.23)
499
and the constraints (2.8) to

1
Lx = G + yGy ,
2
1
Ly = G xGx .
2
(2.24)
The dimension-8 harmonicity equations for L(x, y) and Gi aA are reduced to

(a)
xLxx + yLyy + 2(Lx + Ly ) = 0,
(b)
xGxx + yGyy + 3(Gx + Gy ) = 0.
(2.25)
For any L satisfying (2.25a) it is easy to find the corresponding function G satisfying
Eqs. (2.24) and their corollary (2.25b). As an example, we explicitly present a few polynomial
solutions (up to arbitrary renormalization factors)
L1 = x y,
G1 = 1,
L2 = x 2 + y 2 3xy,
G2 = 2(x y),

G3 = 3 x 2 + y 2 8xy.
L3 = x 3 y 3 + 6xy 2 6x 2 y,
(2.26)
The first solution corresponds to the free N = 8 invariant action, while the higher-order N = 8
invariants produce non-trivial sigma model bosonic metrics.
For the SO(4) SO(4) invariant case the conformal factor G(8) in (2.17) is related to L as
(8) = 4(xLxx + 2Lx ) = 4(yLyy + 2Ly ).
G
(2.27)
The non-trivial polynomial Lagrangians in (2.26) produce the following target bosonic metrics
(8) = 24(x y),
G
2

(8) = 48 x 2 + y 2 3xy .
G
3
(2.28)
It is interesting that the maximally symmetric (SO(8) invariant) singular solution of (2.18),
(8) = a0 + a1
G
so(8)
1
,
(x + y)3
corresponds to the following choice of the N = 4 superfield Lagrangian

a0
a1 1 1
1
.
Lso(8) = (x y) +
8
16 x y x + y
(2.29)
(2.30)
The first term is the free Lagrangian while the second one produces a non-trivial N = 8, d = 1
sigma-model. Note that (2.30) is defined up to terms which satisfy the dimension-4 harmonicity
conditions (2.14). The general function (q, ) preserving the manifest SO(4) SO(4) symmetry is as follows
so(4) = c0 +
c 1 c2
c3
+
+
,
x
y
xy
(2.31)
where ci , i = 0, . . . , 3, are arbitrary real constants. Using this freedom, one can equivalently
replace the second term in (2.30) by the expressions
a1
1
8 x(x + y)
or
a1
1
.
8 y(x + y)
(2.32)
500
2.3. SO(8) covariance

Actually, the SO(8) symmetry is present already at the superfield level since it is the hidden
symmetry of the defining constraints (2.2). Two SO(4)s from this SO(8) are manifest in the
N = 4 superfield description. They are formed just by two automorphism SU(2)s realized on the
indices i, a and two PauliGrsey type SU(2)s realized on the superfield indices , A. It turns out
that the constraints (2.2) respect the covariance also under the following hidden SO(8)/[SO(4)
SO(4)] transformations
1
(8) i = i aA qaA lbA lb D ic qcA ,
2
1
(8) q aA = k aA k + l bA lb D ia i .
(2.33)
2
Here i aA satisfying the reality conditions with respect to its four doublet indices encompasses
just 16 parameters of the coset SO(8)/[SO(4)SO(4)], and the additional -dependent terms are
necessary for preserving the constraint (2.2). It is straightforward to check that the Lie bracket
of two transformations (2.33) yields just the transformations of the above mentioned manifest
SU(2) symmetries, e.g.,

k ,
[(8)(2) , (8)(1) ] k = ( ) k + (k l) l (j t) tb + (b d) j d
(2.34)
j b
where

1 t aA
(tk) = (2) k(1) aA (1 2) ,
2
etc. Also, one can check that (2.33) takes the manifest N = 4 supersymmetry transformations
into the hidden ones (2.3) and vice versa. Defining the manifest supersymmetry transformations
as
i = kb Qkb i , q aA = kb Qkb q aA , Qkb = kb ikb t ,

(2.35)
it is easy to find
1 A ia
[(8) , ] i = (br)
D qaA = (br) i ,
2
kb
Qkb i = (br) i ,
[(8) , ] i = (br)
(2.36)
where
A
= l aA la ,
(br)
kb
(br)
= k bA A .
It is not too easy to directly check the SO(8) invariance of the full N = 4 superfield action
corresponding to the Lagrangian (2.30), since its invariance is implicit (it holds modulo total
spinor derivatives under the Berezin integral), and essential use of the defining constraints (2.2)
is needed when checking it. The proof of the SO(8) invariance for the free part of (2.30) is rather
simple, and it is essentially based on the identity

dt d 4 q aA i = 0,
(2.37)
which follows from the constraints (2.2). The proof for the second part of (2.30) (or its equivalent
forms (2.32)) is rather intricate, and the more direct way to check the SO(8) invariance of the
related action is to make use of the component form of it.
501
2.4. The general component action of the multiplet (8, 8, 0)

A direct calculation with the help of the constraints (2.2) and their corollaries (2.7), (2.15),
as well as the definition of the Grassmann integration measure (2.5), yields the following exact
answer for the component form of the general N = 4 superfield action (2.4):

gen
S(8) = dt Lcomp (q, , ), Lcomp = Lbos + L2ferm + L4ferm .
(2.38)
Here, Lbos was defined in (2.16) while two-fermion and four-fermion terms are given by the
expressions

i
t kA kA ((q) L) t a a (() L) ,
16
(q) L
c (q) L
t kA = t kA + kD t q(D
((q) L)1 + Ai t (i
((q) L)1
cA)
q
k)
(q) L
() L
a t qaA
((q) L)1 a t k
((q) L)1 ,
k
q aA
i () L
A () L
(() L)1 + d t q(d
(() L)1
t a = t a + a t (
i)
q a)A
(q) L
() L
lB t qaB
(() L)1 lB t l
(() L)1 ,
l
q aB

1 1 ()
1
L4ferm =
() 2() L (AB) (AB) 2(q) L
48 2
2
L2ferm =
+ () c iD
+ 3 () (ik)
(2.39)
(2.40)
2
2 () L
(AB) k b (q) L
B
q cD i
q bA k

2
2 (q) L
(AB) (cd) () L
.
i k
q cA q dB
(2.41)
Here,
1 a
,
4 a

1
(AB) = iA iB ,
4
() =
1 a b
,
4

1
(ik) = Ai kA ,
4
(ab) =
(2.42)
the spinors a and iA are the lowest components of the spinor superfields defined in (2.7),
and we omitted the vertical bar of (q) L and () L (now q aA and i denote the first bosonic
components of the relevant superfields).
It is rather tedious though straightforward to check that (2.38) is invariant under the manifest
N = 4 supersymmetry
1
q aA = ia iA ,
2
1
i = ia a ,
2
iA = 4i ib t qbA ,
b = 4i ib t i ,
(2.43)
502
without any restrictions on the functions () L and (q) L, and under the transformations of
hidden N = 4 supersymmetry
1
q aA = A a ,
2
1
i = A Ai ,
2
iA = 4iA t i ,
b = 4iA t qAb ,
(2.44)
provided only that the dimension-8 harmonicity condition holds, i.e.,

(q) L = () L G(8) .
(2.45)
The component transformations (2.43), (2.44) follow from the superfield transformation laws
(2.35) and (2.3). It can be directly checked that each of these transformations closes off-shell on
t -translations and that (2.43), (2.44) commute with each other.
The SO(4) SO(4) invariant case corresponds to the choice L = L(x, y) in (2.16), (2.39),
(2.40) and (2.41), where one should substitute the derivatives with respect to q aA and i as
= 2qcA x ,
= 2i y ,
cA
q
i

(q) = 4 x + 2xx2 ,
() = 4 y + 2yy2 .
(2.46)
The superfield SO(8) transformations (2.33) induce the following transformations of the component fields:
(8) i = i aA qaA ,
(8) a = k aA kA ,
(8) q aB = i aB i ,
(8) iA = i aA a .
(2.47)
While the SO(8) invariance of the free version of (2.38), with (q) L = () L = const (it
corresponds to the first term in the N = 8 supersymmetric Lagrangian (2.30)), is immediately
seen, the check of the SO(8) invariance of the component action corresponding to the second
term in (2.30) requires some effort. Nevertheless, using the properties that in the SO(8) invariant
case
(x y )G(8) = 0,
1
Lx + (x + y)Lxx = 0,
4
(2.48)
one can check that (2.38) is SO(8) invariant in each order in the fermionic fields. It is amusing
that for this check it suffices to use only (2.48), without resorting to the explicit form of the factor
G(8) (recall that G(8) (x + y)3 in the non-trivial SO(8) invariant case).
In what follows we shall argue that the (8, 8, 0) component action (2.38) is the generating one
for the sigma-model type off-shell component actions of all other N = 8 multiplets described
in [20]: these actions can be obtained from (2.38) by an algorithmic procedure preserving N = 8
supersymmetry. In this sense (2.38) is the true analog of the general sigma-model type action of
the single root N = 4 supermultiplet (4, 4, 0) [33]. Thus the class of general actions of two
mutually mirror N = 4 root multiplets (4, 4, 0) involves a subclass of N = 8 supersymmetric actions. The general action of this subclass given above can be called the parent N = 8
mechanics action.
503

3.1. Hidden N = 4 supersymmetry and bosonic metric
In the N = 4 approach the multiplet (7, 8, 1) is a sum of the N = 4 off-shell multiplets (3, 4, 1)
and (4, 4, 0) which are described by the N = 4 superfields v ik = v ki and q aA subjected to the
following constraints
(a)
D a(i v kl) = 0,
(b)
D i(a q b)A = 0.
(3.1)
As in the previous model, the doublet index A corresponds to one external PauliGrsey type
SU(2) group. The transformations of the hidden N = 4 supersymmetry completing the manifest
one to the full N = 8 supersymmetry are as follows [20],
1 (i k) aA
2
v ik = A
(3.2)
Da q ,
q aA = iA D aj vij .
2
3
The constraint (3.1b) implies the same corollaries (2.7), (2.15) as in the previous case. We shall
also need some consequences of the constraint (3.1a), namely

1 j i ak
+ ki aj , ak D ia vik ,
3
3i
i
ia bk
D = 3i ab t v ik ab ik F, F Dia ia ,
2
6
2
D ia F = t ai .
3
D ia v j k =
(3.3)
Starting from the most general action of the superfields v ik , q aA in N = 4 superspace,

gen
S(7) = dt d 4 L(v, q),
(3.4)
and considering its variation under the hidden N = 4 supersymmetry transformations (3.2), one
can deduce the general conditions of the N = 8 invariance of the action by requiring the variation
of L to be a total spinor derivative:
L
L
1 (i
2
L = A Dak) q aA ik iA D aj vij aA = iA Dak GaA ik
2
v
3
q
(3.5)
(cf. (2.6)). Equating the coefficients of the independent spinors ak and iA in both sides of this
condition, we find
(a)
L
GaA ik
= BA
,
q aB
vik
(b)
L
GaA ki
= ji
.
v kj
qaA
(3.6)
As the compatibility condition of these equations, L should satisfy the dimension-7 harmonicity equation
((v) + (q) )L(v, q) = 0,
(3.7)
where
(v) =
2
.
v ik vik
(3.8)
504
This constraint is the only integrability condition for L following from Eqs. (3.6). From the same
general N = 8 supersymmetry conditions it follows that
((v) + (q) )GaA ik = 0.
(3.9)
For any L satisfying (3.7) the function

can be restored, up to an unessential solution of
the appropriate homogeneous equations, from (3.6) and its corollary (3.9).
Using the relations (2.7), (2.15), (3.3) and the definition (2.5), one can compute the bosonic
part of the most general v, q action (3.4) as

1
1 2
gen
2
2
dt ((q) L|)(t q) ((v) L|) (t v) + F
S(4+3)bos =
(3.10)
.
2
2
GaA ik
The basic condition of N = 8 supersymmetry (3.7) requires two independent conformal factors
in (3.10) to coincide, implying that the general target bosonic metric in the (7, 8, 1) case is
conformally flat, like in the previous (8, 8, 0) case, i.e.,

1
1 2
gen
(7)
2
2
dt G
S(7)bos =
(3.11)
(t q) + (t v) + F ,
2
2
with
G(7) = (q) L| = (v) L|.
(3.12)
As we shall see later, further descending to the multiplets (6, 8, 2) and (5, 8, 3) preserves the
fundamental properties of the conformal flatness of the bosonic target metrics and the harmonicity of the relevant conformal factors. It is natural to assume that these two properties are retained
also upon descending to the multiplets with a smaller number of physical bosons. For the case
of the single (4, 8, 4) multiplet this was independently demonstrated in [35] (see also [31] where
the case with few such multiplets was considered). In the extreme case of the multiplet (1, 8, 7)
the harmonicity condition should degenerate into its dimension-1 version which implies the corresponding conformal factor to be a linear function of the single physical boson. This agrees with
the results of [41].
To complete this subsection, we present the particular form of the N = 8 supersymmetry
conditions (3.6) for the case when all three SU(2) symmetries realized on the superfields v ik and
q aA are preserved. In this case L is a function of two independent invariants x = q aA qaA and
y = v ik vik , and the matrix GaA ij is reduced to the form3
GaA ij = q aA v ij G(x, y).
Then (3.6) is reduced to the simple set of two equations
1
3
1
Ly = G + xGx ,
Lx = G yGy ,
2
4
2
the integrability condition of which is the equation
3
xLxx + yLyy + 2Lx + Ly = 0,
2
(3.13)
(3.14)
3 One could also add the term q aA ij G(x,
the homogeneous equations 2G

+
y). However, Eqs. (3.6) imply for G
x =G
y = 0, so this extra term is unessential and can be omitted.
xG
505
which is just the dimension-7 Laplace equation (3.7) for the considered particular case. It is of
interest to present a few first polynomial solutions of this equation, together with the relevant
(7) = 4(xLxx + 2Lx ) = 4(yLyy + 3 Ly ):
conformal factors G
2
4
(7) = 8,
L1 = x y,
G
1
3

8 2
4
(7)
2
G2 = 24 x y ,
L2 = x + y 4xy,
5
3

64
48
8 2
(7)
3
3
2
2
2
L3 = x y 8x y + xy ,
G3 = 48 x + y 4xy .
35
5
5
(3.15)
Thus, there exist non-trivial sigma-model type interactions of the multiplet (7, 8, 1). The appear (7) is due to the fact that both these quantities satisfy
ance of the same expressions for L and G
the dimension-7 Laplace equation and, hence, Eq. (3.14) in the [SU(2)]3 invariant case.
3.2. SO(7) symmetry
An analog of the metric (2.29) in the considered case is the SO(7) invariant metric
(7) = a0 + a1
G
so(7)
1
.
(x + y)5/2
The corresponding N = 4 superfield Lagrangian reads

a0
4
a1
1
x y +
.
Lso(7) =
8
3
3 x x +y
(3.16)
(3.17)
The (7, 8, 1) analog of the freedom (2.31) is that of adding to (3.17) the terms
= b0 +
b1
b2
b3
+ + .
x
y x y
(3.18)
They do not contribute to the action since they simultaneously satisfy both the dimension-4 and
dimension-3 harmonicity conditions. In distinction to the SO(8) case, this gauge freedom is
not too useful since it does not help to bring the action (3.17) to a simpler form.
The SO(7) analogs of the SO(8) superfield transformations (2.33) can be also found from
requiring the SO(7) covariance of the defining constraints (3.1),
1
(7) v ik = ik aA qaA l(i aA la D k)c qcA ,
2
2
aA
ik aA
vik li bA lb D ta vti .
(7) q =
3
(3.19)
Here ik aA = ki aA form 12 real parameters corresponding to the coset SO(7)/[SO(4) SU(2)].

