Nucl - Phys.B v.624

Nuclear Physics B 624 (2002) 344
www.elsevier.com/locate/npe
Implications of the Higgs boson searches on

different soft SUSY-breaking scenarios
S. Ambrosanio a,b , A. Dedes c , S. Heinemeyer d , S. Su e ,
G. Weiglein a,f
a CERN, TH Division, CH-1211 Geneva 23, Switzerland
b Banca di Roma, Direzione Generale, Linea Finanza, viale U. Tupini 180, I-00144 Roma, Italy
c Physikalisches Institut der Universitt Bonn, Nuallee 12, D-53115 Bonn, Germany
d HET, Physics Department, Brookhaven Natl. Lab., Upton, NY 11973, USA
e California Institute of Technology, Pasadena, CA 91125, USA
f Institute for Particle Physics Phenomenology, University of Durham, Durham DH1 3LR, UK
Received 28 June 2001; accepted 19 December 2001
Abstract
We investigate the Higgs boson sector of the Minimal Supersymmetric Standard Model (MSSM)
in the framework of the three most prominent soft SUSY-breaking scenarios, mSUGRA, mGMSB
and mAMSB. For each scenario, we determine the parameters at the electroweak scale from the
set of input variables at higher energy scales (depending on the specific scenario) and evaluate the
Higgs boson properties. The latter are based on results obtained within the Feynman-diagrammatic
approach by taking into account the complete one-loop and the dominant two-loop contributions. The
maximum value of the mass of the lightest neutral CP-even MSSM Higgs boson, mh , is determined
in the three scenarios, and the behavior of the Higgs couplings to fermions and gauge bosons is
investigated. Restrictions on tan and on the set of higher-energy scale parameters are derived
from the lower limits arising from the Higgs search at LEP2. We furthermore discuss the regions
of parameter space in the three scenarios compatible with interpreting the excess observed at LEP2
as a Higgs signal, mh = 115+1.3
0.9 GeV. The case where the events observed at LEP2 could originate
from the production of the heavier neutral CP-even Higgs boson is also considered. The implications
of a possible Higgs signal at 115 GeV for SUSY searches at future colliders are briefly discussed for
each of the three scenarios. 2002 Elsevier Science B.V. All rights reserved.
E-mail addresses: sandro.ambrosanio@bancaroma.it (S. Ambrosanio), dedes@th.physik.uni-bonn.de

(A. Dedes), sven.heinemeyer@bnl.gov (S. Heinemeyer), shufang@theory.caltech.edu (S. Su),
georg.weiglein@cern.ch (G. Weiglein).
0550-3213/02/$ see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 5 4 - X
S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344
PACS: 12.60.Jr; 12.60.Fr; 14.80.Ly
1. Introduction
The search for the light neutral Higgs boson is a crucial test of Supersymmetry
(SUSY) that can be performed with the present and the next generation of high-energy
colliders. The prediction of a relatively light Higgs boson is common to all supersymmetric
models whose couplings remain in the perturbative regime up to a very high energy
scale [1]. Finding the Higgs boson is thus one of the main goals of
todays high-energy
physics. The data taken during the final year of LEP running at s 206 GeV, while
establishing a 95% C.L. exclusion limit for the Standard Model (SM) Higgs boson of
mH > 113.5 GeV, showed at about the 3 level an excess of signal-like events over the
background expectation which is in agreement with the expectation for the production of
a SM Higgs boson of mH = 115+1.3
0.9 GeV [2]. A Higgs mass value of about 115 GeV
would indicate that the SM can only be valid up to a scale 106 GeV (or the vacuum
must me meta stable), since new physics contributions are necessary in order to prevent
the effective Higgs potential from becoming unstable [3]. In the Minimal Supersymmetric
Standard Model (MSSM), on the other hand, the mass of the lightest CP-even Higgs boson,
mh , is bounded from above by mh 135 GeV [4] (taking into account radiative corrections
up to two-loop order [413]). The effective Higgs potential is stabilized by contributions
of the SUSY partners of the SM particles [14]. Within the MSSM, the LEP excess can
be interpreted as the production of the lightest CP-even Higgs boson, which over a wide
parameter range has SM-like couplings, or of the heavier CP-even Higgs boson, in a region
of parameter space where the CP-odd Higgs boson A is light and the ratio of the vacuum
expectation values of the two Higgs doublets, tan , is relatively large.
In the MSSM no specific assumptions are made about the underlying SUSY-breaking
mechanism, and a parameterization of all possible SUSY-breaking terms is used. This
gives rise to the huge number of more than 100 new parameters in addition to the
SM, which in principle can be chosen independently of each other. A phenomenological
analysis of this model in full generality would clearly be very involved, and one usually
restricts to certain benchmark scenarios [15]. On the other hand, models in which all
the low-energy parameters are determined in terms of a few parameters at the Grand
Unification scale (or another high-energy scale), employing a specific soft SUSY-breaking
scenario, are much more predictive. The most prominent scenarios in the literature are
minimal Supergravity (mSUGRA) [16,17], minimal Gauge Mediated SUSY Breaking
(mGMSB) [18] and minimal Anomaly Mediated SUSY Breaking (mAMSB) [1921].
Analyses of the Higgs sector in these scenarios have been performed [2228], mostly
focusing only on the maximum value of mh . Within the mSUGRA scenario more recently
some implications of the LEP2 results on the Higgs search have been investigated in the
context of further constraints arising from the requirement that the lightest Supersymmetric
particle (LSP) should give rise to an acceptable dark matter relic density, and that the
predictions of the model should be in agreement with the experimental results on b s
and the anomalous magnetic moment of the muon.
In this paper we investigate in detail the predictions in the Higgs sector arising
from the three SUSY-breaking scenarios mSUGRA, mGMSB and mAMSB. We relate
the input from these scenarios in a uniform way to the predictions for the low-energy
phenomenology in the Higgs sector, allowing thus a direct comparison of the predictions
arising from the different scenarios. The high-energy parameters given in the three
scenarios are related to the low-energy SUSY parameters via renormalization group (RG)
running, taking into account contributions up to two-loop order. After transforming the
parameters obtained in this way into the corresponding on-shell parameters [2931], they
are used as input for the program FeynHiggs [32], which contains the complete oneloop and dominant two-loop corrections in the MSSM Higgs sector evaluated in the
Feynman-diagrammatic (FD) approach [4,5,33]. Further restrictions such as from precision
observables and the non-observation of SUSY particles are also taken into account. Based
on these predictions for the Higgs sector phenomenology, we analyze the consequences
of the results obtained from the Higgs search at LEP on the parameter space of the
three scenarios. This is done by considering both the LEP exclusion bound [34] and the
interpretation of the LEP excess as a possible signal. For the latter case we furthermore
discuss the corresponding spectra of the SUSY particles in view of the SUSY searches at
the next generation of colliders.
The rest of the paper is organized as follows. In Section 2 the three soft SUSY-breaking
scenarios as well as the evaluation of the CP-even Higgs boson sector of the MSSM are
briefly reviewed. Details about the combination of renormalization-group equation (RGE)
and FD calculation are given, and the parameter restrictions used are listed. The description
of our data sets and the numerical analyses for the three scenarios is given in Section 3.
The conclusions can be found in Section 4.
2. The Higgs sector in soft SUSY-breaking scenarios

The fact that no SUSY partners of the SM particles have so far been observed means
that low-energy SUSY cannot be realized as an unbroken symmetry in nature, and
SUSY models thus have to incorporate extra supersymmetry breaking interactions. This
is achieved by adding to the Lagrangian (defined by the given SU(3)C SU(2)L U(1)Y
gauge symmetry and the superpotential W ) some extra interaction terms that respect the
gauge symmetry but break supersymmetry. This breaking, however, should be such that
no quadratic divergences appear and the technical solution to the hierarchy problem
is not spoiled. Such terms are generally called soft SUSY-breaking terms. The most
general supersymmetry breaking interaction Lagrangian resulting from spontaneously
broken Supergravity in the flat limit (MP , where MP is the Planck mass) contains just
four types of soft SUSY-breaking terms [35], i.e., gaugino masses, -scalar masses,
-scalar cubic superpotential interactions and -scalar quadratic superpotential
interactions. Assuming that R-parity [36,37] is conserved, which we do in this paper for all
SUSY-breaking scenarios, reduces the amount of new soft terms allowed in the Lagrangian.
Choosing a particular soft SUSY-breaking pattern allows further reduction of the number
of free parameters and the construction of predictive models.
In this section, we first explain how we employ the principle of radiative electroweak
symmetry breaking (REWSB). Then we introduce the three most commonly studied soft
SUSY-breaking scenarios and describe the general method used to derive predictions for
the low-energy Higgs sector, which applies to all scenarios.
2.1. Radiative electroweak symmetry breaking
The investigation of REWSB in the MSSM [38] relies on a RG analysis. The Higgs
boson running mass-squared matrix, although positive definite at large energy scales
of the order of the Grand Unification scale MGUT , yields a negative eigenvalue at low
energies causing the spontaneous breakdown of the electroweak (EW) symmetry. The
result can be interpreted as a prediction of MZ in terms of parameters at a large energy
scale. Alternatively, one can consider MZ as being determined by experiment and derive
in this way the absolute value of the -parameter (which defines the coupling of the two
Higgs doublets) as well as the value of the bilinear soft-SUSY breaking parameter B at
a scale in the vicinity of the EW scale, from the minimization conditions of the effective
potential,
2 (Q) =
2H2 tan2
m
2H1 m
B(Q) =
tan2
1
MZ2 (Q),
2
(m
21 + m
22 ) sin 2
,
2(Q)
(1)
(2)
where Q is derived from the scalar fermion sector. It is usually chosen such that radiative
corrections to the effective potential are rather small compared to other scales. In Eqs. (1),
(2) tan v2 /v1 is the ratio of the two vacuum expectation values of the Higgs fields H2
and H1 responsible for giving masses to the up-type and down-type quarks, respectively.
In Eqs. (1), (2), tan is evaluated at the scale Q, from the scale MZ , where it is considered
as an input parameter.1 By m
2Hi = m2Hi + vi in Eqs. (1), (2) we denote the radiatively
corrected running Higgs soft-SUSY breaking masses and
m
2i = m2Hi + 2 + vi m
2Hi + 2
(i = 1, 2),
(3)
where vi are the one-loop corrections based on the 1-loop ColemanWeinberg effective
potential V , vi = 2v1 i V
vi ,

Ma2 Ma2
1
Ma2
2Ja
vi =
() (2Ja + 1)Ca a
ln 2 1 .
64 2 a
vi vi
Q
(4)
Here Ja is the spin of the particle a, Ca are the color degrees of freedom, and a = 1(2)
for real scalar (complex scalar), a = 1(2) for Majorana (Dirac) fermions. Q is the energy
scale and the Ma are the field dependent mass matrices. Explicit formulas of the vi are
given in Appendices of Refs. [40,41]. In our analyses contributions from all SUSY particles
1 See, for example, the discussion in Appendix of Ref. [39] or in Ref. [6].
at the one-loop level are incorporated.2 With MZ2 here we denote the tree level running
Z boson mass, MZ2 (Q) = 12 (g12 + g22 )v 2 (v 2 v12 + v22 ), extracted at the scale Q from its
physical pole mass MZ = 91.187 GeV. The REWSB is fulfilled, see Section 2.6.3, if and
only if there is a solution to the Eqs. (1), (2).3
2.2. mSUGRA
A dramatic simplification of the structure of the SUSY-breaking interactions is
provided either by Grand Unification assumptions or by Superstrings. For example, SU(5)
unification implies at tree level equality relations between the scalar soft-SUSY breaking
masses mQ
= mU
c = mE
c , and mL = mD
c , equality between the soft breaking gaugino
masses M1 = M2 = M3 and for two of the trilinear soft breaking couplings, Ad = Ae . On
the other hand, SO(10) unification implies equality of all scalar particle masses, equality
of Higgs masses and equality of the three types of trilinear couplings. The simplest
possible choice at tree level is to take all scalar particle and Higgs masses equal to a
common mass parameter M0 , all gaugino masses are chosen to be equal to the parameter
M1/2 and all trilinear couplings flavor blind and equal to A0 . This situation is common
in the effective Supergravity theories resulting from Superstrings but there exist more
complicated alternatives. Interestingly, the contribution of the family-anomalous U(1)
universal D-term to the scalar quark masses may be intra-family non-universal, and may
differ from the usually assumed universal boundary conditions [42]. Another alternative
are, for example, Superstrings with massless string modes of different modular weights
that lead to different scalar particle masses at tree level [43]. Thus, it seems that in almost
all the realistic models motivated by Grand Unified Theories (GUTs) or Superstrings the
universality assumption is broken and each of these models has to be addressed separately
in order to study its phenomenology at low energies. On the other hand, one should
note that such non-minimal alternatives like flavor-dependent scalar particle masses are
constrained by limits on Flavor Changing Neutral Currents (FCNC) processes. In what
follows, we shall therefore consider the simplest (and most commonly used) case of the
three parameters at the GUT scale, namely, M0 , M1/2 and A0 , which is usually called the
mSUGRA scenario.
In order to solve Eqs. (1), (2), i.e., in order to impose the constraint of REWSB, one
needs as input tan (MZ ) and sign(). The running soft SUSY-breaking Higgs mass
parameters, mH1 and mH2 , are defined at the EW scale after their evolution from the
GUT scale where we assume that they have a common value, M0 . In addition the radiative
corrections vi to the minimization conditions Eq. (4) are defined from the low-energy
2 The corresponding two-loop corrections are not available yet. Assuming the size of these unknown higher-
order corrections to be of the same size as for the Higgs-boson mass matrix, see Section 2.5, the resulting values of
and B could change by 510%. These parameters will serve as input for our numerical analysis in Section 3.
The possible changes would hardly affect our results obtained in the Higgs-boson sector and only mildly affect
the analysis of SUSY particle spectra in Section 3.5.
3 Sometimes in the literature, the requirement of the REWSB is described by the inequality m2 (Q)m2 (Q)
1
2
|(Q)B(Q)|2 < 0. This relation is automatically satisfied here from Eqs. (1), (2) and from the fact that the
physical squared Higgs masses must be positive.
SUSY spectrum and the masses of the SM particles, which in turn means knowledge
of M0 , M1/2 and A0 at the GUT scale. Thus, apart from the SM masses provided by
the experimental data [44], 4 parameters and a sign are required to define the mSUGRA
scenario:
{M0 , M1/2 , A0 , tan , sign()}.
(5)
In our numerical procedure we employ a two-loop renormalization group analysis for

all parameters involved, i.e., all couplings, dimensionful parameters and VEVs. We start
with the MS values for the gauge couplings at the scale MZ , where for the strong coupling
constant s a trial input value in the vicinity of 0.120 is used. The MS values are converted
into the corresponding DR ones [45]. The MS running b and masses are run down to
mb = 4.9 GeV, m = 1.777 GeV with the SU(3)C U(1)em RGEs [46] to derive the
running bottom and tau masses (extracted from their pole masses). This procedure includes
all SUSY corrections at the one-loop level and all QCD corrections at the two-loop level
as given in Ref. [41]. Afterwards by making use of the two-loop RGEs for the running
masses mb , m , we run upwards to derive their MS values at MZ , which are subsequently
converted to the corresponding DR values. This procedure provides the bottom and tau
Yukawa couplings at the scale MZ . The top Yukawa coupling is derived from the topquark pole mass, mt = 175 GeV, which is subsequently converted to the DR value, mt (mt ),
where the top Yukawa coupling is defined. The evolution of all couplings from MZ running
upwards to high energies now determines the unification scale MGUT and the value of the
unification coupling GUT by
1 (MGUT )|DR = 2 (MGUT )|DR = GUT .
(6)
At the GUT scale we set the boundary conditions for the soft SUSY breaking parameters,
i.e., the values for M0 , M1/2 and A0 are chosen, and also 3 (MGUT ) is set equal to GUT .
All parameters are run down again from MGUT to MZ . For the calculation of the soft
SUSY-breaking masses at the EW scale we use the step function approximation [47].
Thus, if the equation employed is the RGE for a particular running mass m(Q), then Q0 is
the corresponding physical mass determined by the condition m(Q0 ) = Q0 . After running
down to MZ , the trial input value for s has changed. At this point the value for tan is
chosen and fixed. As described in Section 2.1, the parameters || and B are calculated from
the minimization conditions (1) and (2), respectively. Only the sign of the -parameter is
not automatically fixed and thus chosen now. This procedure is iterated several times until
convergence is reached.
2.3. mGMSB
A very promising alternative to mSUGRA is based on the hypothesis that the soft
SUSY breaking (SSB) occurs at relatively low energy scales and it is mediated mainly
by gauge interactions through the so-called messenger sector (GMSB) [18,48,49].
This scheme provides a natural, automatic suppression of the SUSY contributions to
flavor-changing neutral currents and CP-violating processes. Furthermore, in the simplest
versions of GMSB (denoted hereafter with mGMSB), the MSSM spectrum and most of
the observables depend on just 4 parameters and a sign,

{Mmess , Nmess , , tan , sign()},
(7)
where Mmess is the overall messenger mass scale; Nmess is a number called the messenger
index, parameterizing the structure of the messenger sector; is the universal soft SUSYbreaking mass scale felt by the low-energy sector; tan is the ratio of the vacuum
expectation values of the two Higgs doublets; sign() is the ambiguity left for the SUSY
higgsino mass after imposing a correct REWSB (see Section 2.1 and, e.g., Refs. [22,50
53]).
The phenomenology of GMSB (and more in general of any theory with low-energy
with mass [54] given by
SSB) is characterized by the presence of a very light gravitino G

2
F
F
m3/2 = mG
=

2.37 eV,

100 TeV
3 MP
where F is the fundamental scale of SSB and MP = 2.44 1018 GeV is the reduced
is always the LSP in

Planck mass. Since F is typically of order 100 TeV, the G
these theories. Hence, ifR-parity is conserved, any MSSM particle will decay into the
gravitino. Depending on F , the interactions of the gravitino, although much weaker than
gauge and Yukawa interactions, can still be strong enough to be of relevance for collider
physics. In most cases, the last step of any SUSY decay chain is the decay of the next-tolightest SUSY particle (NLSP), which can occur either outside or inside a typical detector,
possibly close to the interaction point. The nature of the NLSPor, more precisely, of the
SUSY particle(s) having a large branching ratio for decaying into the gravitino and the
relevant SM partnerdetermines four main scenarios giving rise to qualitatively different
phenomenology [51].
The low-energy parameter sets for this scenario have been calculated by using the
program SUSYFIRE 4 and adopting the phenomenological approach of Refs. [5153],
see also Ref. [27]. The origin of is not specified, nor the assumption B = 0 is made at
the messenger scale. Instead, correct REWSB is imposed to trade and B for MZ and
tan , leaving the sign of undetermined, see Section 2.1. However, note that to build a
fully coherent GMSB model, one should also find a more fundamental solution to the latter
problem, perhaps providing a dynamical mechanism to generate and B, possibly with
values of the same order of magnitude. This might be accomplished radiatively through
some new interaction. In this case, the other soft terms in the Higgs potential, namely,
m2H1,2 , will be also affected and this will in turn change the values of || and B coming
from REWSB conditions. We have checked this circumstance in detail in the case where
one has an extra term in the Higgs potential of the type + (see the parameterization of
Ref. [50]) and also performed some checks in the general case. In all cases we did not find
any big changes in mh .
4 An updated, generalized and Fortran-linked version of the program used in Ref. [51]. It generates minimal
and non-minimal GMSB and SUGRA models. For inquiries about this software package, please send e-mail to
sandro.ambrosanio@bancaroma.it.
10
To determine the MSSM spectrum and low-energy parameters, the RGE evolution is
solved with boundary conditions at the Mmess scale, where

a (a = 1, 2, 3),
Ma (Mmess ) = Nmess g
Mmess
2

a
2
2
Ca ,
m
(Mmess ) = 2Nmess f
(8)
Mmess a 4
for the gaugino and the scalar masses, respectively. The exact expressions for g and f at the
one- and two-loop level can be found, e.g., in Ref. [51], and Ca are the quadratic Casimir
invariants for the scalar fields. As usual, the scalar trilinear couplings Af are assumed to
vanish at the messenger scale, as suggested by the fact that they (and not their square) are
generated via gauge interactions with the messenger fields at the two loop-level only.
The interesting region of the GMSB parameter space is selected as follows. Barring
the case where a neutralino is the NLSP and decays outside the detector (large F ),
the GMSB signatures are very spectacular and the SM background is generally negligible
or easily subtractable. Therefore, also in accordance with negative results in the LEP2
searches [55], only models where the NLSP mass is larger than 100 GeV are considered.
Other requirements are: Mmess > 1.01, to prevent an excess of fine-tuning of the
messenger masses; the mass of the lightest messenger scalar be at least 10 TeV; Mmess >
MGUT exp(125/Nmess), to ensure the perturbativity of gauge interactions up to the
GUT scale; Mmess 105 , for simplicity. As a result, the messenger index Nmess , which
is assumed to be an integer independent of the gauge group, cannot be larger than 8. To
prevent the top Yukawa coupling from blowing up below the GUT scale, tan > 1.5 is
required. Models with tan 55 (with a mild dependence on ) are forbidden by the
REWSB requirement, see Section 2.6.3, and typically fail to give MA2 > 0.
The models are generated using SUSYFIRE with the following prescriptions for the
high-energy input parameters. Logarithmic
steps have been used for (between about
45 TeV/Nmess and about 220 TeV/ Nmess ), Mmess / (between about 1.01 and 105 ) and
tan (between 1.5 and about 60), subject to the constraints described above. SUSYFIRE
starts from the values of particle masses and gauge couplings at the weak scale and then
evolves them up to the messenger scale through RGEs. At the messenger scale, it imposes
the boundary conditions (8) for the soft particle masses and then evolves the RGEs back
to the electroweak scale. The decoupling of each SUSY particle at the proper threshold is
taken into account. Two-loop RGEs are used for gauge couplings, third generation Yukawa
couplings and gaugino soft masses. The other RGEs are taken at the one-loop level.5 At
the scale Q, derived from the scalar fermion sector, REWSB conditions are imposed by
means of the one-loop effective potential approach. For the vi in Eq. (4) all dominant
corrections from the stop, sbottom and stau sector are included. The program then evolves
5 Contrary to the mSUGRA scenario in Section 2.2 the scalar masses are treated only at the one-loop level in
mGMSB and the mAMSB scenario in Section 2.4. However, the main effects arise from the Yukawa couplings,
which are consistently treated at the two-loop level. The two-loop effects on the scalar masses have been shown
to be at the 5% level [47].
11
up again to Mmess and so on. Three or four iterations are usually enough to get a good
approximation for the MSSM spectrum.
2.4. mAMSB
The most recently proposed Anomaly Mediated SUSY Breaking (AMSB) scenario
[19,20] provides an alternative way to give mass to all the SUSY particles. In this model,
SUSY-breaking happens on a separate brane and is communicated to the visible world via
the super-Weyl anomaly. The overall scale of SUSY particle masses is set by maux , which
is the VEV of the auxiliary field in the supergravity multiplet. In the AMSB scenario, the
low-energy soft supersymmetry breaking parameters Mi (gaugino masses, i = 13), m2scalar
and Ay at the GUT scale are given by [19,21]
gi
maux ,
gi

1
2
g +
y m2aux + m20 ,
mscalar =
4 g
y
y
Ay = maux .
y
Mi =
(9)
(10)
(11)
Notice that the slepton squared-masses would be negative if m0 were absent. There
have been several proposals to solve this tachyonic slepton problem: bulk contributions
[19], non-decoupling effects of ultra-heavy vectorlike matter fields [56], coupling of extra
Higgs doublets to the leptons [57], contributions from the R-parity violating couplings in
Eq. (10) (with m0 = 0) [58], and a heavy mass threshold contribution at higher orders [59].
Here we have adopted a phenomenological approach and have introduced an additional
mass scale m0 at the GUT scale in order to keep the slepton masses positive [21]. For
simplification, we choose m0 to be the same for all the super scalar particles. Therefore, in
the minimal case (mAMSB), the particle spectrum can be determined by 3 parameters and
a sign:
{maux , m0 , tan , sign()}.
(12)
Eqs. (9), (10) and (11) would hold at all scales if m0 were absent. However, once
m0 is introduced at the GUT scale, the above definitions of Mi , m2scalar and Ay set the
boundary conditions and the entire SUSY spectrum can be obtained via the running of
supersymmetric RGEs down to a lower scale.
Once the squark threshold is crossed, the squarks decouple and one is left with an
effective field theory with two Higgs doublets and all the standard model particles.6 The
two unknown parameters || and B are determined by the minimization of the Higgs
effective potential as explained in Section 2.1. Therefore, the low-energy spectrum is fixed
once the values of maux , m0 , tan and the sign of are known.
6 Gluinos are also decoupled since their masses are close to the squark masses. The contributions of Bino and
Winos to the Higgs sector can be neglected since the U(1)Y and SU(2) gauge couplings are small.
12
2.5. Evaluation of predictions in the Higgs boson sector of the MSSM

The most relevant parameters for Higgs boson phenomenology in the MSSM are the
mass of the CP-odd Higgs boson, MA , the ratio of the two vacuum expectation values,
tan , the scalar top masses and mixing angle, mt1 , mt2 , t, for large tan also the scalar
bottom masses and mixing angle, mb1 , mb2 , b , the Higgs mixing parameter, , the gluino
mass, mg , and the U(1) and SU(2) gaugino masses, M1 and M2 . The way in which these
low-energy parameters are derived in each of the soft SUSY-breaking scenarios has been
described in Sections 2.22.4. Since the RG running employed in the three scenarios is
based on the DR scheme, the corresponding low-energy parameters are DR parameters. In
order to derive predictions for observables, i.e., particle masses and mixing angles, these
parameters in general have to be converted into on-shell parameters.
For the predictions in the MSSM Higgs sector we use results obtained in the Feynmandiagrammatic (FD) approach (see below) within the on-shell renormalization scheme.
Since they incorporate two-loop contributions in the tt sector, the parameters in the scalar
top sector (which enter at one-loop order in the predictions for the Higgs boson masses)
have to be appropriately converted from DR to on-shell parameters [2931]. We perform
this conversion using the full O(s ) contributions.
In the FD approach the masses of the two CP-even Higgs bosons, mh and mH , are
derived beyond tree level by determining the poles of the hH -propagator matrix whose
inverse is given by

2
hH q 2
H q 2
q m2H,tree +
1
(Higgs) = i
(13)

,
hH q 2
h q 2
q 2 m2h,tree +
denote the renormalized Higgs boson self-energies. Determining the poles of
where the
the matrix Higgs in Eq. (13) is equivalent to solving the equation

2
hH q 2 2 = 0.
h q 2 q 2 m2H,tree +
H q 2
q m2h,tree +
(14)
We use the result for the Higgs boson self-energies consisting of the complete oneloop result for the Higgs boson self-energies in the on-shell scheme [33] combined with
the dominant two-loop contributions of O(s ) [4,5] and further subdominant corrections
[7,8], see Ref. [4] for details. The matrix Eq. (13) therefore contains the renormalized
Higgs boson self-energies

2
(1)
(2)
s q 2 =
(15)
s q + s (0), s = h, H, hH,
where the momentum dependence is neglected only in the two-loop contribution.
An effective mixing angle,

1 2
(MA2 + MZ2 ) sin cos
eff = arctan
, < eff < ,
2
2
2
2
2

2
2
MZ cos + MA sin 1 mh
(16)
can furthermore be obtained from diagonalizing the mixing matrix in the basis of
the unrotated neutral CP-even fields 1 , 2 , neglecting the momentum dependence
everywhere (alternatively, the mass matrix in the h, H basis is diagonalized by the angle
, where eff = tree + , see, e.g., Ref. [60]). Inserting eff in the tree-level formulas
13
for Higgs production and decay, the dominant universal corrections in the Higgs sector are
taken into account [60,61].
The results for the Higgs boson masses and eff as well as the conversion from DR to
on-shell parameters using the full O(s ) contributions for both the parameters in the tt
and bb sector are implemented in the Fortran code FeynHiggs [32].
For the investigation of the mSUGRA scenario FeynHiggs has been interfaced to the
program SUITY [62], used for the evaluation of the low-energy spectrum of the mSUGRA
scenario. A combined program, FeynSSG [63], has been created on the basis of FeynHiggs
and SUITY, in which the two subprograms run automatically.
As a further check of our results for the Higgs boson sector, we have (in addition to the
FeynHiggs calculation) evaluated all results with a code based on an independent one-loop
calculation [64], but using the two-loop routines of FeynHiggs. The difference of the oneloop calculation based on Ref. [33], used in FeynHiggs, and the ones given in Ref. [64]
are only due to different renormalization prescriptions and thus of higher order [65]. The
results we found in both approaches are as expected very similar and lead to the same
conclusions.
2.6. Other constraints
While our main focus in this paper is on the physics in the Higgs sector, we also take into
account some further (relatively mild) constraints when determining the allowed parameter
values. These constraints are discussed in the following.
2.6.1. Precision observables
The electroweak precision observables are affected by the whole spectrum of SUSY
particles. The main SUSY contributions to the W boson mass, MW , the effective leptonic
weak mixing angle, sin2 eff , and other Z boson observables usually arise from t/b
contributions. They enter via the leading contribution to the -parameter [66]. In our
analysis we take into account the corrections arising from t/b loops up to two-loop
order [67]. A value of outside the experimentally preferred region of SUSY
3 103 [44] indicates experimentally disfavored t and b masses.7 The evaluation of
SUSY is implemented in FeynHiggs.
We have verified that in our analysis below the constraint does not play a significant
role, i.e., nearly all generated model points give rise to an acceptable contribution to the
electroweak precision observables. As a conservative approach, we do not apply any further
constraints from g 2 or b s .
2.6.2. Experimental bounds on SUSY particle masses
The search for SUSY particles has been one of the main tasks pursued at Run I of the
Tevatron and at LEP. The searches all turned out to be negative, thus lower limits on the
SUSY particle masses have been set. In order to restrict the allowed parameter space in
7 Since the SUSY evaluation involves scalar bottoms at the two-loop level, also the parameters in the
b sector have to be transformed from DR to on-shell.
14
the three soft SUSY-breaking scenarios we employed the following constraints on their
low-energy mass spectrum [44,55,6871]:
me > 95 GeV (mSUGRA, AMSB),
m > 85 GeV (mSUGRA, AMSB),
mt > 95 GeV (mSUGRA, AMSB),
mb > 85 GeV (mSUGRA, AMSB),
mg > 190 GeV (mSUGRA, AMSB),
m > 103 GeV (mSUGRA, m > 300 GeV),
m 84.6 GeV (mSUGRA, m < 300 GeV),
m 45 GeV (AMSB),
m 0 36 GeV (mSUGRA),
1
m 0 45 GeV (AMSB),
1
mNLSP > 100 GeV (GMSB).
(17)
Note that the NLSP condition in the GMSB scenario (applying either to the lightest
neutralino or to the lighter stau) automatically imposes stronger bounds on the other
particle masses than the purely experimental bounds.
2.6.3. Other phenomenological restrictions
Besides constraints from precision observables and from unsuccessful direct search
for SUSY particles, we also take into account the following restrictions (if not indicated
otherwise):
For the top-quark mass, throughout this paper we use the value mt = 175 GeV.
A variation of mt directly affects the result for mh , while its influence on the other
quantities studied here is more moderate. As a rule of thumb, a change in mt by
1 GeV also results in a change in mh of about 1 GeV [72].
The GUT or high-energy scale parameters are taken to be real, no SUSY CP-violating
phases are assumed.
In all models under consideration the R-parity symmetry [36,37] is taken to be
conserved.
Parameter sets that do not fulfill the condition of radiative electroweak symmetry
breaking (REWSB), i.e., the one-loop minimization conditions of Eqs. (1), (2), are
discarded (already at the level of model generation).
Within all soft SUSY-breaking scenarios considered here, the condition of REWSB
leads to restrictions on tan . For example, it is almost impossible (or a huge fine tuning
is required) to find very large values of tan , tan 60 which pass this constraint.
This is because in that region both the RGEs of the Higgs soft breaking masses receive
15
large corrections not only from the top Yukawa coupling but also from the bottom and
the tau Yukawa couplings. This drives the numerator (and thus 2 ) of Eq. (1) negative,
thus excluding this parameter set from our analysis. Another possibility of not fulfilling
the REWSB condition is a very heavy soft SUSY-breaking spectrum, so that m2H2 does
not reach negative values and thus does not trigger REWSB.
Parameter sets that do not fulfill the strong CCB constraints are discarded (already
at the level of model generation), i.e., models for which the physical vacuum would be
charge or color breaking. In our analysis this corresponds to cases where the squared
scalar quark or charged lepton masses are becoming negative at the scale Q, where Q
is the energy scale at which the low-energy parameters are decoupled. However, we
do not test the models for local or global charge or color breaking minima in general.
In most cases the tunneling time from our charge and color conserving vacuum to the
charge or color breaking minimum is much longer than the present age of the universe,
and thus they are in practice not dangerous [7376].
The original motivation for the introduction of SUSY into particle physics was the
solution of the hierarchy problem. This sets a natural upper bound on the SUSY
particle masses, which of course depends on how much fine tuning one is willing
to accept. In our analysis we have imposed a (rather mild) naturalness bound. This
upper bound on the SUSY particle masses has been chosen to be equal for all three soft
SUSY-breaking scenarios. Thus the low-energy mass spectra are directly comparable.
We have imposed
mq 1.5 TeV,
mg 2 TeV.
(18)
These bounds give rise to upper bounds also on the other scalar masses and on the
electroweak gaugino masses (depending on the specific scenario).
The bounds imposed in Eq. (18) can of course not be considered as strict upper bounds
(although constraints from cosmology, b s and g 2 in general also favor a
relatively light particle spectrum [7780]), but carry a certain degree of arbitrariness.
In particular, we do not consider here the scenario of focus point supersymmetry [81],
in which squarks and sleptons in the multi-TeV range can occur. It should be noted,
however, that in all three soft SUSY-breaking scenarios an upper bound of O(5 TeV)
is obtained by the requirement of REWSB. On the other hand, saturating this upper
bound of O(5 TeV) requires severe fine tuning to satisfy the minimization conditions
given in Eqs. (1), (2).
The imposed upper bounds on SUSY masses also naturally result in a limit for the
soft SUSY-breaking parameters at the high energy scale, M0 , M1/2 in mSUGRA,
in mGMSB and m0 , maux in mAMSB. If the bounds in Eq. (18) were relaxed,
heavier particle spectra would be allowed. The effect on the Higgs boson sector is
only logarithmic and thus rather small. Concerning the collider phenomenology as
presented in Section 3.5, the detection of SUSY particles would become more difficult.
We demand that the lightest SUSY particle is uncolored and uncharged. In the GMSB
scenario the LSP is always the gravitino, so this condition is automatically fulfilled.
Within the mSUGRA and mAMSB scenario, the LSP is required to be the neutralino.
Parameter sets that result in a different LSP are excluded.
16
We do not impose further restrictions arising from BR(b s ) [77] and g 2 [78],
which could lead to additional constraints on the three soft SUSY-breaking scenarios.
Restrictions of this kind depend on the experimental errors of these observables and
the uncertainties in their theoretical prediction and could change considerably if the
experimental central values change in the future. Moreover, slight modifications of the
SUSY-breaking scenarios which would have only a minor impact on the phenomenology of the models discussed here could have a strong influence on constraints from
BR(b s ) and g 2. This could happen, for instance, in the case of BR(b s )
via the presence of small flavor mixing terms in the SUSY Lagrangian [82].
As a conservative approach, we therefore do not discard parameter sets which do not
fulfill the constraints from BR(b s ) [77] and g 2. It should be noted, however,
that if these constraints are imposed and the AMSB scenario is taken at face value, i.e.,
without any additional contributions, the parameter space allowed by the experimental
values of BR(b s ) and g 2 is rather restricted. As for the mSUGRA scenario,
the effect of BR(b s ) would disallow a region with small M1/2 for large tan ,
whereas the effect of g 2 would be to set an upper bound on the combination of M0
and M1/2 , see, e.g., Ref. [80].
In the same spirit, we also do not apply any further cosmological constraints, i.e.,
we do not demand a relic density in the region favored by dark matter constraints, see
Ref. [83] and references therein. As in the case of BR(b s ) and g 2, slight modifications of the scenario which do not concern collider phenomenology could have a
strong impact on the bounds derived from cosmology. In the case where the LSP relic
abundance in the scenarios discussed here is too small to explain the observed amount
of cold dark matter (CDM), a further mechanism could provide the additionally required amount of CDM (this would certainly apply to the case of mGMSB). If on the
other hand the amount of CDM appears to be too large in a given scenario, thermal
inflation [84] could offer a mechanism for bringing the CDM density into agreement
with the cosmological bounds. Furthermore, the neutralino could turn out to be the
NLSP and decay (outside of collider detectors) into a very weakly interacting LSP
(e.g., the axino [85]), or there could be a small amount of R-parity violation present in
the model.
3. Numerical analyses
3.1. Experimental bounds from the MSSM Higgs sector
The results from the Higgs search at LEP have excluded a considerable part of the
MSSM parameter space [34]. On the other hand, an excess at about the 3 level has been
observed which is compatible with the production of a SM Higgs boson with a mass of
about 115 GeV [2]. For our numerical analysis we will focus on three different cases
implying different restrictions on the MSSM parameter space. In case (I) we investigate
the full parameter space which is allowed in the three scenarios when taking into account
the exclusion bounds from the Higgs search and the further constraints discussed in the
previous section. In case (II) and case (III), on the other hand, we specifically focus on
17
the interpretation of the excess observed in the Higgs search at LEP as production of the
lightest CP-even Higgs boson of the MSSM (case (II)) and of the heavier CP-even Higgs
boson of the MSSM (case (III)).
Case (I)
The results of the search for the MSSM Higgs bosons are usually interpreted in three
different benchmark scenarios [15]. The 95% C.L. exclusion limit for the SM Higgs boson
of mH > 113.5 GeV applies also for the lightest CP-even Higgs boson of the MSSM in the
parameter region of large MA and/or small tan . In the unconstrained MSSM this bound
is reduced to mh > 91.0 GeV [34] for MA 150 GeV and tan 8 as a consequence
of a reduced coupling of the Higgs to the Z boson. For the CP-odd Higgs boson a lower
bound of MA > 91.9 GeV has been obtained [34]. In order to correctly interpolate between
the parameter regions where the SM lower bound of mH > 113.5 GeV and the bound
mh > 91.0 GeV apply, we use the result for the Higgs-mass exclusion given with respect
to the reduced ZZh coupling squared (i.e., sin2 ( eff ), see Eq. (16)) [86]. We have
compared the excluded region with the theoretical prediction obtained at the two-loop level
for mh and sin2 ( eff ) for each parameter set (using mt = 175 GeV).
Another important constraint on mh comes from the searches in pp collisions at Run I
b bb
b has been
of the Tevatron [87]. No evidence of a signal of the type pp bbh
found. This leads to an improvement of the LEP limits in the region of large tan ,
tan 50. Since the bounds obtained at the Tevatron are given only in the no-mixing and
scenario [15,87], they do not necessarily apply to all cases of our present analysis. As
mmax
h
a conservative treatment, we therefore do not use the Tevatron bound for excluding models
in the high tan region. It should be noted, however, that owing to the REWSB constraint
in our analysis below we do not find allowed models for tan 60.
Case (II)
In this scenario the LEP excess is interpreted as production of the lightest CP-even
Higgs boson of the MSSM, and we thus focus on the parameter regions in the three soft
SUSY-breaking scenarios where
mh = 115 2 GeV.
(19)
The assumed error of 2 GeV is somewhat larger than the region favored by the LEP
data, in order to allow for some theoretical uncertainties from unknown higher order
corrections in the Higgs boson mass calculation (note that the dominating theoretical
uncertainty is related to the experimental error of the top-quark mass; we focus in our
analysis on the value mt = 175 GeV, the corresponding mh values for different values of
mt can be obtained from the approximate relation mh /mt = O(1)). In order to allow
an interpretation of the LEP excess in terms of the lightest MSSM Higgs boson, it is
furthermore necessary that the production and decay rates of h are similar to those of
the SM Higgs boson. We therefore in addition demand sin2 ( eff ) 0.8, which results
in a production cross section for the Higgs strahlung process, e+ e Zh sin2 ( eff )
close to the SM cross section. Furthermore, we require that the hbb coupling in the MSSM
is not strongly suppressed compared to the SM case. The hbb coupling differs from the
corresponding SM coupling in two ways. Firstly, it has an additional factor sin eff / cos
18
(appearing squared in the branching ratio). We demand that sin2 eff / cos2 0.8.
Secondly, the hbb vertex can be affected by gluino loop corrections (and less importantly
also gaugino loop corrections) [88,89]. Usually they are parameterized via mb ,
mb
2s
Yt
mg tan I (mb1 , mb2 , mg ) +
At tan I (mt1 , mt2 , ),
3
4
(20)
where Yt = h2t /(4) and

I (a, b, c) =
a 2 b2 ln(a 2 /b2 ) + b2 c2 ln(b 2 /c2 ) + c2 a 2 ln(c2 /a 2 )

.
(a 2 b2 )(b2 c2 )(a 2 c2 )
(21)
The main correction to the hbb coupling is proportional to 1/(1 + mb ). In our analysis
of case (II) we additionally demand that |mb | < 0.5.
Case (III)
In this scenario we investigate whether the LEP excess can be interpreted as the
production of the heavy CP-even Higgs boson in the MSSM. In order to allow this
interpretation, H has to have SM-like couplings to the Z, i.e., cos2 ( eff ) 0.8. In
this case the h production via Higgs strahlung, e+ e Zh, is highly suppressed, whereas
the associated production e+ e Ah could be beyond the kinematic reach of LEP. We
apply a bound of mh + MA > 206 GeV in this scenario. It should be noted that this bound
is very conservative, since values of mh + MA as low as about 190 GeV are not excluded
from the Higgs search at LEP [34]. As in case (II) we also require that the decay of the
heavy CP-even Higgs boson is SM like, i.e., the dominating decay channel is H bb.
Therefore, we demand cos2 eff / cos2 > 0.8.
3.2. mSUGRA
For the numerical analysis we have scanned over about 50000 models, where the
parameters have been randomly chosen in the intervals
50 GeV M0 1 TeV,
50 GeV M1/2 1 TeV,
3 TeV A0 3 TeV,
1.5 tan 60,
sign = 1.
(22)
Although we have scanned over about 50000 models, we show in the figures of this paper
a subset of around 5000 (randomly) selected points to keep the density of the points at
a reasonable level. This has been done for all three soft SUSY-breaking scenarios. No
reduction of the data points is applied for the cases (I)(III) in parameter regions with a
small density of points, i.e., in particular, for mh < 113 GeV and sin2 ( eff ) < 0.99.
We first analyze the allowed parameter region in the Higgs sector of the mSUGRA
scenario. In Fig. 1 we show the variation of the light Higgs boson mass with respect to tan
for the three cases defined in Section 3.1. Fig. 2 shows the allowed parameter space in the
19
Fig. 1. The light CP-even Higgs boson mass mh as a function of tan in the mSUGRA scenario. The three
cases as discussed in Section 3.1 are displayed together with the rejected models. Case (I) corresponds to the
models that have passed all theoretical and experimental constraints. Case (II) is the subset of case (I) with mh
values in the region favored by recent LEP Higgs searches, 113 GeV mh 117 GeV, and SM like couplings of
the h. In case (III), whose parameter points are indicated by an arrow for better readability, the heavier CP-even
Higgs boson lies in the region 113 GeV mH 117 GeV, while the lighter one has a suppressed coupling to the
Z boson and is too heavy to be produced in associated production.
MA tan plane. Case (I), corresponding to the models that have passed all experimental
and theoretical constraints, is indicated in the figures by big light shaded points. Big dark
shaded points indicate case (II), i.e., the subset of case (I) in which h has SM like couplings
and its mass lies within 113 GeV mh 117 GeV. The points corresponding to case (III),
for which an interpretation of the LEP excess in terms of production of the heavier CP-even
Higgs boson is possible, are displayed as blue stars (indicated by arrows in the plots).
The little black dots indicate parameter points which, while in principle possible in the
mSUGRA scenario, are rejected because of the experimental and theoretical constraints
discussed above.
As a general feature, Fig. 1 shows that mh sharply increases with tan in the region
of low tan , while for tan 10 the mh values saturate. Values of tan 60 are not
allowed due to the REWSB constraint. The LEP2 Higgs boson searches exclude the
models with mh 113 GeV and tan 50. This is contrary to the general LEP2 Higgs
20
Fig. 2. Allowed parameter space within the mSUGRA scenario in the MA tan plane for the three cases defined
in Section 3.1.
boson searches in the mmax

scenario [15,34,72], where the exclusion bound on the SM
h
Higgs boson mass applies to mh only for tan 8. For larger values of tan and small
MA in the unconstrained MSSM a suppression of the hZZ coupling is possible, giving
rise to a reduced production rate compared to the SM case. In the mSUGRA scenario a
significant suppression of sin2 ( eff ) (i.e., the hZZ coupling) occurs only for a small
allowed parameter region with tan 50, see Fig. 3. This feature can be understood
from the correlation between MA and tan shown in Fig. 2. Small values of MA with
MA 150 GeV, which are necessary for values of sin2 ( eff ) 1, are only possible
for tan 50. For tan 45 we find that MA is always larger than about 300 GeV, giving
thus rise to a SM like behavior of the hZZ coupling.
As one can see in Figs. 1, 2 case (III) can indeed be realized in the mSUGRA scenario
in a small parameter region where 50 tan 55, 103 GeV mh , MA 113 GeV and
mH = 115 2 GeV. It should be noted, however, that this parameter region is close to
the exclusion bounds obtained at Run I of the Tevatron [87] (which, as discussed above,
we have not imposed in the present analysis). With the upcoming results from Run II
of the Tevatron it should be possible to fully cover the parameter space compatible with
case (III).
21
Fig. 3. The values for sin2 ( eff ) realized in the mSUGRA scenario are given in the M1/2 tan plane.
From Fig. 1 one can read off an upper bound8 on the light CP-even Higgs boson mass
in the mSUGRA scenario of
mmax
124 GeV
h
(mSUGRA).
(23)
Values close to this upper limit on mh are reached in a large region of moderate and large
values of tan , 20 tan 50.
A lower bound on tan is inferred in the mSUGRA scenario,
tan 3.3 (mSUGRA).
(24)
It should be noted that the two bounds quoted here refer to mt = 175 GeV.
The upper bound on mh is about 6 GeV lower than the one in the unconstrained
MSSM [4,34], and the limit on tan is also more restrictive. This is caused by the fact
that not all parameter combinations of the unconstrained MSSM can be realized in the
mSUGRA scenario. In order to obtain the largest values for mh , in particular, large values
of the parameter Xt ,
Xt At / tan ,
(25)
are necessary, which appears in the off-diagonal element of the t mass matrix. Nonlogarithmic genuine two-loop contributions to mh give rise to an asymmetry with respect to
8 This bound is 3 GeV lower than the one obtained in Ref. [26] due to additional constraints imposed in the
present analysis, see Section 2.6.
22
Fig. 4. Allowed parameter space within the mSUGRA scenario in the plane of the mixing parameter in the t
sector, Xt , and tan for the three cases defined in Section 3.1.
the sign of Xt , and the maximum value obtained for mh is about 5 GeV higher for Xt > 0
than for Xt < 0 [4,90].
In Fig. 4 the allowed parameter space in the Xt tan plane of the mSUGRA scenario is
depicted for the three cases discussed above. The figure shows that the mSUGRA scenario
strongly favors negative values of Xt . The absence of models with large positive Xt is the
main reason for the decrease in the upper bound within the mSUGRA compared to the
unconstrained MSSM.
Concerning the lower bound on mh , we find the same bound as in the unconstrained
MSSM, i.e., mh 91 GeV. As discussed above, mh values below 113 GeV being
compatible with the LEP exclusion bounds are only possible in a small parameter region
with tan 50 in the mSUGRA scenario.
We now turn to the restrictions on the parameter space of the underlying mSUGRA
parameters, M0 , M1/2 and A0 , which are obtained if the LEP excess is interpreted as a
signal, i.e., for the cases (II) and (III). Figs. 5, 6 show that the cases (II) and (III) result in
similar allowed regions of parameter space. While for M0 the whole range up to 1 TeV is
allowed, M1/2 is restricted to M1/2 650 GeV, see Fig. 5. In Fig. 6 the cases (II) and (III)
are shown in the M0 A0 plane. |A0 | is restricted to |A0 | 2M0 . For the special case
A0 = 0 we find that M0 is bounded from above by M0 700 GeV.
23
Fig. 5. Cases (II) and (III) in the mSUGRA scenario are shown in the M1/2 A0 plane.
We find that M1/2 values as low as 100 GeV (these values are only obtained for not
too small M0 , M0 300 GeV; the lower bound on M1/2 for smaller values of M0 is about
M1/2 200 GeV) are compatible with the interpretation of the LEP excess as a signal of
the lightest CP-even Higgs boson in the MSSM. This result is in contrast to the analysis in
Ref. [91], where for mt = 175 GeV and mh 113 GeV a lower bound of M1/2 310 GeV
has been found for A0 = 0, while we obtain a lower bound of M1/2 200 GeV, see Fig. 5.
The main difference between our result and the one obtained in Ref. [91] can be traced to
the inclusion of genuine non-logarithmic two-loop corrections in the result for mh in the
present paper (see also Ref. [80]). The lower values found for M1/2 in the present analysis
give rise to a different low energy spectrum for the SUSY particles. As an exemplary case,
we find that M0 M1/2 200 GeV and 400 GeV A0 400 GeV is compatible with
mh = 115 2 GeV together with all the constraints listed in Section 2.6.3. In this example
the heaviest SUSY particle is the gluino with a mass of 590 GeV and all the scalar quarks
have masses about 500 GeV or lower, depending on A0 and on tan . The SUSY spectrum
in the three soft SUSY-breaking scenarios will be analyzed in more detail in Section 3.5.
Finally, we investigate the hbb coupling in the mSUGRA scenario in comparison with
the SM case, where the decay into b quarks is the dominant decay channel of the Higgs
boson. The hbb coupling is mainly altered in two ways compared to the SM: it has an extra
factor sin eff /cos and it receives a correction 1/(1 + mb ), see Eq. (20).
24
Fig. 6. Cases (II) and (III) in the mSUGRA scenario are shown in the M0 A0 plane.
Fig. 7 shows the different values of sin2 eff /cos2 (which enters the h bb decay
rate) realized within the mSUGRA scenario in the M1/2 tan plane. The figure shows
that a significant enhancement of the hbb coupling is possible over a wide range of the
mSUGRA parameter space. In these parameter regions the sensitivity in the Higgs search
via the h bb channel is in general slightly increased compared to the SM case. On the
other hand, the increase in the h bb partial width leads in general to a reduced branching
ratio of h , making thus the search via this channel at the LHC more difficult [80].9
It can furthermore be seen in Fig. 7 that a significant suppression of the hbb
coupling is only possible in a small fraction of the mSUGRA parameter space. Values
of sin2 eff /cos2 < 0.7 are only obtained in the parameter region tan 50.
In Fig. 8 the quantity mb is analyzed in the mSUGRA scenario. The maximal values
obtained for mb in the mSUGRA scenario are about 0.4 (where values of |mb | > 0.3
are only realized for tan 40). The figure shows that values of sin2 ( eff ) 1
(corresponding to a suppressed coupling of the Higgs to vector bosons and lower allowed
values for mh ) are always correlated in the mSUGRA scenario with negative values
of mb , giving rise to an enhancement of the hbb coupling. Positive values for mb
9 For a similar analysis for the charged Higgs bosons, see Ref. [93].
25
Fig. 7. The values for sin2 eff / cos2 realized in the mSUGRA scenario are given in the M1/2 tan plane.
(corresponding to a suppression of the hbb coupling) are only possible if the Higgs boson
couples with full strength to W and Z.
3.3. mGMSB
For this paper, about 40000 mGMSB models were generated under well defined
hypotheses described in Section 2.3, using the program SUSYFIRE and adopting the
phenomenological approach of Refs. [5153], see also Ref. [27]. Concerning the variation
of the high-energy parameters one should keep in mind that the lower bound on arises
from the requirement that mNLSP 100 GeV, while its upper bound as well as the upper
bound on Mmess originate mainly from the upper bound imposed on SUSY particle masses,
see Section 2.6.3. The upper bounds on and Mmess automatically restrict Nmess from
above, as explained in Section 2.3. The above restrictions yield the following variations of
the high-energy parameters:
104 GeV 2 105 GeV,
1.01 Mmess 105 ,
1 Nmess 8,
26
Fig. 8. The allowed values in the mSUGRA scenario for sin2 ( eff ) and the quantity mb (see Eq. (20)). The
corresponding values for mh are also indicated.
1.5 tan 55,

sign = 1.
(26)
As above, we first analyze the allowed parameter region in the Higgs sector in this
scenario. In Fig. 9 we show the variation of the light Higgs boson mass with respect to
tan for cases (I) and (II) defined in Section 3.1. Case (III), where the LEP excess is
interpreted as a signal of the heavier CP-even Higgs boson in the MSSM, is not realized in
the mGMSB scenario. Fig. 10 shows the allowed parameter space in the MA tan plane.
It can be seen from the two figures that the experimental and theoretical constraints
discussed in Section 2.6 have a bigger effect on the parameter space than in the case
of the mSUGRA scenario. In particular, they significantly influence the upper bound
on mh , which is reduced in this way by about 3 GeV. Like in the mSUGRA case, the
LEP2 Higgs boson searches exclude the models with mh 113 GeV and tan 50.
A significant suppression of sin2 ( eff ) (i.e., the hZZ coupling) occurs only for a small
allowed parameter region with tan 50 (the lower density of points with tan 50
and mh < 113 GeV as compared to Fig. 1 has no direct physical meaning; it is mainly
due to the fact that tan has been varied on a logarithmic scale in the mGMSB scenario,
27
Fig. 9. The light Higgs boson mass mh as a function of tan in the mGMSB scenario. Cases (I) and (II) as
discussed in Section 3.1 are displayed together with the rejected models. Case (I) corresponds to the models that
have passed all theoretical and experimental constraints. Case (II) is the subset of case (I) with mh values in
the region favored by recent LEP Higgs searches, 113 GeV mh 117 GeV, and SM like couplings of the h.
Case (III), where the LEP excess is interpreted as a signal of the heavier CP-even Higgs boson in the MSSM, is
not realized in the mGMSB scenario.
while a linear scale has been chosen in the mSUGRA scenario). This can be understood
from Fig. 10, which shows that values of MA 300 GeV are only realized for tan 50.
Values of tan 55 are not allowed due to the REWSB constraint.
For the upper bound on the light CP-even Higgs boson mass in the mGMSB scenario
we obtain10
mmax
119 GeV
h
(mGMSB).
(27)
values of tan , 20 tan 50.
A lower bound on tan is inferred in the mGMSB scenario,
tan 4.6 (mGMSB).
(28)
10 This bound is 4 GeV lower than the one obtained in Ref. [27], mainly due to additionally imposed
constraints like the naturalness bound on the scalar quark masses, see Section 2.6.3.
28
Fig. 10. Allowed parameter space within the mGMSB scenario in the MA tan plane for cases (I) and (II)
defined in Section 3.1.
As above, the two bounds quoted here refer to mt = 175 GeV.

As in the mSUGRA scenario, the restrictions on the mixing parameter Xt (see Eq. (25))
arising from the parameter correlations in the mGMSB scenario (and the upper bound
imposed in Eq. (18)) are the main effect causing the decrease in the upper bound on mh as
compared to the unconstrained MSSM. Fig. 11 shows that only negative values for Xt are
allowed in the mGMSB scenario. Compared to the mSUGRA case (see Fig. 4) the allowed
parameter region for Xt is smaller and is furthermore shifted towards smaller values of
|Xt |.
Concerning the underlying GMSB parameters, Mmess , Nmess and , no severe restrictions can be deduced for the cases (I) and (II), see Figs. 12, 13. In Fig. 12 we show the
allowed regions in the Mmess plane. The experimental and theoretical constraints imposed in our analysis affect in particular the region of low Mmess and low . In the left
plot of Fig. 13 the allowed regions in the Nmess plane are presented. Lower values of
Nmess correspond to higher values of . This is a consequence of the boundary values
imposed on the physical masses in Eq. (8). The Higgs boson mass constraints cut away a
significant part of the range for each value of Nmess . We only find allowed parameter regions for Nmess 7 (although higher values of Nmess might be allowed if the upper bound
of Mmess / is relaxed). The right plot of Fig. 13 shows the Nmess tan plane. Case (II)
29
Fig. 11. Allowed parameter space within the mGMSB scenario in the plane of the mixing parameter in the t
sector, Xt , and tan for cases (I) and (II) defined in Section 3.1.
corresponds to about the same allowed region as case (I), apart from the values Nmess 4,
where the highest values of tan are not allowed in case (II).
We also investigate the hbb coupling within the mGMSB scenario. In Fig. 14 the
different values of sin2 eff /cos2 realized within the mGMSB scenario are shown in
the tan plane. In contrast to the mSUGRA case, no values of sin2 eff /cos2 < 1
exist, i.e., no suppression of the hbb coupling occurs in this way. As above, a significant
enhancement of the hbb coupling is possible. This applies in particular to the region of the
highest values of tan .
We have also analyzed the quantity mb (see Eq. (20)) within the mGMSB scenario.
The absolute value of mb is smaller in the mGMSB scenario than in the mSUGRA case
and does not exceed |mb | = 0.2. Values of |mb | > 0.1 are only realized for tan 35.
3.4. mAMSB
According to the description presented in Section 2.4 about 50000 models have been
created. The GUT scale parameters have been varied in the ranges
20 TeV maux 100 TeV,
30
Fig. 12. Cases (I) and (II) in the mGMSB scenario are shown in the Mmess plane.
Fig. 13. Cases (I) and (II) in the mGMSB scenario are shown in the Nmess (tan ) plane in the left (right) plot.
The legend shown in the left plot applies to both plots.
31
Fig. 14. The values for sin2 eff /cos2 realized in the mGMSB scenario are given in the tan plane.
0 m0 2 TeV,
1.5 tan 60,
sign = 1.
(29)
The general behavior of the Higgs boson sector has already been described in Ref. [25].
Both mh and MA increase with maux , which determines the SUSY mass scale. mh depends
only very weakly on m0 and tan except for large m0 and small tan . MA gets larger for
larger m0 and tan , although the dependence on tan is rather weak.
In Fig. 15 we show the variation of the light Higgs boson mass with respect to tan for
cases (I) and (II) defined in Section 3.1. Case (III), where the LEP excess is interpreted as
a signal of the heavier CP-even Higgs boson in the MSSM, is not realized in the mAMSB
scenario (as will be explained below). Fig. 16 shows the allowed parameter space in the
MA tan plane.
We find the highest values for mh at tan values of about 35. The experimental and
theoretical constraints discussed in Section 2.6 reduce the upper bound on mh by 12 GeV.
32
Fig. 15. The light CP-even Higgs boson mass mh as a function of tan in the mAMSB scenario. Cases (I)
and (II) as discussed in Section 3.1 are displayed together with the rejected models. Case (I) corresponds to the
models that have passed all theoretical and experimental constraints. Case (II) is the subset of case (I) with mh
values in the region favored by recent LEP Higgs searches, 113 GeV mh 117 GeV, and SM like couplings
of the h. Case (III), where the LEP excess is interpreted as a signal of the heavier CP-even Higgs boson in the
MSSM, is not realized in the mAMSB scenario.
In the mAMSB scenario we do not find a significant suppression of sin2 ( eff ) (i.e.,
the hZZ coupling). As can be seen in Fig. 16, values of MA below 300 GeV are only
realized in the interval 30 tan 40, and MA stays always above about 150 GeV, giving
thus rise to a SM like hZZ coupling. As a consequence, the LEP2 Higgs boson searches
exclude all models with mh 113 GeV. This affects mainly the region tan 10, while
for larger values of tan hardly any mAMSB model results in a Higgs boson mass lower
than 113 GeV, see Fig. 15. Values larger than tan > 60 are not allowed due to the REWSB
constraint.
Fig. 16 furthermore shows that the experimental and theoretical constraints discussed
in Section 2.6 exclude a significant fraction of the parameter space in the MA tan
plane. In general larger values of MA correspond to smaller tan (giving rise to a smaller
bottom Yukawa coupling yb ) because m2Hd , which contributes mainly to MA2 , is larger at
low energies due to the RGE running. For larger tan , smaller MA cannot be realized.
Otherwise the corresponding m0 would be too small to avoid negative slepton masses. The
33
Fig. 16. Allowed parameter space within the mAMSB scenario scenario in the MA tan plane for cases (I)
and (II) defined in Section 3.1.
relatively large values required for MA in the mAMSB scenario, on the other hand, exclude
the possibility of case (III).
The upper bound on the light CP-even Higgs boson mass in the mAMSB scenario is
(for mt = 175 GeV)
mmax
122 GeV
h
(mAMSB).
(30)
values of tan , tan 10.
A lower bound on tan is inferred in the mAMSB scenario (for mt = 175 GeV),
tan 3.2 (mAMSB).
(31)
As above, we have analyzed the allowed values of the mixing parameter in the scalar top
sector, Xt . Fig. 17 shows the allowed parameter space in the plane of the heavier t mass,
mt2 , and Xt . In contrast to the mSUGRA and the mGMSB scenarios positive values for
Xt are preferred. The experimental and theoretical constraints discussed in Section 2.6 are
seen to have a significant effect, limiting the allowed values of Xt to Xt 1.5 TeV. Fig. 17
shows that in the mAMSB scenario Xt is bounded from above, Xt mt2 . This is the main
reason for the decrease in the upper bound on mh compared to the unconstrained MSSM,
34
Fig. 17. Allowed parameter space within the mAMSB scenario in the plane of the heavier scalar top mass, mt ,
2
and the mixing parameter in the t sector, Xt , for cases (I) and (II) defined in Section 3.1.
since the highest values for mh are reached for values of Xt significantly larger than the
heavier t mass (see, e.g., Refs. [4,90]).
In Fig. 18 we present the allowed regions in the plane of the high energy input
parameters m0 and maux . The experimental and theoretical constraints imposed in our
analysis affect, in particular, the region of large m0 and maux . We find no allowed models
with maux 70 TeV. On the other hand, even the smallest values of m0 giving rise to
acceptable slepton masses lead to allowed parameter points in cases (I) and (II).
Concerning the hbb coupling within the mAMSB scenario, we have analyzed
the possible values for sin2 eff /cos2 and mb . We do not find any models with
sin2 eff /cos2 0.9 (values of sin2 eff /cos2 < 1 only occur for tan 40), i.e., the
SUSY contributions entering via eff do not give rise to a significant reduction of the
h b b decay rate in the mAMSB scenario. Values of sin2 eff /cos2 > 10 are possible
for large tan .
The quantity mb (see Eq. (20)) receives large contributions in the mAMSB scenario,
in particular, for large tan and relatively small m0 . This is shown in Fig. 19, where
the different values for |mb | are indicated in the m0 tan plane. We find that positive
values of mb , leading to a suppression of the hbb coupling, are bounded from above by
35
Fig. 18. Cases (I) and (II) in the mAMSB scenario are shown in the m0 maux plane.
mb 0.5. On the other hand, we obtain negative contributions as large as mb 0.8,
giving rise to a strongly enhanced hbb Yukawa coupling.
3.5. The SUSY mass spectra in the three SUSY-breaking scenarios compatible with a
possible Higgs signal at LEP
We finally compare the mass spectra in the three soft SUSY-breaking scenarios
assuming that the LEP excess is due to the production of the h or H boson in the MSSM
(cases (II) and (III)) and briefly discuss possible implications for SUSY searches at the next
generation of colliders. 11 In Figs. 20, 21 we show the spectra of the lightest neutralinos,
the charginos, the scalar top and bottom quarks, the scalar leptons and of the gluino
in the mSUGRA, mGMSB and mAMSB scenarios. The points shown for the mGMSB
and mAMSB scenarios correspond to case (II), while for the mSUGRA models we do not
distinguish in these models between cases (II) and (III) (in general case (III) results in
about the same mass ranges as case (II)).
11 Phenomenological differences as well as characteristic signatures at future experiments between the three
models have also recently discussed in [92].
36
Fig. 19. The allowed values for the quantity |mb | (see Eq. (20)) for different values of m0 and tan in the
mAMSB scenario.
For the lightest neutralino, in the mSUGRA and mAMSB scenarios values as low as
m1 50 GeV are compatible with case (II), while in the mGMSB scenario the lower
bound of m1 100 GeV holds in accordance with Eq. (17). The upper bounds on m1
in Fig. 20 are about 200, 300 and 350 GeV in the mAMSB, mSUGRA and mGMSB
scenario, respectively. For m2 , values as low as about 100 GeV are possible in the
mSUGRA scenario, while we find upper bounds between about 550 and 650 GeV in the
three scenarios.
The lightest chargino is bounded from above by about 200 GeV in case (II) for the
mAMSB scenario, by about 550 GeV in the mSUGRA scenario, and by about 650 GeV in
the mGMSB scenario. For m + we find a lower bound of about 250 GeV in the mSUGRA
2
and mAMSB scenarios, while the lower bound in the mGMSB scenario is about 350 GeV.
Since within the GMSB scenario the LSP is always the gravitino,
detection of the
lightest neutralino via 10 10 production is possible in this scenario if F is not too large,
while the search in the mSUGRA and mAMSB scenarios has to focus on 10 20 associated
production.
Within the mSUGRA scenario, the neutralino and chargino searches at Run II of the
Tevatron and the LHC will be sensitive to a significant part of the parameter space of the
models shown in Fig. 20 [9496]. A future e+ e linear collider (LC) with a center of mass
37
Fig. 20. The allowed mass ranges for the lightest neutralinos (upper left plot), the charginos (upper right), the
scalar top quarks (lower left) and the scalar bottom quarks (lower right) are shown for the cases (II) and (III) of
the mSUGRA, mGMSB and mAMSB scenarios.
(CMS) energy of s 1 TeV will have a very good chance to observe both the associated
production of 10 20 and the production of the lightest chargino, 1+ 1 [97,98].
In the mGMSB scenario the discovery potential at the next generation of colliders
for gauginos (and to some extent also for squarks) strongly depends on the lifetime and
other properties of the NLSP. The charginos are in general heavier than the two lightest
neutralinos, following the mass relation originating from the condition of a unified gaugino
mass at the high energy scale, m + 2 m 0 m 0 . The Tevatron and the LHC will cover
1
38
Fig. 21. The allowed mass ranges for the gluino (left plot, shown in the mt mg plane) and for the scalar leptons
1
(right plot) are shown for the cases (II) and (III) of the mSUGRA, mGMSB and mAMSB scenarios.
at least part of the parameter space in Fig. 20, and the prospects at a LC for neutralino and
chargino production in the GMSB scenario are very promising.
A peculiar feature of the mAMSB scenario is that the Wino is always lighter than the
Bino [21,99]:
M1 : M2 : M3 2.8 : 1 : 8.3.
(32)
In most of the parameter space, the neutral Wino is the LSP. The NLSP, the charged Wino,
is generically extremely mass degenerate with the LSP and decays, after its production at
a collider, after centimeters into an LSP plus a very soft lepton or pion. The detection of
such a charged Wino poses novel experimental challenges since such events escape the
conventional triggers. The search for SUSY in the Wino LSP scenario has been studied by
several groups [21,100,101]. It was pointed outin Ref. [100] that hundreds of Wino pairs
can be produced at Run II of the Tevatron with s = 2 TeV and L = 2 fb1 . Tens of Wino
pairs can be produced in association with a jet. The accompanying high energy jet works
as a trigger and the detection of 5 events is possible
for Wino masses up to 180 GeV at the
Tevatron. At an LC with a CMS energy of s 1 TeV both the associated production of