Like in the SO(8) case, one can easily check that Eq. (3.19) has a correct closure and transforms
the manifest supersymmetry into the hidden one and vice versa. Its realization on the component
fields can be also easily found, but we do not explicitly quote it here. In particular, (7) F = 0.
3.3. The component structure and relation to the (8, 8, 0) multiplet
The hidden N = 4 supersymmetry (3.2) has the following realization on the irreducible N = 4
sets of the component fields (q aA , iA ) and (v ik , ai , F ), where now we denote by the same
506
letters the lowest components of the corresponding superfields:

1 (i k)A
,
v ik = A
2
2
q aA = iA ia ,
iA = 4ikA t v i k 2iiA F,
3
1 i
k
ak = 3iA
(3.20)
t q aA ,
F = A
t iA .
2
On the other hand, let us come back to the hidden supersymmetry transformations (2.44)
of the multiplet (8, 8, 0). Let us identify there two SU(2) groups acting on the doublet indices
, , . . . and i, k, . . . and split the field i into its trace-free and trace part with respect to this
diagonal SU(2),
1
i=k = (ik) + ik .
2
Now it is easy to check that, upon identification
(3.21)
4
a
=k
= ka ,
t = F,
(3.22)
3
the transformation rules (2.44) precisely imply (3.20). Hence, if in the N = 8 supersymmetric
component action (2.38) (with the condition (2.45)) we make the identification (3.22) and assume
that the conformal factor G(8) does not depend on , we obtain the general off-shell N = 8
supersymmetric component action of the multiplet (7, 8, 1). The restricted conformal factor will
satisfy the dimension-7 harmonicity condition as a consequence of the dimension-8 equation
(2.45). This consideration is the direct N = 8 analog of a similar procedure in [33] where the
general off-shell component action of the N = 4 multiplet (3, 4, 1) was obtained from the general
action of the root multiplet (4, 4, 0). An essential difference is, however, that the conformal
factors in the N = 4 cases generally do not satisfy any harmonicity condition, while in the N = 8
cases they necessarily do. The automorphic duality between the multiplets (7, 8, 1) and (8, 8, 0)
in the considered N = 4 superfield formulation can be depicted as
(ik) = v ik ,
(4, 4, 0) (4, 4, 0)
(4, 4, 0) (3, 4, 1).
(3.23)
In the cases of the other N = 8 multiplets considered below one can also establish this automorphic duality with the components of the root N = 8 multiplet (8, 8, 0) and explicitly
obtain the full component sigma-model type actions from (2.38).
3.4. Potential terms
To close this section, we note that for the (7, 8, 1) multiplet one can construct the potential
terms by adding to the superfield action for this multiplet a term which is linear in the superfield v (ik) ,

i
pot
S(7) = m dt d 4 ab ia kb v (ik) ,
(3.24)
3
where m is a real parameter of mass dimension. It is easy to see that, due to the v ik defining
constraint (3.1a), this term is invariant both under the manifest and hidden N = 4 supersymmetries and hence it is N = 8 supersymmetric. Then, integrating over s, one finds its component
off-shell form as

pot
S(7) = m dt F.
(3.25)
507
Eliminating the auxiliary field F in the sum of (3.25) and (3.11), F = 2m(G(7) )1 , we find the
scalar on-shell potential term

1
sc
S(7)
(3.26)
= m2 dt G(7) .
Thus, the single conformal factor G(7) specifies both the target bosonic metric and the scalar
potential. A similar mechanism of generating potential terms exists in the case of the other N = 8
multiplets which feature auxiliary fields. For the multiplet (4, 8, 4) it was demonstrated in [31],
while for the multiplets (6, 8, 2) and (5, 8, 3) it will be shown below.4
4.1. Definitions and N = 8 invariant action
The N = 8 multiplet (6, 8, 2) admits two equivalent off-shell N = 4 superfield formulations [20]:
(a)
(6, 8, 2) = (3, 4, 1) (3, 4, 1),
(b)
(6, 8, 2) = (4, 4, 0) (2, 4, 2).
(4.1)
Most convenient for our purposes is the first N = 4 splitting in (4.1), when this multiplet is
described by two real isotriplet N = 4 superfields v ik = v ki and w ab = w ba subjected to the
constraints
Da(n v ik) = 0,
(c
Dk w ab) = 0.
(4.2)
The superfield v ik is actually the same as the one used in the N = 4 description of the N = 8
multiplet (7, 8, 1) in the previous section. Like in the previous case, using (4.2) and its corollaries
given in Appendix A, one can prove that the irreducible set of the superfield projections includes,
besides these superfields themselves, two quartets of spinors,
ia = D ka vki ,
ia = D ib wba ,
(4.3)
and two isosinglet auxiliary superfields,

i
i
H = Dia ia .
F = Dia ia ,
(4.4)
6
6
All other superfield projections are expressed as time derivatives of these basic ones. The hidden
N = 4 supersymmetry mixes the superfields v (ik) and w (ab) as
2
2 (a b)
k)
w (ab) = i Dk v (ik) .
v (ik) = a(i Db w (ab) ,
3
3
The free action invariant under (4.5) reads

free
S(6) = dt d 4 v 2 w 2 .
(4.5)
(4.6)
As in the previous cases, in order to find the general N = 8 supersymmetric action in the
N = 4 superfield formalism, one starts with the most general action of the superfields w ab
4 A similar phenomenon takes place in the case of nonlinear (4, 8, 4) multiplet [36,37].
508
and v ik ,
gen
S(6) =

dt d 4 L(v, w),
(4.7)
varies the Lagrangian L(v, w) with respect to (4.5) and requires the variation to be a total spinor
derivative:
L
L
2
2 (a b)
k)
L(v, w) = a(i Db w (ab) ik + i Dk v (ik) ab = ia Dkb Gik ab (v, w).
3
v
3
w
(4.8)
From (4.8), using the defining constraints (4.2), one finds the conditions on the function Gik ab :
L
Gik ab
= ni
,
v kn
wab
L
Gik ab
= da
.
w bd
vik
(4.9)
From these conditions it follows that the Lagrangian L and function Gik ab should satisfy the
six-dimensional Laplace equations
(6) L = 0,
(6) Gik ab = 0,
(4.10)
where
(6) = (v) + (w) ,
(v) =
2
2
,
=
.
(w)
v ik vik
w ab wab
(4.11)
In the case of unbroken manifest R-symmetry SU(2) SU(2) the function Gik ab is reduced
to
Gik ab = v ik w ab G(x, y),
where now
x = v ik vik ,
y = w ab wab ,
(4.12)
and the conditions (4.9) are reduced to

3
1
Lx = G yGy ,
4
2
3
1
Ly = G + xGx .
4
2
(4.13)
The corresponding dimension-6 Laplace equations for the Lagrangian and the function Gik ab in
this case are reduced to the equations
3
xLxx + yLyy + (Lx + Ly ) = 0,
2
5
xGxx + yGyy + (Gx + Gy ) = 0.
2
The first few polynomial solutions of Eq. (4.14) are:
(4.14)
(4.15)
10
(4.16)
L3 = x 3 7x 2 y + 7xy 2 y 3 .
xy + y 2 ,
3
It also bears no problem to find the relevant functions G by solving Eq. (4.15).
The general bosonic component action corresponding to the superfield action (4.7) is easily
found to be
L1 = x y,
L2 = x 2

gen
S(3+3)bos
509

1 2
ik
dt ((v) L|) t v t vik + B
2

1
ab
2
.
((w) |)L t w t wab + H
2
1
=
2
(4.17)
The bosonic part of the N = 8 supersymmetric action corresponds to the choice

G(6) = (v) L = (w) L,
(4.18)
and so is given by

1
1 2
gen
(6)
ik
ab
2
dt G
,
t v t vik + t w t wab + B + H
S(6)bos =
2
2
(6) G(6) = 0.
(4.19)
In the SO(4) invariant case we have

(6) = 4 xLxx + 3 Lx = 4 yLyy + 3 Ly .
G
2
2
(4.20)
The conformal factors corresponding to the SO(4) invariant solutions (4.16) are

10
(6)
(6)
(6)
2
2
G1 = 4,
G2 = 20(x y),
G3 = 42 x xy + y .
3
(4.21)
4.2. SO(6) SO(2) symmetry

The constraints (4.2) reveal covariance under the two hidden mutually commuting internal
symmetry transformations
2
(6) v ik = ik ab wab j (i cb j c D k)a wab ,
3
2
ab
ik ab
vik + j i c(b j c D ka) vki ,
(6) w =
3
1
1
(2) w ab = i(a D kb) vki .
(2) v ik = b(i D k)a wab ,
3
3
(4.22)
(4.23)
Here, ik ab = ki ab = ik ba and comprise, respectively, nine and one real parameters. The
transformations (4.22) belong to the coset SO(6)/SO(4), while (4.23) form an Abelian SO(2)
group. The SO(2) transformations also commute with the manifest SO(4) symmetry.
Thus, the full R-symmetry in the considered case is SO(6) SO(2), as expected. The transformations (4.22), (4.23) transform the manifest N = 4 supersymmetry into the implicit one and
vice versa, as in the previously studied cases. It is instructive to show how these transformations are realized on the component fields (v ik , ia , F ) and (w ab , ia , H ) defined as the lowest
components of the corresponding superfield projections:
(6) v ik = ik ab wab ,
(6) F = 0,
(6) =
kb ,
(6) H = 0,
1
1
(2) ia = ia ,
(2) ia = ia ,
(2) v ik = (2) w ab = 0,
2
2
(2) H = F.
(2) F = H,
(6) w
ab
(6) ia = ik ab kb ,
ik ab
vik ,
ia
ik ab
(4.24)
(4.25)
510
We see that the SO(6) transformations act on the physical bosonic and fermionic fields, leaving
the auxiliary fields invariant, while the SO(2) ones are realized on the fermionic and auxiliary
fields which form two SO(2) doublets. The physical bosonic fields are SO(2) singlets.
The SO(6) invariant solution of the dimension-6 harmonicity condition for the conformal
factor G(6) in the bosonic action (4.19) is
(6) = a0 + a1
G
so(6)
1
,
(x + y)2
(4.26)
where now x and y are the lowest bosonic components of the superfields defined in (4.12). Using
the relations (4.20), it is not difficult to restore the N = 4 superfield Lagrangian giving rise to
(4.26) (as in the previous cases, it is defined up to adding mutual solutions of two dimension-3
harmonicity equations). The first term in (4.26) corresponds to the free action (4.6), while the
second term is reproduced from the Lagrangian

x
a1 1
Lso(6) = arctan
(4.27)
.
2 xy
y
Note that this Lagrangian, like the free one (4.6), alters its sign under the interchange x y,
modulo terms const 1xy which satisfy the dimension-3 harmonicity conditions and therefore
do not contribute to the action.
4.3. Relation to the multiplets (7, 8, 1) and (8, 8, 0)
It is instructive to give the realization of the hidden N = 4 supersymmetry (4.5) on the component fields:
2
v ik = a(i k)a ,
3
2 (a
w ab = i b)i ,
3
3i ia
3i
ia
i
ab
ia = 3ika t v ik + ia F,
= 3ib t w H,
2
2
2 ia
2 ia
F = t ia ,
(4.28)
H = t ia .
3
3
Now let us compare these transformation laws with the hidden N = 4 supersymmetry transformations (3.20) of the components of the multiplet (7, 8, 1). The N = 4 multiplet (v ik , ia , F )
is common for both N = 8 multiplets. Then let us identify the SU(2) group acting on the doublet indices A, B, . . . of the (4, 4, 0) multiplet (q aA , iA ) with the SU(2) acting on the doublet
indices a, b, . . . , which implies identifying these two sets of doublet indices as well. Then we
decompose
1
q bA=a = q (ab) + ba q
2
and identify
(4.29)
4
ka = ka ,
t q = H.
(4.30)
3
It is easy to check that, for the fields defined by (4.29) and (4.30), Eqs. (3.20) imply just the transformation properties (4.28). In other words, we proved that the multiplets (7, 8, 1) and (6, 8, 2)
are related to each other by an automorphic duality of the same type as the one discussed in
q (ab) = w ab ,
511
Section 3.3 and relating the multiplets (8, 8, 0) and (7, 8, 1). This chain of two dualities implies
a third one directly relating the multiplets (8, 8, 0) and (6, 8, 2). This chain can be depicted as
(4, 4, 0) (4, 4, 0)
(4, 4, 0) (3, 4, 1)
(3, 4, 1) (3, 4, 1).
(4.31)
Thus, we conclude that the general d = 1 sigma-model off-shell component action of the multiplet (6, 8, 2) can be obtained from the generating (8, 8, 0) multiplet action (2.38), (2.16),
(2.39)(2.41) just via (i) the proper identification of the doublet indices i, a, , A; (ii) substituting the decompositions (3.21) and (4.29); (iii) assuming that the conformal factor G(8) defined
by (2.45) bears no dependence on the fields and q and so can be identified with G(6) in (4.19);
(iv) identifying t = F , t q = H .
Let us make a comment on the alternative N = 4 formulation (4.1b). It is related with the
root multiplet (8, 8, 0) via the following chain of automorphic dualities
(4, 4, 0) (4, 4, 0)
(4, 4, 0) (3, 4, 1)
(4, 4, 0) (2, 4, 2)
(4.32)
and can also be easily treated within our approach, giving rise to the same component Lagrangians. Let us here present the corresponding N = 4 superfield Lagrangian yielding the SO(6)
invariant metric (4.26) in components. It now can be determined from the equations
(6) ,
4(xLxx + 2Lx ) = 4(yLyy + Ly ) = G
so(6)
(4.33)
and is a chiral N = 4 superfield describing the multiplet (2, 4, 2).

where x = q aA qaA , y = ,

The Lagrangian Lso(6) obtained as a solution of these equations has the surprisingly simple form
a0
a1 log(x + y)
(x 2y)
.
8
4
x
By construction it satisfies the dimension-6 harmonicity condition, with the Laplacian
Lso(6) =
(6) = (q) + 4
(4.34)
2
= 4(xLxx + yLyy + 2Lx + Ly ).