10 20 and of 1+ 1 should be observable.
We now turn to the mass spectra of the third generation squarks compatible with
cases (II) and (III). Within the mSUGRA scenario the lowest mass values for the third
generation squarks are possible, around 150 and 450 GeV for mt1 and mt2 , respectively,
and around 300 and 450 GeV for mb1 and mb2 , respectively, see Fig. 20. Within the GMSB
scenario, the low-energy masses generated for colored particles are s , but only g2 for
uncolored particles. Thus, scalar quarks are in general heavier than sleptons or electroweak
gauginos in this scenario. Fig. 20 shows that in the mGMSB and mAMSB scenarios no
third generation squark below about 400 GeV is possible in case (II). Similarly, a gluino
as light as about 300 GeV is possible in the mSUGRA scenario, while the lower bound
on mg is about 200 GeV higher in the mGMSB and mAMSB scenarios, see Fig. 21.
39
As a consequence, the prospects for the production of third generation squarks and the
gluino in these scenarios
are not very high both at Run II of the Tevatron and a LC with
a CMS energy of s 1 TeV. At the LHC, on the other hand, the production of the third
generation squarks and the gluino is guaranteed (it should be noted, however, that this
conclusion relies on the naturalness bound imposed in our analysis, see Section 2.6.3).
Finally we analyze the mass spectrum of the scalar leptons in the three scenarios,
see Fig. 21. In the mGMSB scenario, stringent upper limits on the scalar masses of
m1 400 GeV and m2 600 GeV apply, giving rise to good prospects for production of
scalar at the LHC and a future LC (at Run II of the Tevatron a discovery reach of only
up to 150 GeV is expected [94]). In the mSUGRA and the mAMSB scenarios, on the
other hand, much larger masses of the scalar leptons are possible, and their discovery in
the scenario considered here is not guaranteed at the next generation of colliders.
4. Conclusions
We have analyzed the three most prominent soft SUSY-breaking scenarios, mSUGRA,
mGMSB and mAMSB, regarding their phenomenology in the Higgs sector. We have
discussed the constraints arising from the exclusion limits in the Higgs sector recently
obtained at LEP and the possible implications in the situation where the excess of events
observed at LEP is interpreted as a signal of the light or heavy neutral CP-even MSSM
Higgs boson with mass of about 115 GeV.
In order to obtain the predictions for the Higgs sector in the three scenarios, we
have combined the two-loop RG calculations, employed to derive the low-energy mass
spectrum from the high-energy parameters in mSUGRA, mGMSB and mAMSB, with
Feynman-diagrammatic results up to two-loop order for the Higgs boson spectrum and
the effective mixing angle in the Higgs sector. The possible values for the low-energy mass
spectrum have been obtained by scanning over the fundamental (high-energy) parameters
in the scenarios. In addition to the constraints on the MSSM Higgs sector from the Higgs
search at LEP, which is the main issue in this paper, we have taken into account some
further phenomenological constraints on the low-energy mass spectrum. The lower bounds
on SUSY particle masses obtained from the searches at LEP2 and the Tevatron have
been incorporated as well as the constraints from electroweak precision observables. We
have assumed CP-invariance and conservation of R-parity and have discarded models
giving rise to charge and color breaking minima in the scalar potential or violating the
condition for radiative electroweak symmetry breaking. We have furthermore imposed
a mild naturalness upper bound on the masses of the scalar quarks and the gluino of
1.52 TeV.
As upper bound on the mass of the lightest CP-even Higgs boson (for mt = 175 GeV)
we have found mh 124, 119 and 122 GeV in the mSUGRA, mGMSB and mAMSB
scenario, respectively. In these scenarios the tan values are excluded up to tan
3.3, 4.6 and 3.2, respectively. The upper bound on mh in the three scenarios is significantly
reduced compared to the unconstrained MSSM. This decrease in the upper bound on mh
is, in particular, related to the restrictions imposed on the mixing in the scalar top sector by
the underlying structure of the three scenarios. We have furthermore investigated the Higgs
40
couplings to vector bosons and fermions (in particular, the b quark). Using these results, we
have discussed in which parameter regions a non SM-like behavior of the Higgs production
and decay processes is possible in the three scenarios.
The set of models that passed all constraints (called case (I) in our terminology), was
then further analyzed in view of whether they permit the interpretation of the excess of
events observed at LEP2 as a signal of the light CP-even Higgs boson (case (II)) or
the heavy CP-even Higgs boson (case (III)) with a mass of 115 2 GeV and SM-like
couplings.
While the interpretation of the LEP excess as production of the light CP-even Higgs
boson is possible in all three scenarios, the interpretation as a signal of the heavy CPeven Higgs boson is only possible in the mSUGRA scenario in a small parameter region
with 50 tan 55, which is constrained from the Higgs search results of Run I of the
Tevatron and is also close to the region where no radiative electroweak symmetry breaking
occurs.
Assuming the interpretation of the LEP excess as a Higgs signal in the MSSM
(according to cases (II) and (III)), we have analyzed the restrictions on the parameter space
of the fundamental parameters in the three scenarios. We have furthermore investigated
the corresponding spectra of the SUSY particles in these scenarios in view of the SUSY
searches at Run II of the Tevatron, the LHC and an e+ e LC with center of mass energy of
up to 1 TeV. While for the scenario studied here the Tevatron has only a limited chance to
observe SUSY particles, the LHC can always cover the scalar tops and bottoms (the latter
is related to the naturalness condition imposed on the models in our analysis). A LC offers
very good prospects for gaugino and slepton production. We find that at least a part of the
gaugino and slepton spectrum should be accessible at at LC with center of mass energy of
up to 1 TeV in all three scenarios.
Acknowledgements
We thank K. Olive for helpful communications concerning the comparison of our
results. We furthermore thank L. Covi for helpful discussion on astro-physical constraints.
S.S. has been supported by the DOE grant DE-FG03-92-ER-40701. A.D. would like to
acknowledge financial support from the Network RTN European Program HPRN-CT2000-0014 Physics Across the Present Energy Frontier: Probing the Origin of Mass.
References
[1] G. Kane, C. Kolda, J. Wells, Phys. Rev. Lett. 70 (1993) 2686, hep-ph/9210242;
J. Espinosa, M. Quirs, Phys. Lett. B 302 (1993) 51, hep-ph/9212305.
[2] ALEPH Collaboration, R. Barate et al., Phys. Lett. B 495 (2000) 1, hep-ex/0011045;
L3 Collaboration, M. Acciarri et al., Phys. Lett. B 495 (2000) 18, hep-ex/0011043;
DELPHI Collaboration, P. Abreu et al., Phys. Lett. B 499 (2001) 23, hep-ex/0102036;
OPAL Collaboration, G. Abbiendi et al., Phys. Lett. B 499 (2001) 38, hep-ex/0102036;
For a preliminary compilation of the LEP data presented on November 3rd, 2000, see: P. Igo-Kemenes, LEP
Higgs working group, lephiggs.web.cern.ch/LEPHIGGS/talks/index.html.
41
[3] G. Altarelli, G. Isidori, Phys. Lett. B 337 (1994) 141;

J.A. Casas, J.R. Espinosa, M. Quirs, Phys. Lett. B 342 (1995) 171, hep-ph/9409458;
T. Hambye, K. Riesselmann, hep-ph/9708416;
G. Isidori, G. Ridolfi, A. Strumia, hep-ph/0104016.
[4] S. Heinemeyer, W. Hollik, G. Weiglein, Eur. Phys. J. C 9 (1999) 343, hep-ph/9812472.
[5] S. Heinemeyer, W. Hollik, G. Weiglein, Phys. Rev. D 58 (1998) 091701, hep-ph/9803277;
S. Heinemeyer, W. Hollik, G. Weiglein, Phys. Lett. B 440 (1998) 296, hep-ph/9807423.
[6] M. Carena, J. Espinosa, M. Quirs, C. Wagner, Phys. Lett. B 355 (1995) 209, hep-ph/9504316.
[7] M. Carena, M. Quirs, C. Wagner, Nucl. Phys. B 461 (1996) 407, hep-ph/9508343.
[8] H. Haber, R. Hempfling, A. Hoang, Z. Phys. C 75 (1997) 539, hep-ph/9609331.
[9] R. Hempfling, A. Hoang, Phys. Lett. B 331 (1994) 99, hep-ph/9401219.
[10] R.-J. Zhang, Phys. Lett. B 447 (1999) 89, hep-ph/9808299.
[11] J. Espinosa, R. Zhang, Nucl. Phys. B 586 (2000) 3, hep-ph/0003246.
[12] J. Espinosa, I. Navarro, hep-ph/0104047.
[13] G. Degrassi, P. Slavich, F. Zwirner, hep-ph/0105096.
[14] J. Ellis, D. Ross, Phys. Lett. B 506 (2001) 331, hep-ph/0012067.
[15] M. Carena, S. Heinemeyer, C. Wagner, G. Weiglein, hep-ph/9912223.
[16] H.P. Nilles, Phys. Lett. B 115 (1982) 193;
H.P. Nilles, Nucl. Phys. B 217 (1983) 366;
A.H. Chamseddine, R. Arnowitt, P. Nath, Phys. Rev. Lett. 49 (1982) 970;
R. Barbieri, S. Ferrara, C.A. Savoy, Phys. Lett. B 119 (1982) 343;
H.P. Nilles, M. Srednicki, D. Wyler, Phys. Lett. B 120 (1983) 346;
E. Cremmer, P. Fayet, L. Girardello, Phys. Lett. B 122 (1983) 41;
S. Ferrara, L. Girardello, H.P. Nilles, Phys. Lett. B 125 (1983) 457;
L. Hall, J. Lykken, S. Weinberg, Phys. Rev. D 27 (1983) 2359;
S.K. Soni, H.A. Weldon, Phys. Lett. B 126 (1983) 215;
R. Arnowitt, A.H. Chamseddine, P. Nath, Nucl. Phys. B 227 (1983) 121;
For more details see: S. Weinberg, Supersymmetry, The Quantum Theory of Fields, Vol. 3, Cambridge Univ.
Press, 2000.
[17] For reviews see also: H.-P. Nilles, Phys. Rep. 110 (1984) 1;
H.E. Haber, G.L. Kane, Phys. Rep. 117 (1985) 75;
A.B. Lahanas, D.V. Nanopoulos, Phys. Rep. 145 (1987) 1;
S.P. Martin, hep-ph/9709356, zippy.physics.niu.edu/primer.shtml.
[18] For a review, see: G.F. Giudice, R. Rattazzi, Phys. Rep. 322 (1999) 419, hep-ph/9801271.
[19] L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 79, hep-th/9810155.
[20] G.F. Giudice, M.A. Luty, H. Murayama, R. Rattazzi, JHEP 9812 (1998) 027, hep-ph/9810442.
[21] T. Gherghetta, G.F. Giudice, J.D. Wells, Nucl. Phys. B 559 (1999) 27, hep-ph/9904378.
[22] J.A. Bagger, K. Matchev, D.M. Pierce, R. Zhang, Phys. Rev. D 55 (1997) 3188, hep-ph/9609444.
[23] K.T. Matchev, D.M. Pierce, Phys. Lett. B 445 (1999) 331, hep-ph/9805275;
W. de Boer, hep-ph/9808448.
[24] T.A. Kaeding, S. Nandi, hep-ph/9906342.
[25] S. Su, Nucl. Phys. B 573 (2000) 87, hep-ph/9910481.
[26] A. Dedes, S. Heinemeyer, P. Teixeira-Dias, G. Weiglein, J. Phys. G 26 (2000) 582, hep-ph/9912249.
[27] S. Ambrosanio, S. Heinemeyer, G. Weiglein, hep-ph/0002191;
S. Ambrosanio, S. Heinemeyer, G. Weiglein, hep-ph/0005142.
[28] J. Ellis, G. Ganis, D.V. Nanopoulos, K.A. Olive, Phys. Lett. B 502 (2001) 171, hep-ph/0009355.
[29] M. Carena, H. Haber, S. Heinemeyer, W. Hollik, C. Wagner, G. Weiglein, Nucl. Phys. B 580 (2000) 29,
hep-ph/0001002.
[30] S. Heinemeyer, W. Hollik, G. Weiglein, hep-ph/9910283.
[31] J. Espinosa, R. Zhang, JHEP 0003 (2000) 026, hep-ph/9912236.
[32] S. Heinemeyer, W. Hollik, G. Weiglein, Comput. Phys. Commun. 124 (2000) 76, hep-ph/9812320;
S. Heinemeyer, W. Hollik, G. Weiglein, hep-ph/0002213;
The codes are accessible via www.feynhiggs.de.
42
[33] A. Dabelstein, Nucl. Phys. B 456 (1995) 25, hep-ph/9503443;

A. Dabelstein, Z. Phys. C 67 (1995) 495, hep-ph/9409375.
[34] LEP working group for Higgs boson searches, LHWG Note 2001-2, lephiggs.web.cern.ch/LEPHIGGS.
[35] L. Girardello, M.T. Grisaru, Nucl. Phys. B 194 (1982) 65.
[36] For a review, see: H. Dreiner, hep-ph/9707435.
[37] G. Bhattacharyya, Nucl. Phys. Proc. Suppl. A 52 (1997) 83, hep-ph/9608415.
[38] L.E. Ibez, G.G. Ross, Phys. Lett. 110 (1982) 215;
K. Inoue, A. Kakuto, H. Komatsu, S. Takeshita, Progr. Theor. Phys. 68 (1982) 927;
K. Inoue, A. Kakuto, H. Komatsu, S. Takeshita, Progr. Theor. Phys. 71 (1984) 413;
J. Ellis, D.V. Nanopoulos, K. Tamvakis, Phys. Lett. B 121 (1983) 123;
L.E. Ibez, Nucl. Phys. B 218 (1983) 514;
L. Alvarez-Gaum, J. Polchinski, M. Wise, Nucl. Phys. B 221 (1983) 495;
J. Ellis, J.S. Hagelin, D.V. Nanopoulos, K. Tamvakis, Phys. Lett. B 125 (1983) 275;
L. Alvarez-Gaum, M. Claudson, M. Wise, Nucl. Phys. B 207 (1982) 96.
[39] M. Drees, M.M. Nojiri, Phys. Rev. D 45 (1992) 2482.
[40] V. Barger, M.S. Berger, P. Ohmann, Phys. Rev. D 49 (1994) 4908, hep-ph/9311269.
[41] D.M. Pierce, J.A. Bagger, K. Matchev, R. Zhang, Nucl. Phys. B 491 (1997) 3, hep-ph/9606211.
[42] The scalar particle spectrum and phenomenology of this model has been studied in: A. Dedes, A.E. Faraggi,
Phys. Rev. D 62 (2000) 016010, hep-ph/9907331.
[43] D. Matalliotakis, H.-P. Nilles, Nucl. Phys. B 435 (1995) 115, hep-ph/9407251;
A. Lleyda, C. Muoz, Phys. Lett. B 317 (1993) 82, hep-ph/9308208;
N. Polonsky, A. Pomarol, Phys. Rev. D 51 (1995) 6532, hep-ph/9410231.
[44] Particle Data Group, Eur. Phys. J. C 15 (2000) 1.
[45] S. Martin, M. Vaughn, Phys. Lett. B 318 (1993) 331, hep-ph/9308222.
[46] H. Arason, D.J. Castano, B. Keszthelyi, S. Mikaelian, E.J. Piard, P. Ramond, B.D. Wright, Phys. Rev. D 46
(1992) 3945.
[47] A. Dedes, A.B. Lahanas, K. Tamvakis, Phys. Rev. D 53 (1996) 3793, hep-ph/9504239.
[48] M. Dine, W. Fischler, M. Srednicki, Nucl. Phys. B 189 (1981) 575;
S. Dimopoulos, S. Raby, Nucl. Phys. B 192 (1981) 353;
M. Dine, W. Fischler, Phys. Lett. B 110 (1982) 227;
M. Dine, M. Srednicki, Nucl. Phys. B 202 (1982) 238;
M. Dine, W. Fischler, Nucl. Phys. B 204 (1982) 346;
L. Alvarez-Gaum, M. Claudson, M.B. Wise, Nucl. Phys. B 207 (1982) 96;
C.R. Nappi, B.A. Ovrut, Phys. Lett. B 113 (1982) 175;
S. Dimopoulos, S. Raby, Nucl. Phys. B 219 (1983) 479.
[49] M. Dine, A.E. Nelson, Phys. Rev. D 48 (1993) 1277, hep-ph/9303230;
M. Dine, A.E. Nelson, Y. Shirman, Phys. Rev. D 51 (1995) 1362, hep-ph/9408384;
M. Dine, A.E. Nelson, Y. Nir, Y. Shirman, Phys. Rev. D 53 (1996) 2658, hep-ph/9507378.
[50] S. Dimopoulos, S. Thomas, J.D. Wells, Phys. Rev. D 54 (1996) 3283, hep-ph/9604452;
S. Dimopoulos, S. Thomas, J.D. Wells, Nucl. Phys. B 488 (1997) 39, hep-ph/9609434.
[51] S. Ambrosanio, G.D. Kribs, S.P. Martin, Phys. Rev. D 56 (1997) 1761, hep-ph/9703211.
[52] S. Ambrosanio, G.A. Blair, Eur. Phys. J. C 12 (2000) 287, hep-ph/9905403.
[53] S. Ambrosanio et al., JHEP 0101 (2001) 014, hep-ph/0010081.
[54] P. Fayet, Phys. Lett. B 70 (1977) 461;
P. Fayet, Phys. Lett. B 86 (1979) 272;
P. Fayet, Phys. Lett. B 175 (1986) 471;
P. Fayet, in: S. Ferrara, J. Ellis, P. van Nieuwenhuizen (Eds.), Unification of the Fundamental Particle
Interactions, Plenum, New York, 1980, p. 587.
[55] K. Jacobs, Talk given at the XXXVI Rencontres de Moriond, 11th of March 2001, Moriond, France.
[56] A. Pomarol, R. Rattazzi, JHEP 9905 (1999) 013, hep-ph/9903448.
[57] Z. Chacko, M.A. Luty, E. Ponton, JHEP 0004 (2000) 001, hep-ph/9905390.
[58] B.C. Allanach, A. Dedes, JHEP 0006 (2000) 017, hep-ph/0003222.
[59] E. Katz, Y. Shadmi, Y. Shirman, JHEP 9908 (1999) 015, hep-ph/9906296.
[60] S. Heinemeyer, W. Hollik, G. Weiglein, Eur. Phys. J. C 16 (2000) 139, hep-ph/0003022.
43
[61] S. Heinemeyer, W. Hollik, J. Rosiek, G. Weiglein, Eur. Phys. J. C 19 (2001) 535, hep-ph/0102081.
[62] A. Dedes, unpublished.
[63] A. Dedes, S. Heinemeyer, G. Weiglein, FeynSSG: A program for the minimal supergravity and Higgs
spectrum, in preparation. The code will be accessible via www.feynhiggs.de.
[64] P. Chankowski, S. Pokorski, J. Rosiek, Phys. Lett. B 286 (1992) 307;
P. Chankowski, S. Pokorski, J. Rosiek, Nucl. Phys. B 423 (1994) 423, hep-ph/9303309.
[65] M. Frank, S. Heinemeyer, W. Hollik, G. Weiglein, in preparation.
[66] M. Veltman, Nucl. Phys. B 123 (1977) 89.
[67] A. Djouadi, P. Gambino, S. Heinemeyer, W. Hollik, C. Jnger, G. Weiglein, Phys. Rev. Lett. 78 (1997) 3626,
hep-ph/9612363;
A. Djouadi, P. Gambino, S. Heinemeyer, W. Hollik, C. Jnger, G. Weiglein, Phys. Rev. D 57 (1998) 4179,
hep-ph/9710438;
S. Heinemeyer, G. Weiglein, hep-ph/0102317.
[68] LEP Fest, October 911, 2000, Talks given by F. Cerutti, ALEPH Collaboration, M. Kienzle, L3
Collaboration, R. Hemingway, OPAL Collaboration, V. Hedberg, DELPHI Collaboration.
[69] C. Pagliarone, SUSY Searches at the Tevatron Collider, Proceedings of 13th Topical Conference on Hadron
Collider Physics, Mumbai, India, 1420 January 1999, hep-ex/0011016.
[70] I. Iashvili, hep-ex/0007001.
[71] J.B. de Vivie, hep-ex/9911032.
[72] S. Heinemeyer, W. Hollik, G. Weiglein, JHEP 0006 (2000) 009, hep-ph/9909540.
[73] A. Riotto, E. Roulet, Phys. Lett. B 377 (1996) 60, hep-ph/9512401.
[74] A. Kusenko, P. Langacker, G. Segre, Phys. Rev. D 54 (1996) 5824, hep-ph/9602414.
[75] S. Abel, C.A. Savoy, Phys. Lett. B 444 (1998) 119, hep-ph/9809498;
S. Abel, T. Falk, Phys. Lett. B 444 (1998) 427, hep-ph/9810297;
S. Abel, B.C. Allanach, JHEP 0007 (2000) 037, hep-ph/9909448.
[76] For an overview on bounds see, for example, S. Sarkar, Rep. Prog. Phys. 59 (1996) 1493, hep-ph/9602260.
[77] CLEO Collaboration, M.S. Alam et al., Phys. Rev. Lett. 74 (1995) 2885, as updated in S. Ahmed et al.,
CLEO CONF 99-10;
Belle Collaboration, K. Abe et al., hep-ex/0103042.
[78] H.N. Brown et al., Muon g 2 Collaboration, Phys. Rev. Lett. 86 (2001) 2227, hep-ex/0102017;
A. Czarnecki, W.J. Marciano, Phys. Rev. D 64 (2001) 013014, hep-ph/0102122.
[79] G. Kane, S. King, L.-T. Wang, hep-ph/0010312;
A. Bottino, N. Fornengo, S. Scopel, hep-ph/0012377;
J. Ellis, D.V. Nanopoulos, K.A. Olive, hep-ph/0102331.
[80] J. Ellis, S. Heinemeyer, K.A. Olive, G. Weiglein, hep-ph/0105061, to appear in Phys. Lett. B.
[81] J.L. Feng, T. Moroi, Phys. Rev. D 61 (2000) 095004, hep-ph/9907319;
J.L. Feng, K.T. Matchev, T. Moroi, Phys. Rev. Lett. 84 (2000) 2322, hep-ph/9908309;
J.L. Feng, K.T. Matchev, T. Moroi, Phys. Rev. D 61 (2000) 075005, hep-ph/9909334.
[82] F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321, hep-ph/9604387;
F. Borzumati, C. Greub, T. Hurth, D. Wyler, Phys. Rev. D 62 (2000) 075005, hep-ph/9911245.
[83] J. Ellis, J. Hagelin, D. Nanopoulos, K. Olive, M. Srednicki, Nucl. Phys. B 238 (1984) 453;
H. Goldberg, Phys. Rev. Lett. 50 (1983) 1419;
J. Ellis, T. Falk, G. Ganis, K. Olive, M. Srednicki, Phys. Lett. B 510 (2001) 236, hep-ph/0102098.
[84] D. Lyth, E. Stewart, Phys. Rev. D 53 (1996) 1784, hep-ph/9510204.
[85] L. Covi, J. Kim, L. Roszkowski, Phys. Rev. Lett. 82 (1999) 4180, hep-ph/9905212.
[86] R. Barate et al., ALEPH Collaboration, Phys. Lett. B 499 (2001) 53, hep-ex/0010062.
[87] T. Affolder et al., CDF Collaboration, Phys. Rev. Lett. D 86 (2001) 4472, hep-ex/0010052.
[88] L.J. Hall, R. Rattazzi, U. Sarid, Phys. Rev. D 50 (1994) 7048, hep-ph/9306309.
[89] M. Carena, S. Mrenna, C. Wagner, Phys. Rev. D 60 (1999) 075010, hep-ph/9808312;
M. Carena, S. Mrenna, C. Wagner, Phys. Rev. D 62 (2000) 055008, hep-ph/9907422;
H. Eberl, K. Hidaka, S. Kraml, W. Majerotto, Y. Yamada, Phys. Rev. D 62 (2000) 055006, hep-ph/9912463;
M. Carena, D. Garcia, U. Nierste, C. Wagner, Nucl. Phys. B 577 (2000) 88, hep-ph/9912516.
[90] S. Heinemeyer, W. Hollik, G. Weiglein, Phys. Lett. B 455 (1999) 179, hep-ph/9903404.
44
[91] J. Ellis, G. Ganis, D.V. Nanopoulos, K.A. Olive, hep-ph/0009355;

J. Ellis, hep-ex/0011086.
[92] I. Hinchliffe, P. Richardson, hep-ph/0106212.
[93] A. Belyaev, D. Garcia, J. Guasch, J. Sola, hep-ph/0105053.
[94] S. Abel et al., SUGRA Working Group Collaboration, Report of the SUGRA working group for Run II of
the Tevatron, hep-ph/0003154, and references therein;
R. Culbertson et al., hep-ph/0008070, and references therein.
[95] ATLAS Collaboration, Detector and Physics Performance Technical Design Report, CERN/LHCC/99-15
(1999), see: atlasinfo.cern.ch/Atlas/GROUPS/PHYSICS/TDR/access.html.
[96] CMS Collaboration, see: cmsinfo.cern.ch/Welcome.html/CMSdocuments/CMSplots/.
[97] TESLA TDR Part 3, in: R.D. Heuer, D. Miller, F. Richard, P.M. Zerwas (Eds.), Physics at an e+ e Linear
Collider, hep-ph/0106315, see: tesla.desy.de.
[98] T. Abe et al., American Linear Collider Working Group Collaboration, Resource book for Snowmass 2001,
hep-ex/0106055;
T. Abe et al., American Linear Collider Working Group Collaboration, hep-ex/0106056;
T. Abe et al., American Linear Collider Working Group Collaboration, hep-ex/0106057;
T. Abe et al., American Linear Collider Working Group Collaboration, hep-ex/0106058.
[99] J.L. Feng, T. Moroi, Phys. Rev. D 61 (2000) 095004, hep-ph/9907319.
[100] J.L. Feng, T. Moroi, L. Randall, M. Strassler, S. Su, Phys. Rev. Lett. 83 (1999) 1731, hep-ph/9904250.
[101] C.H. Chen, M. Drees, J. Gunion, Phys. Rev. Lett. 76 (1996) 2002, hep-ph/9512230;
C.H. Chen, M. Drees, J. Gunion, Phys. Rev. D 55 (1997) 330, Erratum, hep-ph/9902309;
J. Gunion, S. Mrenna, Phys. Rev. D 62 (2000) 015002, hep-ph/9906270.

One-loop N-point superstring amplitudes with

manifest d = 4 supersymmetry
Nathan Berkovits a , Brenno Carlini Vallilo a,b
a Instituto de Fsica Terica, Universidade Estadual Paulista, Rua Pamplona 145, 01405-900,
So Paulo, SP, Brazil

b California Institute of Technology, Pasadena, CA 91125, USA
Received 13 November 2001; accepted 19 December 2001
Abstract
The hybrid formalism for the superstring is used to compute one-loop amplitudes with an arbitrary
number of external d = 4 supergravity states. These one-loop N-point amplitudes are expressed
as KobaNielsen-like formulas with manifest d = 4 supersymmetry. 2002 Published by Elsevier
Science B.V.
PACS: 11.30.Pb
1. Introduction
Although some nonperturbative aspects of superstring theory can now be studied
through duality symmetries, there are still many perturbative aspects of superstring
theory which are not well understood. For example, perturbative finiteness of superstring
amplitudes has only been proven by patching together amplitude computations using
the light-cone GreenSchwarz (GS) formalism with amplitude computations using the
RamondNeveuSchwarz (RNS) formalism [1].1 This patching is necessary since the
light-cone GS amplitudes have contact term problems whereas the RNS amplitudes are
only spacetime supersymmetric after summing over spin structures.
E-mail addresses: nberkovi@ift.unesp.br (N. Berkovits), vallilo@ift.unesp.br, vallilo@theory.caltech.edu

(B. Carlini Vallilo).
1 The proof of [2] using the twistor-string formalism is incorrect since the unphysical poles come from the
chiral boson correlation function were not treated in a BRST-invariant manner.
0550-3213/02/$ see front matter 2002 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 5 3 - 8
46
N. Berkovits, B. Carlini Vallilo / Nuclear Physics B 624 (2002) 4562
Another aspect of perturbative superstring theory which is not well developed is the
computation of scattering amplitudes involving more than four external fermions. After
extracting their low-energy contributions, these amplitude computations could be useful
for simplifying analogous computations in QCD [3]. However, contact term interactions in
the light-cone GS formalism and the complicated nature of Ramond vertex operators in the
RNS formalism have made these amplitudes difficult to compute. So to study perturbative
finiteness and to obtain amplitude expressions for more than four external fermions, it
would be nice to have a superstring formalism which does not suffer from the problems of
the GS and RNS formalisms.
Over the last seven years, a hybrid formalism has been developed which combines
the advantages of the GS and RNS formalisms without including their disadvantages.
The hybrid formalism is manifestly spacetime supersymmetric and therefore does not
require summing over spin structures and can easily handle an arbitrary number of external
fermions. Furthermore, in a flat target-space background, the hybrid worldsheet action is
quadratic so scattering amplitudes can be computed using free field OPEs.
In this paper, one-loop scattering amplitudes will be computed using the d = 4 version
of the hybrid formalism which describes the type II superstring2 compactified on any sixdimensional manifold which preserves at least d = 4 supersymmetry [4]. These one-loop
amplitudes will be computed for an arbitrary number of external N = 2, d = 4 supergravity
states which are independent of the compactification. The amplitudes will be expressed as
KobaNielsen-like formulas with manifest N = 2, d = 4 supersymmetry which closely
resemble the formulas computed in [5] for tree amplitudes. Although these one-loop
amplitudes could be computed in principle using the RNS or light-cone GS formalisms,
the problems described earlier have up to now prevented such computations. However, it
will be shown that when all external states are NSNS, the amplitudes agree with the RNS
result.
There are various generalizations of these one-loop computations which might be
possible. One possible generalization would be to compute one-loop amplitudes involving
external compactification-dependent states. Such computations would be interesting for
anomaly analysis but require a better understanding of correlation functions involving
the chiral boson in the hybrid formalism. A second possible generalization would be
to compute one-loop amplitudes using the d = 6 [6] or d = 10 [7] pure spinor versions
of the hybrid formalism which manifestly preserve more spacetime symmetries. And a
third possible generalization would be to use the hybrid formalism to compute multiloop
amplitudes which might be useful for studying perturbative finiteness. Up to now, only
special topological amplitudes [8] which involve trivial correlation functions of the chiral
boson have been computed at higher loops using the hybrid formalism [9].
Section 2 of this paper will review the d = 4 hybrid formalism. Section 3 will use
the correlation functions of the hybrid worldsheet variables on a torus to compute explicit
d = 4 supersymmetric expressions for one-loop N -point amplitudes with external massless
d = 4 states. And Section 4 will prove that these one-loop expressions are gauge-invariant,
2 Although only the type II superstring will be discussed here, all results are easily generalized to one-loop
open or heterotic superstring amplitudes.
47
single-valued, modular invariant, and agree with the RNS one-loop amplitudes when all
external states are in the NSNS sector.
2. Review of d = 4 hybrid formalism
2.1. Worldsheet action
After embedding the RNS superstring in a c = 2 N = 2 superstring and performing a
field redefinition on the worldsheet variables, the superstring can be described in a d = 4
super-Poincar covariant manner using the d = 4 hybrid formalism [4]. This formalism
can be used to describe either the uncompactified superstring or any compactification of
the superstring which preserves at least d = 4 supersymmetry. For the type II superstring,
the worldsheet variables in this formalism consist of the N = 2 d = 4 superspace variables
= 1 to 2, the fermionic conjugate momenta
[x m , L , L , R , R ] for m = 0 to 3 and (, )
[pL , p L , pR , p R ], a left- and right-moving chiral boson [L , R ], and a worldsheet
N = (2, 2) c = 3 superconformal field theory which represents the compactification
manifold.
The worldsheet action in conformal gauge for these fields is

1 m
dz d z x xm + pL L + p L L + pR R + pR R
2
+ SL + SR + SC ,
(2.1)
where SL and SR are the actions for the left- and right-moving chiral bosons, and SC is
the action for the compactification-dependent variables. Note that z versus z is correlated
with L versus R, and not with versus .
As y z, the free-field OPEs of the four-dimensional variables are
x m (y)x n (z) mn log |y z|2 ,
L (y)L (z) log(y z),
pL (y)L (z)
,
y z
.
y z
p L (y)L (z)
,
y z
p R (y)R (z)
y z
pR (y)R (z)
R (y)R (z) log(y z ),
(2.2)
Note that all worldsheet variables are periodic and that the chiral boson cannot be
e R (z) (y z ). It has the same
fermionized since eL (y) eL (z) (y z) and eR (y)
behavior as the negative-energy field that appears when bosonizing the RNS ghosts.
2.2. Worldsheet N = 2 superconformal generators
Using RNS variables, one can construct the twisted c = 2, N = 2 superconformal
J ] = [Tmatter + Tghost, jBRST , b, bc + ] where the L/R index is
generators [T , G, G,
being suppressed in this subsection. In terms of the d = 4 hybrid variables, these generators
48
are mapped under the field redefinition to [4]

1
1
T = x m xm p p + 2 + T C ,
2
2
= e (d)
2+G
C ,
G = e (d)2 + GC ,
G
J = + J C ,
(2.3)
where
i
1 2
1
2,
d = p + x ()
+ ()
2
4
8
i
1
1
d = p + x ( )2 + ( )2 ,
2
4
8
C , J C ] are the twisted c = 3, N = 2 generators of the
x = xm m , and [T C , GC , G
superconformal field theory used to describe the compactification manifold.
As was shown by Siegel [10], d and d satisfy the OPEs

,
d (y)d (z) i
y z
d (y)U (z)
D U
,
y z
U
D
d (y)U (z)
,
y z
,
(y z)2
m
d (y) m (z) i
,
yz
mn
m (z) n (z)
,
(y z)2
d (y) (z)
d (y)d (z) regular,
d (y)d (z) regular,
d (y) (z)
,
(y z)2
d (y) m (z) i
m
,
yz
(2.4)
where

i
m = x m m + ,
2
i
= i ,
D = ,
D
2
2
(2.5)
and U (z) = U (x(z), (z), (z))

is a scalar superfield. The advantage of working with
the variables d , d and m is that they commute with the spacetime supersymmetry
generators

i
1 2
q = dz p x ()
,
2
8

q =

i
1 2
dz p x ( ) .
2
8
49
2.3. Massless compactification-independent vertex operators

Since the hybrid formalism is a critical N = 2 superconformal field theory, physical
vertex operators can be described by U(1)-neutral N = 2 primary fields with respect to the
superconformal generators of (2.3). For massless states of the type II superstring which
are independent of the compactification, the unintegrated vertex operator only depends on
the zero modes of [x m , L , L , R , R ] and is therefore described by the scalar superfield
U (x, L , L , R , R ), which is the prepotential for an N = 2, d = 4 supergravity and tensor
multiplet [11]. The NSNS fields for the graviton, antisymmetric tensor, and dilaton are in

the L L R L component of U , i.e.,

L ][DR , D
]U = m n (hmn + bmn + mn ),
[DL , D
R

==0
(2.6)
and the RR field strengths for the U(1) vector and complex scalar are in the (L )2 L
(R )2 and (L )2 (R )2 components of U , i.e.,

R

L (DR )2 D
U = mn Fmn ,
(DL )2 D
R
==0

L )2 (DR )2 D
U = m m y,
DL (D
R
==0
L are defined as in (2.5) with and variables, and DR and D

R
where DL and D
L
L
are defined as in (2.5) with R and R variables. Note that in standard SU(2) notation for
N = 2 superspace, L = + , L = + , R = , and R = .
For U to be an N = 2 primary field, it must satisfy the constraints
L )2 U = (DR )2 U = (D
R )2 U = m m U = 0.
(DL )2 U = (D
The first four constraints are the N = 2, d = 4 supersymmetric generalization of the usual
polarization conditions, and the last constraint is the equation of motion in this gauge.
Using the OPEs of (2.2), one can compute that the integrated form of the closed superstring
vertex operator is

d 2z V =

R GR (U ) =
L GL G
d 2z G

2
d 2 z H (z) U (z, z ),
(2.7)
where
L
L )2 DL + d (z)(DL )2 D
H (z) = dL (z)(D
L

L + i L (z) D , D
,
+ L (z)DL L (z)D
L
L
2
(2.8)
act on the
| |2 signifies the leftright product, and all superspace derivatives D and D
superfield U .
50
Using the field redefinition to write (2.7) in terms of RNS variables, one can check that
the NSNS fields of (2.6) couple in (2.7) as

p n
q
2
d z V = d 2 z x m + iLm L kp x
+ iRn R kq (hmn + bmn + mn ),
(2.9)
2
which is the usual RNS vertex operator. Furthermore, d z V is invariant up to a surface
term under the linearized gauge transformation
L )2 L + (DR )2 R + (D
R )2 R ,
U = (DL )2 L + (D
(2.10)
which is the N = 2, d = 4 supersymmetric generalization of the gauge transformation

(hmn + bmn + mn ) = m Ln + n Rm . For example, under U = (DL )2 L ,

m

L + i L
z)(DL )2 L
D
V = L D
,
D
L
L H (
2

(z)(DL )2 L
L m m H
= DL D
L
(z)(DL )2 L ).
= z (H
(2.11)
3. One-loop N -point scattering amplitude

3.1. Topological prescription
In [9], a topological prescription was given for computing scattering amplitudes for
any c = 2, N = 2 superconformal field theory. This prescription uses the twisted version
of the N = 2 superconformal field theory in which the worldsheet N = 2 superconformal
ghosts contribute zero central charge and decouple from the scattering amplitudes. For the
c = 2, N = 2 superconformal field theory representing the self-dual string on T 2 R 2 , this
prescription was used by Ooguri and Vafa [12] to explicitly compute the g-loop partition
function dependence on the T 2 moduli.
For the c = 2, N = 2 superconformal field theory which represents the superstring
using the d = 4 hybrid formalism reviewed in Section 2, there is a subtlety in the
prescription caused by the negative energy chiral boson . Like the chiral boson in
the RNS formalism, correlation functions of can have unphysical poles which need to
be treated carefully. Fortunately, as in the special multiloop amplitudes computed in [8,9],
this subtlety can be ignored here because the chiral boson will decouple from the other
worldsheet variables.
The topological prescription for the one-loop type II superstring amplitude is

2

JL JR V1 (z1 , z 1 ) VN (zN , z N ) ,
A = d 2 (2 )2 d 2 z1 d 2 zN

(3.1)
R +J C ) is constructed from the U(1) currents
d 2 w (L +JLC )(
R
where JL JR =
of [4] and signifies the two-dimensional correlation function on a torus. As discussed
51
in [9], this prescription reproduces (up to picture-changing subtleties) the standard oneloop RNS prescription
in the large
Hilbert space. Note that when written in terms of

RNS variables, ( JL JR )2 = ( d 2 w (bL cL + L L )(bR cR + R R ))2 , and this term
is necessary for providing the [bL , cL , L , L ] and [bR , cR , R , R ] zero modes in the large
RNS Hilbert space.
When the external states are d = 4 supergravity states represented by (2.7), the vertex
operators are independent of the variable and the compactification variables. So the
two-dimensional correlation function factorizes into a d = 4 contribution coming from
the [x m , , , p , p ] worldsheet variables which depends on the external states, and a
d = 6 contribution coming from the variable and compactification variables which does
not depend of the external states. Since the d = 4 worldsheet variables are free fields, one
can easily compute their correlation functions on a torus. Although the compactification
variables are not necessarily free fields, they only contribute an overall factor through their
partition function which is independent of the external momenta and polarizations.
3.2. Partition functions on a torus
Since the d = 4 worldsheet variables [x m , , , p , p ] all are periodic free fields, it
is straightforward to compute their partition functions. From the four x m s, the partition

1/2
2in ) is
function is (|( )|2 2 )4 = |( )|8 22 , where ( ) = ei/12
n=1 (1 e
the Dedekind eta function. And the partition functions for the fermions [L , L , R , R ] and
[pL , p L , pR , p R ] contribute |( )|16 for correlation functions involving all sixteen
fermionic zero modes. Note that these sixteen fermionic zero modes will come from the
external vertex operators.
The partition functions of the remaining variables depend on the twisted c = 3,
N = 2 superconformal field theory which describes the compactification manifold. For
j
j
example, for the uncompactified superstring, the remaining variables are [x j , x j , , ,
j
L
j
R , R , L , R ], where j = 1 to 3, L/R are fermions of zero conformal weight and L/R

are fermions of +1 conformal weight.
The partition function for the six uncompactified xs provides the factor (|( )|2
j
j
1/2 6
j
j
2 ) = |( )|12 23 . Naively, the partition function for [L , L , R , R ] is zero

j j
j j
since ( JL JR )2 = ( (L + L L )(R + R R ))2 provides at most eight
of the twelve fermion zero modes. However, thepartition function for the field diverges
in the absence of e factors, so the expression ( JL JR )2 needs to be regularized. This
can be easily done by writing
j
1
1 = lim L1 (y)R1 (y)e
(3.2)
L (y)R (y)
L (z)R1 (z)eL (z)+R (z)
yz

and computing 1( JL JR )2 . With this regularization, the partition function for
[L2 , L3 , L2 , L3 ] and [R2 , R3 , R2 , R3 ] gives |( )|8 22 where the eight fermion zero
modes come from (JL JR )2 and the 22 factor comes from integrating (JL JR ) twice
over the torus.
Finally, the partition function for the [L1 , L1 , L ] and [R1 , R1 , R ] variables can be
computed using the same method as was used in [13] for the [L , L , L ] and [R , R , R ]
52
variables in the large RNS Hilbert space. It was shown in [13] that functional integration
over [, , ] variables reproduces the VerlindeVerlinde
prescription of [14] for [, ]

correlation functions if one inserts L (v)R (v)(
L R )g into the g-loop correlation

function. For example, the one-loop [, ] correlation function

L (y) R (y)
L (z) R (z)
(3.3)
is reproduced by the [, , ] correlation function

L (z)+R (z)
.
L (v)R (v)
d 2 w L (w)R (w)
eL (y)R (y)
e
(3.4)
Since only the zero modes of and contribute to (3.4), this correlation function can
be written as

L (z)+R (z)
2 L (y)R (y)
L (z)R (z)eL (y)R (y)
e
,
(3.5)
where the 2 factor comes from the d 2 w integration. But

4

L (y) R (y)
L (z) R (z) = ( )
(3.6)
for odd spin structure, which is the relevant spin structure for the periodic variable.
So by comparing with (3.5) one finds that 1 = |( )|4 21 for the [L1 , L1 , L ] and
[R1 , R1 , R ] partition functions where 1 is defined in (3.2).

Multiplying together the above partition functions, one finds for the uncompactified
superstring that all ( ) factors cancel and

2
6
2
A = d (2 )
(3.7)
d z1 d 2 zN V1 (z1 , z 1 ) VN (zN , z N ) ,
where signifies the correlation function on a torus divided by the partition function,
and the correlation function must include all sixteen zero modes of the d = 4 fermionic
variables. Using techniques similar to those used in [5] for tree amplitudes, (3.7) will now
be explicitly computed.
3.3. KobaNielsen-like formula for one-loop amplitude
To evaluate (3.7) for external massless compactification-independent states, it is
convenient to use (2.7) to write
N
N

2
Vr (zr , z r ) =
|Hr | Ur (zr , z r )
r=1
r=1
N

s=1

9s 9s
9s =9s =0

2 N
N

9r Hr
Ut (zt , z t ) ,
exp

r=1
t =1
(3.8)
53
r are fermionic derivatives which act only on

where Hr is defined in (2.8) and Dr and D
Ur (zr , z r ). For example,
D2
N

Ur (zr , z r ) = U1 (z1 , z 1 )D U2 (z2 , z 2 )
r=1
N

Ur (zr , z r ).
r=3
Furthermore, it will be convenient to introduce the notation

Hr = d (zr )wr + d (zr )w r + (zr )ar + (zr )a r + m (zr )brm ,
(3.9)
where
r )2 Dr ,
r ,
w r = (Dr )2 D
wr = (D

i
r .
brm = m Dr , D
2
ar = Dr ,
r ,
a r = D
(3.10)
Note that the L/R index has been suppressed in these formulas and that | |2 signifies the
leftright product.
The first step in evaluating (3.8) is to eliminate the d variables using the OPEs of
of d , any correlation function involving
(2.4).3 After separating off the zero mode d(0)
d (z) is determined by the conditions that it is a periodic function of z and has the pole
structure determined by the OPEs of (2.4). So (3.8) is equal to

2 N

N

:exp M +
(3.11)
9r d(0) wr :
Ur (zr , z r ) ,

9s 9s
9s =9s =0
s=1
where
M=
r=1
9r d (zr )w r + (zr )a r + m (zr )brm

r

9r wr F (zs zr ) Ds + 9s i (zs )w s i (zs )bs
r,s
9r 9s F (zr zs )wr as
r,s
9r 9s 9t F (zr zs )F (zt zs ) (zs )w s wt wr ,
(3.12)
r,s,t
the function F (z) in (3.12) is the Dirac propagator for odd spin structure
F (z) = z log 1 (z, ),
and the normal ordering symbol : : signifies that Dr is always ordered to the left of
[wr , w r , ar , a r , brm ]. The term proportional to 9r 9s 9t in (3.12) comes from the pole of
d (zt ) with the residue of the pole of d (zr ) at zs . Note that all (z) variables in (3.11)
3 Although one could instead have used the free field OPEs of (2.2) to eliminate the p variables of (3.8), this
would break manifest d = 4 supersymmetry.
54
since there are no more d variables to contract

can be set equal to their zero mode (0)
with.
Although F (z) F (z) 2in under z z + m + n , the correlation function of
. To
(3.11) is single-valued on the torus after integrating out the fermionic zero modes d(0)
see this, suppose that zu zu + m + n for some u. Then M M 2inM, where

M =
9r wr
Du + 9u i (zu )w u i (zu )bu
r

+ 2 (zu )w u
9t wt F (zt zu )
9u wu
Ds + 9s i (zs )w s i (zs )bs
+ 2 (zs )w s

9t wt F (zt zs )
brings down the term ( 9 w )2 , the first term in M does not

Since integrating over d(0)
r r r

contribute since it is proportional to r 9r wr . One can check that the second term in M
contributes

2
Ut (zt , z t )
:M| exp M| :
t

=

:9u wu

2
Ut (zt , z t ) ,
dy d | exp M| :
(3.13)

where the contour integration in dy d goes around all N external vertex operator
locations zr . Deforming this contour off the back of the torus, one finds that (3.13) is
zero, so the correlation function of (3.11) is single-valued on the torus.
can similarly use the OPEs of (2.4) to eliminate d in (3.11) and write
One
N
r=1 Vr (zr , z r ) as

2 N
N

(0)
(0)
9r d wr + d w r + L :
Ur (zr , z r ) ,
:exp

9s 9s
9s =9s =0
s=1
r=1
(3.14)
where
L=

r
9r m (zr )brm +

s w r
9r F (zr zs ) Ds wr + D
r,s

9r 9s F (zr zs )i (zs )w s wr + F (zr zs ) as wr + a s w r

rs

r,s,t
9r 9s 9t bt ws w r F (zt zs )F (zr zt )
9r 9s 9t 9u w t wu ws w r F (zu zr )F (zs zr )F (zt zr ),
55
(3.15)
r,s,t,u
, and the normal ordering symbol : :
all (z) variables are set equal to their zero mode (0)
r is always ordered to the left of Dr , which is always ordered to the
now signifies that D
left of [wr , w r , ar , a r , brm ].

The final step in evaluating N
r=1 Vr (zr , z r ) is to perform the correlation function
over the x m variables. Note that m in (3.15) is equal to x m since and have been
set equal to their zero modes. The correlation function over x m is easily performed using
the scalar Greens function x m (y, y)x
n (z, z ) mn G(y z), where
2 2

2
Im(y z) .
G(y z) = log1 (y z, ) +
2
(3.16)
For example,

2

m

m
iks xm (zs )
cr xm (zr )
e
exp

s
r

1
= exp
krm ksm G(zr zs ) + icrm ksm G(zr zs ) + i crm ksm G(z
r zs )
2
r,s
1
1
crm csm 2 G(zr zs ) crm csm 2 G(zr zs )
2
2

crm csm G(z

r zs )

2
m
m
Im
Im
(csm + iksm zs )
= exp
cr + ikr zr
2
r
s

1
exp icrm ksm F (zr zs ) + crm csm F (zr zs )
2
2

1 m
+ kr ksm log 1 (zr zs , ) ,
2

(3.17)
m
m
where krm in (3.17) is defined by krm = krm so that eikr xm (zr ) = eikr xm (zr ) .
So after performing the correlation function over x m , integrating out the zero modes
, d , , ], and plugging into (3.7), one finally obtains the N = 2, d = 4
of [d(0)
(0) (0) (0)
supersymmetric KobaNielsen-like formula for the one-loop amplitude

N

2
6
2
2
A = d (2 )
d z1 d zN

9s 9s
s=1
9s =9s =0

2
2
2
2
2
d L d L d R d R :| exp C| :

r
9r wLr
2

s
2
9s w Ls

r
9r wRr
2

s
2
9s w Rs
56

2

m
2
m
m
1 (zr zs , )kr ksm
Im K + i
exp
kr zr
2
r
r,s

t (k, L , L , R , R ),
U

(3.18)
where
C=

s w r iksm brm
9r F (zr zs ) Ds wr + D
r,s

9r 9s F (zr zs )F (zs zt )kt w s wr
+
r,s,t

1 m
+ F (zr zs ) as wr + a s w r + br bsm
2

i
9r 9s 9t bt ws w r F (zt zs ) F (zr zs ) F (zr zt )

r,s,t
r,s,t,u
Km =

9r 9s 9t 9u w t wu ws w r F (zu zr ) F (zu zt )
F (zs zr )F (zr zt ),
9r brm im
9r 9s wr w s F (zr zs ),
(3.19)
(3.20)
r,s
m
t (k, , )
is the Fourier transform of Ut (x, , ),
krm is defined by krm = krm in K
U
m

and C, F (z) = z log 1 (z, ), [wr , w r , ar , a r , br ] is defined in (3.10), and the normal
r is always ordered to the left of Dr , which is always
ordering symbol : : signifies that D
ordered to the left of [wr , w r , ar , a r , brm ].
4. Consistency of the one-loop formula

Although the derivation of (3.18) was completely straightforward, the consistency of
this one-loop formula will now be checked by showing that it is invariant under gauge
transformations of the external states, single-valued as a function of the vertex operator
locations, modular invariant, and agrees with the RNS one-loop amplitude when all
external states are in the NSNS sector.
4.1. Gauge invariance
To check the invariance of (3.18) under the gauge transformation of (2.10), first consider
Uu = (DL )2 Lu for the external state at zu . Since DL Uu = 0, one sees from (3.10) that
=w
= a = 0, a = D
, and bum = ikum . So from (3.19) and (3.20),
Lu
wLu
Lu
Lu
Lu

9u
57

2

2
m
m
Im K + i
:| exp C| : exp
kr zr
2
r

9u =0
= : i
F (zu zr )krm bum +

9r F (zr zu ) a u w r + brm bum
9r 9s bu ws w r

F (zu zs ) F (zr zs ) F (zr zu )
r,s

2i
m
m
bum Im K + i
kr zr
+
2
r

2

2
2
m
m
| exp C| exp
Im K + i
:
kr zr
2
r

u w r ibrm kum
=:
F (zu zr )krm kum +
9r F (zr zu ) D
r
9r 9s ku ws w r F (zr
zs )F (zr zu )
r,s

2
m
m
+
kum Im K + i
kr zr
2
r

2

2
2
m
m
| exp C| exp
Im K + i
:,
kr zr
2
r
(4.1)
u , one needs to be careful with normal ordering since

where when changing a u to D

u = :eC D
u : : D
u , eC :
:eC :D

u : :ku
= :eC D
9r F (zr zu )wr eC :
(4.2)
when 9u = 0. One can easily check that (4.1) is equal to

2

2
:| exp C|2 exp

krm zr
Im K m + i
:
zu 9u =0
2
r

m
1 (zr zs , )kr ksm ,
(4.3)
r,s
so the amplitude of (3.18) is gauge invariant up to a surface term. Similarly, one can show
L )2 L + (DR )2 R + (D
R )2 R only change
that the gauge transformations Uu = (D
(3.18) by a surface term.
58
4.2. Single-valued function of vertex operator locations

Under the transformation zu zu + m + n for integer m and n, the function
F (z) = z log 1 (z, ) F (z) 2in.
Since the points zu and zu + m + n are identified on the torus, one needs to check that the
integrand of (3.18)
under
this transformation.

is single-valued
=
=
m
m
9
w
9
w
Using that
r
r
r
r
r
r
r kr = 0 from the [d , d , x ] zero mode
2
integration, one finds that :| exp C| : transforms into

s
D
Ds + w u
:exp C + 2in 9u wu

s
s

9r brm + 9u wu
9s F (zs zt )kt w s
ikum
r
+ ku
s,t
9r 9s F (zr zs )w s wr
r,s
2

:

(4.4)

2
= :exp C + 2nkum K m : ,

s are total derivatives which can be
where the terms proportional
to s Ds and s D

C
ignored after ordering s Ds to the left of e using

:eC
(4.5)
Ds : =
Ds :eC : :
9r F (zr zs )ks w r eC : .
s
r,s
Since the term

2

m
2
m
1 (zr zs , )kr ksm
Im i
exp
kr zr
2
r
r,s
(4.6)
is invariant under zu zu + m + n ,

2

m
2
m
m
1 (zr zs , )kr ksm
exp
Im K + i
kr zr
2
r
r,s
transforms into

2

2
m
m
m
Im K + i
exp 4inkum Im K exp
kr zr
2
r

k m k
1 (zr zs , ) r sm .
r,s
So the product

2

m
2
1 (zr zs , )kr ksm
krm zr
Im K m + i
:| exp C|2 : exp
2
r
r,s
is invariant under zu zu + m + n , implying that the integrand of (3.18) is single-valued.
59
4.3. Modular invariance

Under the modular transformation = 1 and z z = 1 z, the amplitude
of (3.18) should remain invariant. To check this, it is useful to define 9r = 1 9r . Using
2 = | |2 2 ,
F (z, ) = 1 F (z , ) 2iz ,
F (z, ) = 2 F (z , ) 2i 1 ,
(4.7)
one finds that

N

2 6
8
2
2
A = d (2 ) | |
d z1 d zN

9s 9s
s=1
9s =9s =0

2
2
2
2
2
d L d L d R d R :| exp C| :

2
9r wLr
exp
2
9s w Ls

2
krm zr
Im K m + i
2
r
r
2
2
9r wRr
2
9s w Rs

m
1 (zr zs , )kr ksm
r,s
t (k, L , L , R , R ),
U
(4.8)
where C and K m are defined in (3.19) and (3.20).

One can write :exp C: in terms of 9r , zr , and using
:exp C(9r , zr , ):

= :exp
C(9r , zr , ) + 2i
i
9r brm

9r zr
wr

s
Ds + w r

s
D
ksm zs
9r 9s zr F (zs

zt ) zt F (zr zs ) kt w s wr
r,s,t

1
1
9r 9s brm bsm i 1
9r 9s 9t bt ws w r F (zr zs )
2
r,s
r,s,t

1

9r 9s 9t 9u w t wu ws w r F (zs zr )F (zt zu ) :

= :exp
r,s,t,u
C(9r , zr , ) + 2i

iKm

t
where
K m = 1 K m =

r
9r brm im

r,s
ktm zt
1 m
Km
K
2

9r 9s wr w s F (zr zs , )
:,
(4.9)
(4.10)
60

and (4.5) has been used to order s Ds to the left of :exp C: .
Since (4.6) is modular invariant,

2

m
2
m
m
1 (zr zs , )kr ksm
Im K + i
exp
kr zr
2
r
r,s

2

2

(2 )1 (2 )1 Km
= exp
Km
+ 2iKm
ksm zs

2
s

2

m
2
m
m
1 (z z , )kr ksm
exp Im K + i
kr zr
r
s
2
r
r,s

2

= exp i Km
Km
+ 2iKm
ksm zs

s

2

m
2
1 (z z , )kr ksm .
exp Im K m + i
krm zr
r
s
2
r
r,s
So

2

2
m
2
1 (zr zs , )kr ksm
:exp C(9r , zr , ) : exp
krm zr
Im K m + i
2
r
r,s

2

2
2
= :exp C(9r , zr , ) : exp Im K m + i
krm zr
2
r

m

kr ksm

1 (zr zs , )
,
r,s
which implies using (4.8) that A is modular invariant.

4.4. Equivalence with RNS amplitude for external NSNS states
In this final subsection, the one-loop amplitude of (3.18) will be shown to be equivalent
to the RNS prescription when all external d = 4 states are in the NSNS sector. The method
used for evaluating the RNS amplitude will be an N -point generalization of the four-point
one-loop computation in [15].
In the RNS formalism, the one-loop amplitude for N external massless d = 4 NSNS
states can be written as

N

2
2
2
2
d w bL cL (w)bR cR (w)
ARNS = d (2 )
d zr
r=1
spin
N

r=1

VrRNS (zr , z r )
(4.11)
61

where VrRNS
is defined in (2.9), d 2 w bL cL (w)bR cR (w)
provides the fermionic ghost zero
modes, and spin signifies the two-dimensional correlation function summed over the
four possible spin structures on the torus. Since Vr only involves m in four of the ten
directions, the odd spin structure does not contribute to (4.11).
The first step to evaluating (4.11) is to compute the partition functions of the various
worldsheet fields. For the uncompactified superstring, the [, ] partition function cancels
the [ 8 , 9 ] partition function, the x partition function contributes |( )|20 25 , and the
4
[b
L2, cL , bR , cR ] partition function contributes |( )| 2 where the 2 factor comes from the
d w integration in (4.11).
To compute the partition function for for = 0 to 7, it is convenient to Wick-rotate
to Euclidean space and then use SO(8) triality to map to a MajoranaWeyl SO(8)
spinor variable a for a = 1 to 8. As in the GS formalism, a satisfies periodic boundary
conditions on the torus so there is no need to sum over spin structures. As discussed in [15],
the only subtlety in this SO(8) triality map comes from the odd spin structure because of
the missing zero modes of 8 and 9 . Note that the odd spin structure contributes parityviolating amplitudes involving odd powers of the spacetime 9 tensor. However, as in the
one-loop four-point amplitude computed in [15], there is no parity violating contribution
to one-loop N -point amplitudes when all external states contain d = 4 polarizations and
momenta. So the above subtlety coming from odd spin structures can be ignored.
Since the partition function for eight left- and right-moving periodic fermions
contributes |( )|16 , the scattering amplitude of (4.11) is
N

N

2
6
2
RNS
Vr (zr , z r ) ,
ARNS = d (2 )
(4.12)
d zr
r=1
r=1
where

n

+ i c nq d kq (hmn + bmn + mn ),
VrRNS = x m + iLa mp ab Lb kp x
R
cd R
(4.13)
( mn )ab are constructed from the SO(8) Pauli matrices, signifies the correlation
function divided by the partition function, and the correlation function must include all
sixteen fermionic zero modes of La and Ra .
To compare (4.12) with the one-loop amplitude of (3.7), divide the MajoranaWeyl

, ) = 1 to 2
SO(8) spinor a into the SO(4) SO(4) spinors and for (, ,
where the first SO(4) acts on the unprimed spinor indices and = 0 to 3, while the second
SO(4) acts on the primed spinor indices and = 4 to 7. Then by defining

= + ,

= ,
p = ,
p = + ,
(4.14)
one can equate the correlation function and zero modes of a with the correlation functions
and zero modes of [ , , p , p ]. Furthermore, one can use

a mn ab b = p mn + p mn
to show that the hybrid vertex operator of (2.7) for NSNS states coincides with the RNS
vertex operator of (4.13). So the RNS and hybrid one-loop amplitudes for external NSNS
states have been proven to be equivalent.
62
Acknowledgements
N.B. would like to thank CNPq grant 300256/94-9, Pronex grant 66.2002/1998-9,
and FAPESP grant 99/12763-0 for partial financial support. B.C.V. would like to thank
FAPESP grant 00/02230-3 for financial support and the California Institute of Technology
for hospitality. This research was partially conducted during the period that N.B. was
employed by the Clay Mathematics Institute as a CMI Prize Fellow.
References
[1] S. Mandelstam, The N loop string amplitude: explicit formulas, finiteness and absence of ambiguities, Phys.
Lett. B 277 (1992) 82.
[2] N. Berkovits, Finiteness and unitarity of Lorentz-covariant GreenSchwarz superstring amplitudes, Nucl.
Phys. B 408 (1993) 43, hep-th/9303122.
[3] Z. Bern, D. Kosower, The computation of loop amplitudes in gauge theories, Nucl. Phys. B 379 (1992) 452.
[4] N. Berkovits, Covariant quantization of the GreenSchwarz superstring in a CalabiYau background, Nucl.
Phys. B 431 (1994) 258, hep-th/9404162;
N. Berkovits, A new description of the superstring, in: Proc. of VIII J.A. Swieca Escola de Vera, hepth/9604123.
[5] N. Berkovits, Super-Poincar invariant KobaNielsen formulas for the superstring, Phys. Lett. B 385 (1996)
109, hep-th/9604120.
[6] N. Berkovits, Quantization of the type II superstring in a curved six-dimensional background, Nucl. Phys.
B 565 (2000) 333, hep-th/9908041.
[7] N. Berkovits, Super-Poincar covariant quantization of the superstring, JHEP 0004 (2000) 018, hepth/0001035.
[8] M. Bershadsky, S. Cecotti, H. Ooguri, C. Vafa, Holomorphic anomalies in topological field theories, Nucl.
Phys. B 405 (1993) 279, hep-th/9302103;
I. Antoniadis, E. Gava, K.S. Narain, T.R. Taylor, Topological amplitudes in string theory, Nucl. Phys. B 413
(1994) 162, hep-th/9307158;
M. Bershadsky, S. Cecotti, H. Ooguri, C. Vafa, KodairaSpencer theory of gravity and exact results for
quantum string amplitudes, Commun. Math. Phys. 165 (1994) 311, hep-th/9309140.
[9] N. Berkovits, C. Vafa, N = 4 topological strings, Nucl. Phys. B 433 (1995) 123, hep-th/9407190.
[10] W. Siegel, Classical superstring mechanics, Nucl. Phys. B 263 (1986) 93.
[11] N. Berkovits, W. Siegel, Superspace effective actions for 4D compactifications of heterotic and type II
superstrings, Nucl. Phys. B 462 (1996) 213, hep-th/9501016.
[12] H. Ooguri, C. Vafa, All loop N = 2 string amplitudes, Nucl. Phys. B 451 (1995) 121, hep-th/9505183.
[13] U. Carow-Watamura, Z.F. Ezawa, K. Harada, A. Tezuka, S. Watamura, Chiral bosonization of superconformal ghosts on Riemann surface and path integral measure, Phys. Lett. B 227 (1989) 73.
[14] E. Verlinde, H. Verlinde, Multiloop calculations in covariant superstring theory, Phys. Lett. B 192 (1987)
95.
[15] J. Polchinski, Superstring Theory and Beyond, String Theory, Vol. 2, Cambridge Univ. Press, 1998.