4.4. Potential terms in superfields and components

In the considered case there exist two independent FayetIliopoulos-type terms preserving
both the manifest and hidden N = 4 supersymmetries,

i
i
pot
dt d 4 ik ia kb w(ab) ,
S(6) = m dt d 4 ab ia kb v (ik) + m
(4.35)
3
3
where m and m
are two mass parameters. In components, (4.35) amounts to

pot
dt H.
S(6) = m dt F m
(4.36)
The invariance of (4.35) under both N = 4 supersymmetries can be checked using the constraints (4.2). In fact it is simpler to check the invariance in the component formulation (4.36),
using the transformation rules (4.28) (and the corresponding transformation rules of the explicit
N = 4 supersymmetry). After elimination of the auxiliary fields F and H from the sum of the
bosonic actions (4.19) and (4.36), one obtains the on-shell scalar potential term

1
sc
dt (6) .
= m2 + m
2
S(6)
(4.37)
G
512
Note that (4.35), (4.36) break the SO(2) R-symmetry but respect the SO(6). Therefore, there
exists a unique SO(6) invariant scalar potential corresponding to the choice (4.26),5

2
(x + y)2
sc
2
dt
Sso(6) = m + m
.
(4.38)
a1 + a0 (x + y)2
5.1. N = 8 invariant action
The off-shell N = 8 multiplet (5, 8, 3) admits two equivalent N = 4 superfield splittings [20],
(a)
(5, 8, 3) = (4, 4, 0) (1, 4, 3),
(b)
(5, 8, 3) = (3, 4, 1) (2, 4, 2).
(5.1)
For reasons to be explained below we choose the option (a) and will describe the multiplet in
question by the N = 4 superfields q aA and u subjected to the constraints
b)
D k(b q a)A = 0,
D k(a Dk u = 0.
(5.2)
The superfield q aA is the same as in Sections 2 and 3, and we will use the same notation for
its irreducible superfield projections. Thus, besides the superfields q aA and u themselves, the
constraints (5.2) single out the following independent projections, the lowest components of
which constitute the irreducible field content of the multiplet (5, 8, 3).
Spinors:
kb = D kb u,
kA = D kb qbA .
(5.3)
Auxiliary fields:
i
A(ik) = Db(k i)b .
(5.4)
2
To prove that any other superfield projection is expressed as a time derivative of these basic
ones, one can use the identities (2.7), (2.15) together with those for the superfield u,
D ia D kb u = D ia kb = i ik ab t u + i ab A(ik) ,
la
(ik)
D A
= i t
kl
ia
+ i t .
il
ka
(5.5)
(5.6)
The hidden N = 4 supersymmetry is realized by the transformations

1
u = kA D ka qaA .
2
The invariant free action is given by

free
S(5)
= dt d 4 q 2 2u2 .
q aA = kA D ka u,
The generic action of the two N = 4 superfields considered,

gen
S(5) = dt d 4 L(q, u),
(5.7)
(5.8)
(5.9)
5 To prevent possible confusion, let us mention that the SO(2) symmetry is broken in fermionic Yukawa-type couplings
which complete the scalar potentials to the full N = 8 invariant and which are omitted in (4.37), (4.38).
513
possesses N = 8 supersymmetry if, under the transformations (5.7), it transforms into a total
spinor derivative,
L(q, u) = kA D ka u
L
L
1
+ kA D ka qaA
= iA Dkb Gik bA (q, u).
aA
q
2
u
(5.10)
By comparing the coefficients in the left- and right-hand sides of this equality, one finds the
general conditions of N = 8 supersymmetry in the system under consideration, namely
L
Gik aA
,
= ik
u
qaA
L
Gik aA
= ik BA
.
q aB
u
(5.11)
These conditions can be essentially simplified because they imply that, up to an unessential
constant,
Gik aA = ik GaA .
(5.12)
Then, (5.11) are equivalently rewritten as

L
GaA
,
=
u
qaA
L
GaA
= BA
.
aB
q
u
(5.13)
As in the previous cases, the only constraint on the Lagrangian L, following from the set of
Eqs. (5.13) as their compatibility condition, is the dimension-5 harmonicity equation
(5) L = 0,
(5.14)
where now
(5) = (q) + (u) ,
(q) =
2
2
,
=
2
.
(u)
q aA qaA
u2
(5.15)
Eqs. (5.13) also imply the same equation for the function GaA , i.e.,
(5) GaA = 0.
(5.16)
As in the previous cases, all these equations are drastically simplified for R-symmetric Lagrangians. If we wish to preserve all three involved SU(2) symmetries, we are led to choose the
ansatz
GaA = q aA G(x, u),
x = q aA qaA ,
for which Eqs. (5.13) are reduced to

Gu = 2Lx ,
2G + xGx = Lu .
(5.17)
The corresponding equations for L and G, derived as the integrability conditions of the system (5.17), have the following form:
1
1
xGxx + 3Gx + Guu = 0,
(5.18)
xLxx + 2Lx + Luu = 0.
2
2
The second equation is just the dimension-5 Laplace equation for L in the variables x, u. The
first few polynomial solutions of this linear equation read
L1 = x 2y,
L2 = x 2 6xy + 2y 2 ,
8
L3 = x 3 12x 2 y + 12xy 2 y 3 , (5.19)
5
514
where we defined, by analogy with the previously studied cases, y = u2 . The first solution is the
free action, the remaining ones generate non-trivial sigma-model type interactions.
The bosonic component action obtained from the general N = 8 supersymmetric action of the
superfields q aA , u reads

1
1
gen
dt G(5) t q aB t qaB + t ut u + A(ik) A(ik) ,
S(5)bos =
(5) G(5) = 0, (5.20)
2
2
where
G(5) = (q) L = (u) L.
(5.21)
(5) is related to L(x, y) by

For the SO(4) invariant case (5.18) the conformal factor G
(5) = 4(xLxx + 2Lx ) = 4(Ly + 2yLyy ).
G
(5.22)
5.2. SO(5) SU(2) symmetry

The N = 4 superfield constraints (5.2) are preserved under the following hidden internal symmetry transformations [10],
(5) q aA = aA u cA kc D ka u,
1
(5) u = 2aA qaA + cA kc D kb qbA .
2
(5.23)
Here, aA is a quartet of real parameters. It is straightforward to check that these transformations

close on SO(4) = SU(2)R SU(2) where the first SU(2) factor acts on the doublet indices a of
the superfield q aA and the Grassmann coordinate ka , i.e., it belongs to the manifest R-symmetry
group SO(4)R , while the second SU(2) factor acts on the extra doublet index A of q aA and so
commutes with the manifest N = 4 supersymmetry. Thus the transformations (5.23), together
with the two SU(2) groups just mentioned, form the 10-parameter group SO(5), and these transformations belong to the coset SO(5)/SO(4). It can be also checked that the second manifest
R-symmetry group, SU(2)R acting on the doublet indices i, k, commutes with (5.23). Thus the
full R-symmetry of the considered case is SO(5) SU(2)R . It is instructive to give the realization of (5.23) on the component fields, using the definitions (5.3), (5.4), the relation (5.5) and the
constraint (2.2) with its corollary (2.15):
(5) q aA = aA u,
(5) u = 2aA qaA ,

1
(5) ia = aA Ai ,
(5) iA = 2bA bi ,
2
(5.24)
(5) Aik = 0.
(5.25)
We observe, in particular, that the auxiliary field Aik is inert under the SO(5) group and is
transformed only under the automorphism SU(2) group which acts on the indices i, k, . . . and
commutes with SO(5). The physical bosons are transformed only under SO(5). On the fermions,
like in the previous cases, the whole R-symmetry group SO(5) SU(2) is effective.
As in the previous models, it is interesting to establish the explicit form of the N = 4 superfield
Lagrangians corresponding to the bosonic metric with maximal internal symmetry. In the case
under consideration, this is SO(5) symmetry, and the SO(5) invariant solution of the dimension-5
harmonicity equation for the conformal factor G(5) in (5.20) reads
(5) = a0 + a1
G
so(5)
1
,
(2x + y)3/2
x = q 2 , y = u2 ,
(5.26)
515
where, as in the previous cases, a0 , a1 are some real constants. From the transformation
rule (5.24) it follows that 2x + y = 2q aA qaA + u2 is indeed SO(5) invariant. Using the definition (5.22), one finds that the first term in (5.26) corresponds to the free Lagrangian (5.8),
while the second term is reproduced from the Lagrangian
a1 2x + y
Lso(5) =
(5.27)
.
4
x
As in the previous cases, this Lagrangian is defined modulo
1
u
+ c3 ,
(5.28)
x
x
which simultaneously solves the dimension-3 and dimension-1 Laplace equations and so does
not contribute into the action. Choosing, e.g.,
a1
c0 = c1 = c2 = 0,
(5.29)
c3 = ,
4
one can cast the Lagrangian (5.27) into the alternative form
a1 u 2x + y
a1
1
Lso(5) =
(5.30)
.
=
4
x
2 u + 2x + y
c 0 + c1 u + c 2
It coincides with the N = 8 superconformally invariant Lagrangian found in [10] by a rather

complicated recurrence procedure, starting from some general N = 4 superconformally invariant action for the superfields q and u and then requiring invariance under the hidden N = 4
supersymmetry. We see that the same action can be recovered much more simply from the
requirement of SO(5) invariance. Also, it follows from our consideration that the N = 8 supersymmetric Lagrangians (5.27) and (5.30) automatically satisfy the dimension-5 harmonicity
condition, a property which was not noticed in [10]. We remark that this paper also constructed
an N = 8 superconformal Lagrangian for the alternative N = 4 splitting of the multiplet (5, 8, 3)
corresponding to the option (b) in (5.1). It can be written as

log( W 2 + W 2 + )
conf
L(b)
(5.31)
,
W2
where W 2 = W ab Wab and W ab represents the multiplet (3, 4, 1), while and are chiral
and antichiral N = 4 superfields describing the multiplet (2, 4, 2). We have checked that this
Lagrangian can also be recovered from the requirement of SO(5) invariance, and that it satisfies
the appropriate dimension-5 harmonicity condition

2
2
Lconf
+4
(b) = 0.
W ab Wab

This confirms the universal character of the harmonicity conditions for the N = 4 Lagrangians
as the conditions of N = 8 supersymmetry.
5.3. Relation to the multiplet (8, 8, 0)
Let us now demonstrate that the N = 8 multiplet (5, 8, 3) in the N = 4 superfield description (5.1a) is, by a variant of the automorphic duality, directly related to the root N = 8
multiplet (8, 8, 0). This implies that its general off-shell sigma-model component action can
516
be restored by simple rules from the parent (8, 8, 0) action, as in the cases considered in the
previous sections.
To this end, let us first write the realization of the hidden N = 4 supersymmetry (5.7) on the
component fields:

1
u = kA kA ,
kA = 2i kA t u + lA Alk ,
2
(i
k
t q aA ,
Aik = A t k)A .
ka = 2iA
q aA = kA ka ,
(5.32)
As the second step, let us identify, in the hidden supersymmetry transformations of the component fields of the (8, 8, 0) multiplet (2.44), two SU(2) groups realized on the indices i and of
the second (4, 4, 0) multiplet, and decompose the field i =k just on the pattern of (3.21),
1
ik = (ik) + ik , = ki ik .
2
Now it is easy to check that after the formal identifications
(5.33)
1
t (ik) = Aik
(5.34)
2
the transformations (2.44) precisely yield (5.32).
It immediately follows from this remarkable fact that the general sigma-model type off-shell
action of the multiplet (5, 8, 3) can be obtained from the (8, 8, 0) multiplet action (2.38), (2.16),
(2.39)(2.41) just by making the substitutions (5.34) in the latter and by eliminating the dependence on the triplet (ik) in the general conformal factor G(8) defined by (2.45). The reduced
factor satisfies the dimension-5 harmonicity condition, as expected, and hence it can be identified
with G(5) (recall (5.20)).
The direct automorphic duality between the off-shell N = 8 multiplets (8, 8, 0) and (5, 8, 3)
can be depicted as
= u,
=ka = 2 k a ,
(4, 4, 0) (4, 4, 0)
(4, 4, 0) (1, 4, 3).
(5.35)
In contrast to the chain of dualities (4.31), this chain does not go through any intermediate step.
Nevertheless, there exists another option, also finally resulting in the multiplet (5, 8, 3), but in
the alternative N = 4 superfield description (5.1b) including the chiral N = 4 supermultiplet:
(4, 4, 0) (4, 4, 0)
(4, 4, 0) (3, 4, 1)
(3, 4, 1) (2, 4, 2).
(3, 4, 1) (3, 4, 1)
(5.36)
This is just a continuation of the chain (4.31). The final component action is of course equivalent
to the one produced by the chain (5.35).
5.4. Potential terms
For the case under consideration, the off-shell N = 8 supersymmetric potential terms in superfield and component form read

pot
S(5) = m dt d 4 ab ia kb C (ik) u,
(5.37)

pot
S(5)comp = m dt C(ik) A(ik) .
(5.38)
517
They clearly break the SU(2) factor in the full R-symmetry group SO(5) SU(2) down to U(1)
while preserving the SO(5) factor. After eliminating Aik from the full action by its algebraic
equations of motion there arises an on-shell scalar potential term accompanied by some Yukawatype fermionic couplings. The scalar potential can be found just from the sum of (5.20) and
(5.38),

1
sc
S(5)
(5.39)
= m2 C 2 dt G(5) .
Though this term does not exhibit any breaking of the R-symmetry SU(2) (because only the
square C 2 = C ik Cik appears), the suppressed fermionic terms explicitly involve C ik , and so this
SU(2) is broken in the total on-shell action.
Finally, we would like to point out that the general N = 4 superfield action for the N = 8
multiplet (5, 8, 3) in the splitting (3, 4, 1) (2, 4, 2) was constructed previously and studied in
[21,22,24]. At the same time, the alternative splitting (4, 4, 0) (3, 4, 1) was not elaborated on
too much. In this paper, we have filled this gap and have shown that it yields equivalent results
for the action and potentials. We also established the precise relation of the multiplet (5, 8, 3)
to the root N = 8 multiplet (8, 8, 0) through an automorphic duality, which was not explicitly
done before.
6. Summary, further examples and outlook
In this paper we have studied several so far unexplored models of N = 8 supersymmetric
mechanics with 8 physical fermions [20] in the off-shell N = 4 superfield approach: those associated with the N = 8 multiplets (8, 8, 0), (7, 8, 1) and (6, 8, 2). Also, the model associated with
the multiplet (5, 8, 3) was studied for the N = 4 splitting (5, 8, 3) = (4, 4, 0) (1, 4, 3) which
was not addressed in full generality before. We derived general conditions of the existence of the
second hidden N = 4 supersymmetry in these models, gave the corresponding general superfield
actions and their component bosonic cores (and the full component action for the (8, 8, 0) case),
considered some instructive examples including the actions with the maximal internal symmetry SO(b) SO(a) where b and a are the corresponding numbers of the physical and auxiliary
bosonic fields, and found the precise realization of this maximal symmetry on the superfields and
component fields. Three main characteristic common features of the models studied here can be
summarized as follows.
The general condition ensuring N = 8 supersymmetry of the corresponding N = 4 superfield
Lagrangians is that the latter should satisfy the harmonicity condition with respect to the
involved bosonic N = 4 superfields as arguments.
The bosonic target metric is always conformally flat, with the conformal factor being related
in a simple way to the superfield Lagrangian and obeying the same dimension-b harmonicity
condition. For the multiplets with auxiliary fields, the same conformal factor specifies the
on-shell scalar potential of physical bosonic fields.
The N = 8 mechanics models reveal an interesting hierarchical structure. All these are
related to the root (8, 8, 0) model through the automorphic duality which is the opportunity to replace, without affecting N = 8 supersymmetry, time derivatives of some physical
bosonic fields in a given off-shell multiplet by the auxiliary field with the same transformation rule, thus producing a new off-shell multiplet. Using this duality, the off-shell component
sigma-model type actions of all considered multiplets can be recovered by simple rules from
518
the general component action of the multiplet (8, 8, 0) which is thus the generating action
for all models.
All these properties seem to extend also to the N = 8 mechanics models associated with the
rest of the multiplets classified in [20], viz. the multiplets (4, 8, 4), (3, 8, 5), (2, 8, 7), (1, 8, 7)
and (0, 8, 8). Actually, our analysis can be directly applied to any N = 4 splitting of these multiplets, such that the involved N = 4 superfields possess manifestly N = 4 supersymmetric actions
in the ordinary N = 4, d = 1 superspace. The latter criterion fails to be valid for the purely
fermionic N = 4 multiplet (0, 4, 4) the actions of which can be formulated in the ordinary N = 4
superspace only with the inclusion of explicit Grassmann coordinates. Therefore, the N = 4 splittings involving this multiplet, e.g., (4, 8, 4) = (4, 4, 0) (0, 4, 4), fall beyond our analysis. Since
the N = 4 splittings of the multiplets (1, 8, 7) and (0, 8, 8) are unique and necessarily include
the multiplet (0, 4, 4) [20], our analysis cannot be directly applied to these special cases. Nevertheless, it is reasonable to assume that at least two last characteristic features from the above list
are also shared by the models related to these two multiplets. Indeed, it was argued in Ref. [41]
in a different (component) approach that the bosonic target metric in the (1, 8, 7) case should be
at most linear in the physical bosonic field, which is just the general solution of the dimension-1
Laplace equation (1) G(1) = 0. As for the extreme fermionic multiplet (0, 8, 8), its general
action should coincide with the free one; however, containing 8 auxiliary fields, this multiplet
could produce new non-trivial scalar potentials while being coupled to N = 8 multiplets featuring physical bosonic fields.
Concerning the multiplet (4, 8, 4) in the splitting (4, 4, 0) (0, 4, 4), it is worth noting that
in Ref. [35] it was considered in the harmonic N = 4, d = 1 superspace [42] where both its
N = 4 constituents (including their off-shell actions) admit a manifestly N = 4 supersymmetric
description.6 The general component action of the relevant N = 8 mechanics model was found
to show the second property from the list above.
In fact, this N = 8 multiplet admits the alternative N = 4 splittings
(a)
(4, 8, 4) = (3, 4, 1) (1, 4, 3)
and (b)
(4, 8, 4) = (2, 4, 2) (2, 4, 2),
(6.1)
which nicely match with our approach and naturally continue the automorphic duality chains
(5.35) and (5.36). Though we did not fully explore the N = 8 systems associated with these
options, it is obvious that they are well inscribed into the hierarchical structure of the N = 8
mechanics models. Therefore, they have to exhibit all the basic features listed above and to give
rise to actions equivalent to the ones constructed in [35] (see also [31]).
As an example of how our approach works in this case, let us present the N = 8 supersymmetric Lagrangian of the (3, 4, 1) and (1, 4, 3) superfields v ik = v ki and u which yields the
maximally symmetric target metric satisfying the dimension-4 harmonicity condition, this time
the SO(4) invariant one,
(4) = a0 + a1
G
so(4)
1
,
x +y
x = v ik vik , y = u2 .
(6.2)
Here, the term a0 corresponds to the free Lagrangian, while the term a1 belongs to a
sigma-model Lagrangian with a non-trivial SO(4) invariant self-interaction. The appropriate La6 Actually, all N = 4 off-shell multiplets admit natural formulations in the N = 4, d = 1 harmonic superspace [4244]
which so seems to supply the most adequate arena for treating these multiplets.
grangian Lso(4) is defined by the equation

3
(4)
Gso(4) = 4 xLxx + Lx = 2(Ly + 2yLyy )

2
519
(6.3)
and, up to the freedom of adding solutions of the dimension-3 and dimension-1 Laplace equations, is given by the expression

y
y 1
a0
Lso(4) = (x 3y) a1
(6.4)
arctan
log(x + y) .
6
x
x 2
Note that the N = 8 bi-harmonic superspace action of the multiplet (4, 8, 4) corresponding to the
SO(4) invariant metric (6.2) was found in Ref. [31].
Analogously, for the multiplet (3, 8, 5) there also exists an N = 4 superfield splitting which
allows treatment by our method, namely
(3, 8, 5) = (2, 4, 2) (1, 4, 3).
(6.5)
It continues the automorphic duality chain (5.36) and (6.1). For the corresponding metric with
the maximal symmetry of the SO(3),
(3) = b0 + b1 1 ,
G
so(3)
x +y
y = u2 ,
x = ,
(6.6)
where is the lowest component of the N = 4 chiral superfield representing the multiplet
(2, 4, 2), one can also find the relevant N = 4 superfield Lagrangian from the relation
(3) = 4(xLxx + Lx ) = 2(Ly + 2yLyy ).
G
so(3)
(6.7)
The appropriate solution is

b0
(x 2y) + b1 x + y y log( y + x + y ) .
(6.8)
4
The piece b1 in (6.8) is nothing but the N = 8 superconformal action of the multiplet (3, 8, 5)
found in [10] by a rather involved recurrence procedure. We see how simple is its derivation in
our framework.
A final example supporting the general character of the hierarchical structure of the N = 8
mechanics model is the multiplet (2, 8, 6), which has a unique N = 4 splitting appropriate for
effecting our procedure,
Lso(3) =
(2, 8, 6) = (1, 4, 3) (1, 4, 3).
(6.9)
Clearly, this splitting is the next link in the automorphic duality chain (5.36) continued to (6.5).
The corresponding maximally symmetric metric solving the dimension-2 harmonicity condition
is the SO(2) invariant one

(2) = c0 + c1 log u2 + v 2 ,
G
(6.10)
so(2)
where u and v are the first components of the two (1, 4, 3) superfields (which are mirror to each
other [20]). The corresponding N = 4 Lagrangian Lso(2) is determined from the set of equations
(2)
Luu = Lvv = G
so(2)
(6.11)
520
and, modulo harmless solutions of two dimension-1 Laplace equations, is given by the expression

c1 2

u
(c0 3c1 ) 2
Lso(2) =
(6.12)
u v2 +
u v 2 log u2 + v 2 + 4vu arctan .
2
2
v
Note that both terms in this expression are odd with respect to the permutation u v (up to
terms vanishing under the N = 4 superspace integral).
The results of this paper raise some questions and suggest several further directions of study.
First of all, it would be interesting to work out in more detail some of the models studied here,
both at the classical and quantum level, for specific choices of the metric factor, and to establish
their links with systems of current interest, e.g., superextensions of integrable CalogeroMoser
models. In connection with the AdS2 /CFT1 version of the general string/gauge correspondence
and the supersymmetric black-hole story it is also of primary interest to study superconformally invariant N = 8, d = 1 models. Clearly, the non-trivial N = 4 superfield Lagrangians with
maximal internal symmetry constructed here for many multiplets and N = 4 splittings are just
candidates for the appropriate superconformal mechanics actions. This is supported by the fact
that the nonlinear Lagrangians in (5.30), (5.31) and (6.8) are superconformal. However, whereas
it is more or less clear that the superconformal actions should simultaneously enjoy the maximal internal R-symmetry (appearing as the important part of the relevant superconformal group)
the converse is not always true: e.g., the SO(4) invariant action of the multiplet (4, 8, 4), in the
manifestly N = 8 supersymmetric description in the N = 8 bi-harmonic superspace [31], is not
superconformally invariant, the same property should be shared by the N = 4 superfield counterpart (6.4) of this action. Thus the issue of the superconformal invariance of the actions respecting
maximal R-symmetries requires a special analysis.
In the examples throughout the paper we constructed N = 8 supersymmetric scalar potentials
by adding, to the sigma-model type superfield actions, the linear FayetIliopoulos-type terms
which at the component level amount to terms linear in the auxiliary fields. On the other hand,
some N = 4 multiplets (e.g., (3, 4, 1) and (4, 4, 0)) admit more general superfield potential terms
yielding in components, apart from the scalar potentials, also some minimal couplings to the
background target gauge fields [8,42,43]. It is interesting to inquire whether some of such N = 4
generalized potential terms can be completed to N = 8 invariants.
The hierarchical relationships between different N = 8 mechanics models with 8 physical
fermions were established here at the off-shell component level. It would be desirable to see
this hierarchy structure at the manifestly supersymmetric superfield level, at least in the N = 4
superfield approach. For the case of N = 4 mechanics model it was shown in [43,44] that the
corresponding hierarchical structure is directly related to gauging, by non-propagating gauge
N = 4, d = 1 multiplets, triholomorphic (commuting with supersymmetry) isometries of the
parent N = 4 mechanics model associated with the N = 4 root multiplet (4, 8, 4). This
superfield framework made it possible to reveal some new possibilities and features which were
not seen in the component approach. It would be desirable to extend this gauging procedure to the
N = 8 mechanics models as well. Though all homogeneous compact triholomorphic isometries
used in [43,44] cease to be triholomorphic with respect to the full N = 8 supersymmetry (they
do not commute with the hidden N = 4 supersymmetry and become a part of the relevant full Rsymmetries), there still remain some non-compact isometries, the shift and rescaling ones, which
commute with both N = 4 supersymmetries, manifest and hidden. So these isometries can be
appropriate candidates for gauging. In the N = 4 case, the gauging is most naturally performed
521
in the N = 4, d = 1 harmonic superspace [42]. Hence, as a prerequisite, it would be helpful to

reformulate the models studied here within the N = 4 harmonic superspace framework.
The N = 8, d = 1 off-shell multiplets classified and studied in [10,20] and in the present paper are defined by linear constraints. On the other hand, in the N = 4 case most of analogous
off-shell linear multiplets have their nonlinear cousins [9,3840,42,43]. This fact raises the question whether similar nonlinear counterparts exist for the linear N = 8 multiplets and which new
properties (including the geometric ones) the relevant N = 8 mechanics models could have. In
particular, it seems that the only possibility to obtain models with more general target geometries
(not reducing to the conformally flat ones) is to involve nonlinear multiplets. Until now not too
many nonlinear N = 8 multiplets are known: in Ref. [34] a nonlinear version of the multiplet
(2, 8, 6) was discussed and in Refs. [36,37] a nonlinear version of the multiplet (4, 8, 4) was
described. It seems important to have the full list of nonlinear N = 8 multiplets and the N = 8
mechanics models associated with them, like the one existing now for linear N = 8 multiplets.
The N = 4 superfield approach could be appropriate for this purpose. One could try to pair into
irreducible N = 8 multiplets nonlinear N = 4 multiplets or, say, nonlinear multiplet with a linear one, and to study general conditions under which the corresponding actions with manifest
N = 4 supersymmetry possess also the hidden N = 4 supersymmetry, as we did here for the
linear multiplets.
At last, it is imperative to work out the appropriate manifestly N = 8 supersymmetric superfield approach which would allow one to construct and study N = 8 mechanics models without
imposing any constraints on the relevant superfield actions. One such approach exists, this is the
N = 8 bi-harmonic superspace approach [31], but it perfectly works only for the linear and nonlinear (4, 8, 4) multiplets and seems to be not too helpful in the other cases. Another approach
is just the d = 1 reduction of the standard harmonic superspace [28], also with a limited scope
of applications to N = 8, d = 1 supermultiplets. It is worth to seek for some alternative more
universal N = 8, d = 1 superfield approaches.
Acknowledgements
The work of E.I. and A.S. was supported in part by the RFBR grant 06-02-16684, the RFBRDFG grant 06-02-04012-a, the grant DFG, project 436 RUS 113/669/0-3, the grant INTAS
05-7928 and a grant of HeisenbergLandau program. They thank Institute of Theoretical Physics
of the Leibniz University of Hannover for the kind hospitality during the course of this study.
E.I. thanks Laboratoire de Physique, UMR5672 of CNRS and ENS Lyon, for the warm hospitality at the final stage of this work.
Appendix A
We quote some useful identities for the N = 4 superfields describing two different (3, 4, 1)
multiplets:

1 ni lk
l n ik
ni
lk
nk
li
nk li
Da Db v = iab t v + t v + F ,
2
3i
D ia kb = 3i ab t v ik ik ab F,
2
2
ia
ia
D F = t ,
3
522