Models of ScherkSchwarz symmetry breaking

in 5d: classification and calculability
Riccardo Barbieri a , Lawrence J. Hall b,c , Yasunori Nomura b,c
a Scuola Normale Superiore and INFN, Piazza dei Cavalieri 7, I-56126 Pisa, Italy
b Department of Physics, University of California, Berkeley, CA 94720, USA
c Theoretical Physics Group, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA
Received 11 July 2001; accepted 17 December 2001
Abstract
The form of the most general orbifold breaking of gauge, global and supersymmetries with a
single extra dimension is given. In certain theories the Higgs boson mass is ultraviolet finite due
to an unbroken local supersymmetry, which is explicitly exhibited. We construct: a 1 parameter
SU(3)SU(2)U (1) theory with 1 bulk Higgs hypermultiplet, a 2 parameter SU(3)SU(2)U (1)
theory with 2 bulk Higgs hypermultiplets, and a 2 parameter SU(5) SU(3) SU(2) U (1) theory
with 2 bulk Higgs hypermultiplets, and demonstrate that these theories are unique. We compute the
Higgs mass and compactification scale in the SU(3) SU(2) U (1) theory with 1 bulk Higgs
hypermultiplet. 2002 Elsevier Science B.V. All rights reserved.
PACS: 12.60.-i; 12.60.Jv; 12.10.-g
1. Introduction
The origin of symmetry breaking, one of the key questions of particle physics, is largely
unknown. In four dimensions, symmetries can be spontaneously broken by scalar fields,
fundamental or composite. In higher-dimensional theories, a very different geometrical
view is possible: symmetries can be broken by the boundary conditions on a compact space.
While this idea has been known for many years [1], its early application was restricted to
string motivated theories with certain six-dimensional compact spaces [2].
It is remarkable that, until recently, there was no attempt to discover the simplest
extensions of the standard model, or the minimal supersymmetric standard model (MSSM),
E-mail address: yasunori@thsrv.lbl.gov (Y. Nomura).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 4 9 - 6
64
R. Barbieri et al. / Nuclear Physics B 624 (2002) 6380
in which symmetries, such as supersymmetry, PecceiQuinn symmetry and grand unified

gauge symmetry, were broken by this ScherkSchwarz mechanism. While several such
theories now exist [314], the question of their uniqueness is unknown, and is addressed
in this paper.
We consider theories based on a single compact extra dimension. We study the
spacetime symmetries of this dimension in Section 2.1, and construct the most general
form for the breaking of supersymmetry, global symmetry and gauge symmetry in
Section 2.2.
In Section 3 we exhibit the form of the local 5d gauge symmetry and supersymmetry
which are unbroken by orbifolding. These unbroken symmetries depend on the location in
the bulkfor example they are different at the two fixed points of the orbifold. These
unbroken symmetries are crucial since they dictate the form of both bulk and brane
interactions. There has been considerable recent debate [15] about whether the mass of
the Higgs boson in certain theories of this type is finite. In Section 3.5 we argue that these
unbroken local 5d symmetries ensure that there are no quadratic ultraviolet divergences in
the Higgs mass.
In Section 4 we make a complete classification of theories with 5d local supersymmetry
with one or two Higgs doublets in the bulk, with gauge group either SU(3) SU(2) U (1)
or SU(5). There are very few such theories, and we briefly describe some possible locations
for the matter fields. Only a single SU(3) SU(2) U (1) theory with a single bulk Higgs
has been constructed in the literature [7], and this theory is found to be an important special
case of a 1 parameter family of such theories. We explore electroweak symmetry breaking
in this family in Section 5, with particular attention to the Higgs boson mass and the
compactification scale.
The two Higgs theories, with gauge group SU(3) SU(2) U (1) and SU(5), are
shown in Section 5 to each form unique two parameter families of models. The form
of SU(5) breaking is unique, and the form for supersymmetry breaking involves a
single free parameter, and is therefore also highly constrained. Conclusions are drawn in
Section 6.
2. The classification
In this section we construct a classification of supersymmetric field theories in 5
dimensions, where the physical space of the fifth dimension is an orbifold of finite size.
2.1. 5-dimensional spacetime
We begin by considering the fifth dimension to be the infinite line R 1 . What are the most
general spacetime transformations acting on this line, which can be used to compactify the
spacetime by identifying points transforming into each other under these operations? One
of them is a translation T (2R) which induces y y + 2R. When we identify the points
connected by this transformation, that is y + 2R with y, it compactifies R 1 to the circle
S 1 = R 1 /T . The other possibility is a parity Z(y0 ) which reflects the line about y = y0 .
An identification using this operation, that identifies (y y0 ) with y y0 , produces
65
an orbifold which is the half line, R 1 /Z2 . This identification involves the choice of a
special point, y0 , which is a fixed point under the transformation. This parity alone does not
compactify the space. No further independent spacetime identifications can be made on the
line (if there are several commensurate translations, we take T to be the one of lowest R).
In this paper, we are interested in the case that both translation and parity identifications are
made. In this case the physical space can be taken to be 0 (y y0) R, corresponding
to the orbifold S 1 /Z2 .
Let be a column vector representing all fields of the theory. The action of these
transformations on the fields can be written as
T (2R)[(y)] = T 1 (y + 2R),
(1)
Z(0)[(y)] = Z(y),
(2)
where we have chosen y0 = 0. Acting with Z twice produces the identity, so that this is
a Z2 transformation, with Z 2 = 1. An identification under these operations are made by
imposing the conditions T (2R)[(y)] = (y) and Z(0)[(y)] = (y), that is
(y + 2R) = T (y),
(3)
(y) = Z(y).
(4)
This identification makes sense only when the bulk action is invariant under the operations
(y) T (2R)[(y)] and (y) Z(0)[(y)], since otherwise physics is not the same
on all equivalent pieces of the line of length R.
The simultaneous imposition of both T (2R) and Z(0) is not automatically consistent,
because the spacetime motion induced by T (2R)Z(0) is identical to that induced by
Z(0)T 1 (2R). Consistency therefore requires that the field transformation is the same
nomatter which choice is made; thus we require T Z = ZT 1 , or
ZT Z = T 1 .
(5)
Thus the most general spacetime symmetries can be taken to be a reflection y y, under
which the fields transform as a Z2 , and a translation, under which the fields transform as
Eq. (1) with any symmetry T of the action, as long as Eq. (5) is satisfied.
The compound transformation T (2R)Z(0), induces the spacetime motion y R
(y R), which is a reflection about the point y = R. Its action on the fields is
Z = T Z, and from Eq. (5) we discover that Z 2 = 1, so that T (2R)Z(0) = Z(R)
is a reflection about R which also induces a Z2 transformation on the fields. One can
choose to describe the compactification in terms of the identifications Z(0) and T (2R)
or equivalently by Z(0) and Z(R); the orbifolds S 1 /Z2 and R 1 /(Z2 , Z2 ) are equivalent
[7]. While the physical space is the line segment 0 < y < R, we have found it convenient
to assemble four such equivalent neighboring segments into a circle of circumference 4R,
as shown in Fig. 1. The utility of this construction is to provide a diagrammatic view of
Z(0) and Z(R) as reflections about orthogonal axes with fixed points O and O .
In general Z and Z do not commute. In the special case that they do, T 2 = 1, so that T
is also a Z2 transformation. Acting twice with T (2R) induces a complete revolution of
the circle of Fig. 1, so that, in this commuting case, the eigenfunctions of Eqs. (3), (4) are
66
Fig. 1. A diagrammatic representation of Z(0) and Z( R) as reflections about y = 0 and y = 0, with

y = y R.
single-valued on this circle:

(+, +) : cos[ny/R],
T = +1
(, ) : sin[ny/R],

(+, ) : cos[(n + 1/2)y/R],
T = 1
(, +) : sin[(n + 1/2)y/R],
(6)
where (, ) refer to the (Z, Z ) parities.

Given a specific field content of a theory, a complete list of the possible forms for Z and
Z can be obtained. As an illustrative example, consider a theory with N complex scalars,
assembled into a vector . A basis can be chosen such that Z = P is a diagonal matrix.
However, in this basis Z is in general non-diagonal:
Z (i ) = U P U ,
(7)
where P is diagonal. The N N unitary matrix U (i ) describes the relative orientation

of the field bases which diagonalize Z and Z , and depends on a set of continuous parameters i .
The number of physical parameters i is less than N 2 , and depends on the numbers of
positive and negative eigenvalues in P and P . Two cases will be of particular importance
to us. If either P or P is proportional to the identity, then U can be rotated away, and there
is no need to introduce any i parameters. Next, consider the case of N = 2. The only
non-trivial case is when neither P or P is proportional
to the unit matrix: P = P = 3 .
A general U matrix would have the form U = exp(i 3i=0 i i ), where 0 is the unit
matrix and 1,2,3 are the Pauli spin matrices. However, the parameters 0,3 drop out of Z ,
while a basis rotation which preserves Z = 3 allows 1 to be rotated away. Hence, the
only non-trivial 2 2 case is described by a single parameter :
Z = 3 ,
Z () = ei2 3 ei2 .
(8)
= e2i2
The description in terms of (Z, T ) is somewhat simpler, since T

= R(2), the
2 2 rotation matrix for angle 2. In this case the field can be expanded in a set of
67
KaluzaKlein (KK) eigenfunctions of Eqs. (3), (4):

y cos[ny/R]+n (x)
(x, y) = R
sin[ny/R]n (x)
R
=

n=
n=0
cos[(n + )y/R]
sin[(n + )y/R]
where n is given by
1
2 (+n (x) + n (x)),
n (x) = +0 (x),
1
2 (+n (x) n (x)),
n (x),
(9)
for n > 0,
for n = 0,
for n < 0.
(10)
The special cases = 0 (1/2) give T = 0 (0 ), so that T 2 = 1. In these cases Z and Z

commute, so that the above eigenfunctions Eq. (9) reduce to Eq. (6).
2.2. General form for orbifold symmetries
We consider N = 1 supersymmetric gauge theories in 5d with gauge group G. The
vector multiplet contains components V = (AM , , , ) and the theory contains a set
of hypermultiplets with components H = (, c , , c ). There may be multiple copies
of a hypermultiplet of given gauge charge, and therefore, since supersymmetry allows
only kinetic terms in the bulk, the bulk Lagrangian can possess some flavor symmetry
H . From the 4d viewpoint the theory possesses two supersymmetries, with transformation
parameters = (1 (y), 2 (y)), which we take to be local transformations. The bulk
Lagrangian possesses a global SU(2)R symmetry under which = (1 , 2 ), = (, )
and = (, c ) form doublets.
The symmetry Z(0) induces y y and the supersymmetric kinetic terms then force
relative signs for the parities P of the components inside V or H. In particular one
discovers that, from the 4d viewpoint, Z necessarily breaks N = 2 supersymmetry to
N = 1 supersymmetry. The 5d supersymmetric multiplets are then conveniently assembled
into 4d supersymmetric multiplets with the P charges: V = (V (+), ()) and H =
(H (+), H c()), where V (A , ) is a 4d vector multiplet, whereas ( + iA5 , ),
H (, ) and H c ( c , c ) are chiral multiplets. This action of Z within an N = 2 multiplet
we define as the set of charges 3 . However, this does not give the complete action
of Z. There may be different overall phase rotations for different hypermultiplets, PH .
Finally, even within a hypermultiplet, Z can act differently on different components of an
irreducible gauge multiplet. We label this by PG , which we take to be an element of the
gauge group. If PG is the unit matrix there is no gauge symmetry breaking, otherwise there
is. Hence, we write
Z = 3 PH PG .
(11)
The most general possibilities for PH and PG are given by PH2 = PG2 = 1 so that Z 2 =
(PH PG )2 = 1 (not just PH2 = PG2 = 1). An example of the case with PH2 = PG2 = 1
is provided by G = SU(2) with two iso-doublet hypermultiplets H = (H , Hc ). In this
68
case, to have a non-trivial boundary condition in the gauge space (PG = 1), we have to
take PG = i3 in the space of fundamental representation. (PG = 3 is not an element
of SU(2).) Therefore, PG2 = 1. Then, to have Z 2 = 1, we also have to assign PH2 = 1
for H , for instance, as H iH and Hc iHc . The combined transformation,
PH PG , is written as H 3 H and Hc 3 Hc , which cannot be reproduced in
terms of the parity matrices PH and PG satisfying PH2 = PG2 = 1, since with PH2 = PG2 = 1
the induced transformation is always the same for H and Hc . Indeed, in this SU(2) case,
the transformations for H and Hc must be opposite under PH PG , in spite of the fact
that H and Hc belong to the same representation, 2, under the SU(2). Similar situations
also occur, for instance, in the case of G = SO(10) with vector representations. However,
all explicit models we discuss in this paper are described by PH2 = PG2 = 1, because the
gauge breaking considered in these models are only SU(5) SU(3) SU(2) U (1).
Therefore, we call PH and PG as parity matrices, but it should be understood that the
eigenvalues for them can be i in general.
The argument for Z(0) applies identically to the symmetry Z(R), but the basis
which diagonalizes Z is in general different from that which diagonalizes Z. Choosing
Z diagonal, we immediately find that Z must take the form
Z = U 3 U V PH V W PG W ,
(12)

where PH,G
are diagonal matrices with PH 2 = PG 2 = 1, U, V are unitary matrices, and
W is an appropriate matrix making the action invariant under the operation Z(R) (for
instance, W is a unitary (orthogonal) matrix for G = SU(n) (SO(n))). Note that U is an
element of SU(2)R . Thus the action of 3 and U 3 U is the same on the SU(2)R singlets:
(, c ) (, c ) and (A , + iA5 ) (A , ( + iA5 )), but differs on the SU(2)R
doublets and . From the discussion preceding Eq. (8), there is no loss of generality in
taking 3 = 3 and U = ei2 acting on the SU(2)R doublet space. On the other hand,
the forms for V and W are highly model-dependent. For example, if the flavor group
H = U (1)N , then V = 1. If H contains SU(2) factors, then in the corresponding 2 2
blocks, V is either 0 or ei2 , depending on whether the two hypermultiplets of the same
gauge charge have equal or opposite P parities.
Throughout this paper, we consider theories with local supersymmetry in the bulk.
However the above classification is unchanged in the non-supersymmetric case, as long as
the action of 3 is reinterpreted. It acts as (A , A5 ) (A , A5 ) and (, ) (, ),
where and are the components of any 5d Dirac fermion.
3. Symmetries and symmetry breaking

Every non-trivial entry in Eqs. (11), (12) causes symmetry breaking. One of the two
supersymmetries is broken by 3 , and the other is broken by a non-trivial U . The flavor
symmetry H is broken by PH , PH , V and the gauge symmetry G by PG , PG , W .
69
3.1. Supersymmetry breaking

If the extra dimension is not compactified, the theory is invariant under the local
supersymmetry transformations (x, y) = T (x, y) / , . . . with (x, y) = (1 (x, y),
2 (x, y)) an arbitrary function of spacetime. Compactification with the orbifold boundary
conditions of Eqs. (11), (12) reduces the set of local supersymmetry transformations. The
action of Z and Z in SU(2)R space is given by Eq. (8) so that the theory is invariant under
supersymmetry transformations with the form of Eq. (9)

y cos[ny/R] +n (x)
1 (x, y)
=R
.
(x, y) =
(13)
2 (x, y)
sin[ny/R] n (x)
R
n=0
Although this is a significant restriction, the theory still possesses local 5d supersymmetry.
From the low energy viewpoint, the number of 4d supersymmetries is the number of
independent modes of Eq. (13) having = (x) independent of y. There is at most a
single zero-mode, since the action of both Z and Z necessarily involves 3 . In fact, an
unbroken 4d supersymmetry only results if = 0, in which case it is the mode:

+0 (x)
(x) =
(14)
.
0
For = 0, Eq. (13) has no zero-mode and hence no 4d supersymmetry survives into the
infrared. One supersymmetry is broken by 1/R, the other by /R. For any = 0, the nth
mode is proportional to (cos[(n + )y/R], sin[(n + )y/R])n (x), and has an axis which
rotates with y. For = 1/2, the 4d supersymmetries on the branes at y = 0 and y = R
are orthogonal.
3.2. Global symmetry breaking
The global symmetry H arises from a repetition of hypermultiplets with the same gauge
quantum numbers. If H is generated by T , then the identification by Z(0) breaks those
generators for which [PH , T ] = 0. Similarly, an identification by Z(R) breaks those
generators having [V PH V , T ] = 0. The unbroken global group H is generated by the
set of generators which commutes with both PH and V PH V .
3.3. Gauge symmetry breaking
With a non-compact fifth dimension, the theory is invariant under gauge transformations
of G: AaM = M . a (y) + , with arbitrary gauge transformation parameters . a (y).
Compactification with non-trivial PG implies that the gauge fields split up into two
sets Aa = (Aa+ , Aa ), which are (+, ) under y y. The + modes have generators
which commute with PG : [PG , T a+ ] = 0. Hence the compactified theory possesses only a
restricted gauge symmetry with gauge parameters constrained to satisfy [12]
. a (y) = . a (y).
(15)
Aa+
Aa
On making a KK mode expansion,

have zero-modes while
do not, so that
the low energy 4d gauge group is G+ , generated by T a+ . We frequently say that
70
compactification using the parities PG has induced the gauge symmetry breaking G
G+ . An alternative viewpoint is that the theory on the compact space possesses a restricted
set of gauge transformations, Eq. (15), which are not broken. They do not include zeromode transformations of G/G+ .
A precisely analogous argument applies for the gauge symmetry breaking induced by
PG : G G+ . If W = 1, so that PG and PG are simultaneously diagonalizable, then the
zero-mode gauge bosons correspond to the generators which commute with both PG and
PG , and are, therefore, (+, +) modes. The lightest gauge boson mode for other generators
have masses of order 1/R. These modes are either (+, ), (, +) or (, ). For W = 1,
the mode eigenfunctions are given by Eq. (6).
In the case that W ( ) has a non-zero Euler angle, , further gauge symmetry generators
are broken [W PG W , T a ] = 0, with some previously massless gauge bosons acquiring
mass /R. In this case the KK modes of the local gauge transformations have forms
which depend on the continuous parameter . Thus in general the total structure of gauge
symmetry breaking, G G , is very rich.
3.4. The brane action
The action has both bulk and brane contributions

S = d 4 x dy L5 + (y)L4 + (y R)L4 .
(16)
The form of the bulk action is very tightly constrained by the unbroken local 5d gauge
and supersymmetry transformations discussed above. What interactions are allowed on the
branes at y = 0, R?
The constraints imposed by the local symmetries are found in the following way:
the brane actions L4 , L4 are the most general allowed by the gauge and supersymmetry
transformations that act at y = 0, R. At y = 0 these transformations are:

(x)
. a+ (x)T a+ ,
(17)
(x, 0) =
,
0
while, at y = R, for the case W = 1, the transformations are

(x)
. a+ (x)T a+ ,
(x, R) = R()
.
0
(18)
For W = 1, the form of the gauge transformations at y = R may be more complicated.

For = 0 the supersymmetries on the two branes are differentfor = 1/2 they are
orthogonal.
What restrictions are imposed on the brane actions L4 , L4 by the global symmetries
SU(2)R and H ? These symmetries may be accidental symmetries of the bulk
a consequence of 5d local supersymmetryso that the brane actions need not respect them.
However, if orbifolding leaves some part of the global symmetry unbroken, we may choose
to impose this on both the bulk and brane actions.
The orbifold transformations Z(0, R) may contain non-trivial contributions from the
global symmetries H and SU(2)R . However, this does not restrict the form of L4 , L4 . It
71
only says that H and/or SU(2)R must be symmetries of the bulk action L5 . Of course, the
brane action at any point can only involve fields that are even about that point (or derivatives
of odd fields). However, a brane interaction at y = 0 can transform non-trivially under Z
or, equivalently, T . This transformation simply serves to fix the brane action at y = 2nR
in terms of that at y = 0it does not constrain the action at y = 0.
We conclude that the global symmetries H and SU(2)R do not necessarily place any
restrictions on L4 , L4 , which may be taken to be the most general set of interactions
invariant under the gauge and supersymmetry transformations at y = (0, R).
3.5. Calculability
The short distance divergence structure of the theory must reflect the unbroken local
gauge symmetry and supersymmetry. Since the action was taken as the most general
respecting the local symmetries, all short distance radiative corrections must take the form
of local operators which are already present in Eq. (16). Thus any quantity which is forced
by the local symmetries to vanish at tree level will have finite UV radiative corrections.
Such quantities need not vanish; they may be generated by IR physics.
As an example, consider the case of G = SU(3) SU(2) U (1) with a single Higgs
doublet hypermultiplet in the bulk [7]. The zero-mode structure is precisely that of the
one Higgs doublet standard model. There has been considerable debate recently about the
radiative structure of the zero-mode Higgs boson mass [15], with some arguing that it is
finite and some that it is quadratically divergent. The exact unbroken local supersymmetry,
given by Eq. (13) with = 1/2, is sufficient to guarantee that the mass of the zeromode Higgs boson is radiatively UV finite to all orders in perturbation theory. Those who
claim divergent behavior have apparently not realized that there is an unbroken 5d local
supersymmetry in the theory. The Higgs mass is non-zero because at distances larger than
1/R there are non-local IR contributions. In the low energy effective theory (in this case
the standard model) these contributions appear to be quadratically divergent (the usual
top loop contribution to the Higgs mass). However, at shorter distances the locality of the
fifth dimension becomes operative and removes the divergence. In calculating explicit loop
diagrams, this is seen most transparently by going to position space for the extra dimension,
as in Ref. [8]. When using momentum space, the internal propagators are expanded in KK
modes, and the sum of these KK modes must be done in a way which preserves the local
supersymmetry. A simple way of doing this is to include the contributions from all modes.
Whether a quantity is finite and calculable simply depends on whether a local operator
can be written which contributes to it. In any theory with the local supersymmetry Eq. (13),
all magnetic dipole moment operators vanish at tree level and will, therefore, be radiatively
finite. Recently a finite one-loop contribution to b s has been computed [16] in the
theory of Ref. [7]. On the other hand,
the electroweak parameter arises at tree level from
the supersymmetric operator (y) d 4 (H egV H )2 , and hence is subject to quadratically
divergent radiative corrections.
In calculating radiative corrections to such quantities as the Higgs boson mass and
b s one may wonder whether contributions from gravitino exchange are important.
Since we have local supersymmetry in the bulk, such interactions are certainly present.
They cannot change the above arguments about finiteness, and now we argue that they
72
contribute only very small amounts to the finite quantities. The local symmetry structure
of the theory becomes apparent at distances smaller than 1/R, and hence all contributions
from shorter distances are cut off. However, the gravitino interactions are weak at scale 1/R
and do not make a substantial contribution. At some higher energy scale the gravitational
interaction becomes strong, but these local interactions cannot contribute significantly to
the finite quantities.
4. Simple models with bulk Higgs

In this section we consider simple supersymmetric models with G = SU(5) or G =
SU(3) SU(2) U (1), with the Higgs doublet(s) in the bulk. The breaking of electroweak
symmetry is then linked to the physics of the bulk. In SU(5) theories, a crucial role of the
bulk is to accomplish doublettriplet splitting. In the non-unified case, the Higgs is also a
near zero-mode. In all cases we consider the role the bulk plays in breaking supersymmetry.
In general, we are interested in non-trivial U, V , W so that Z and Z are not
simultaneously diagonalized. However, some of the simple theories do have [Z, Z ] = 0,
while in other theories the lack of commutativity is small, so that it is convenient to think
first about the commuting case.
After global and gauge symmetry breaking, there are a collection of H G irreducible
hypermultiplets (, ; c , c ). Given the gauge and global parities PG , PG , PH and PH
of Eqs. (11), (12), the fermion has four possibilities for its (P , P ) parities: (p, p )
with p, p = 1. The (P , P ) parities of all other components of the hypermultiplet are
now fixed in terms of PR exp(2i2 ) = (+1, 1) for = (0, 1/2):

(p, PR p ), (p, p ); c (p, PR p ), c (p, p ) .
(19)
There are four different types of hypermultiplet according to whether p = and t pp =
. If PR = +1 supersymmetry is unbroken, and there is a zero-mode chiral multiplet only
for parities of equal signs (t = +1). If PR = 1 supersymmetry is broken, and for parities
of equal signs (t = +1) the zero-mode is a fermion, while for parities of opposite signs
(t = 1) the zero-mode is a scalar.
Similarly, after gauge symmetry breaking, a gauge boson may have any combination of
parities, A (p, p ), but the other components of the 5d vector multiplet are then given:

A (p, p ), (p, PR p ); + iA5 (p, p ), (p, PR p ) .
(20)
The KK mode expansion for any of these fields is given by its (P , P ) quantum numbers
according to Eq. (6). If t = +1 (1) the KK modes have mass mn = n/R ((n + 1/2)/R).
The fermion masses of the tower are Dirac type.
It is remarkable that in 5d the most general possible supersymmetry breaking is
described by just a single parameter . For arbitrary , but keeping = = 0, the
eigenfunctions of the SU(2)R doublets (, c ) and (, ) pass from Eq. (6) to Eq. (9)
with the eigenvalues shifted by /R

mn /R, non-zero-mode,
mn
(21)
/R,
zero-mode.
73
The gauginos become Majorana and are shifted in mass relative to their gauge boson
partners. Similarly the hypermultiplet scalars are shifted in mass relative to their fermionic
partners. In both cases the mass of the zero-mode is lifted by /R, while the excited
members of the SU(2)R doublets get split in mass by /R relative to the corresponding
SU(2)R singlet states.
4.1. Models with G = SU(3) SU(2) U (1)
We choose the orbifold symmetries to preserve the gauge group, so that PG and PG are
trivial. All the vector multiplets therefore have p = p = 1 in Eq. (20).
The simplest possibility is that there is a single Higgs hypermultiplet in the bulk. From
Eq. (19) we see that if t = pp = 1 for this hypermultiplet, there is a single zero-mode
Higgsino, so that this case is forbidden by anomalies. For t = 1 and supersymmetry
unbroken, RP = 1, there is no zero-mode Higgs boson. Such a situation is hard to reconcile
with observation: supersymmetry is unbroken and the Higgs mass squared has a large
positive value comparable to the masses of the KK excitations of the standard model
gauge particles. The unique theory with 1 Higgs hypermultiplet has t = 1 and = 0.
The case of = 1/2 was studied in Ref. [7]. In this theory the Higgs potential depends
on only 1 unknown parameter, 1/R, and since the Higgs mass is finite it can be predicted:
mh = 127 8 GeV. A deformation of this theory is possible by allowing to deviate from
1/2, so that [Z, Z ] = 0. We study this deformation in Section 5. The unique 1 Higgs
hypermultiplet theory may, therefore, be described by Z, T in the supersymmetry and
Higgs flavor spaces as
Z = 3 1,
T =e
(22)
2i(1/2+)2
1.
(23)
In many theories it is useful to consider the Z, T basis, since the symmetry breaking
is transparently summarized by T . The simplest assignment of matter is for quark and
lepton superfields to all be in the bulk with positive T parity so that they all contain
a single zero-mode fermion. Thus the orbifold quantum numbers in the matter flavor
space are (ZM , TM ) = (+1, +1). Indeed, the requirement that all charged fermions have
Yukawa coupling to the zero-mode Higgs and that the KK modes would not yet have been
discovered makes this all but unique. The only other possibility known to us has uR and
dR superfields located on the branes at y = 0 and y = R, respectively, and the rest of the
matter in the bulk.
The most general theory with two Higgs hypermultiplets is conveniently described in
the supersymmetry and Higgs flavor spaces by
Z = 3 3 ,
T =e
2i2
(24)
2i2
(25)
and involves two free parameters: , .1 This theory was written down by Pomarol and
Quiros [5], who took the view that and were of order unity. At the compactification
1 For Z = the two lightest Higgs modes have the same hypercharge.
3
0
74
scale, 1/R, supersymmetry is broken, so that the theory below 1/R is non-supersymmetric
and must contain a Higgs zero-mode. This happens only for the case = , which was
the focus of their work [5,6]. Such a light Higgs requires a relation between the breaking
of supersymmetry, , and the breaking of PecceiQuinn symmetry, . We have recently
advocated an alternative view [14] where and are taken to be extremely small. In
this case the effective theory below 1/R is the MSSM. The parameters and force
a non-trivial y dependence for the zero-mode Higgs, hu,d , and gauginos, , so that on
compactification they lead to the mass terms

2
( + h.c.) 2 hu hu + hd hd
h u h d + h.c.
2R
R
R

2
2
2 hu hu + hd hd + 2 (hu hd + h.c.).
(26)
R
R
The first line gives the supersymmetry breaking soft masses determined by alone, while
the second line gives the PecceiQuinn breaking terms induced by alone. The third
term is proportional to both supersymmetry and PecceiQuinn symmetry breaking. It is
remarkable that these mass terms correspond precisely to those of the MSSM. The common
scalar and gaugino mass is /R, the parameter is /R and the soft parameter B is
predicted to be 2/R. The signs of the two PecceiQuinn breaking terms are correlated
such that the conventional parameter is negative. It is remarkable that the most general
2 Higgs hypermultiplet theory in 5d, Eqs. (24), (25), leads to the MSSM soft operators,
with a unified origin for both supersymmetry breaking and the parameter. The smallness
of supersymmetry breaking and the parameter are both due to the smallness of the
commutator [Z, T ]. Quarks and leptons can be on either brane, or, if they are in the bulk,
they have orbifold quantum numbers in the matter flavor space of (ZM , TM ) = (+1, +1).
L=
4.2. Models with G = SU(5)

The weakest aspects of conventional 4d grand unified theories are the breaking of the
unified gauge symmetry and arranging the mass splitting of Higgs triplets from Higgs
doublets. Assigning a non-trivial action for the orbifold symmetries in the gauge space, and
taking 1/R to be the scale of gauge coupling unification, opens up new, higher-dimensional
possibilities for grand unification, with orbifold breaking of the gauge group and orbifold
doublettriplet splitting. In the case of 5d, there are two parities Z and Z available for
gauge symmetry breaking. If the gauge group is SO(10), there is no choice for these parities
which gives a set of zero-modes for the 5d vector multiplet corresponding to a successful
weak mixing angle prediction. We therefore confine our attention to the case that the gauge
group is SU(5).
To obtain a theory below 1/R with (approximate) 4d supersymmetry and two Higgs
doublets, one should start with two Higgs hypermultiplets in the 5 of SU(5). The most
general orbifold symmetry which breaks SU(5) to SU(3) SU(2) U (1), and does
not give unwanted zero-modes from the 5d vector multiplet, or from the two Higgs
hypermultiplets, is [14]
Z = 3 3 I5 ,
(27)