1 da cb
da
cb
db
ca
db ca
= iik t w + t w + H ,
2
3i
D ia kb = 3i ik t w ab ik ab H,
2
2
D ia H = t ia .
3
Dic Dkd w ab
References
[1] R. de Lima Rodrigues, The quantum mechanics SUSY algebra: An introductory review, hep-th/0205017.
[2] R. Britto-Pacumio, J. Michelson, A. Strominger, A. Volovich, Lectures on superconformal quantum mechanics and
multi-black hole moduli spaces, hep-th/9911066.
[3] A.V. Smilga, Low dimensional sisters of SeibergWitten effective theory, in: M. Shifman, et al. (Eds.), From Fields
to Strings, vol. 1, World Scientific, 2004, pp. 523558, hep-th/0403294.
[4] D. Bak, K. Lee, P. Yi, Phys. Rev. D 62 (2000) 025009, hep-th/9912083.
[5] V. Akulov, A. Pashnev, Teor. Mat. Fiz. 56 (1983) 344.
[6] S. Fubini, E. Rabinovici, Nucl. Phys. B 245 (1984) 17.
[7] E. Ivanov, S. Krivonos, V. Leviant, J. Phys. A: Math. Gen. 22 (1989) 345.
[8] E. Ivanov, S. Krivonos, O. Lechtenfeld, JHEP 0303 (2003) 014, hep-th/0212303.
[9] E. Ivanov, S. Krivonos, O. Lechtenfeld, Class. Quantum Grav. 21 (2004) 1031, hep-th/0310299.
[10] S. Bellucci, E. Ivanov, S. Krivonos, O. Lechtenfeld, Nucl. Phys. B 684 (2004) 321, hep-th/0312322.
[11] D.Z. Freedman, P.F. Mende, Nucl. Phys. B 344 (1990) 317.
[12] G.W. Gibbons, P.K. Townsend, Phys. Lett. B 454 (1999) 187, hep-th/9812034.
[13] S. Bellucci, A. Galajinsky, S. Krivonos, Phys. Rev. D 68 (2003) 064010, hep-th/0304087.
[14] A. Galajinsky, O. Lechtenfeld, K. Polovnikov, Phys. Lett. B 643 (2006) 221, hep-th/0607215.
[15] S.J. Gates Jr., L. Rana, On extended supersymmetric quantum mechanics, Maryland Univ. Preprint UMDPP 93-24,
October 1994.
[16] S.J. Gates Jr., L. Rana, Phys. Lett. B 342 (1995) 132, hep-th/9410150.
[17] A. Pashnev, F. Toppan, J. Math. Phys. 42 (2001) 5257, hep-th/0010135.
[18] R.A. Coles, G. Papadopoulos, Class. Quantum Grav. 7 (1990) 427;
G. Papadopoulos, Class. Quantum Grav 7 (2000) 3715, hep-th/0002007;
C.M. Hull, The geometry of supersymmetric quantum mechanics, hep-th/9911001.
[19] G.W. Gibbons, G. Papadopoulos, K.S. Stelle, Nucl. Phys. B 508 (1997) 623, hep-th/9706207.
[20] S. Bellucci, E. Ivanov, S. Krivonos, O. Lechtenfeld, Nucl. Phys. B 699 (2004) 226, hep-th/0406015.
[21] D.-E. Diaconescu, R. Entin, Phys. Rev. D 56 (1997) 80458052, hep-th/9706059.
[22] A.V. Smilga, Nucl. Phys. B 652 (2003) 93, hep-th/0209187.
[23] B.M. Zupnik, Nucl. Phys. B 554 (1999) 365, hep-th/9902038.
[24] E.A. Ivanov, A.V. Smilga, Nucl. Phys. B 694 (2004) 473, hep-th/0402041.
[25] S. Bellucci, S. Krivonos, A. Nersessian, Phys. Lett. B 605 (2005) 181, hep-th/0410029.
[26] S. Bellucci, S. Krivonos, A. Nersessian, A. Shcherbakov, 2k-dimensional N = 8 supersymmetric quantum mechanics, hep-th/0410073.
[27] S. Bellucci, S. Krivonos, A. Shcherbakov, Phys. Lett. B 612 (2005) 283, hep-th/0502245.
[28] A. Galperin, E. Ivanov, V. Ogievetsky, E. Sokatchev, Pisma Zh. Eksp. Teor. Fiz. 40 (1984) 155, JETP Lett. 40
(1984) 912;
A.S. Galperin, E.A. Ivanov, S. Kalitzin, V.I. Ogievetsky, E.S. Sokatchev, Class. Quantum Grav. 1 (1984) 469.
[29] A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E.S. Sokatchev, Harmonic Superspace, Cambridge Univ. Press, Cambridge, 2001, 306p.
[30] S. Bellucci, S. Krivonos, A. Sutulin, Phys. Lett. B 605 (2005) 406, hep-th/0410276.
[31] S. Bellucci, E. Ivanov, A. Sutulin, Nucl. Phys. B 722 (2005) 297, hep-th/0504185.
[32] M. Faux, S.J. Gates Jr., Phys. Rev. D 71 (2005) 065002, hep-th/0408004.
[33] S. Bellucci, S. Krivonos, A. Marrani, E. Orazi, Phys. Rev. D 73 (2006) 025011, hep-th/0511249.
[34] S. Bellucci, A. Beylin, S. Krivonos, A. Shcherbakov, Phys. Lett. B 633 (2006) 382, hep-th/0511054.
[35] S. Bellucci, S. Krivonos, A. Shcherbakov, Phys. Rev. D 73 (2006) 085014, hep-th/0604056.
[36] S. Bellucci, S. Krivonos, A. Marrani, Phys. Rev. D 74 (2006) 045005, hep-th/0605165.
[37] E. Ivanov, Phys. Lett. B 639 (2006) 579, hep-th/0605194.
[38]
[39]
[40]
[41]
523
C. Burdik, S. Krivonos, A. Shcherbakov, Czech. J. Phys. 55 (2005) 1357, hep-th/0508165.

S. Krivonos, A. Shcherbakov, Phys. Lett. B 637 (2006) 119, hep-th/0602113.
C. Burdik, S. Krivonos, A. Shcherbakov, Hyper-Khler geometries and nonlinear supermultiplets, hep-th/0610009.
Z. Kuznetsova, M. Rojas, F. Toppan, JHEP 0603 (2006) 098, hep-th/0511274;
F. Toppan, Irreps and off-shell invariant actions of the N -extended supersymmetric quantum mechanics, in: Fifth
International Conference on Mathematical Methods in Physics IC2006, 2428 April 2006, CBPF, Rio de Janeiro,
Brazil, hep-th/0610180.
[42] E. Ivanov, O. Lechtenfeld, JHEP 0309 (2003) 073, hep-th/0307111.
[43] F. Delduc, E. Ivanov, Nucl. Phys. B 753 (2006) 211, hep-th/0605211.
[44] F. Delduc, E. Ivanov, Nucl. Phys. B 770 (2007) 179, hep-th/0611247.
Functional relations from the YangBaxter algebra:

Eigenvalues of the XXZ model with non-diagonal
twisted and open boundary conditions
W. Galleas
Universidade Federal de So Carlos, Departamento de Fsica, C.P. 676, 13565-905, So Carlos-SP, Brazil
Received 6 August 2007; received in revised form 31 August 2007; accepted 5 September 2007
Abstract
In this work we consider a functional method in the theory of exactly solvable models based on the
YangBaxter algebra. Using this method we derive the eigenvalues of the XXZ model with non-diagonal
twisted and open boundary conditions for general values of the anisotropy and boundary parameters.
2007 Published by Elsevier B.V.
PACS: 05.50.+q; 02.30.Ik
Keywords: Functional equations; Lattice models; Open boundary conditions
1. Introduction
Functional equations methods appeared in the theory of exactly solvable lattice models intimately connected with Baxters commuting transfer matrix method [1]. In the early seventies
Baxter introduced in his pioneer work [1] the concept of Q-operators and T Q relations determining the eigenvalues of the transfer matrix of the corresponding vertex model.
The complex calculations involved in Baxters construction of Q-operators seems to have
restricted its use and other functional methods, such as the Reshetikhins analytical Bethe ansatz,
were employed instead to obtain the spectrum of transfer matrices related with quantum Kac
E-mail address: wgalleas@df.ufscar.br.

0550-3213/$ see front matter 2007 Published by Elsevier B.V.
doi:10.1016/j.nuclphysb.2007.09.011
W. Galleas / Nuclear Physics B 790 [PM] (2008) 524542
525
Moody algebras [2]. However we remark here the recent progresses in the construction of Qoperators by employing quantum algebras representation theoretic methods [3,4].
On the other hand, the advent of the algebraic Bethe ansatz in the late seventies provided a
systematic approach to find the eigenvalues and eigenvectors of transfer matrices of integrable
vertex models. This method, originally proposed by Takhtadzhan and Faddeev [5], is based on
the existence of a pseudovacuum or reference state and appropriate commutation rules arising
from the YangBaxter algebra, which is a common algebraic structure associated with integrable
vertex models.
Although the algebraic Bethe ansatz method is a powerful tool exhibiting a rich mathematical
structure, its implementation for models that do not possess a trivial reference state is still an
obstacle to be overcomed. The aim of this paper is to show that the YangBaxter algebra can
be explored in order to generate functional relations determining the spectrum of transfer matrices. In order to illustrate that we consider the XXZ model with non-diagonal twists and open
boundaries for general values of the anisotropy and boundary parameters. Besides the relevance
of studying models with general boundary conditions in the context of statistical mechanics [6],
the XXZ model with non-diagonal twists and open boundaries are both included in a class of
models where a trivial reference state is absent.
This paper is organized as follows. In Section 2 we describe the XXZ model with general
toroidal boundary conditions. In particular, we discuss the case of non-diagonal twists and we derive functional relations for its eigenvalues making use of the YangBaxter algebra. In Section 3
we approach the eigenvalue problem for the XXZ model with non-diagonal open boundaries
using the algebraic-functional method devised in Section 2. Concluding remarks are discussed
in Section 4 and in Appendices AD we give some extra results and technical details.
2. The XXZ model with non-diagonal twisted boundary conditions
The advent of the quantum inverse scattering method [5,7] was an important stage in the
development of the theory of exactly solvable quantum systems. This method unveiled a deep
connection between solutions of the YangBaxter equation, quantum integrable systems and
exactly solvable lattice models of statistical mechanics in two dimensions [8]. In statistical
mechanics an important role is played by vertex models whose respective transfer matrix is
constructed from local Boltzmann weights contained in an operator LAj . Let V be a finite dimensional linear space, the integrability of the vertex model is achieved when the operator valued
function L : C End(V V ) is a solution of the YangBaxter equation, namely
L12 ( )L13 ()L23 () = L23 ()L13 ()L12 ( ),
(1)
defined in the space V1 V2 V3 . Here we use the standard notation Lij End(Vi Vj ). The
complex valued operator LAj () can be viewed as a matrix in the space of states A denoting
for instance the horizontal degrees of freedom of a
square lattice, while its matrix elements are
operators acting non-trivially in the j th position of L
i=1 Vi . In its turn the space Vj represents
the space of states of the vertical degrees of freedom at the j th site of a chain of lengh L.
The transfer matrix of the corresponding vertex model can be written in terms of the monodromy matrix TA () defined by the following ordered product
TA () = LAL ()LAL1 () LA1 ().
(2)
526
As a consequence of the YangBaxter equation, the monodromy matrix satisfies the following
quadratic relation
R( )TA () TA () = TA () TA ()R( ),
(3)
usually denominated YangBaxter algebra. The R-matrix appearing in (3) follows from the solution of the YangBaxter equation through the relation R() = P L() where P denotes the usual
permutation operator. The above R-matrix plays the role of structure constant for the Yang
Baxter algebra and it consist of an invertible complex valued matrix acting on the tensor product
A A.
The invariance of the YangBaxter algebra plays an important role in the description of integrable spin chains with general toroidal boundary conditions. One can easily verify the invariance
of (3) under the transformation TA () GA TA () provided that the c-number matrix GA is a
symmetry of the R-matrix, i.e.

(4)
R(), GA GA = 0.
Consequently we can define the operator

T () = TrA GA TA ()
(5)
which constitutes an one parameter family of commuting transfer matrices, i.e. [T (), T ()] = 0.
Now restricting ourselves to the XXZ model, Vi C2 and the corresponding L-operator is
that of the anisotropic six vertex model
a()
0
0
0
b() c()
0
0
L() =
(6)
,
0
c() b()
0
0
0
0
a()
whose Boltzmann weights are given by a() = sinh( + ), b() = sinh() and c() = sinh( ).
In Refs. [9,10] the authors discuss the possible classes of twist matrices GA compatible with
the R-matrix associated with (6). These twist matrices turn out to be

0
0
,
(ii) GA =
(i) GA =
(7)
0
0
where and are arbitrary complex parameters. The above matrices GA are non-singular for
, = 0 and the XXZ model with generalized toroidal boundary conditions is obtained as the
logarithmic derivative of the transfer matrix (5) at the point = 0 [9,10].
The understanding of physical properties of vertex models demands the exact diagonalization
of their respective transfer matrices, which can provide us information about the free energy
behaviour and the nature of the elementary excitations. When the twist matrix of type (i) is
considered, a trivial reference state is available and the eigenvalues of the corresponding transfer
matrix can be obtained through the algebraic Bethe ansatz [5,7]. By way of contrast, the twist
matrix of type (ii) breaks the U (1) symmetry of the system leaving only a Z2 invariance.
Although a trivial reference state in the framework of the algebraic Bethe ansatz is no longer
available when the matrix GA of type (ii) is considered, the eigenvalues of the corresponding
transfer matrix were obtained in [9] by means of the Baxters T Q method. It is worthwhile to
remark here that the isotropic case associated with the XXX spin chain admits any 2 2 twist
matrix and interesting enough the algebraic Bethe ansatz solution can be obtained by exploring
the GL(2) symmetry [11].
527
From the historical point of view, the T Q method was introduced in Baxters remarkable
works on the eight vertex model [1] and more recently it has found applications in many areas
such as the study of integrable systems [12], conformal field theory [3,13], correlations functions [14] and efficient description of finite temperature properties [15]. Motivated by the ideas
of Baxters T Q method [1,8] and the algebraic Bethe ansatz [5,7], in what follows we shall
demonstrate how we can use the YangBaxter algebra to obtain functional relations determining
the spectrum of the transfer matrix T ().
Let us consider the eigenvalue problem for the transfer matrix (5),
T ()| = ()|,
(8)
taking into account the type (ii) GA matrix given in (7). As discussed in the Ref. [10] and shown
in Appendix A through the use of similarity transformations, we can set = = 1 without loss
of generality for the purposes of this paper. Now considering the definition (2), the monodromy
matrix TA () consist of a 2 2 matrix whose elements are operators that we denote

A() B()
TA () =
(9)
.
C() D()
Therefore, the transfer matrix (5) reads
T () = B() + C().
(10)
In contrast to the case when GA is diagonal [16], the state |0 defined as
L

1
|0 =
0
(11)
j =1
is not an eigenstate of the transfer matrix (10). Nevertheless, the state |0 is still of great utility.
Considering the definition (2) together with (6), the elements of TA () satisfy the following
relations
A()|0 = a()L |0,
B()|0 = ,
D()|0 = b()L |0,

C()|0 = 0,
(12)
where the symbol stands for a non-null value. Notice the relations (12) imply in
T ()|0 = B()|0
(13)
T ()B()|0 = B()B()|0 + C()B()|0.
(14)
and
Now we have reached a point of fundamental importance. The term C()B()|0 present in
the right-hand side of (14) can be evaluated with the help of the YangBaxter algebra (3). In
order to show that we collect the following commutation rule among the ones encoded in the
relation (3),
C()B() = B()C() +

c( )
A()D() A()D() .
b( )
(15)
The commutation rule (15), together with the relations (12), results in the following identity
C()B()|0 =

c( )
a()L b()L a()L b()L |0.
b( )
(16)
528
Therefore, as we are interested in C()B()|0, the next step is to consider the limit in
the above relation which can be evaluated using LHopitals rule. Thus we are left with
C()B()|0 = M()|0
(17)
where M() = Lc()2 a()L1 b()L1 .