T = e2i2 e2i2
I3
0
0
I2
75

,
(28)
where In is the n n unit matrix. This theory has unbroken, local 5d supersymmetry
transformations of the form Eq. (13). It also has unbroken, local 5d gauge transformations.
Those corresponding to the generators of the standard model gauge group, . 321 (y),
have (Z, Z ) = (+, +), while the remaining transformation parameters, . X (y), have
(Z, Z ) = (+, ). These transformation parameters, therefore, have the appropriate KK
mode expansions of Eq. (6). Notice that the full SU(5) gauge transformations are operative
at the brane at y = 0, while only those of SU(3) SU(2) U (1) act at the brane at y = R.
In the case that = = 0 the orbifold does not break 4d supersymmetry or the Peccei
Quinn symmetry. This is the case introduced by Kawamura [10] and extended to include
matter and the unification of gauge couplings [12]. It is important to stress that the orbifold
symmetries in Higgs flavor space must take the form
(ZH , TH ) = (3 , 0 ).
(29)
Other assignments do not lead to zero-mode Higgs doublets. For example, the 3 ensures
that the light doublets are vector-like with respect to the unbroken gauge group.
) can be placed on the brane at y = 0.
Three generations of grand unified matter (T , F
The Yukawa couplings to the bulk Higgs fields are also located at this point, and should,
therefore, be SU(5) invariant.2 These Yukawa interactions do not induce d = 5 proton
decay, because the form of the masses for the Higgs triplets, generated by the orbifolding,
possesses a symmetry which sets the amplitude to zero [12]. The Yukawa couplings lead
to the successful b/ mass relation for the third generation. Similar relations for the lighter
generations can be avoided by mixing with heavy bulk matter [12].
Alternatively, matter may be placed in the bulk [12,13]. A single generation requires
, F
. On the two-dimensional space, (T , T )
two sets of 10 + 5 hypermultiplets: T , T , F

, F
), the orbifold symmetry acts as
and (F
(ZM , TM ) = (0 , 3 ),
(30)
for each generation. The 0 ensures that the light matter is chiral under the unbroken gauge
group, while the 3 ensures that an entire generation, q, u, d, l, e, is massless. Strictly
(d), F
(l).
speaking, the unification of quarks and leptons is largely lost: T (u, e), T (q), F
However, the SU(5) understanding of the quantum numbers of a generation is preserved.
This combination of bulk matter is the smallest which leads to anomaly-free, chiral zeromodes, and it automatically gives the quantum numbers of a generation. There is no proton
decay from SU(5) gauge exchange, and there are no SU(5) fermion mass relations [12].
Whether matter is placed on the brane or in the bulk, the theory beneath 1/R is the MSSM
without supersymmetry breaking or PecceiQuinn symmetry breaking.
The only remaining freedom in the structure of the orbifold symmetry Eqs. (27), (28)
is a non-zero value for and [14]. Since these parameters break 4d supersymmetry and
the PecceiQuinn symmetry, they must be extremely small. As in the case of the standard
model gauge group, they lead to the zero-mode mass terms of Eq. (26).
2 Those of Ref. [11] are not invariant under the SU(5) gauge transformations discussed above.
76
The color triplet Higgsino mass matrix from orbifolding turns off dimension 5 proton
decay, and bulk matter turns off proton decay from the SU(5) gauge interactions and from
scalar Higgs triplet exchange. Hence 1/R could be reduced to the TeV scale.3 While the
precise weak mixing angle prediction is lost, KK modes of standard model particles, as
well as those of X and Y gauge bosons and fermions, could be produced at high energy
colliders. In such grand unified theories at the TeV scale, supersymmetry could be broken
by the orbifold via a large parameter, and there could be one or two Higgs quasi-zeromodes. Above the compactification scale the running of the gauge couplings is dominated
by SU(5) symmetric power law running [12,18]. Thus, for this scheme to be viable, there
must be some large new exotic contributions to the gauge couplings either at or beneath
the compactification scale.
5. SU(3) SU(2) U (1) model with one Higgs doublet
In this section, we investigate radiative electroweak symmetry breaking in one Higgs
doublet theories with G = SU(3) SU(2) U (1). We consider the case where all three
generations of matter and a single Higgs propagate in the 5d bulk. The most general
orbifold boundary conditions are given by Eqs. (22), (23), so that we have a 1 parameter
family of theories parameterized by a real number (0 < 1/2). For any member of
this family, the Higgs effective potential depends on only one free parameter 1/R, so that
the physical Higgs boson mass mh and the compactification scale 1/R are calculable. The
= 0 case corresponds to the model in Ref. [7].
The KK mass spectrum for the = 0 case is given by
n
n+1
h c , , : n + 1/2 ,
,
m,
m
c , h,
mc , hc , :
,
(31)
R
R
R
where n = 0, 1, 2, . . . and m represents q, u, d, l, e. A non-zero value for modifies
the above mass spectrum such that the scalar and gaugino masses are shifted by /R.
In particular, the Higgs boson h obtains a tree-level mass /R, and the two linear
combinations of m
and m
c have split masses of (n + 1/2 )/R. As a result, the Higgs
effective potential depends on and the values for mh and 1/R also depend on .
Radiative electroweak symmetry breaking occurs only when is small. Since the treelevel Higgs mass squared is given by 2 /R 2 and one-loop negative contribution through
the top Yukawa coupling is (1/ 4 )(yt2 /R 2 ), 1/ 2 is required to break electroweak
symmetry. This small perturbs the field dependent masses of the top and stop KK towers
as

1
1
,
mFn = n + arctan(yt H R)
(32)
R

1
1
mBn = n + arccot(yt H R)
(33)
,
R
m, h, A :
3 Grand unified theories at the TeV scale were considered in Ref. [17] with a different mechanism of
suppressing proton decay from the SU(5) gauge interactions.
77
Fig. 2. The physical Higgs boson mass mh as a function of .
where n = , . . . , +, H |h|, and there are one Dirac fermion (Fn ) and two
complex scalars (Bn ) at each level n. With these masses, we can calculate the one-loop
Higgs effective potential Vt (H ) from the topstop loop, using calculational techniques in
Ref. [19]:

cos[2k arctan(yt RH )]
9
Vt (H ) =
16 6 R 4
k5
k=1

cos[2k arccot(yt RH )]
cos[2k ]
(34)
.
k5
Then, together with the tree-level Higgs potential
2 2 g2 + g 2 4
H ,
H +
(35)
8
R2
we can derive the values for mh and 1/R by requiring that H = 175 GeV.
In Figs. 2 and 3, we have plotted the predicted values for mh and 1/R as a function of ,
including one-loop gauge contributions to the quadratic term in the potential. They are both
monotonically increasing functions with respect to . The = 0 case reproduces the values
obtained in Ref. [7]: mh = 127 GeV and 1/R = 731 GeV. Note that the definition of R here
is different from that in Ref. [7] by a factor of 2, so that it corresponds to 1/R = 366 GeV in
the notation of Ref. [7]. (A slight increase of 1/R compared with the previous value comes
from an improved treatment of higher order effects. This also changes the central value for
the estimate of the lightest stop mass to mt = 211 GeV.) As is increased to 0.1, 1/R
approaches infinity meaning that electroweak symmetry breaking does not occur beyond
that value of . It is interesting that we can obtain larger values for 1/R by perturbing the
model of Ref. [7] with small non-zero . It reduces the amount of tuning required to obtain
VH,0 (h) =
78
Fig. 3. The compactification scale 1/R as a function of .
phenomenologically acceptable value of the parameter, since the contribution from KK

towers to the parameter scales as (1/R)2 .
6. Conclusions
In this paper we have given the most general form for the orbifold breaking of
symmetries from a single extra dimension. While the structure of gauge and flavor
symmetry breaking is very rich, the breaking of supersymmetry is described by a single
free parameter. The supersymmetry breaking from all 5d theories is therefore guaranteed
to have a simple form. We have explicitly exhibited the form of the local supersymmetry
transformations which are left unbroken by the orbifolding. All ultraviolet divergences
of the theory must correspond to local operators which respect this unbroken local
supersymmetry. It is this symmetry that results in the ultraviolet finiteness of the Higgs
mass in certain theories.
We have explicitly constructed the most general orbifold symmetries for N = 1
supersymmetric, 5d models with gauge group SU(3) SU(2) U (1) and SU(5), having
either one or two Higgs hypermultiplets in the bulk. There are very few such theories.
There is a unique one parameter family of SU(3) SU(2) U (1) theories with a single
Higgs hypermultiplet. We have studied radiative electroweak symmetry breaking in this
family of theories, and the Higgs boson mass and the compactification scale are shown as
a function of this parameter, , in Figs. 2 and 3, respectively. The special case = 0 gives
a central value for the Higgs mass of 127 GeV [7].
There is a unique two parameter family of SU(3) SU(2) U (1) theories with two
Higgs hypermultiplets. One parameter breaks supersymmetry and the other breaks Peccei
Quinn symmetry. This family was first constructed by Pomarol and Quiros [5] where
79
the compactification scale was taken to be in the TeV region. To obtain a zero-mode
Higgs boson, the two orbifold parameters were taken equal, giving a one-dimensional
parameter space. After radiative electroweak symmetry breaking, the Higgs boson was
found to be lighter than 110 GeV throughout this one-dimensional parameter space, almost
excluding the theory. Theories with a heavier Higgs boson might result when the two
parameters are allowed to differ by a small perturbation. Another possibility is that the
compactification scale is taken much larger than the TeV scale, and both parameters are
taken very small [14]. In this case the theory below the compactification scale is the
MSSM with a constrained form for the soft supersymmetry breaking operators as shown
in Eq. (26). Radiative electroweak symmetry breaking requires that the compactification
scale be in the interval 106 109 GeV.
There is a unique two parameter family of SU(5) theories with two Higgs hypermultiplets, where the orbifolding breaks SU(5) SU(3) SU(2) U (1). In the special case
that the two free parameters vanish, the orbifolding corresponds to that introduced by
Kawamura [10] and developed in Ref. [12]. The theory including the two free orbifold
parameters gives a unified origin for SU(5), supersymmetry and PecceiQuinn breaking
[14].
These three families of theories are the only ones in 5d with the stated gauge groups
and bulk Higgs modes. While each family has variants depending on the location of the
quarks and leptons, we have stressed the tightly constrained form of orbifold symmetry
breaking. In all cases, the group theoretic structure of the symmetry breaking is given by
Eqs. (27), (28), where Z is the orbifold parity and T the translation symmetry. If there is
no SU(5) unification the last space is removed in these equations with appropriate sign
changes in the second space. If there is a single Higgs doublet, then in the second space Z
is +1 and T is 1.
Acknowledgements
We thank Massimo Porrati, Riccardo Rattazzi and Neal Weiner for useful conversations.
Y.N. thanks the Miller Institute for Basic Research in Science for financial support. This
work was supported by the ESF under the RTN contract HPRN-CT-2000-00148, the
Department of Energy under contract DE-AC03-76SF00098 and the National Science
Foundation under contract PHY-95-14797.
References
[1] J. Scherk, J.H. Schwarz, Phys. Lett. B 82 (1979) 60;
J. Scherk, J.H. Schwarz, Nucl. Phys. B 153 (1979) 61.
[2] P. Candelas, G.T. Horowitz, A. Strominger, E. Witten, Nucl. Phys. B 258 (1985) 46;
E. Witten, Nucl. Phys. B 258 (1985) 75;
L.J. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 261 (1985) 678;
L.J. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 274 (1986) 285;
J.D. Breit, B.A. Ovrut, G.C. Segre, Phys. Lett. B 158 (1985) 33;
A. Sen, Phys. Rev. Lett. 55 (1985) 33;
L.E. Ibanez, J.E. Kim, H.P. Nilles, F. Quevedo, Phys. Lett. B 191 (1987) 282.
80
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
I. Antoniadis, Phys. Lett. B 246 (1990) 377.

I. Antoniadis, C. Munoz, M. Quiros, Nucl. Phys. B 397 (1993) 515, hep-ph/9211309.
A. Pomarol, M. Quiros, Phys. Lett. B 438 (1998) 255, hep-ph/9806263.
A. Delgado, A. Pomarol, M. Quiros, Phys. Rev. D 60 (1999) 095008, hep-ph/9812489.
R. Barbieri, L.J. Hall, Y. Nomura, Phys. Rev. D 63 (2001) 105007, hep-ph/0011311.
N. Arkani-Hamed, L.J. Hall, Y. Nomura, D.R. Smith, N. Weiner, Nucl. Phys. B 605 (2001) 81, hepph/0102090.
A. Delgado, M. Quiros, Nucl. Phys. B 607 (2001) 99, hep-ph/0103058.
Y. Kawamura, Prog. Theor. Phys. 105 (2001) 999, hep-ph/0012125.
G. Altarelli, F. Feruglio, Phys. Lett. B 511 (2001) 257, hep-ph/0102301.
L.J. Hall, Y. Nomura, Phys. Rev. D 64 (2001) 055003, hep-ph/0103125.
A. Hebecker, J. March-Russell, hep-ph/0106166.
R. Barbieri, L.J. Hall, Y. Nomura, hep-ph/0106190.
D.M. Ghilencea, H. Nilles, Phys. Lett. B 507 (2001) 327, hep-ph/0103151;
A. Delgado, G. von Gersdorff, P. John, M. Quiros, Phys. Lett. B 517 (2001) 445, hep-ph/0104112;
R. Contino, L. Pilo, Phys. Lett. B 520 (2001) 357, hep-ph/0104130;
Y. Nomura, hep-ph/0105113;
N. Weiner, hep-ph/0106021;
H.D. Kim, hep-ph/0106072.
R. Barbieri, G. Cacciapaglia, A. Romito, hep-ph/0107148.
K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Lett. B 436 (1998) 55, hep-ph/9803466;
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47, hep-ph/9806292.
Y. Nomura, D.R. Smith, N. Weiner, Nucl. Phys. B 613 (2001) 147, hep-ph/0104041.
I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quiros, Nucl. Phys. B 544 (1999) 503, hep-ph/9810410.

Non-universal gaugino masses

in supersymmetric SO(10)
Nidal Chamoun a,b , Chao-Shang Huang a , Chun Liu a , Xiao-Hong Wu a
a Institute of Theoretical Physics, Chinese Academy of Sciences, P.O. Box 2735, Beijing 100080, PR China
b Department of Physics, Higher Institute for Applied Sciences and Technology, P.O. Box 31983,
Damascus, Syria
Received 26 October 2001; accepted 19 December 2001
Abstract
We consider SUSY SO(10) models in which SUSY-breaking occurs via an F -term which does
not transform as an SO(10) singlet. This results in non-universal GUT-scale gaugino masses leading
to a different pattern of sparticle masses from what is expected in the minimal supergravity model
(mSUGRA). We study three breaking chains of SO(10) down to the standard model through SU(4)
SU(2) SU(2), SU(2) SO(7) and flipped SU(5) achieved by the representations 54 and 210
which appear in the symmetric product of two SO(10) adjoints. We examine the phenomenological
implications of the different boundary conditions corresponding to the different breaking chains and
present the sparticle spectrum. 2002 Elsevier Science B.V. All rights reserved.
PACS: 12.10.Kt; 14.80.Ly; 12.10.Dm
Keywords: SO(10) grand unification; Supersymmetry; Gaugino mass
1. Introduction
Grand unification theories (GUT) are among the most promising models for physics
beyond the standard model (SM). Supersymmetry (SUSY) is necessary to make the
huge hierarchy between the GUT scale and the electroweak scale stable under radiative
corrections. There are some experimental evidences for the SUSY GUT. One is the
apparent unification of the measured gauge couplings within the minimal SUSY SM
(MSSM) at scale MGUT 2 1016 GeV [1]. Another is the small neutrino masses
extracted from recent observation of neutrino oscillations [2], which imply that the next
scale of new physics is the GUT scale.
The most simple GUT model is the SU(5). The next simple one is the SO(10) [3] which
will be studied in this paper. Whenever there is no intermediate scale of new physics
E-mail address: wuxh@itp.ac.cn (X.-H. Wu).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 5 2 - 6
82
N. Chamoun et al. / Nuclear Physics B 624 (2002) 8194
between the GUT and electroweak scales, the gauge coupling unification is guaranteed.
SO(10) models have additional desirable features over SU(5) ones. All the matter fermions
in one generation fit into one spinor representation of SO(10). The representation contains
the right-handed neutrino and, thus, provides an interesting framework in view of the
small neutrino masses [4]. The R-parity can be automatically conserved as a consequence
of some gauge symmetry breaking [5]. Higgs fields can be put into any irreducible
representation (irrep) we want since AdlerBellJackiw anomalies are absent for all
representations of SO(10). In addition, SO(10) has more attractive subgroups, such as
SU(5) U (1), SU(4) SU(2) SU(2), etc., thus it has more interesting breaking
patterns than SU(5) has. Practically, it seems that the SUSY SU(5) is not favored by
experiments [6].
Because of SUSY breaking, the MSSM has over hundred soft parameters, such as
gaugino masses, which are determined by the SUSY-breaking mechanism. The minimal
supergravity model (mSUGRA) provides an attractive and economical framework to fix
the soft parameters in the MSSM. In mSUGRA, SUSY is broken in a hidden sector, then
gravitational-strength interactions automatically transmit SUSY breaking to the visible
sector which contains all the SM fields and their superpartners. Furthermore, one assumes
that the Khler potential takes a certain canonical form; as a result, all scalar fields
get the same contribution m20 to their squared scalar masses and all trilinear parameters
have the same value A0 . In addition, one assumes that the gauge kinetic function is a
function of the gauge singlet so that gaugino masses have a universal value m1/2 at
scale MGUT . The resulting weak scale spectrum of superpartners and their couplings can
then be derived in terms of the SM parameters in addition to four continuous plus one
discrete parameters m0 , m1/2 , A0 , tan and sign() provided that the radiative breaking
mechanism of the electroweak symmetry is assumed.
However, these universal boundary conditions adopted in the mSUGRA are simple assumptions about the nature of high-scale physics and may remove some interesting degrees of freedom. Indeed, there exist interesting classes of mechanism in
which non-universal soft SUSY breaking terms can be derived. For instance, stringinspired supergravity or models in extra dimensions can lead to non-universality for
SUSY-breaking parameters at the string unification scale or compactification scale [7].
There exists interesting phenomenology in SUSY models with non-universal gaugino
masses [8].
Non-universal gaugino masses may arise in supergravity models in which a non-minimal
gauge field kinetic term is induced by the SUSY-breaking vacuum expectation value (vev)
of a chiral superfield that is charged under the GUT group [9]. The effect of non-singlet
SUSY preserving vev on gaugino masses was studied in Refs. [10] and [9]. The boundary
conditions for the gaugino masses have been worked out for the case of SU(5) GUT [9] and
their phenomenological implications have been investigated [11]. To our knowledge, there
have not been studies of the non-universal gaugino masses resulting from SUSY-breaking
vev of SO(10) non-singlet chiral superfields and the purpose of this paper is to provide just
such a study.
83
The paper is organized as follows. In Section 2 we present the group-theoretical

results which determine the boundary conditions for the gaugino masses coming from a
condensation of F -component of a chiral superfield of a SO(10) non-singlet which is in the
symmetric Kronecker product of two SO(10) adjoints and we restrict our study to the lowerdimensional representations 54 and 210. Each of these irreducible representations (irreps)
leads to a proper pattern of non-universal gaugino masses depending on which breaking
chain it leads to. For the irrep 54 it can lead to two breaking chains from SO(10) down to the
SM. One chain is through the phenomenologically interesting subgroup SU(4) SU(2)
SU(2) G422 corresponding to PatiSalam model [12] and the other chain is through the
subgroup SU(2) SO(7). We determined the gaugino masses corresponding to these two
chains irrespective of the Higgs multiplet used to break the intermediate subgroup to the
SM. In accordance with the successful MSSM prediction of the gauge coupling unification,
we assume that the breaking of the intermediate stages takes place also around MGUT . As
to the 210 representation, it can lead to many breaking chains [13,14] and we chose to
study the chain through the flipped SU(5) U (1) [15] followed by a breaking via a
10 representation of SU(5) contained in the spinor rep of SO(10) down to the SM. Using
these boundary conditions, one can use the renormalization group equations (RGEs) to
deduce the weak scale values of the sparticle spectrum which we present in Section 3 and
we compare them to mSUGRA case. In calculating the spectrum we take into account all
constraints from the present negative searches of sparticles (superpartners of SM particles
and extra Higgs bosons) at collider experiments and b s as well as the recent data of
the E821 experiment on the muon anomalous magnetic dipole moment. In Section 4 we
end up with concluding remarks.
2. SUSY SO(10) GUTs with non-universal gaugino masses

We discuss the non-universality of soft SUSY-breaking gaugino masses in SUSY SO(10)
GUT.
In this class of models, non-universal gaugino masses are generated by at least a nonsinglet chiral superfield that appears linearly in the gauge kinetic function and whose
auxiliary F -component acquires a vev [9,11]

F ab a b
,
L d 2 fab ()W a W b + h.c.
(1)
M
where the gauge kinetic function is
fAB = f0 (s )AB +

n
fn (s )
n
AB
+
M
with M being a parameter of the mass dimension, s and n are the singlet and nonsinglet chiral superfields, respectively, the a,b are the gaugino fields and the F is the
auxiliary field component of .
In conventional models of supergravity breaking, the assumption that only the singlet
field Fs gets a vev is made so that one obtains universal gauge masses. However,
84
in principle, the chiral superfield which communicates supersymmetry breaking to

the gaugino fields can lie in any representation found in the symmetric product of two
adjoints
(45 45)symmetric = 1 + 54 + 210 + 770,
(2)
where only 1 yields universal masses. Thus the gaugino masses Ma , where the index a =
3, 2, 1, represents the SM SU(3) SU(2) U (1) generators as a whole are, in general,
non-universal at the MGUT scale.
In principle, an arbitrary linear combination of the above representations is also allowed
and here we make two basic assumptions. The first one is that the dominant component
of gaugino masses comes from one of non-singlet F -components. The second one is
that the SO(10) gauge symmetry is broken down at the scale MGUT to an intermediate
group H by a non-zero vev for the scalar component of the non-singlet superfield. In
its turn, H is subsequently broken down to the SM at the same scale MGUT . Only
the non-zero vev of the component of F which is neutral with respect to H yields
gaugino masses since H remains unbroken after SUSY breaking. Depending on which
breaking chain one follows down to the SM, ratios of gaugino masses Ma s at MGUT are
determined by group theoretical factors. We restricted our study to the lower-dimensional
representations 54 and 210 and we discussed several possible breaking chains in the
following subsections.
Before we present our detailed discussions on gaugino masses, a remark is in place.
According to the above recipe that gives gaugino masses at tree level, the SUSY-breaking
vev of the non-singlet superfield is also responsible for the breaking of the gauge symmetry.
Because of the SO(10) breaking down to H , there are heavy gauge supermultiplets which
correspond to the broken generators and receive masses of order of mGUT . However,
the SUSY-breaking effects proportional to the vev of non-singlet F -component split the
heavy gauge supermultiplets so that they behave as messengers which communicate SUSYbreaking to the H gauge supermultiplet (as well as the quark and lepton supermultiplets)
by loop effects. The soft terms (gaugino and squark masses, etc.) generated by the gaugemediated mechanism with gauge messengers have been calculated in Ref. [16]. Applying
their results to our case, the loop-induced soft terms can be neglected compared to those
generated at tree level (i.e., those discussed in the paper) if M MGUT since they are
proportional to (mGUT )/4 which is about 3 103 . In general, the size of M is modeldependent and between mGUT and mPlanck . For example, M mGUT in the M-theory on
S 1 /Z2 , M mstring 4 1017 GeV in the weakly coupled heterotic E8 E8 string theory,
and M mPlanck in general supergravity models. In the paper we limit ourself to the case
of M mGUT .
2.1. The representation 54
Looking at the branching rule for the GUT group SO(10) [17], we see that the
representation 54 can break it into several subgroups (e.g., H = G422 SU(4) SU(2)
SU(2), H = SU(2) SO(7), H = SO(9)). Noting that the choice H = SO(9) would lead to
universal gaugino masses, we choose to study the following breaking chains.
85
2.1.1. SO(10) H = G422 SU(3) SU(2) U (1)

The group G422 corresponds to the phenomenologically interesting PatiSalam model
SU(4)C SU(2)R SU(2)L , where the lepton number is the fourth colour. The branching
rule of the SO(10) represention 54 to G422 is
54 = (20, 1, 1) + (6, 2, 2) + (1, 3, 3) + (1, 1, 1).
(3)
So at the first step of the breaking chain, we assume that the traceless and symmetric 2ndrank tensor 54 representation scalar fields have the non-zero vev
54 = v Diag(2, 2, 2, 2, 2, 2, 3, 3, 3, 3),
(4)
where the indices 1, . . . , 6 correspond to SO(6) SU(4) while those of 7, . . . , 0

(henceforth the index 0 means 10) correspond to SO(4) SU(2) SU(2). To break G422
down to the SM, one may simply choose the 16 Higgs fields since the branching rule of
the representation 16 to SM is
1)4/3 + (1, 2)1 + (1, 1)2 + (1, 1)0 ,
16 = (3, 2)1/3 + (3, 1)2/3 + (3,
(5)
where the number on the lower right denotes the quantum number Y of U (1)Y . The
decomposition of the gauge (super) multiplet 45 of SO(10) under G422 is given by
A(45) = A(15, 1, 1) + A(1, 3, 1) + A(1, 1, 3) + A(6, 2, 2).
(6)
The contents of the gauge multiplets A(15, 1, 1), A(1, 3, 1) and A(1, 1, 3) in SM are,
respectively,
1)4/3 + A(1, 1)0,
A(15, 1, 1) = A(8, 1)0 + A(3, 1)4/3 + A(3,
A(1, 1, 3) = A(1, 3)0,
A(1, 3, 1) = A(1, 1)2 + A(1, 1)2 + A(1, 1)0.
(7)
When the neutral component H (1, 1)0 of the 16 Higgs fields develops a vev H (1, 1)0 =
v G422 will be broken down to SM. Then the gauge multiplets, A(1, 1)0 in A(15, 1, 1)
and A(1, 1)0 in A(1, 3, 1), will mix with each other. That is, we need to identify the weak
hypercharge Y generator as a linear combination of the generators of SU(4)C SU(2)R
sharing the same quantum numbers. Using this, we can determine the U (1)Y term in the
gaugino mass expression in function of the coupling constants g4 , g2 corresponding to
SU(4)C and SU(2)R , respectively. Here, as we mentioned in the introduction, we assume
that the intermediate breakings down to the SM take place all around the MGUT scale
motivated by the MSSM successful unification of gauge couplings, and so we have g2
g4 g leading finally to gaugino masses Ma (a = 3, 2, 1) in the ratio 1 : (3/2) : (1).
2.1.2. SO(10) H = SU(2) SO(7) SU(3) SU(2) U (1)
The first stage of this breaking chain is achieved by the 54 traceless and symmetric
2nd-rank tensor with the non-zero vev
54 = v Diag(7/3, 7/3, 7/3, 1, 1, 1, 1, 1, 1, 1),
(8)
where the indices 1, 2, 3 correspond to SO(3) SU(2) and 4, . . . , 0 correspond to SO(7).

Subsequently, SO(7) is broken to SO(6) SU(4), which in turn is broken to SU(3) U (1).
As a result we get the gaugino masses Ma (a = 3, 2, 1) in the ratio 1 : (7/3) : 1.
86
2.2. The representation 210

The irrep 210 of SO(10) can be represented by a 4th-rank totally antisymmetric
tensor ij kl . It can break SO(10) in different ways [17].
2.2.1. SO(10) G422 SU(3) SU(2) U (1)
If the only non-zero vev is 7890 = v, where the indices 1 to 6 correspond to SU(4)
while those of 7 to 0 correspond to SO(4) then the intermediate stage is H = G422 [18].
We see immediately here that when SU(4) would be broken to SU(3) we shall get massless
gluinos (SU(3) gauginos). One can also see that if the only non-zero vevs are assumed
to be 1234 = 3456 = 1256 = w, then SO(10) is broken to G3221 SU(3) SU(2)
SU(2) U (1) [18] leading, eventually, to SU(2)L massless gauginos. Consequently one
can, in principle, assume both v and w = 0 and get G3221 as an intermediate stage without
getting massless gauginos. We would like to note here that one should not swiftly drop
the massless gluino scenario since it is not completely excluded phenomenologically [19]
and particularly because, as we said earlier, the breaking could be achieved in principle
from any linear combination of the irreps (1, 54, 210, 770). We did not study the case
corresponding to v, w = 0, neither the case where the intermediate stage is G3211
SU(3) SU(2) U (1) U (1) achieved by 1278 = 1290 = 3478 = 5678 = 5690 =
v [13]. Rather we concentrated on the phenomenologically interesting case of flipped
SU(5).
2.2.2. SO(10) H51 SU(5) U (1) SU(3) SU(2) U (1)
The flipped SU(5) model [20] exhibits some very suitable features such as fermion
and Higgs-boson content, the natural doublettriplet mass splitting mechanism and, among
others, no cosmologically embarrassing phase transitions.
The 210 irrep can break SO(10) to H51 when its singlet under H51 takes a non-zero
vev which amounts to its non-zero components as a 4th-rank totally antisymmetric tensor
being [14]
1234 = 1256 = 1278 = 1290 = 3456 = 3478
= 3490 = 5678 = 5690 = 7890 = v.
(9)
Next, we should break H51 SU(5)U (1)X to the SM SU(3)C SU(2)L U (1)Y . The
SU(5) group can be decomposed into SU(3)C SU(2)L U (1)Z . The weak hypercharge Y
must be a linear combination of Z and X. Under SU(5) U (1)X the 16 Higgs decomposes
as
16 = 101 + 5 3 + 15 ,
(10)
where the number on the lower right denotes the quantum number X of U (1)X . Because the
10 representation has the following branching rule under SU(5) SU(3) SU(2) U (1)Z
1)2/3 + (3, 2)1/6 + (1, 1)1,
10 = (3,
if the 101 (more precisely, the neutral component (1, 1) of SU(3) SU(2) in the 101
representation) in the 16 gets a non-zero vev, then H51 will break to the SM with Y/2 =
87
Table 1
Relative masses of gauginos at the GUT scale and at the weak scale achieved by vevs of the F -term
of superfields in representations corresponding to different breaking chains. The case A of the singlet
representation 1 for the F -term corresponds to the minimal supergravity model
Case
Intermediate stage
A
B
C
D
1
54
54
210
G422
SU(2) SO(7)
H51
M1GUT
1
1
1
96/25
M2GUT
1
1.5
7/3
1
M3GUT
M1mZ
M2mZ
M3mZ
1
1
1
1
0.42
0.42
0.42
1.6
0.88
1.3
2.1
0.88
3.0
3.0
3.0
3.0
Z), where Z is the generator of SU(5) which commutes with the generators of
SU(3)C SU(2)L and is normalized (for the five-dimensional representation of SU(5)) as

1 1 1 1 1
Z = Diag , , , ,
(11)
.
3 3 3 2 2
1
5 (X
Note that had we chosen Y/2 = Z corresponding to the GeorgiGlashow SU(5) we would
get universal gaugino masses.

Introducing the properly normalized U (1)Z and U (1)X generators LZ =
3
5
Z, LX =
such that Tr(LZ = 1/2 in the defining represenation of SU(5) and Tr16 (LX )2 = 1
in the 16 spinor representation of SO(10) [17] we could identify the properly normalized
U (1)Y field as a linear combination of the U (1)X and U (1)Z fields. Again, we assume that
the breaking of the intermediate stage H51 happens at the MGUT scale resulting in g1
g5 g for the coupling constants. Then we finally get gaugino masses Ma (a = 3, 2, 1) in
the ratio 1 : 1 : (96/25).
1 X
80
)2
2.3. Summary
We summarize in Table 1 our results for the relative gaugino masses at MGUT scale and
mZ scale recalling that M30 : M20 : M10 at MGUT (Ma0 MaGUT ) evolves approximately to
M3 : M2 : M1 7M30 : 2M20 : M10 at the weak scale mZ . The cases B, C, D correspond to
different breaking chains, respectively. The case A corresponds to the mSUGRA, i.e., the
universal gaugino mass case.
3. Phenomenological analysis and mass spectra

The gaugino mass patterns we have obtained are of phenomenological interest. There
can, in principle, be various terms in the superpotential and Khler potential that may
give rise to non-universal sqaurk and slepton masses, and whether such terms exist is
model-dependent. For simplicity, we assume universal soft sfermion masses and trilinear
couplings in our numerical analysis in order to clarify the phenomenological implications
of non-universal gaugino masses. We use the event generator ISAJET [21] (version 7.48)
88
to simulate models with non-universal gaugino mass parameters at the scale MGUT in this
section. The model parameter space used in our work is expanded by m0 , M30 m1/2 ,
A0 , tan and sign(). M20 and M10 can then be calculated in terms of M30 according to
Table 1. ISAJET calculates an iterative solution to the 26 RGEs and imposes the radiative
electroweak symmetry breaking constraint. This determines all the sparticle masses and
mixings and can calculate the branching fractions for all sparticles, particles and Higgs
bosons.
The constraints of lower bounds of sparticle and Higgs bonson masses [22] are included.
And we require the gauge coupling unification at the scale MGUT = 2.0 1016 GeV.
Throughout the work we take mt = 175 GeV.
A check for the compatibility of the models with the b s constraint is included.
The prediction of the b s decay branching ratio [23] should be within the current
experimental bounds [24]
2 104 < BRexp (b s ) < 4.2 104 .
Because there is no full next-to-leading order (NLO) formula available in SUSY models
we use the leading order (LO) calculation with about 30% theoretical uncertainty
included. This constraint is very strong for negative mu-term (sign() = 1) 1 due to the
constructively interference of SUSY contributions with the SM contributions [26]. It leads
to a rather large (but still smaller than 1 TeV) sparticle mass spectrum. For sign() = +1 ,
there are regions of the parameter space where the mass spectrum is low while tan is large
since the SUSY contributions destructively interfer with the Higgss and Ws contributions
in this case.
The g 2 constraint of the muon anomalous magnetic dipole moment a 12 (g 2)
[27] is also considered. The current data of the E821 experiment [28] give the following
Table 2
Mass spectra in the four models (A, B, C, D) for m1/2 = 300 GeV, m0 = 400 GeV, A0 = 350 GeV
and tan = 20. All the masses are shown in GeV and evaluated at the scale mZ
Model
me1,2
m1,2
Mg
mH1,2
mu 1,2
mt1,2
md
A
726
211/375
114/529
117/212/351/374
730/716
448/416
558/704
394/445
735/715
+348/535
657/710
B
731
257/404
111/521
123/259/289/404
766/717
502/416
576/715
389/493
770/716
286/527
684/708
C
743
101/579
111/572
86/101/147/579
843/718
616/416
547/781
385/606
847/716
106/577
696/767
D
729
208/358
113/553
208/324/355/493
734/765
494/588
598/707
482/572
738/726
+328/558
656/721
1,2
m 0
1,2,3,4
1 We follow the conventional definition of the sign() [25].
1,2
/mH
mb
1,2
89
bound on the supersymmetry contribution to a

11 1010 < aSUSY < 75 1010.
(12)
It is well-known that the diagram with charginosneutrino in the loop gives a dominant
contribution to aSUSY for general SUSY mass parameters and the muon chirality can be
flipped by the Yukawa coupling of muon which is proportional to 1/ cos ( tan in the
large tan case) [29,30]. Therefore, even with a relatively large mass spectrum, the bound
on the supersymmetry contribution to a , Eq. (12), can be satisfied in the large tan case.
On the contrary, the bound requires scenarios of small charginos and sneutrino masses
when tan is small. So the combined consideration of (g 2) and b s leads to that
Fig. 1. The neutralino and the chargino masses as a function of tan for the cases in Table 1. Also
plotted is ||. We have taken m0 = 400 GeV, A0 = 600 GeV and Ma0 = m1/2 times the number
appearing in Table 1, with m1/2 = 300 GeV.
90
the regions of large tan and low mass spectrum which are allowed by b s alone
decrease significantly.
Table 2 illustrates the numerical results of the mass spectra evaluated at the mass scale
mZ for the values m1/2 = 300 GeV, m0 = 400 GeV, A0 = 350 GeV and we have taken
a large value for the tan = 20 since this would enhance the g 2 constraint. We see
from the table that the mass spectrum is relatively heavy, which comes in order to satisfy
the b s constraint in the sign() = 1 case, as pointed out above. All cases have
neutralino LSP. Cases A, C and D have chargino NLSP, and case B an stau NLSP. The four
cases are experimentally distinguishable, because the sparticle mass splitting patterns are
quite different among the four.
In Fig. 1 the tan dependence of the ||, neutralinos and charginos masses for the four
cases in Table 1 is presented, where we have taken m1/2 = 300 GeV, m0 = 400 GeV, A0 =
Fig. 2. The neutralino and the chargino masses as a function of m1/2 for the cases in Table 1. Also
plotted is ||. We have taken tan = 8, m0 = 200 GeV and A0 = 300 GeV.
91
600 GeV. We noted that for far larger values of m0 , m1/2 resulting in larger masses for
the smuon and charginos, the (g 2) constraint would be violated. Also we have chosen
a rather large value for the trilinear scalar coupling A0 which appears in the off-diagonal
elements of the squark mass matrix in order to favor a large stop mass splitting. One can
see from the figure that the ||, neutralinos and charginos masses are insensitive to tan
when tan is relatively large (say, larger than 10).
Figs. 2 and 3 present the m1/2 dependence of the ||, neutralinos and charginos masses
for the four cases with tan being taken to be 8 and 25, respectively. In the cases B and C
corresponding to the representation 54, the (g 2) constraint was not respected for > 0.
This is in agreement with the anlysis in Refs. [29,31]. As pointed out in Refs. [29,31], in
most of the parameter space the sign of the SUSY contributions to a is directly correlated
Fig. 3. The neutralino and the chargino masses as a function of m1/2 for the cases in Table 1. Also
plotted is ||. We have taken tan = 25, m0 = 200 GeV and A0 = 300 GeV.
92
with the sign of the product M2 such that it is positive (negative) for M2 > 0 (M2 < 0.
Thus, the latest result of the E821 experiment, Eq. (12), suggests that M2 > 0 so that one
has < 0 since M2 < 0 in the cases B and C. In all four cases we see that the LSP is the
neutralino but in case D corresponding to the 210 representation the lightest chargino and
neutralino are approximately degenerate while for the other three cases the approximate
degeneracy happens for the heaviest ones.Thus, in the case D the lightest chargino is longlived. Therefore, the experimental signals for the case D are different from those expected
from conventional R-parity conserved SUSY modles, e.g., mSUGRA (i.e., the case A)
which have been studied in Ref. [32].
4. Conclusion
We have studied SUSY SO(10) models in which the gaugino masses are not universal
at the GUT scale and we have performed the group theory methods required to calculate
their ratio. Then, for some specific values of the soft mass parameters which are chosen to
respect the experimental constraints coming from the direct search of sparticles, b s
and a , we compared phenomenologically these models. The mass spectrum in the case
D is particularly interesting due to the presence of the approximately degenerate lightest
chargino and neutralino. All the breaking chains allow for boundary conditions compatible
with current experimental data on the b s branching ratio and the g 2 measurement.
However, these two constraints show a strong correlation and aSUSY becomes very large
for the large tan region and is expected to become the powerful tool in order to constrain
the SUSY parameter space and so to decide which breaking chain is preferable.
The pattern of non-universal gaugino masses at MGUT is determined only by the
breaking chain from SO(10) down to SM if the scale at which the breaking of the
intermediate subgroup happens is the same as that at the first step of the breaking chain.
Otherwise, it also depends on the scale at which the breaking of the intermediate subgroup
happens. However, the dependence is normally weak as long as the intermediate scale
is not too low. 2 Besides the irreps 54 and 210 of SO(10) necessary to get non-universal
masses, we use only one more irrep of SO(10), the spinor representation 16, to realize the
next step of breaking chains. This is economical in constructing a SUSY SO(10) GUT. It is
important to give an explicit form of a superpotential for a SUSY SO(10) GUT model with
non-universal gaugino masses to construct a specific model, which is beyond the scope of
the paper and left to future work.
Acknowledgements
This work was supported in part by the National Natural Science Foundation of China.
N.C. acknowledges financial support from COMSATS.
2 The intermediate scale is larger than about 1010 GeV in most of the model building studies (see, e.g.,
Refs. [13,20]).
93
References
[1] P. Langacker, M.-X. Luo, Phys. Rev. D 44 (1991) 817;
C. Giunti, C.W. Kim, U.W. Lee, Mod. Phys. Lett. A 6 (1991) 1745;
U. Amaldi, W. de Boer, H. Furstenau, Phys. Lett. B 260 (1991) 447;
J. Ellis, S. Kelley, D. Nanopolous, Phys. Lett. B 260 (1991) 131.
[2] Super-Kamiokande Collaboration, Y. Fukuda et al., Phys. Rev. Lett. 81 (1998) 1562;
SNO Collaboration, Q.R. Ahmad et al., Phys. Rev. Lett. 87 (2001) 071301.
[3] H. Georgi, in: Particles and Fields, AIP, New York, 1975, p. 575;
H. Fritzsch, P. Minkowski, Ann. Phys. 93 (1975) 193;
Z.-Y. Zhao, J. Phys. G 8 (1982) 1019.
[4] M. Gell-Mann, P. Ramond, R. Slansky, in: Supergravity, North-Holland, Amsterdam, 1979;
T. Yanagida, in: Proceedings of the Workshop on Unified Theory and Baryon Number of the
Universe, KEK, Tsukuba, 1979;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912;
C.H. Albright, S.M. Barr, Phys. Rev. Lett. 85 (2000) 244;
C.H. Albright, S.M. Barr, Phys. Rev. D 62 (2000) 093008;
C.H. Albright, S.M. Barr, Phys. Rev. D 64 (2001) 073010;
C.H. Albright, hep-ph/0110259;
N. Maekawa, hep-ph/0110276;
B. Bajc, G. Senjanovic, F. Vissani, hep-ph/0110310.
[5] See, for an example, S.P. Martin, Phys. Rev. D 46 (1992) R2769.
[6] H. Murayama, A. Pierce, hep-ph/0108104.
[7] A. Brignole, L.E. Ibez, C. Muoz, Nucl. Phys. B 422 (1994) 125;
A. Brignole, L.E. Ibez, C. Muoz, Nucl. Phys. B 436 (1995) 747, Erratum;
A. Brignole, L.E. Ibez, C. Muoz, C. Scheich, Z. Phys. C 74 (1997) 157;
T. Kobayashi, D. Suematsu, K. Yamada, Y. Yamagishi, Phys. Lett. B 348 (1995) 402;
R. Dermisek, A. Mafi, hep-ph/0108139.
[8] C.H. Chen, M. Drees, J.F. Gunion, Phys. Rev. D 55 (1997) 330;
Y. Kawamura, H.P. Nills, M. Olechowski, M. Yamaguchi, J. High Energy Phys. 9806 (1998)
008;
M. Brhlik, G.J. Good, G.L. Kane, Phys. Rev. D 59 (1999) 115004;
M. Brhlik, L. Everett, G.L. Kane, J. Lykken, Phys. Rev. Lett. 83 (1999) 2124;
M. Brhlik, L. Everett, G.L. Kane, J. Lykken, Phys. Rev. D 62 (2000) 035005;
E. Accomando, R. Arnowitt, B. Dutta, Phys. Rev. D 61 (2000) 075010;
T. Ibrahim, P. Nath, Phys. Rev. D 61 (2000) 093004;
S. Khalil, T. Kobayashi, O. Vives, Nucl. Phys. B 580 (2000) 275;
C.-S. Huang, W. Liao, Phys. Rev. D 62 (2000) 016008.
[9] G. Anderson et al., in: New Directions for High Energy Physics, Snowmass 96, SLAC, Menlo
Park, CA, 1997, hep-ph/9609457;
J. Amundson et al., in: New Directions for High Energy Physics, Snowmass 96, SLAC, Menlo
Park, CA, 1997, hep-ph/9609374.
[10] C.T. Hill, Phys. Lett. B 135 (1984) 47;
J. Ellis, K. Enqvist, D.V. Nanopoulos, K. Tamvakis, Phys. Lett. B 155 (1985) 381;
M. Drees, Phys. Lett. B 158 (1985) 409.
[11] G. Anderson, H. Baer, C.-H. Chen, X. Tata, Phys. Rev. D 61 (2000) 095005;
K. Huitu, Y. Kawamura, T. Kobayashi, K. Puolamki, Phys. Rev. D 61 (2000) 035001;
SUGRA Working Group, V. Barger, C.E.M. Wagner et al., hep-ph/0003154.
[12] J.C. Pati, A. Salam, Phys. Rev. D 10 (1974) 275.
[13] J. Sato, Phys. Rev. D 53 (1996) 3884.
[14] X.-G. He, S. Meljanac, Phys. Rev. D 41 (1990) 1620.
94
[15] A. De Rujula, H. Georgi, S.L. Glashow, Phys. Rev. Lett. 45 (1980) 413;
S.M. Barr, Phys. Lett. B 112 (1982) 219;
I. Antoniadis, J. Ellis, J.S. Hagelin, D.V. Nanopoulos, Phys. Lett. B 194 (1987) 231.
[16] G.F. Giudice, R. Rattazzi, Nucl. Phys. B 511 (1998) 25.
[17] R. Slansky, Phys. Rep. 79 (1981) 1.
[18] C.S. Aulakh, R.N. Mohapatra, Phys. Rev. D 28 (1983) 217.
[19] For a recent study, see: E.L. Berger et al., Phys. Rev. Lett. 86 (2001) 4231;
E.L. Berger, Talk in the 3rd Circum-Pan-Pacific Symposium High Energy Spin Physics,
Beijing, 2001.
[20] S.M. Barr, in [15];
K. Tamvakis, Phys. Lett. B 201 (1988) 95.
[21] H. Baer, F.E. Paige, S.D. Protopopescu, X. Tata, BNL-HET-99/43, hep-ph/0001086.
[22] Particle Data Group, D.E. Groom et al., Eur. Phys J. C 15 (2000) 1.
[23] For a recent analysis, see G. Degrassi, P. Gambino, G.F. Giudice, JHEP 0012 (2000) 009.
[24] CLEO Collaboration, S. Ahmed et al., CLEO-CONF-99-10, hep-ex/9908022.
[25] H.E. Haber, G.L. Kane, Phys. Rep. 117 (1985) 75.
[26] C.-S. Huang, T. Li, W. Liao, Q.-S. Yan, S.-H. Zhu, Eur. Phys. J. C 18 (2000) 393;
F. Borzumati, C. Greub, T. Hurth, D. Wyler, Phys. Rev. D 62 (2000) 075005, and references
therein;
Chao-shang Huang, Qi-shu Yan, Phys. Lett. B 442 (1998) 209;
J.L. Lopez, D.V. Nanopoulos, X. Wang, A. Zichichi, Phys. Rev. D 51 (1995) 147;
T. Goto, Y. Okada, Prog. Theor. Phys. 94 (1995) 407;
M.A. Diaz, Phys. Lett. B 322 (1994) 207;
F. Borzumati, Z. Phys. C 63 (1994) 291;
R. Garisto, J.N. Ng, Phys. Lett. B 315 (1993) 372;
R. Barbieri, G.F. Giudice, Phys. Lett. B 309 (1993) 86;
S. Bertolini, F. Borzumati, A. Masieso, G. Ridolfi, Nucl. Phys. B 353 (1991) 591.
[27] A. Czarnecki, W.J. Marciano, Phys. Rev. D 64 (2001) 013014, and references therein;
S. Martin, J. Wells, Phys. Rev. D 64 (2001) 035003.
[28] Muon g 2 Collaboration, H.N. Brown et al., Phys. Rev. Lett. 86 (2001) 2227.
[29] J.L. Lopez, D.V. Nanopoulos, X. Wang, Phys. Rev. D 49 (1994) 366.
[30] U. Chattopadhyay, P. Nath, Phys. Rev. D 53 (1996) 1648;
T. Moroi, Phys. Rev. D 53 (1996) 6565;
T. Moroi, Phys. Rev. D 56 (1997) 4424, Erratum.
[31] S. Komine, T. Moroi, M. Yamaguchi, Phys. Lett. B 506 (2001) 93.
[32] C.-H. Chen, M. Drees, J.F. Gunion, Phys. Rev. Lett. 76 (1996) 2002;
J.F. Gunion, S. Mrenna, Phys. Rev. D 64 (2001) 075002.

An exact elliptic superpotential for N = 1

deformations of finite N = 2 gauge theories
Nick Dorey, Timothy J. Hollowood, S. Prem Kumar
Department of Physics, University of Wales Swansea, Swansea, SA2 8PP, UK
Received 5 September 2001; accepted 12 December 2001
Abstract
We study relevant deformations of the N = 2 superconformal theory on the world-volume of N
D3-branes at an Ak1 singularity. In particular, we determine the vacuum structure of the massdeformed theory with N = 1 supersymmetry and show how the different vacua are permuted by
an extended duality symmetry. We then obtain exact, modular covariant formulae (for all k, N and
arbitrary gauge couplings) for the holomorphic observables in the massive vacua in two different
ways: by lifting to M-theory, and by compactification to three dimensions and subsequent use of
mirror symmetry. In the latter case, we find an exact superpotential for the model which coincides
with a certain combination of the quadratic Hamiltonians of the spin generalization of the elliptic
CalogeroMoser integrable system. 2002 Published by Elsevier Science B.V.
PACS: 12.60.Jv
1. Introduction and summary

Conformal field theories (CFTs) play a key role in quantum field theory as fixed points
of the renormalization group which classify possible universality classes. They can also
give rise to massive theories of direct phenomenological interest after deformation by
relevant operators. In general, little is known about the dynamics of interacting CFTs in
four dimensions. Superconformal field theories (SCFTs) provide an important exception to
this. In particular, strong evidence for several remarkable properties of these theories has
emerged. Firstly these models often have marginal operators which give rise to continuous
families of superconformal field theories. They also exhibit exact dualities which act
E-mail addresses: n.dorey@swan.ac.uk (N. Dorey), t.hollowood@swan.ac.uk (T.J. Hollowood),
s.p.kumar@swan.ac.uk (S.P. Kumar).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 4 7 - 2
96
N. Dorey et al. / Nuclear Physics B 624 (2002) 95145
on the marginal parameters relating regions of strong and weak coupling. Finally, these
theories have a dual description in terms of superstring theory on a ten-dimensional space
whose non-compact part is AdS5 [1]. The example which is best understood is N = 4
supersymmetric YangMills (SYM) theory with gauge group U(N) or SU(N). This theory
describes the world-volume dynamics of N D3-branes in type IIB string theory. The model
has a single marginal dimensionless coupling, = 4i/g 2 + /2 , formed from the gauge
coupling g and the vacuum angle . In the D-brane realization, this coincides with the
complexified IIB string coupling. The theory has an exact SL(2, Z) duality which acts
by modular transformations of and is inherited from the S-duality of the IIB theory.
Finally, at large N , the SU(N) theory has a dual description in terms of weakly-coupled
IIB superstring theory on the near horizon geometry of the D3-branes which is AdS5 S 5 .
In this paper, we will consider a more general class of superconformal field theories
with N = 2 supersymmetry known as Ak1 quiver theories. These theories arise in IIB
string theory when N D3-branes are placed at an elliptic Ak1 singularity in spacetime
[2]. The N = 4 theory is naturally thought of as the k = 1 member of this family.
We will focus on relevant mass-deformations which preserve either N = 2 or N = 1
supersymmetry. Following the convention of the N = 4 case, we will refer to these as
the N = 2 and N = 1 quiver theories, respectively. Our main results, which generalize
previous work in the k = 1 case [3,4], concern the vacuum structure and holomorphic
observables of the N = 1 quiver theories. We study the massive vacua of these theories
using two different methods. Firstly, we analyse the maximal degenerations of the complex
curve which governs the Coulomb branch of the corresponding N = 2 theory [5].
Secondly, we propose an exact superpotential for the theory which coincides with the
Hamiltonian of a certain classical integrable system known as the elliptic spin Calogero
Moser system [68]. The N = 1 vacua can then be analysed directly by stationarizing
this superpotential. The two approaches are shown to be in complete agreement. In the rest
of this introduction we will outline our results. The detailed calculations are provided in
subsequent sections.
The Ak1 quiver theory has gauge group G = U(1) SU(N)k with matter in bifundamental representations as determined by a quiver construction based on the Dynkin
diagram of the Lie group Ak1 . The matter content ensures that the beta functions of each
of the k non-Abelian gauge couplings gi , i = 1, . . . , k, vanish exactly and that the theory
therefore has k exactly marginal complex couplings
i =
4i
i
,
+
2
2
gi
(1.1)
i = 1, . . . , k, where i are the vacuum angles of each SU(N) factor in G. As mentioned

above, these theories arise on the world-volume of N D3-branes placed at an elliptic Ak1
singularity in
spacetime. In this construction, the IIB string coupling is identified with
the coupling ki=1 i of the diagonal SU(N) subgroup of G. An alternative IIA brane
configuration of N D4-branes intersecting k NS5-branes wrapped on a circle arises after
T-duality in the compact direction of the elliptic singularity. As we review below, the strong
coupling properties of the theory can then be analysed, as in [5], by lifting the IIA branes
to M-theory five-branes in eleven dimensions. The M-theory spacetime includes a torus
of complex structure with k marked points whose relative positions on the torus encode
97
the remaining k 1 independent gauge couplings. In the N = 4 case (k = 1) modular

transformations of the spacetime torus yield the familiar geometrical realization of IIB
S-duality in M-theory. For k > 1, the theory has an enlarged S-duality group corresponding
to the modular group of a torus with k marked points
which encode the positions of the
NS5s. In addition to modular transformations of = ki=1 i , this includes shifts of the
individual marked points by periods of the torus, physically realized as the movement of
NS5-branes around non-trivial cycles, and other non-trivial dualities. The IIB set-up of
branes at an orbifold singularity also yields a large-N closed string dual of the quiver
theories. In this case the relevant near-horizon geometry is the orbifold AdS5 S 5 /Zk [9].
As usual, a useful way to study a conformal theory is via its relevant perturbations. The
only such perturbations which preserve the full N = 2 supersymmetry are mass-terms for
the k bi-fundamental hypermultiplets. After the deformation the resulting N = 2 theory
has a non-trivial Coulomb branch which can be studied by the methods of Seiberg and
Witten [10]. As usual the Coulomb branch of the Ak1 theory can be described as the
moduli-space of a complex curve whose genus is equal to the rank of the gauge group,
which is r = k(N 1) + 1. The periods of a certain holomorphic one-form on the curve
then yield the exact mass formula for BPS states of the theory. In fact, hypermultiplet
masses are easily included in the IIA brane construction described above and lifting the
IIA branes to M-theory yields a single M5-brane wrapped on R4 [5]. This provides
an explicit construction of as an N -fold cover of the torus in spacetime with prescribed
singularities at the k marked points. The action of the extended S-duality group described
above is manifest in this approach.
As for any N = 2 theory in four dimensions, the resulting Coulomb branch is a
special Khler manifold of complex dimension r, with singular submanifolds where
BPS states become massless. For generic values of the masses, there are isolated points
on the Coulomb branch at which the maximal number of mutually-local BPS states
become massless. These points are special because they survive soft-breaking of N = 2
SUSY down to N = 1 and yield N = 1 vacua where the theory is realized in different
massive phases. Most famously, there are points on the Coulomb branch where magnetic
monopoles become massless and, after soft-breaking, condense in the vacuum leading
to the confinement of electric charges. When combined with the known holomorphy
properties, this also provides a powerful approach for obtaining exact results for the
vacuum properties of N = 1 theories. In this paper we will present an exact analysis of the
maximally singular points on the Coulomb branch of the Ak1 quiver theories for all k and
N and the resulting massive N = 1 vacua. These results generalize the existing knowledge
about massive deformations of the N = 4 theory [3,4] which we will now briefly review. In
terms of N = 2 supersymmetry, the N = 4 theory contains a vector multiplet and a single
hypermultiplet in the adjoint representation of the gauge group. The most general relevant
deformation which preserves N = 2 supersymmetry consists of introducing a mass term
for the adjoint hypermultiplet. Following [11], the resulting theory will be referred to as
N = 2 SUSY YangMills. The Coulomb branch of the N = 2 theory with gauge group
SU(N) is described by a branched N -fold cover of the standard flat torus E with complex
structure parameter which is part of the spacetime of the M-theory brane construction.
The points on the Coulomb branch which yield massive N = 1 vacua correspond to the
maximal degenerations of the curve. These are unbranched (unramified) N -fold covers of
98
the torus E [3], which are themselves tori with complex structure parameter of the form
=
q + l
,
p
with pq = N and l = 0, . . . , p 1.
(1.2)

The total number of massive N = 1 vacua is therefore equal to p|N p. As each of these
vacua is associated with the condensation of a BPS state with definite electric and magnetic
charges, the theory in each vacuum is not invariant under S-duality. Instead, S-duality
permutes the massive N = 1 vacua, relating the physics of the theory in one ground-state
at one value of the coupling to that of the theory in another ground-state at a different value
of the coupling. Hence, one might say that the S-duality of the underlying N = 4 theory is
spontaneously broken in the N = 1 theory. The action of SL(2, Z) on the vacua is the
same as the natural action of SL(2, Z) on the set of N -fold covers of the torus E [3].
Despite the spontaneous breaking of S-duality described above, the theory in each
massive N = 1 vacuum has a novel kind of duality named
S-duality in [12]. This duality
reflects the fact that the degenerate curve describing each massive vacuum is a torus with
complex structure parameter given above. The low-energy physics of the N = 2 theory
(and its N = 1 deformation) only depends on the complex structure of the curve and
therefore is invariant under modular transformations acting on .
S-duality therefore has
the same status as the IR electricmagnetic duality of SeibergWitten theory and is not
expected to be valid at all length-scales. Nevertheless, this new duality leads to interesting
predictions for the behaviour of the theory in the limit of large t Hooft coupling which
can be compared directly with the IIB dual background studied by Polchinski and Strassler
[11].
One of the main aims of this paper is to extend our understanding of supersymmetrypreserving relevant deformations to the Ak1 quiver models. In fact we find natural
generalizations of the phenomena described above which are familiar from the N = 4
case. Specifically, we study the maximally singular curves of the N = 2 quiver theories
which are again unbranched N -fold covers of the spacetime torus E . We find:
1. The extended S-duality group described above is spontaneously broken and has a
non-trivial action on the set of massive N = 1 vacua. In particular, shifts of the k
individual marked points by periods of E now generate an additional degeneracy
of N k1 massive vacua for each
unbranched N -fold cover E . The total number of
massive vacua is therefore N k1 p|N p.
2. The
S-duality group of each massive vacuum is extended in the obvious way to the
modular group of a torus of complex structure parameter with k marked points. In
particular shifts of the individual marked points by periods of E are exact dualities of
the IR physics in each vacuum.
One of the most fascinating and mysterious aspects of N = 2 supersymmetric gauge
theories in four dimensions is their relation to finite-dimensional classical integrable
systems [3,13,14]. In this paper, we will provide results which extend this correspondence
in new directions and harness it to provide a practical method of computing physical
quantities. We will now review the connection between N = 2 SUSY and classical
99
integrability in the context of the mass-deformed N = 4 theory and briefly summarize

our new results.
As discussed above, the N = 2 theory with gauge group U(N) has a Coulomb branch
of complex dimension N which is the moduli-space of a Riemann surface of genus N .
In [3], Donagi and Witten gave a concrete recipe for constructing as the spectral curve
of a certain classical integrable system. The phase space of the system in question is the
moduli space of solutions of a set of two-dimensional field equations for a U(N) gauge
field and adjoint Higgs known as Hitchins equations. The equations are obtained from the
dimensional reduction of the self-dual U(N) YangMills equation in four dimensions to
the two-dimensional torus E by including certain -function source terms. As usual for
self-duality equations, the resulting moduli-space is a hyper-Khler manifold. One of the
three linearly-independent complex structures of this manifold will play a special role in
the following discussion. In particular, the resulting complex symplectic manifold has a
very concrete realization as the complexified phase space of the elliptic CalogeroMoser
model [8].
The CalogeroMoser model is an integrable system of k non-relativistic particles with
positions Xa and momenta pa , a = 1, . . . , N , interacting via the pairwise potential [15],

V=
(1.3)
(Xa Xb ),
a>b
where (X) is the Weierstrass function. The system is integrable because of the existence
of N Poisson-commuting Hamiltonians Ha , a = 1, . . . , N . The explicit integration of
Hamiltons equations (see Appendix of [15]) proceeds by identifying the appropriate set of
action-angle variables with respect to which the equations of motion become linear. While
the action variables are just the k commuting Hamiltonians themselves, the corresponding
angle variables are the flat coordinates on a k-dimensional torus in phase space. It is a
highly non-trivial fact that this torus is precisely the Jacobian variety of an appropriate
Riemann surface of genus N , known as the spectral curve. Changing back to the original
variables, an explicit solution for the N positions Xa (t), for any set of initial data, can then
be given in terms of -functions on the Jacobian. The connection to N = 2 supersymmetry
in four dimensions starts from the observation that the complex curve which governs
the Coulomb branch of the N = 2 theory with gauge group U(N) is precisely the spectral
curve of the CalogeroMoser system. In particular, the curves coincide if we promote the
N positions and momenta to complex numbers and identify the resulting holomorphic
Hamiltonians Ha with the N order parameters ua = Tr a of the N = 2 theory. (Here,
is the adjoint scalar in the U(N) vector multiplet.) In particular, the first non-trivial
Hamiltonian
H=
N

p2
a
a=1
m2
(Xa Xb ),
(1.4)
a
=b
is identified with the quadratic order parameter u2 .