At this stage we have already gathered the basic ingredients to obtain functional relations
determining the eigenvalues of T (). Operating with the dual eigenvector | on the left side of
Eqs. (13) and (14) we are left with the relations
()F0 = F1 (),
(18)
()F1 () = F2 () + M()F0 ,
(19)
where F0 = |0, F1 () = |B()|0 and F2 () = |B()B()|0.

Here we remark one of the roles played by integrability in this approach. Since the transfer
matrix T () belongs to a commuting family, the eigenvectors | are independent of the spectral
parameter . Thus the dependence of the functions Fi () with is solely determined by the
operator B(). In Appendix C we have computed that dependence making use of Eqs. (2), (6)
and (9).
From Eq. (18) we can see that the functions () and F1 () differ only by the constant factor
F0 , thus they possess the same zeroes. Now we can eliminate () from Eq. (18) and substitute
the result into Eq. (19) which yields the following relation involving only the functions Fi (),
F1 ()2 = F2 ()F0 + M()F02 .
(20)
Considering Eqs. (2), (6) and (9), in Appendix C we have demonstrated that the functions
F1 () and F2 () can be written as
F1 () = F1 (0)
L1

sinh(i )
i=1
sinh((1)
i )
(1)
and F2 () = F2 (0)
2(L1)

sinh(i )
i=1
sinh((2)
i )
(2)
(21)
(i)
where j denote the zeroes of the function Fi (). Now a closer look in Eqs. (18) and (19)
reveals that we can obtain the eigenvalue () by determining the variables (i)
j together with
the ratios
Fi (0)
F0 .
(i)
In order to obtain analogues of Bethe ansatz equations determining the variables j , we first
(1)
consider Eq. (20) at the points = j . Then we find

(1)
(1)
F2 j + M j F0 = 0.
(22)
(2)
Next we consider the points = j in Eq. (20) which yields

2
2
M (2)
F1 (2)
j
j F0 = 0.
(23)
can be
Furthermore, we observe by setting = 0 in Eqs. (18) and (19) that the ratios FFi (0)
0
written in terms of (0) which can be easily evaluated. Therefore we are left with the following
,
expression for the ratios FFi (0)
0
F1 (0)
= (0),
F0
F2 (0)
= (0)2 .
F0
(24)
529
In order to avoid an overcrowded section we have performed the diagonalization of T (0) in

Appendix B. The eigenvalue (0) turns out to be
(0) = sinh( )L e
ir
L
r = 0, . . . , 2L 1.
(25)
Gathering our results so far, from Eqs. (18), (22), (23), (24) we obtain the following expression
for the transfer matrix eigenvalues (),
() = sinh( )L e
ir
L
L1

sinh(i )
i=1
sinh(i )
(1)
(26)
(1)
(i)
recalling that r = 0, . . . , 2L 1 and provided that the variables j satisfy the following system
of non-linear algebraic equations

(1)
sinh((1)
j + ) sinh(j )
sinh( )

L1
=
sinh( )
(2)
(2)
sinh(j + ) sinh(j )
sinh( )
L1
=
sinh( )
2ir
L
(1)
sinh(i j )
i=1
sinh(i )
(2)
2ir
L
(2)
2(L1)

L1
sinh((1)
i
(2)
j )
(1)
i=1
sinh(i )
j = 1, . . . , L 1,
(27)
2
,
j = 1, . . . , 2(L 1).
(28)
Now we shall examine some aspects of Eqs. (27) and (28) taking into account the crossing
properties discussed in Appendix D. Considering Eq. (D.9), it follows that
t () = (1)L+1 ( )
(29)
where t () denotes the eigenvalue of the transposed transfer matrix T t (). Although the relation (29) does not imply in () = (1)L+1 ( ), we have verified the existence of
eigenvalues satisfying that relation through the direct diagonalization of T () for small chain
lengh L.
Let us suppose the relation () = (1)L+1 ( ) holds for some eigenvalue. Hence,
from Eq. (18) we can conclude that the function F1 () also satisfy
F1 () = (1)L+1 F1 ( ).
(30)
Since the function M() enjoys the property M() = M( ), from Eq. (19) we also find
that
F2 () = F2 ( ).
(31)
Assuming that the relations (30) and (31) hold they have remarkable implications concerning
the solutions of Eqs. (27) and (28). Let us analyze first the odd L case. Thus L 1 is an even
(1)
number and Eq. (30) implies, for instance, in the following relation for the variables j ,
(1)
i+ L1
2
(1)
= i ,
Thus altogether we have only
i = 1, . . . ,
L1
2
L1
.
2
(32)
(1)
independent variables j .
530
Now we turn our attention to the L even case. In that case Eq. (30) implies for instance in the
relation
(1)
i+ L2
2
(1)
= i ,
i = 1, . . . ,
L2
,
2
(33)
while the remaining root (1)

L1 is fixed at the value 2 .
(2)
The above analysis can also be carried out for the roots j
Since 2(L 1) is an even number, the variables
(2)
(2)
i+L1 = i ,
(2)
j
taking into account Eq. (31).
must be related for instance by
i = 1, . . . , L 1.
(34)
These implications of the crossing symmetry have been verified by a numerical analysis of
Eqs. (27) and (28) for small values of L, corresponding to a set of four eigenvalues for L > 2
which contains the doubly degenerated ground state energy of the XXZ model with non-diagonal
twists [9]. By comparing the eigenvalues given by Eq. (26) with the direct diagonalization of
T (), we have also verified that Eqs. (26), (27), (28) indeed describes a complete spectrum.
We close this section by remarking the dependence with L1 in the right-hand side of Eqs. (27)
and (28). As far as we know, this kind of dependence with the chain lengh had not appeared
previously in the context of Bethe ansatz, which may be of relevance for the description of the
thermodynamical limit L .
3. The XXZ model with non-diagonal open boundaries
In Sklyanins pioneer work [17], the author has been able to generalize the quantum inverse
scattering method to accommodate integrable spin chains with open boundaries. The Yang
Baxter equation is still the corner stone of Sklyanins approach and it turns out that the XXZ
model with general open boundary conditions can be obtained from the following double-row
transfer matrix
+
(u)TA (u)KA
(u)TA (u) ,
t (u) = TrA KA
(35)
where TA (u) = LAL (u)LAL1 (u) LA1 (u) and TA (u) = LA1 (u)LA2 (u) LAL (u) are the
standard monodromy matrices that generate the corresponding closed spin chain with L sites.
+
(u) and KA
(u), each one
The integrability at the boundaries is governed by the matrices KA
describing the reflection at one of the ends of an open chain.
Moreover, the boundary conditions compatible with the bulk integrability are constrained by
the so-called reflection equations, which for KA

(u) reads
L21 (u v)K2 (u)L12 (u + v)K1 (v) = K1 (v)L21 (u + v)K2 (u)L12 (u v),
(36)
+
(u).
while a similar equation should also hold for the matrix KA
Turning our attention to the XXZ model, the bulk Hamiltonian is described by the L-operator
(6) and the most general boundary matrices satisfying the reflection equations are of the form

+
+
k (u) k12
k11 (u) k12
(u)
(u)
+
KA
(37)
,
K
,
(u) = 11
(u)
=
A
+
+
k21
k21
(u) k22
(u)
(u) k22
(u)
531
whose matrix elements, obtained previously in [18,19], are given by

(u) = sinh h
k12
(u) = h
k11
1 +u ,
2 sinh(2u),

k22 (u) = sinh h

k21 (u) = h3 sinh(2u),
1 u ,

+
+
k11
(u) = sinh h+
k12
(u) = h+
1 u ,
2 sinh(2u 2 ),

+
+
+
k22 (u) = sinh h+
k21 (u) = h3 sinh(2u 2 ),
1 +u+ .
(38)
The above K-matrices possess altogether six free boundary parameters {h

i } and we find the
XXZ model Hamiltonian with open boundaries,
H=
L1

z
x
ix i+1
+ i i+1 + cosh( )iz i+1
i=1
x

z
sinh( )
y
h2 + h
3 1 + i h2 h3 1 + cosh h1 1
sinh(h1 )
x
+
+ z
sinh( ) +
+ y
h2 + h+
3 L + i h2 h3 L + cosh h1 L ,
+
sinh(h1 )
+
related to the former by

+

2L1
t
(0) = 2 sinh h
cosh( )H + L cosh( )2 + sinh( )2 ,
1 sinh h1 sinh( )
ix ,
y
i
(39)
(40)
iz
where
and
denote the usual Pauli matrices acting on the ith site.
For general values of the boundary parameters h
i the double-row transfer matrix (35) does
not exhibit U (1) invariance, which makes the application of the algebraic Bethe ansatz quite
complex due to the lack of a trivial reference state. By imposing h

2 = h3 = 0 the matrices
KA
() (37) become diagonal and the U (1) symmetry exhibited by the bulk of the system is
restored. In that case the spectrum of (35) was obtained in [17] through the algebraic Bethe
ansatz, and the associated Hamiltonian (39) was diagonalized in [20] by means of the coordinate
Bethe ansatz. However, several progresses concerning non-diagonal open boundaries have been
reported in the literature over the past years when the boundary parameters or the bulk anisotropy
satisfy a given constraint [2134].
In what follows we shall explore the YangBaxter algebra to generate functional relations
determining the complete spectrum of the double-row transfer matrix (35) for general values of
the bulk anisotropy and boundary parameters h
i . In order to tackle the eigenvalue problem
t (u)| = (u)|
(41)
for the double-row transfer matrix (35), we first notice the L-operator (6) satisfies the property
L12 (u)L12 (u) = a(u)a(u) Id Id, where Id denotes the 2 2 identity matrix. Therefore, the
monodromy matrices TA (u) and TA (u) are related by TA (u) = [a(u)a(u)]L TA1 (u) and we
find the following relation,

TA (u) Id R(2u) TA (u) Id = Id TA (u) R(2u) Id TA (u) ,
(42)
with a straightforward manipulation of the YangBaxter algebra (3). Analogously to Eq. (9), we
denote the matrix elements of TA (u) by

A(u) B(u)
TA (u) =
(43)
.
C(u)
D(u)
532
In this way the double-row transfer matrix (35) reads

+
+
+
+
(u)U1 (u) + k22
(u)U2 (u) + k12
(u)U4 (u) + k21
(u)U3 (u)
t (u) = k11
(44)
where
U1 (u) = k11
(u)A(u)A(u)
+ k12
(u)A(u)C(u)
+ k21
(u)B(u)A(u)
+ k22
(u)B(u)C(u),
+ k (u)C(u)D(u)
+ k (u)D(u)B(u)
+ k (u)D(u)D(u),
U2 (u) = k (u)C(u)B(u)
11
12
21
22
11
12
21
22
U3 (u) = k11
(u)A(u)B(u)
+ k12
(u)A(u)D(u)
+ k21
(u)B(u)B(u)
+ k22
(u)B(u)D(u),
U4 (u) = k (u)C(u)A(u)
+ k (u)C(u)C(u)
+ k (u)D(u)A(u)
+ k (u)D(u)C(u).
(45)
Next we consider the action of t (u) on the state |0 defined in (11), which will require the
following relations
A(u)|0
= a(u)L |0,
B(u)|0
= ,
D(u)|0
= b(u)L |0,
C(u)|0
=0
(46)
obtained from the structure of the monodromy matrix TA (u), and the relations (12) from the
previous section.
Then, in order to evaluate t (u)|0, we shall make use of the commutation rules between the elements of TA (u) and TA (u) enclosed in the relation (42). In particular, we shall use the following
ones,
c(2u)
b(2u)
B(u)A(u)
B(u)D(u),
a(2u)
a(2u)
a(2u)
c(2u)
D(u)B(u)
=
B(u)D(u) +
A(u)B(u),
b(2u)
b(2u)
c(2u)
b(2u)
B(u)A(u)
B(u)D(u),
A(u)B(u)
=
a(2u)
a(2u)

c(2u)
C(u)B(u)
=
A(u)A(u) D(u)D(u)
+ B(u)C(u),
a(2u)
B(u)B(u)
= B(u)B(u).
A(u)B(u)
=
(47)
Considering Eqs. (44), (45), (12) and (46) together with the relations (47), a straightforward
calculation leaves us with
t (u)|0 = Y0 (u)|0 + Y1 (u)B(u)|0 + Y1 (u)B(u)|0

+ Y2 (u)B(u)B(u)|0,
where the functions Y0 (u), Y1 (u), Y1 (u) and Y2 (u) are given by

sinh(2u + 2 )
Y0 (u) = a 2L (u) sinh h+
1 u sinh h1 + u sinh(2u + )

sinh(2u)
b2L (u) sinh u + + h+
1 sinh u + h1 sinh(2u + )

+
+
a L (u)bL (u) h
2 h3 + h3 h2 sinh(2u) sinh(2u + 2 ),
+

sinh(2u + 2 )
Y1 (u) = a L (u)h
3 sinh h1 u sinh(2u) sinh(2u + )

sinh(2u + 2 )

+ bL (u)h+
3 sinh u + h1 sinh(2u) sinh(2u + ) ,
(48)
533

sinh(2u + 2 )
Y1 (u) = a L (u)h+
3 sinh h1 + u sinh(2u) sinh(2u + )

sinh(2u + 2 )
+
+ bL (u)h
3 sinh u + + h1 sinh(2u) sinh(2u + ) ,
Y2 (u) = h+
3 h3 sinh(2u) sinh(2u + 2 ).
(49)
We can now follow the scheme devised in the previous section and operating with the dual
eigenvector | on the left side of Eq. (48) we obtain
(u)f0 = Y0 (u)f0 + Y1 (u)f1 (u) + Y1 (u)f1 (u) + Y2 (u)f2 (u),
(50)
and f2 (u) = |B(u)B(u)|0.

where f0 = |0, f1 (u) = |B(u)|0, f1 (u) = |B(u)|0
Before we proceed it is important to examine first Eq. (50) under the light of the results
obtained in Appendix D. Indeed, the crossing symmetry enables us to establish the relation
) (D.8) which implies in

B(u) = (1)L+1 B(u
f1 (u) = (1)L+1 f1 (u ).
(51)
Furthermore, the commutation rule [B(u), B(u)]

= 0 given in (47) together with the relation
(D.8), imply in the crossing invariance of the function f2 (u), i.e.
f2 (u) = f2 (u ).
(52)
Among the results of Appendix D is also the crossing invariance of the double-row transfer
matrix t (u) = t (u ), which implies in
(u) = (u ).
(53)
In their turn the functions Yi (u) given in (49) fulfill the relations Y0 (u) = Y0 (u ), Y1 (u) =
(1)L+1 Y1 (u ) and Y2 (u) = Y2 (u ). Taking into account the above properties, one
can easily verify the crossing invariance of Eq. (50), which simplifies to the following functional
relation,
(u)f0 = Y0 (u)f0 + Y1 (u)f1 (u) + Y1 (u )f1 (u ) + Y2 (u)f2 (u).
(54)
Considering now the results from Appendix C, the function f1 (u) (C.13) can be written as
f1 (u) = f1 (0)
L1

sinh(ui u)
i=1
sinh(ui )
(1)
(1)
(55)
On the other hand, the functions f2 (u) and (u) given by (C.14) and (C.16) will receive extra
simplifications due to the relations (52) and (53). Thus they can be written as
f2 (u) = g2 sinh(u) sinh(u + )
L2

(u) = (0)
(0)
sinh(ui )
(2)
sinh(ui + )
sinh(ui )
(56)
(0)
sinh(ui u) sinh(ui + + u)
(0)
i=1
(2)
(2)
i=1
L+2

(2)
sinh(ui u) sinh(ui + + u)
(0)
sinh(ui + )
(57)
The initial condition (0) has been determined in Appendix B through a direct analysis of
t (0). We remark here that the eigenvalue (0) was discussed previously in [31]. The next step is
(i)
then the determination of the variables uj . In order to do that we consider Eq. (54) at the points
534

(0)
(1)
(2)
u = uj , u = uj and u = uj . By doing so we find the following system of algebraic equations,

(0) (0)
(0)
(0) (0)
(0)

Y0 u(0)
+ Y1 uj f1 uj + Y2 uj f2 uj = 0,
j f0 + Y1 uj f1 uj
(58)
(1)
(1)
(1)
(1) (1)
(1)

Y0 uj uj f0 + Y1 uj f1 uj + Y2 uj f2 uj = 0,
(59)
(2)

(2)
(2) (2)
(2)
(2)
Y0 uj uj f0 + Y1 uj f1 uj + Y1 uj f1 uj = 0.
(60)
A direct inspection of Eqs. (58)(60) shows that the ratios f1f(0)
and fg20 are also required. In
0
contrast to the case of twisted boundary conditions considered in the previous section, where
M(0) = 0, the point u = 0 of Eq. (54) is not very enlightening. However, in order to determine
the ratios f1f(0)
and fg20 we consider, for instance, the non-trivial points u = u1 and u = u2 such
0
that Y1 (u1 ) = Y1 (u2 ) = 0. Then we are left with the following equations

Y0 (ui ) (ui ) f0 + Y1 (ui )f1 (ui ) + Y2 (ui )f2 (ui ) = 0, i = 1, 2,
(61)
which can be solved for the required ratios.
Now our results can be gathered and we are left with the following expression for the eigenvalues (u),
L+2
sinh(u(0) u) sinh(u(0) + + u)

i
i
2L
sinh
h
sinh(
)
(u) = 2 cosh( ) sinh h+
,
1
1
(0)
(0)
sinh(u
)
sinh(u
i=1
i
i +)
(62)
(i)
provided that the variables u1 , u2 (we have performed the shifts ui ui 2 ) and uj satisfy
the following system of non-linear algebraic equations,

+
sinh(ui + 2 ) L h
3 sinh(ui 2 h1 )
(63)
= 1, i = 1, 2,
sinh(ui 2 ) h+
3 sinh(ui + 2 h1 )

Y0 u(0)
j

(0)
sinh(u(1)
(0) L1
i uj )
= Y1 uj
(1)
sinh(ui )
i=1
(0)
sinh(u(1)
(0)
L1
i + + uj ) det(H1 )
+ Y1 uj
(1)
det(H0 )
sinh(ui )
i=1
(0)
(0)
(0)

Y2 uj sinh uj sinh uj +
L2

(2)
(0)
(2)
i=1
(2)
(0)
sinh(ui uj ) sinh(ui + + uj ) det(H2 )

sinh(ui )
(2)
sinh(ui + )
det(H0 )

2L
2 cosh( ) sinh h+
1 sinh h1 sinh( )
L+2

(0)
(1)
(0)
(1)
sinh(ui uj ) sinh(ui + + uj )
(0)
(0)
sinh(ui )
sinh(ui + )

(1)
(1)

= +Y2 uj sinh uj sinh u(1)
j +
i=1
(1)
Y0 uj
j = 1, . . . , L + 2,
(64)
L2

(2)
(1)
(2)
(1)
sinh(ui uj ) sinh(ui + + uj ) det(H2 )

(2)
(2)
sinh(ui + )
sinh(ui )
i=1
(1)
535
det(H0 )
(1)
sinh(ui + + uj ) det(H1 )
(1)
L1
,
+ Y1 uj
(1)
det(H0 )
sinh(ui )
i=1
j = 1, . . . , L 1,
(65)

2L
2 cosh( ) sinh h+
1 sinh h1 sinh( )
(0)
L+2

(2)
(0)
(2)
sinh(ui uj ) sinh(ui + + uj )
(0)
(0)
sinh(ui + )
sinh(ui )
i=1
(1)

Y0 u(2)
j
(2)
sinh(ui uj )

L1
= Y1 u(2)
j
(1)
sinh(ui )
i=1

+ Y1 u(2)
j
(2)
sinh(u(1)
L1
i + + uj ) det(H1 )
,
(1)
det(H0 )
sinh(u )
i=1
j = 1, . . . , L 2,
(66)
where Hi are 2 2 matrices resulting from the solution of the system of equations (61) for the
and fg20 . These matrices turn out to be
ratios f1f(0)
0

1 (u1 ) 2 (u1 )
0 (u1 ) 2 (u1 )
H0 =
,
H1 =
,
1 (u2 ) 2 (u2 )
0 (u2 ) 2 (u2 )

1 (u1 ) 0 (u1 )
H2 =
(67)
1 (u2 ) 0 (u2 )
where

2L
0 (u) = 2 cosh( ) sinh h+
1 sinh h1 sinh( )
L+2

(0)
(0)
2 u) sinh(ui + 2 + u)
(0)
(0)
sinh(ui )
sinh(ui + )
L1
(1)
sinh(ui + 2 + u)
,
2
sinh(u(1)
i=1
i )
sinh(ui +
i=1
1 (u) = Y1 u
,
Y0 u
2
sinh u
sinh u +
2 (u) = Y2 u
2
2
2
L2

sinh(u(2)
i +
2 u)
(2)
sinh(ui )
i=1
sinh(u(2)
i +
(2)
sinh(ui
+ u)
+)
(68)
Taking into account Eq. (40) we are left with the following expression for the eigenenergies E
of the Hamiltonian (39),
E = sinh( )
L+2

i=1
sinh(ui ) sinh(ui + )
(0)
(0)
(0)
L cosh( )
sinh( )2
,
cosh( )
(69)
given in terms of the roots ui . We end this section by remarking that numerical checks performed for L = 2, 3, 4 show that the spectrum generated by Eqs. (62)(66) is complete.
536
4. Concluding remarks
In this paper we have proposed a functional method in the theory of exactly solvable models
based on the YangBaxter algebra. Using this method we were able to derive the eigenvalues of
the XXZ model with non-diagonal twists and open boundaries for general values of the bulk
anisotropy and boundary parameters. Our solution is presented in terms of analogues of Bethe
ansatz equations whose variables involved are precisely the roots of the transfer matrix eigenvalues and roots of auxiliary functions defined in terms of the monodromy matrix elements.
Concerning the XXZ model with non-diagonal twists discussed in Section 2, we stress the
unusual dependence of Eqs. (27) and (28) with the chain lengh L. We hope the computation of
physical properties in the thermodynamical limit L , such as the interfacial tension obtained
in [9], will be benefited by the use of Eqs. (27), (28). However, we remark here that the behaviour
(2)
of the roots (1)
j and j , and the analysis of Eqs. (27), (28) in the thermodynamical limit have
eluded us so far, which we plan to investigate in a future work. As shown in [37], we also remark
that our solution corresponds to the eigenvalues of Baxters eight vertex model with elliptic
modulus = 1 and a particular choice of diagonal twist matrix.
Regarding the case of non-diagonal open boundaries described in Section 3, we notice that a
similar expression for the eigenvalues (62) was presented in [38] when h
2 = h3 and q = e is a
root of unity. We also remark the solution recently presented for general values of the anisotropy
and boundary parameters obtained from the representation theory of q-Onsager algebra [39]
and the multiple reference state structure found in [35,36]. An interesting problem would be
unveiling the connection between these approaches. Since our method is based on the Yang
Baxter algebra, which is a common algebraic structure underlying integrable vertex models, we
expect that our approach can be applied for other models with general open boundary conditions
based on q-deformed Lie algebras [40].
Finally, we observe that our solutions involve more than one kind of variable resembling the
so-called nested Bethe ansatz equations, which are typical of models with higher rank symmetry.
We remark that a similar result had been reported previously in the literature obtained from
generalized T Q relations [30].
Acknowledgements
The author thanks M.J. Martins for having introduced him to these problems and for several
discussions. The author also thanks FAPESP for financial support.
Appendix A. The condition = = 1
Let T () denote the transfer matrix

A LAL ()LAL1 () LA1 ()
T () = TrA G
(A.1)
A is the following twist matrix

where LAj () is the L-operator given in (6) and G
A =
G
.
0
(A.2)
Now considering the matrix M =

GA is given by

0 1
GA =
,
1 0
one can verify that M1

A GA MA =
537
GA where
(A.3)
A with = = 1. By noticing that M1 M1 LAj ()MA Mj =

which corresponds to G
j
A
LAj (), we then find the relation
L

M1
j T ()
j =1
L

Mj =

T (),
(A.4)
i=j
where T () is the transfer matrix (10) considered in Section 2. Thus the eigenvalues ()
of the
transfer matrix T () are related to the ones of T () by

()
= ().
(A.5)
Appendix B. The eigenvalues (0) and (0)
In this appendix we consider the diagonalization of the transfer matrices T (0) and t (0) associated with the XXZ model with twisted and open boundary conditions respectively.
B.1. Twisted boundary conditions
The L-operator (6) consist of a regular solution of the YangBaxter equation, i.e. L(0) =
sinh( )P , where P denotes the permutation operator. In this way the transfer matrix (10) at the
point = 0 can be written as T (0) = sinh( )L O where the operator O is given by
O = TrA [GA PAL PAL1 PA1 ].
Using the permutator algebra,
following expression for O,
P2
(B.1)
= Id Id and PAj PAi = PAi Pij , we can readily obtain the
O = G1 P1L P13 P12 ,
(B.2)
and its n times product

O n = G1 G2 Gn
n

Pj L Pj L1 Pj n+1 .
(B.3)
j =1
From Eq. (B.3) we have

O L = G1 G2 GL ,
(B.4)

which implies, with G = 01 01 , in O 2L = I where I is the identity matrix. Thus the eigenvalues O of the operator O are given by
O =e
ir
L
r = 0, 1, . . . , 2L 1,
(B.5)
leaving us with the following expression for (0),

(0) = sinh( )L e
ir
L
r = 0, 1, . . . , 2L 1.
(B.6)
538
B.2. Open boundary conditions
(u) presented in Eqs. (37) and (38) consist of a regular solution of the reflection
The matrix KA
equation, i.e. KA (0) = sinh(h

1 ) Id. Together with the regularity property of the L-operator (6)
and the permutator property P 2 = Id Id, the double-row transfer matrix (35) at the point u = 0
is given by

+
t (0) = sinh( )2L sinh h
(B.7)
1 TrA KA (0) I.
Therefore, we can see that the transfer matrix t (0) is proportional to the identity matrix I with
the following eigenvalue

2L
(0) = 2 cosh( ) sinh h+
(B.8)
1 sinh h1 sinh( ) .
Appendix C. The functions Fi () and fi ()
In this appendix we derive the form of the functions Fi () and fi () used in Sections 2 and 3,
considering the algebraic properties of the monodromy matrices TA () and TA ().
In order to determine the dependence of the functions Fi () with the spectral parameter , let
us first introduce the following notation for the monodromy matrix (2)

AL () BL ()
(L)
TA () =
(C.1)
,
CL () DL ()
differing from the one used in (9) by the index L inserted to emphasize we are considering the
ordered product of L matrices LAj (). For instance, the matrices LAj () are given by

j () j ()
LAj () =
(C.2)
j () j ()
where

a()
0
0
0
j () =
,
j () =
,
0
b() j
c() 0 j

0 c()
b()
0
,
j () =
(C.3)
j () =
0
0 j
0
a() j
are matrices acting non-trivially in the j th site of the chain.
(L)
Moreover, the monodromy matrix TA () defined by Eq. (2) satisfies the recurrence relation
(L+1)
TA
(L)
() = LAL+1 ()TA (),
(C.4)
corresponding to the following relations for its matrix elements

AL+1 () = L+1 ()AL () + L+1 ()CL (),
BL+1 () = L+1 ()BL () + L+1 ()DL (),
CL+1 () = L+1 ()AL () + L+1 ()CL (),
DL+1 () = L+1 ()BL () + L+1 ()DL ().
(C.5)
At this stage, it is important to keep in mind that the spectral parameter enters in the
monodromy matrix only through the Boltzmann weights a() and b(), since c() is independent (6). For the forthcoming discussion, it is convenient to introduce the weight w() only
539
to denote the Boltzmann weights possessing dependence with . Thus, w() can represent both
a() and b(). The reason for the introduction of w() will become clear in what follows. From
Eqs. (2), (C.2) and (C.3) it is clear that B1 () and C1 () are polynomials of degree 0 in w,
while A1 () and D1 () are polynomials of degree 1 in w. These initial conditions allow us to
use Eqs. (C.5) to determine the degree of the elements of TA(L) () in w for arbitrary lengh L.
Thus we obtain that BL () and CL () are polynomials of degree L 1 in w, while the operators
AL () and DL () are of degree L.
In terms of the variables x = e and q = e the Boltzmann weights a() and b() can be
written as a() = 12 (xq x 1 q 1 ) and b() = 12 (x x 1 ). Thus w() is of the form w()
x + x 1 , and expanding its powers we find
B() x L1 + + x (L1)

L1
L

= x (L1) cL1 x 2
+ cL x 2 + + c 0 .
(C.6)
Now considering the functions F1 () = |B()|0 and F2 () = |B()B()|0 used in Section 2, and the fact that | is independent, we can conclude that
F1 () = x (L1) P1 (x),
F2 () = x 2(L1) P2 (x),
(C.7)
where P1 (x) and P2 (x) are polynomials in the variable x 2 of degrees L 1 and 2(L 1),
respectively. Therefore, the polynomials P1 (x) and P2 (x) can be written as
P1 (x) = p1
L1

(1) 2
x 2 xi
,
i=1
P2 (x) = p2
2(L1)

(2) 2
x 2 xi
(C.8)
i=1
(j )
where pj are constants and xi

(j )
xi
=e
(j )
denote the zeroes of the polynomial Pj (x). Using the variables
we then have
F1 () = F1 (0)
F2 () = F2 (0)
L1

sinh(i )
i=1
sinh(i )
2(L1)

sinh(i )
i=1
sinh(i )
(1)
(1)
(2)
(2)
(C.9)
where the constant pj is absorbed by Fj (0).

Let us now consider the monodromy matrix TA(L) () = LA1 ()LA2 () LAL (), with matrix elements

AL () B L ()
(L)
TA () =
(C.10)
,
C L () D L ()
appearing in the construction of the double-row transfer matrix (35). Compared to the notation
previously used (43), we have also included the index L stressing we are considering the ordered
540
product of L matrices LAj (). The monodromy matrix TA () obeys the recurrence formula
(L)
(L+1)
(L)
TA
() = TA ()LAL+1 (),
(C.11)
corresponding to the following recursion relations for the matrix elements

A L+1 () = A L ()L+1 () + B L ()L+1 (),
B L+1 () = A L ()L+1 () + B L ()L+1 (),
C L+1 () = C L ()L+1 () + D L ()L+1 (),
D L+1 () = C L ()L+1 () + D L ()L+1 ().
(C.12)
(L)
The same arguments previously used for the elements of TA () can now be considered for
(L)
the elements of TA (). Thus we obtain that the operators B L () and C L () are polynomials
of degree L 1 in w while A L () and D L () are of degree L. Hence the function f1 (u) =
|B(u)|0 required in Section 3 can be written as
f1 (u) = f1 (0)
L1

sinh(ui u)
i=1
sinh(ui )
(1)
(1)
(C.13)
requires an extra analysis. Noticing that

On the other hand, the function f2 (u) = |B(u)B(u)|0
1
1
j (0) = c(0) j (0)j (0) and j (0) = c(0) j (0)j (0) together with j2 () = j2 () = 0, one can
easily see from Eqs. (C.5) and (C.12) that B(0)B(0)

= 0. This same analysis can be performed
for the point = yielding B( )B( ) = 0. Thus the function f2 (u) can be written as
f2 (u) = g2 sinh(u) sinh(u + )
2(L2)

sinh(ui u)
i=1
sinh(ui )
(2)
(2)
(C.14)
In Section 3 we also require the function (u) defined as

(u) =
|t (u)|
.
|
(C.15)
The above results together with Eqs. (44), (45), and (38) show that the double-row transfer matrix
consist of a polynomial of degree 2(L + 2) in w. Thus the eigenvalues (u) are of the form
(u) = (0)
(0)
where ui
2(L+2)

sinh(ui u)
i=1
sinh(ui )
(0)
(0)
(C.16)
denote the zeroes of (u).
Appendix D. Crossing symmetry

In this appendix we deduce some important properties used in Sections 2 and 3 which are
consequence of the crossing symmetry.
Besides the YangBaxter equation, the L-operator (6) satisfies other important relations,
namely
541
1 t2
() = L12 (),
Temporal invariance: Lt12
2
Crossing symmetry: L12 () = V1 Lt12
( )V11
(D.1)

0 1
where V = 1
and ti stands for the transposition in the space with index i.
0
In order to establish a relation between the matrices TA () and TA (), we first recall the
definitions
TA () = LAL ()LAL1 () LA1 (),

TA () = LA1 () LAL1 ()LAL (),
1
Lt12
()
(D.2)
(D.3)
2
Lt12
()
and observe that

=
which follows from the temporal invariance (D.1). Next we
consider the transposition in the auxiliary space A,
TAtA () = LtAA1 () LtAAL ()
= LtA1 1 () LtAL L (),
(D.4)
which allows us to use the crossing symmetry (D.1) to obtain

TAtA () = (1)L VA1 LA1 ( )VA VA1 LAL ( )VA
= (1)L VA1 TA ( )VA .
(D.5)
Now we also consider the transposition t = t1 tL . Then we find

t t t
TAtA t () = LAL () LA1 () A 1 L
= LtAA1t1 () LtAALtL ()
= LA1 () LAL (),
(D.6)
which leads immediately to the relation

TAtA t () = TA ().
(D.7)
In terms of the matrix elements, the relations (D.5) and (D.7) read,
A() = (1)L D(
),
L+1
B() = (1)
B( ),
C() = (1)L+1 C(
),
L
D() = (1) A( ),
At () = A(),
t
B () = C(),
C t () = B(),
t
D () = D().
(D.8)
A straightforward calculation taking into account the relations (D.8) reveals that the twisted
transfer matrix (10) satisfies the relation
T t () = (1)L+1 T ( ),
(D.9)
while we have the following identity for the double-row transfer matrix (35)
t (u) = t (u ).
(D.10)
We emphasize here that the crossing invariance of the double-row transfer matrix (D.10) had
been previously established in [32,41,42].
542
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]
[39]
[40]
[41]
[42]
R.J. Baxter, Ann. Phys. 70 (1972) 193.

N.Y. Reshetikhin, Lett. Math. Phys. 14 (1987) 235.
V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, Commun. Math. Phys. 200 (1999) 297.
C. Korff, in: J. Fuchs, et al. (Eds.), Non-Commutative Geometry and Representation Theory in Mathematical
Physics, in: Contemporary Mathematics, vol. 391, Amer. Math. Soc., 2005, p. 199, math-ph/0411034.
L.A. Takhtadzhan, L.D. Faddeev, Russ. Math. Surveys 34 (1979) 11.
F.C. Alcaraz, M.N. Barber, M.T. Batchelor, Ann. Phys. 182 (1988) 280.
V.E. Korepin, G. Izergin, N.M. Bogoliubov, Quantum Inverse Scattering Method and Correlation Functions, Cambridge Univ. Press, Cambridge, 1993.
R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, New York, 1982.
M.T. Batchelor, R.J. Baxter, M.J. ORourke, C.M. Yung, J. Phys. A: Math. Gen. 28 (1995) 2759.
C.M. Yung, M.T. Batchelor, Nucl. Phys. B 446 (1995) 461.
G.A.P. Ribeiro, M.J. Martins, W. Galleas, Nucl. Phys. B 675 (2003) 567.
V. Pasquier, M. Gaudin, J. Phys. A: Math. Gen. 25 (1992) 5243;
I. Krichever, O. Lipan, P. Wiegmann, A. Zabrodin, Commun. Math. Phys. 188 (1997) 267;
V. Kuznetsov, E.K. Sklyanin, J. Phys. A: Math. Gen. 31 (1998) 2241.
V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, Commun. Math. Phys. 177 (1996) 381;
V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, Commun. Math. Phys. 190 (1997) 247.
C. Korff, J. Phys. A: Math. Gen. 38 (2005) 6641.
C. Korff, J. Phys. A: Math. Gen. 40 (2007) 3749.
H.J. de Vega, Nucl. Phys. B 240 (1984) 495.
E.K. Sklyanin, J. Phys. A 21 (1988) 2375.
H.J. de Vega, A.G. Ruiz, J. Phys. A: Math. Gen. 26 (1993) L519.
S. Ghoshal, A. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 3841.
F.C. Alcaraz, M.N. Barber, M.T. Batchelor, R.J. Baxter, G.R.W. Quispel, J. Phys. A: Math. Gen. 20 (1987) 6397.
H. Fan, B.Y. Hou, K.J. Shi, Z.X. Yang, Nucl. Phys. B 478 (1996) 723.
J.P. Cao, H.Q. Lin, K.J. Shi, Y.P. Wang, Nucl. Phys. B 663 (2003) 487.
J. de Gier, P. Pyatov, J. Stat. Mech.: Theor. Exp. (2004) P03002.
W.L. Yang, R. Sasaki, Nucl. Phys. B 679 (2004) 495.
R. Murgan, R.I. Nepomechie, C. Shi, J. Stat. Mech.: Theor. Exp. (2006) P08006.
R.I. Nepomechie, J. Phys. A: Math. Gen 37 (2004) 433.
W. Galleas, M.J. Martins, Phys. Lett. A 335 (2005) 167.
C.S. Melo, M.J. Martins, G.A.P. Ribeiro, Nucl. Phys. B 711 (2005) 565.
R. Murgan, R.I. Nepomechie, J. Stat. Mech.: Theor. Exp. (2005) P05007.
R. Murgan, R.I. Nepomechie, J. Stat. Mech.: Theor. Exp. (2005) P08002.
W.L. Yang, R.I. Nepomechie, Y.Z. Zhang, Phys. Lett. B 633 (2006) 664.
W.L. Yang, Y.Z. Zhang, Nucl. Phys. B 744 (2006) 312.
A. Doikou, Phys. Lett. A 366 (2007) 556.
W. Galleas, Nucl. Phys. B 777 (2007) 352.
W.L. Yang, Y.Z. Zhang, JHEP 0704 (2007) 044.
W.L. Yang, Y.Z. Zhang, hep-th/0706.0772.
M.J. Martins, in: Proceedings of Progress in Solvable Lattice Models, RIMS, Kyoto University, 2004, nlin/0512063.
R.I. Nepomechie, Nucl. Phys. B 622 (2002) 615.
P. Baseilhac, K. Koizumi, hep-th/0703106.
R. Malara, A. Lima-Santos, J. Stat. Mech.: Theor. Exp. (2006) P09013.
L. Mezincescu, R.I. Nepomechie, Nucl. Phys. B 372 (1992) 597.
R.E. Behrend, P.A. Pearce, D.L. OBrien, J. Stat. Phys. 84 (1996) 1.

Nucl - Phys.B v.790

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.790

Загружено:

Авторское право:

Доступные форматы

Nuclear Physics B 790 (2008) 127

Higgs boson production via gluon fusion:

Received 28 March 2007; accepted 24 August 2007

Supported in part by the Swiss Bundesamt fr Bildung und Wissenschaft.

E-mail address: michael.spira@psi.ch (M. Spira).

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

The mass matrix of the sfermions in the leftright-basis is given by [5]1

and the masses of the squark mass eigenstates are given by

g = m2f g1 + MZ2 I3f ef sin2 W g2 ,

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

= m2f g1 + MZ2 ef sin2 W g2 ,

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

mt1 = 156 GeV,

mt1 = 155 GeV,

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

1 log 1+1 i 2 < 1,

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

which agrees with the C

of Eq. (18) in the heavy squark limit.

6 The anomalous dimension of the kinetic photon operator F F

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

E h/H = Evirt + Eggg + Egq q +

which agrees quantitatively with the numerical results.

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

2.3. Gluon fusion

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

thresholds the NLO QCD corrections develop

 g h/H Ah/H ( ) 16 2 log( 1 1) + i + const

which agrees quantitatively with the numerical results.

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

M. Mhlleitner, M. Spira / Nuclear Physics B 790 (2008) 127

Nuclear Physics B 790 (2008) 2841

Random walks of Wilson loops in the screening regime

Received 22 August 2007; accepted 31 August 2007

E-mail addresses: buividovich@tut.by (P.V. Buividovich), polykarp@itep.ru (M.I. Polikarpov).

P.V. Buividovich, M.I. Polikarpov / Nuclear Physics B 790 (2008) 2841

P.V. Buividovich, M.I. Polikarpov / Nuclear Physics B 790 (2008) 2841

P.V. Buividovich, M.I. Polikarpov / Nuclear Physics B 790 (2008) 2841

P.V. Buividovich, M.I. Polikarpov / Nuclear Physics B 790 (2008) 2841

P.V. Buividovich, M.I. Polikarpov / Nuclear Physics B 790 (2008) 2841

P.V. Buividovich, M.I. Polikarpov / Nuclear Physics B 790 (2008) 2841

dS 1 1 (x1 ) dS k k (xk ) Fa11 1 (x1 ) Fakk k (xk )

dS (x) d (y) 4 (x y).

P.V. Buividovich, M.I. Polikarpov / Nuclear Physics B 790 (2008) 2841

dS 1 1 (x1 ) dS k k (xk ) Fa11 1 (x1 ) Fakk k (xk )

P.V. Buividovich, M.I. Polikarpov / Nuclear Physics B 790 (2008) 2841

dS 1 1 (x1 ) dS k k (xk ) Fa11 1 (x1 ) Fakk k (xk )

g h/H Ah/H ( ) 16 2 log( 1 1) + i + const

Teven (k ) = (2)3 3 (k k ) = Todd (k),

A general symmetric tensor field Xij (

hbi (t, r, x) = a(r)

Evaluating the integral and using v = 52 /2 1 v 2 , we obtain