The correspondence as stated above is rather abstract as, so far we only have an N = 2
interpretation for the conserved quantities and not the corresponding angle variables.
Fortunately, a much clearer picture emerges after compactification of the four-dimensional
100
theory to three dimensions on a circle [16]. Now the theory acquires new scalar degrees
of freedom from the dimensional reduction of the N massless Abelian gauge fields on the
four-dimensional Coulomb branch. The component of each four-dimensional gauge field
in the reduced direction gives rise to a Wilson line while the remaining components can be
dualized in three dimensions in favour of another scalar field. As first explained by Seiberg
and Witten [17], these variables naturally lie on a torus whose periods are controlled by the
effective Abelian couplings of the low-energy theory in four dimensions. More precisely,
they take values on the Jacobian variety of and can therefore be identified with the
angle-variables of the complexified CalogeroMoser system.
After including both the N complex moduli ua of the four-dimensional theory and
the 2N new periodic scalars discussed above, the Coulomb branch, which we denote
throughout as M, of the compactified theory is a manifold of real-dimension 4N . The lowenergy effective action is a three-dimensional non-linear -model with target M. As the
theory has eight supercharges this manifold must be hyper-Khler. In fact the hyper-Khler
manifold in question is precisely the phase space of the (complexified) classical integrable
system described above. In particular, Seiberg and Witten [17] have argued that M has a
preferred complex structure which is independent of the radius of compactification. With
respect to this complex structure, the manifold has a description as a toric fibration where
the four-dimensional Coulomb branch parameterized by the moduli ua and the Jacobian of
the curve is the fibre above each point. As above, the moduli ua are identified with the N
commuting Hamiltonians of the integrable system, while the holomorphic coordinates on
the fibre are just the corresponding angle variables. Thus, with respect to the preferred
complex structure, the Coulomb branch of the compactified theory coincides with the
complexified phase space of the elliptic CalogeroMoser model. In fact, one may also
demonstrate directly the equivalence of M and the moduli space of Hitchins equations
as hyper-Khler manifolds. This equivalence turns out to be a generalization of the mirror
symmetry between three-dimensional gauge theories with eight supercharges discovered
in [18].
It is also interesting to consider soft breaking to N = 1 supersymmetry from the point of
view of the compactified theory. As in four dimensions, this is accomplished by introducing
a non-zero mass for the adjoint chiral multiplet in the N = 2 vector multiplet, or in
other words adding the perturbation u2 to the superpotential. This superpotential lifts
the Coulomb branch leaving only isolated vacua as in the uncompactified theory. One of
the main results of [4] was that the integrable systems viewpoint provides a simple and
quantitative description of this effect. Essentially all we need is the identification of the
Coulomb branch of the compactified theory as the phase space of the CalogeroMoser
model with the complex positions Xa and momenta pa providing a convenient choice of
holomorphic coordinates. In terms of these variables, the superpotential u2 is simply given
by the CalogeroMoser Hamiltonian given in (1.4) above. This leads to a new connection
between N = 1 supersymmetric gauge theories and integrable systems: the vacua of the
former can be identified with the equilibrium configurations of the latter. Numerous checks
of this identification were presented in [4]. It also provides a practical method of computing
the condensates of chiral operators for the N = 1 theory [12,19].
In the preceding discussion, we outlined the correspondence between the massive
deformations of N = 4 SUSY YangMills and a certain classical integrable system. One of
101
the main goals of the present paper is to extend this to massive deformations of the Ak1
quiver models for k > 1. The first step is to identify the appropriate integrable model.
Fortunately, several relevant results already exist. In particular, Kapustin has used the
mirror symmetry of [18] to show that the Coulomb branch of the N = 2 quiver theories
coincides with the moduli space of an appropriate system of self-duality equations. For
the Ak1 theory, the relevant equations are the U(N) Hitchin equations on the torus E
with specified behaviour at k punctures. In the following we will confirm that the spectral
curve of this system coincides with the complex curve for the Ak1 elliptic model given by
Witten. However our main interest will be to recast this system as a system of interacting
particles with spin in one dimension, analogous to the elliptic CalogeroMoser system
of the k = 1 case. Again, the required result is available in the existing literature. Nekrasov
has shown that the U(N) Hitchin system on a torus with k punctures can be recast as the
spin generalization of the elliptic CalogeroMoser system [8]. This system consists of N
particles moving on a circle, each carrying k spin variables. The Hamiltonian is given
explicitly in (4.36) below. As in the k = 1 case, we investigate the connection between the
maximally degenerate curves, the vacua of the theory obtained by soft-breaking of N = 2
supersymmetry to N = 1 and the equilibrium configurations of the integrable system. In
line with the arguments presented above, our main result is that the exact superpotential
of the theory is precisely a certain linear combination of the quadratic Hamiltonians
of the elliptic spin-CalogeroMoser system. In particular, we show that the equilibrium
configurations of the elliptic spin system are in one-to-one correspondence with the genusone degenerations of the complex curve . Finally we perform an explicit computation of
condensates in both approaches and demonstrate precise agreement.
The paper is organized as follows. In Section 2, we deduce the vacuum structure of
the mass-deformed quiver theory by analysing its classical superpotential. The resulting
structure has a very nice interpretation in terms of brane configurations in type IIA string
theory. In Section 3, we recall, following Witten [5], how these elliptic models can be
solved by lifting the type IIA brane configurations to M-theory. In this picture, the N = 1
deformation can be described in terms of a rotation whose description is relegated to
Appendix B. The net result is that the massive vacua of the mass deformed theory have
a very simple interpretation in the M-theory picture which allows us to solve for the
holomorphic observables. Section 4 describes a rather different way to probe the vacuum
structure via the compactification scenario described above. The net result is an expression
for the exact superpotential of the low energy three-dimensional -model which encodes
via its minima all the holomorphic structure of the vacua.
2. The N = 1 deformations of the quiver theory
We begin by recalling the matter content of the Ak1 quiver gauge theories. They have
i+1,i ),
N = 2 supersymmetry and gauge group U(N)k with k hypermultiplets (Qi,i+1 , Q
i = 1, . . . , k (the subscripts being defined modulo k) transforming in the bi-fundamental
), (N
, N )) of the ith and (i + 1)th U(N) factors. In terms of N = 1
representation ((N , N
superfields we denote the N = 2 vector multiplets as {Wi , i }; i = 1, 2, . . . , k. As each
U(N) factor sees 2N flavors in the fundamental representation, these N = 2 theories
102
Fig. 1. The type IIA setup of the elliptic model consisting of four-branes suspended between five-branes arranged
on a compact direction.
have vanishing -functions. In particular, in the absence of hypermultiplet masses they are
superconformal theories. Consequently there are k finite and independent gauge couplings
i =
4i
i
,
+
2
2
gi
(2.1)
i = 1, . . . , k, each corresponding to an exactly marginal direction in the space

of quiver
theories [9]. Introducing an overall gauge coupling 4i/g 2 + /2 = ki=1 i to be
thought of as the coupling for the diagonal U(N), the theory has a superpotential
W=
k
1
Tr i [i,i+1 , i+1,i ].
g2
(2.2)
i=1
We emphasize that the labels are defined modulo k. In what follows the dynamics of the
U(1) factors of the U(N) groups is often irrelevant and can be ignored as they decouple
from the interacting theory in the infra-red.
We will consider N = 1 deformations of these theories, i.e., soft breaking to N = 1
with generic hypermultiplet masses mi and generic masses i for the adjoint chiral
multiplets via the following N = 1 tree-level superpotential,
W=
k

1
Tr i [i,i+1 , i+1,i ] mi i,i+1 i+1,i i i2 .
2
g
(2.3)
i=1
We refer to this theory as the N = 1 quiver theory. The theory with i = 0 and mi
= 0
which we dub the N = 2 quiver theory, will play a central rle in our analysis of the
classical and quantum vacuum structure of the N = 1 deformation.
The Ak1 quiver theories with and without hypermultiplet masses (also known as the
elliptic models) are obtained in type IIA/M-theory [5] on the world-volume of N D4-branes
suspended between k parallel NS5-branes arranged on a compact direction in spacetime
as in Fig. 1.1 Following the usual conventions of [5] we choose our chain of k NS5branes to be located at x 7 = x 8 = x 9 = 0 and at specific values {xi6 } of the x 6 -coordinate.
The x 6 direction is compactified on a circle of radius L. There are N (fractional) D4branes between the (i 1)th and ith five-branes, with world-volumes parameterized by
x 0 , x 1 , x 2 , x 3 and x 6 . The (classical) position of a D4-brane may be conveniently described
1 As is well-known, in the T-dual type IIB picture, the quiver gauge theory may be obtained as the low-energy
world volume dynamics of N D3-branes at a Zk orbifold singularity in an Ak1 ALE space [2].
103
Fig. 2. N = 2 SUSY YangMills realized in type IIA theory.
by the complex variable v = x 4 + ix 5 . At low energies this construction actually gives rise
to a 4-dimensional SU(N)k U(1) theory with the desired matter content. The missing
U(1) factors corresponding to the center-of-mass motion in the x 4 x 5 plane of all the fourbranes in a given segment between adjacent NS5-branes are IR-free and decouple.2 The
distance between the ith and (i + 1)th five-branes is directly related to the gauge coupling
6
of the ith SU(N) factor, gi2 = (xi+1
xi6 )/(8 2 gs L) where gs is the type IIA string

coupling. Clearly then, the overall coupling 1/g 2 = i 1/gi2 is proportional to the radius
of the x 6 -dimension and is also the gauge coupling for the U(1) factor. As we have already
mentioned, the difference in positions of the centers-of-mass of two neighbouring stacks
in Fig. 1 is frozen in the infra-red and is therefore a modulus which is identified with the
hypermultiplet mass-parameter mi in the low energy theory.
The
periodicity of the spacetime implies that there are only k 1 independent masses
with ki=1 mi = 0. As explained in [5] this restriction may be avoided by modifying the
M-theory spacetime so that upon going around the x 6 circle the value of v is shifted at
some point by an amount m leading to
k
mi = m.
(2.4)
i=1
The parameter m is often referred to as the global mass. The position of the global shift is
completely unphysical and can be chosen anywhere. The simplest such setup (Fig. 2) with a
single five-brane and N D4-branes yields the SU(N), N = 2 theory with a massive adjoint
hypermultipletthe so-called N = 2 theory, first analyzed by Donagi and Witten [3].
Soft breaking of the above set-up to the N = 1 quiver theory is achieved by introducing
generic masses i for the adjoint chiral multiplets i . This can be realized in the brane
setup by rotating the NS5-branes relative to each other in the (v, w)-plane (w = x 8 + ix 9).
This feature is described separately in Appendix B where we generalize the approach of
[20] to the elliptic models.
2.1. Ground states of the N = 1 quiver theory
In this section, we determine the vacuum states of the N = 1 quiver theories
generalizing the structure in the basic N = 1 case [3,21]. It is worth reviewing the latter
2 Actually this is not quite true since the overall U(1) remains dynamical. But this decouples from the other
fields and we ignore it in what follows.
104
in some detail because this will provide us with the necessary intuition for the quiver
generalization.
In the mass deformed N = 4 theory, the vacuum structure can be uncovered by first
establishing a classical picture which can then be rather easily refined to give the full
quantum description. The classical vacua are governed by the tree-level superpotential (2.3)
for k = 1. As usual in an N = 1 theory, we can avoid solving the D-flatness conditions by
solving the F -flatness conditions modulo complex gauge transformations. In the k = 1
case, these are

, = m ,
(2.5a)
+
, = 2,
(2.5b)
modulo SL(N, C). The independent solutions are associated to the partitions N = q1 +
+ qp with
(q1 )
Jc
(q )
Jc 2
,
c = Mc
(2.6)
..
.
(q )
Jc p
where c = 1, 2, 3, 3 , 1 i2 and Jc are the generators
of the
q-dimensional representation of SU(2). The mass parameters are M1 = M2 = 2m and
M3 = m. The solution generically leaves Abelian factors of the gauge group unbroken
and the vacuum is massless. However, for equipartitions, N = q p, the unbroken gauge
symmetry is SU(p). In these cases, the non-Abelian symmetry leads to confinement in the
infra-red with a mass gap. The number of quantum vacua of an SU(p) gauge theory with
N = 1 supersymmetry is p and so the total number of massive vacua is

# massive vacua of N = 1 =
(2.7)
p,
(q)
p|N
a sum over the divisors p of N .

The type IIA brane picture
Before analyzing the classical vacuum structure of the N = 1 quiver theory, it will be
extremely instructive to see how the standard enumeration of massive vacua of N = 1
theory (mass-deformed N = 4 theory) described above arises from the viewpoint of the
type IIA brane picture. In particular, the massive vacua are associated to special (classical)
brane configurations corresponding to maximally singular points on the Coulomb branch
of the N = 2 theory, where new massless (string) states appear. Importantly, the massless
string states should give rise to charged hypermultiplets for all the classically visible U(1)
factors at the aforementioned Coulomb branch singularities. This ensures that upon further
soft breaking to N = 1 all these U(1) factors get Higgsed and only a non-Abelian gauge
symmetry survives and a mass gap is generated. It is easy to see that these requirements
lead us to configurations where the branes are arranged in helices as in Fig. 3(a),(b).
The helices arise in the following way. With only an N = 2 mass ( = 0) the F -flatness
105
Fig. 3. (a) The Coulomb branch singularity of N = 2 SUSY YangMills which yields the Higgs vacuum of
N = 1 theory. (b) The singular point with SU(p)q U(1)q1 gauge symmetry which descends to the SU(p)
vacuum of N = 1 theory.
conditions (2.5a) and (2.5b) allow a more general solution for :
(q )
c1 1[q1 ][q1 ] + J3 1
(q )
c2 1[q2 ][q2 ] + J3 2
= m
..
(qp )
cp 1[qp ][qp ] + J3
(2.8)
for arbitrary constants cr . Each of the p blocks corresponds to a helix and the parameters
cr encode the centre-of-mass position of each helix. Notice that the pitch of a helix as
one goes around x 6 is m, where m is the N = 2 global mass. In Fig. 3(a), there is a
U(1)N1 gauge symmetry with massless hypermultiplets from strings stretching between
two four-branes on either side of the five-brane, meeting each other at the same spacetime
point. Each such hypermultiplet carries charges under two different U(1) factors. Thus in
the low energy theory there are light states charged under each of the N 1 U(1) factors.
Soft breaking to N = 1 will cause all these states to condense, as is apparent from the
expressions for in (2.6), and therefore Higgses all the U(1) factors. This is the Higgs
vacuum. Fig. 3(b) illustrates a more general situation where the branes are arranged in p
coincident helices, where p is a divisor of N . The gauge symmetry is SU(p)q U(1)q1
with pq = N . The massless strings stretching across the five-brane, between two stacks
of D4-branes meeting each other, are bi-fundamentals under the two SU(p) factors and
charged under the two U(1)s associated with the motion the two stacks. Soft breaking to
N = 1 breaks the U(1) factors and simultaneously Higgses the off-diagonal parts of the
non-Abelian gauge symmetry leaving only an unbroken SU(p) gauge group with N = 1
supersymmetry. Since the low energy effective theory is N = 1 supersymmetric gauge
theory with gauge group SU(p), each such configuration gives p vacua, so that the total
number of massive vacua of N = 1 theory is given by the sum over divisors of N .
We can now extend the above picture fairly straightforwardly to the N = 1 deformation
of the N = 2 elliptic models. As far as the massive vacua are concerned once again we
expect the relevant Coulomb branch singularities to be those where a maximal number of
massless hypermultiplets appear, Higgsing all the classically visible U(1) factors upon soft
breaking to N = 1 supersymmetry. At these positions the branes arrange themselves in k
106
Fig. 4. A Coulomb branch singularity of the elliptic model with SU(p)kq U(1)k(q1) gauge symmetry. Upon
soft breaking to N = 1 SUSY it yields p k massive vacua.
helices as shown in Fig. 4. As in the N = 1 theory, the global mass sets the pitch of a helix
as we go around the x 6 circle. We can choose the shift of m to be anywhere, for instance
between the kth and 1st five-brane. The massive vacua correspond to precisely k helices
one of which begins, and one ends, on each of the k five-branes. Any fewer than k helices
and one cannot accommodate the k hypermultiplet masses. Any more, and there would be
additional moduli in the configuration leading to a massless vacuum upon soft breaking
to N = 1. In general each of the k helices can be composite consisting of a number of
coincident D4-branes, yielding unbroken non-Abelian factors. For a massive vacuum it is
necessary that the number in each helix be the same, namely p a divisor of N : pq = N . The
low energy theory at such a point (as in Fig. 4) on the moduli space of the N = 2 quiver
theory has SU(p)kq U(1)k(q1) gauge symmetry with k(q 1) massless hypermultiplets.
The latter originate, as before, from massless strings ending on two coincident four-brane
stacks, one on either side of an NS5. The N = 1 deformation leads to the condensation
of these massless hypermultiplets. This not only Higgses the classically visible U(1)s, but
also breaks the product gauge group associated with each composite helix to its diagonal
SU(p) subgroup. As there are k composite helices, this leaves a classically unbroken
SU(p)k gauge theory with massive matter and N = 1 supersymmetry which yields pk
massive vacua in the quantum theory. There is an additional degeneracy that arises in the
following way. There are q = N/p stacks of D4-branes intersecting a five-brane from
either side. Since there is one helix that starts and one that ends on each five-brane, q 1
of the stacks are continuous across the five-brane but one has a jump. As the jump may be
chosen to be at any of the q stacks there is an additional degeneracy of q per five-brane.
Actually the total degeneracy turns out to be q k1 since the jump at one of the five-branes
can be fixed without-loss-of-generality. Thus the total number of massive vacua of the
N = 1 quiver theory is

k k1
# massive vacua of N = 1 quiver =
(2.9)
p q
= N k1
p.
p|N
p|N
In the following sections, we will show how the exact solution of the elliptic model can
be used to reproduce this vacuum counting in the quantum theory. In addition the picture
described above in terms of four-brane helices will have a natural origin in the quantum
solution to be described from the M-theory point-of-view.
107
Systematic semiclassical analysis

After the intuitive discussion explored above, we now consider the problem of the
vacuum structure of the quiver theories more systematically. As in the N = 1 case, we
first consider the classical vacua that arise from the tree-level superpotential (2.3). This can
be re-cast as

+ m +
1
2
i i .
W = 2 Tr ( + X) ,
(2.10)
k
g
i
Here, X is a diagonal matrix with k independent parameters xi corresponding to the VEVs

of the U(1)-components of the U(N) gauge groups which are frozen in the infra-red. These
parameters are determined in terms of the masses mi via
m
= xi xi+1 .
(2.11)
k

Choosing i xi = 0 then fixes the xi . and are (kN kN)-dimensional matrices and
if we choose an ordering for the elements so that the ith SU(N) factor of the gauge group
is associated to the i mod kth row and columns, then
mi
X = diag(x1 , . . . , xk , x1 , . . . , xk , . . . , . . . , x1 , . . . , xk ).
(2.12)
With this ordering and can only have non-zero elements in positions (u, v),
1 u, v Nk:
: (u, u 1 + nk),
: (u, u + nk),
n Z.
(2.13)
We also define i , i = 1, . . . , k, as the elements of pertaining to the ith SU(N) gauge

group factor. The matrix i has non-zero elements in position (i + mk, i + nk), m, n Z.
The classical vacuum structure of the N = 1 deformation is obtained from the F -flatness
conditions
m

+ X, = ,
(2.14a)
k

P +, = 2
(2.14b)
i i ,
i
modulo complex SL(N, C)k gauge transformation. In (2.14b) P is a projection onto the
traceless part of each of the U(N) factors.
Complex gauge transformations can be used to diagonalize . This uses up all of the
complex gauge transformations apart from the Abelian transformations that fix diagonal
matrices and the Weyl group acting as permutations. The solutions of (2.14a) are associated
to the partitions of the ordered set of kN objects which are identified with the diagonal
elements of :
{1, 2, . . . , k, 1, 2, . . . , k, . . . , . . . , 1, 2, . . . , k}

1, . . . , i1 i1 + 1, . . . , i2 . . . , . . . in1 + 1, . . . , k .

A1
A2
An
(2.15)
108
Here the labels 1, 2, . . . , k, refer to the k SU(N) factors. The subsets Ar are themselves
ordered sets and we will define
3r = dim Ar ,
n
3r = Nk.
(2.16)
r=1
The labelling on the subsets Ar is irrelevant. In other words, a partition giving subsets Ar is
equivalent to the first if they are some re-labelling of the Ar . This equivalence is due to the
action of the Weyl group of SU(N)k . On the ordered set of kN objects, the Weyl group of
the ith SU(N) gauge group factor acts by permuting the ith, (i + k)th, . . . , (i + (kN 1))th
elements:

1, . . . , i , . . . , k, 1, . . . , i , . . . , k, . . . , . . . , 1, . . . , i , . . . , k .
For example, take SU(2)2 . The partition {{1, 2}, {1}, {2}} is equivalent to {{1}, {2}, {1, 2}}
but inequivalent from {{1}, {2, 1}, {2}}. In addition to these equivalences, there is a
restriction on the allowed partitions, proved below, which requires that the set3
{i1 , i2 , . . . , in } {1, 2, . . . , k}.
(2.17)
So in SU(2)2 , {{1, 2, 1, 2}} and {{1, 2}, {1, 2}} are not allowed partitions, but {{1, 2, 1}, {2}},
for instance, is allowed.
For a given allowed partition
+X
=
c1 1[31 ][31 ] + J3(31)
(32 )
c2 1[32 ][32 ] + J3
..
cn 1[3n ][3n ] + J3(3n )

(2.18)
where J3(3) is the diagonal generator of the 3-dimensional representation of SU(2). The
solution can be visualized in the type IIA brane picture as n helices, the rth having 3r /k
complete rotations around x 6 and a centre-of-mass in the v-plane at cr . Notice that the
pitch of a helix as one goes around x 6 is m, where m is the N = 2 global mass. The
share the same block structure:
K
K2
..
(2.19)
Kn
where the blocks Kr have the same pattern of non-vanishing elements as J1 r iJ2 r ,
the conventional raising and lowering matrices of the 3r -dimensional SU(2) representation.
(3 )
3 Notice i is fixed to be k.
n
(3 )
109
However, unlike in the N = 1 case, the numerical values are not the same. Nevertheless,
we can use the remaining Abelian symmetries to set
1 0
,
Kr =
(2.20)
.. ..
.
.
1
leaving
0 k (r)
1
..
.
Kr+ =
..
0
k3(r)
r 1
(2.21)
for constants kl , l = 1, . . . , 3r 1. Notice that the non-zero elements of and are

consistent with the grading (2.13).
The restriction on the allowed partitions (2.17) means that that there must be at least one
block of the partition that ends on each of the k gauge group factors, i.e., at least one of the
ir equals each of 1, 2, . . . , k. Suppose this were not true for the ith gauge group factor. By
taking the difference of the trace of (2.18) for the ith and (i + 1)th gauge group factors,
one would find
m
xi xi+1 = ,
(2.22)
k
in contradiction with mi being generic in (2.11). In particular, notice that this means the
allowed partitions must have n k; in other words there must be a least k helices. Notice
that in terms of the type IIA diagrams that the condition means that at least one helix must
end (and consequently at least one must start) on each NS5-brane.
Finally, we must solve (2.14b). This equation can be re-cast as

+
i i + A,
, = 2
(2.23)
(r)
where A is accounts for the trace parts of the U(N) factors:

A = diag(a1 , . . . , ak , a1 , . . . , ak , . . . , . . . , a1 , . . . , ak ),
(2.24)
for, as yet, arbitrary constants ai . All in all, we have n + k + kN n complex unknowns

in cr , r = 1, . . . , n, ai , i = 1, . . . , k and kl(r) , l = 1, . . . , 3r 1 and r = 1, . . . , n. These
are subject to the following conditions. Firstly, k complex conditions arise from the
tracelessness of , (2.18), in each of the SU(N) factors. Secondly Eq. (2.23) gives Nk
complex conditions. So there are (N + 1)k complex conditions for (N + 1)k complex
unknowns and so generically there exists a solution for each allowed partition. Generically,
the solutions have no additional moduli and so the low energy theories will be pure N = 1
supersymmetric gauge theory with a gauge group that is the subgroup of SU(N)k fixing
and . A given allowed partition generically leads to a solution with unbroken U(1)
factors and so to a massless, or Coulomb, vacuum. However, for very special partitions,
the unbroken gauge group is empty or non-Abelian giving Higgs and confining vacua with
110
a mass gap. We now identify these massive vacua. Firstly, there are partitions into n = k,
the smallest number of possible, subsets:
Higgs partition = {A1 , A2 , . . . , Ak }.
(2.25)
For these vacua, the unbroken gauge group is empty, so they are Higgs vacua. The number
of them can be determined as follows. Since n = k, there is a single ir in (2.15) associated
to each of the k gauge group factors. A given ir can therefore be situated in one of N places.
However, ik is fixed, so the total degeneracy of Higgs vacua is N k1 . The partition with into
Nk sets Ar with a single element each corresponds to = = 0. This vacuum leaves
unbroken the whole SU(N)k gauge group, and since each SU(N) factor yields N confining
quantum vacua, overall there are N k massive vacua of this type. More generally there are a
number of massive vacua associated to each divisor of N , N = pq. The associated partition
are of the form

Confining partition = A1 , . . . , A1 , A2 , . . . , A2 , . . . , . . . , Ak , . . . , Ak ,
(2.26)

p times
p times
p times
k
where i=1 3i = qk and {A1 , . . . , Ak } is an allowedHiggs vacuumpartition of an

SU(q)k theory. The unbroken gauge group is SU(p)k , giving a quantum degeneracy of pk ,
and since there are q k1 inequivalent Higgs vacua for an SU(q)k theory, the total number
of massive vacua associated to the divisor p of N , is q k1 pk = N k1 p. This gives the total
number of massive vacua as in Eq. (2.9). Hence, our more intuitive discussion in terms of
helices produces the right combinatorics as this more detailed semi-classical analysis.
An example is in order. Consider the SU(2)2 theory. The vacua are associated to the
partitions of the ordered set {1, 2, 1, 2}. The partitions {{1, 2, 1, 2}} and {{1, 2}, {1, 2}} are
ruled out by the restriction (2.17). This leaves: (i) {{1, 2, 1}, {2}} and {{1}, {2, 1, 2}} which
correspond to two inequivalent Higgs vacua; (ii) {{1, 2}, {1}, {2}}, {{1}, {2}, {2, 1}} which
have unbroken gauge group U(1) and so are massless; and (iii) {{1}, {2}, {1}, {2}} which
leads to an unbroken SU(2)2 gauge symmetry and four massive confining vacua in the
quantum theory. Overall there are six massive vacua which is consistent with (2.9) and two
additional massless vacua.
Phase structure
We end this section on the (semi)classical analysis of the N = 1 quiver theory with
some remarks on the phase structure of the massive vacua that we found above. Each
SU(N) factor in the quiver has matter fields in the fundamental representation which
therefore transform under the center of that SU(N) factor. However the matter fields are
neutral under a simultaneous gauge rotation in all the SU(N) factors by an element of the
2 ia
center U = e N 1[N][N] , a = 0, . . . , N 1. This can be interpreted as the center of the
diagonal gauge group SU(N)D under which all fields appear to transform in the adjoint
representation. This may then be used to classify the phases of the N = 1 quiver theory.
However since there is only one kind of ZN symmetrythe center (ZN )D of the diagonal
SU(N), the number of possible phases is much smaller than the number of massive vacua.
As in the case of the basic N = 1 theory [3,11], the possible phases are given by the order
N subgroups of (ZN )D (ZN )D . The latter characterizes the (ZN )D -valued electric (ne , 0)
111
and magnetic (0, nm ) charges, respectively, of the sources used to probe the theory. This
implies, for instance, that all the vacua with unbroken SU(p)k gauge symmetry may be
classified into p distinct phases with N k1 vacua in each phase. These p phases correspond
to the subgroups of (ZN )D (ZN )D , generated by the screened charges (p, 0) and (l, q)
with pq = N and l = 0, 1, . . . , p 1. The N k1 Higgs vacua correspond to p = 1. Apart
from computing the condensates, there is however no physical way to distinguish the N k1
vacua in a given phase of the diagonal gauge group. Recall that there are two sources of
the degeneracy of vacua in a given phase of the diagonal gauge group: (i) the multiplicity
pk1 associated with the Witten index for the remaining k 1 SU(p) factors (excluding
the diagonal one); and (ii) a factor of q k1 associated to the position of the jumps in
the four-brane helices. In the limit of infinitely massive matter fields, the former may be
associated with the emergence of an accidental Z2p symmetry in each SU(p) factor which
is spontaneously broken to Z2 by gaugino condensation in each factor. The additional
degeneracy of q k1 has a more unusual interpretation which we return to later.
3. Lifting to M-theory
3.1. Review of the solution to the elliptic model
We now consider the solution of the elliptic model, i.e., the N = 2 quiver theory with
the goal of extracting exact results for the holomorphic observables and vacuum structure
of the N = 1 quiver theory. To this end, we will follow the prescription outlined in [5]
and begin with a brief review.
The exact low energy effective action for the elliptic model can be obtained from
the type IIA brane setup in the limit of large string coupling gs 1 accompanied by a
simultaneous rescaling of the five-brane separations so that the k gauge couplings {i }
are unchanged. In this limit the compact M-dimension x 10 opens up and the system of
intersecting branes in type IIA lifts to a single smooth M5-brane in M-theory. There
are now two circles in spacetime: the compact x 6 dimension with radius L and the Mtheory dimension x 10 with radius R. The locations of the IIA NS5-branes are conveniently
described by the holomorphic coordinate z = x 6 + ix 10 . The (v, z) space is locally C E ,
with v C, z E and E a genus one Riemann surface. Specifically, the complex
structure of the spacetime torus E is determined by obtaining it as a quotient of the
z-plane by the two equivalences
(i)
x 6 x 6 + 2L,
(ii)
x 10 x 10 + 2R.
combined with x 10 x 10 + R,
(3.1)
Thus E has complex structure = iL/R + /2 which is

the coupling associated
with

the overall U(1) factor in the IIA setup; L/R = 4/g 2 = 4 i 1/gi2 and = i i . The
x 10 -positions of the IIA NS5-branes can be directly related to the theta-angles of each
gauge group factor in the quiver,
i =
10 x 10
xi+1
i
(3.2)
112
Thus the M-theory spacetime contains a torus E with complex structure and k distinct,
unordered marked points zi , i = 1, . . . , k, to be identified with the positions of the NS5branes. Different choices of and variations in the positions of the NS5-branes lead to
different gauge couplings and theta angles for the SU(N) factors in the gauge group;
explicitly
i =
i(zi+1 zi )
,
2R
(3.3)
where the zi s are labelled according to their order around x 6 . In the following, for
convenience, we re-scale the torus E so that the M-theory radius R is unity. This makes
the zi s dimensionless.
In the presence of the overall mass shift m described earlier, the (v, z) space in which
the M5-brane propagates is to be thought of as a C bundle over E which is defined as a
quotient Qm obtained by dividing by the combined operations
x 6 x 6 + 2L,
x 10 x 10 + R,
v v + m.
(3.4)
This twisting of the spacetime can be undone almost everywhere at the expense of a special
point on E , the location of which is arbitrary, at which certain discontinuities will occur.
Although arbitrary, this point (chosen at z = 0 below) will appear in the solution of the
model as we see below.
The M5-brane itself has world-volume R1,3 where is a Riemann surface
embedded in Qm . This is the SeibergWitten curve for the theory. The type IIA setup
of intersecting branes naturally lifts to a Riemann surface with genus (N 1)k + 1
which is precisely the rank of the low-energy gauge group. More importantly, the curve
is a branched N -fold cover of the torus E . This is clear from the type IIA limit wherein
the curve reproduces N D4-branes, each wrapped around E . This curve is described by
the equation
F (z, v) = v N f1 (z)v N1 + f2 (z)v N2 + + (1)N fN (z)
=
N

v va (z) = 0,
(3.5)
a=1
with z E C/ where is the lattice 21 Z 22 Z with 2 /1 = .4 Thus, for every

point z in E , the Eq. (3.5) has N roots representing the v-coordinates of the N D4-branes.
The locations zi of simple poles of the va (z) represent the positions of the NS5-branes. Let
us now recall in detail following Witten [5] the properties of the functions va (z) and fa (z).
The functions fa (z) are: (i) single-valued elliptic functions on E with only simple
poles at z = zi and a pole of order a at z = 0; and (ii) the singularity at z = 0 may be
removed by a linear re-definition of v. These imply that the va (z) which need not be
single-valued on E must have the properties: (i) that near z = zi , exactly one of the va (z)
4 Note that our parameterization of the curve differs from that presented in [5] which was written in terms
of variables x and y specifying a point on E via the Weierstrass equation y 2 = 4x 3 g2 ( ; 1 , 2 )x
g3 ( ; 1 , 2 ). This is in fact solved by x = (z; 1, 2 ) and y = (z; 1 , 2 ) where (z; 1 , 2 ) is the
Weierstrass function.
113
has a simple pole; and (ii) all the va s have a simple pole at z = 0 (the special point of
discontinuity described above) with exactly the same residue.
The parameters of the theory, namely the hypermultiplet bare masses mi and the order
parameters on the Coulomb branch of the N = 2 elliptic model must be encoded in the
solution
F (z, v) = 0. The hypermultiplet bare masses are contained in the singular part of
f1 (z) = a va (z). In particular mi is equal to 1/N times the residue at z = zi (the location
of an NS5-brane) [3,5].5 Since f1 (z) is meromorphic on E , the sum of its residues must
vanish and therefore
it must have yet another simple pole which may be chosen at z = 0
with residue N i mi = Nm incorporating the global mass. The constant part of f1 (z) is
the order parameter for the overall U(1) factor in the elliptic model. Each of the remaining
N 1 functions fa (z) have k 1 independent residues at the simple poles and a constant
piece leading to a total of (N 1)k parameters. This is consistent with the Riemann
Roch theorem which determines the space of meromorphic functions on E with k distinct
simple poles to be k-dimensional. These (N 1)k parameters are to be identified with the
order parameters on the Coulomb branch of the SU(N)k elliptic model. In the following
sections our goal will be to compute these order parameters at the special singular points on
the Coulomb branch which descend to N = 1 vacua upon mass perturbation of the theory.
3.2. Duality symmetries
The N = 2 quiver theories (without hypermultiplet masses) have a non-trivial duality
symmetry that is a generalization of the SL(2, Z) duality group of the N = 4 theory. This
duality symmetry is manifest in the type IIA/M-theory construction of the quiver theory
at its conformal point where all VEVs vanish. At this point the N D4-branes lift to an
M5-brane which is multiply wrapped around the M-theory spacetime torus E . The lowenergy four-dimensional field theory then inherits the duality symmetry of M-theory under
the action of SL(2, Z) on the complex structure of the spacetime torus E . This duality
acts on the overall gauge coupling in the usual way and may be interpreted as modular
invariance of the diagonal SU(N) factor.
However, in addition to this the quiver theory has a much more unusual symmetry at the
origin of the Coulomb branch which we now describe. The k individual gauge-couplings
are encoded in the complex structure of E and the positions of the k indistinguishable
marked points zi , representing the five-brane positions. Moving a given NS5-brane all the
way around one of the cycles of the torus (with the remaining k 1 NS5s fixed), leaves the
brane configuration unaltered and therefore is a symmetry of the low energy theory. But this
operation changes the gauge couplings of the individual SU(N) factors in the quiver as they
are given by separations of the five-branes. The positions of the latter change according to
zi zi + 2m1 + 2n2 ,
zl zl ,
all l
= i.
m, n Z,
(3.6)
5 In the parameterization of [5] the masses are 1/N times the residues of the differential form f (x, y) dx/y
1
which is precisely the residue of f1 (z) dz in our notation.
114
Clearly this transformation includes the familiar shift in the theta angles i i + 2m
and i+1 i+1 + 2m which is a symmetry of the theory. However, in addition the
transformation also implies a periodicity of the inverse gauge couplings 1/gi2 with period
1/g 2 . In the T-dual type IIB picture of the quiver theory, the periodicity of the inverse
gauge couplings arises from the periodicity of the scalar fields obtained by integrating the
NSNS 2-form over the vanishing 2-cycles at the orbifold singularity [22,23].
As in the case of the N = 4 theory the duality group must be spontaneously broken
at a generic point on the Coulomb branch of the Ak1 quiver theory. Specifically, the
Coulomb branch of the quiver theory contains certain special singular points. These
are points which become N = 1 vacua upon perturbing the theory with non-zero
hypermultiplet and adjoint masses, mi
= 0 and i
= 0. At these singularities new massless
BPS states appear in the low energy theory and this is signalled by the degeneration of the
corresponding cycles of the SeibergWitten curve. As in the case of mass deformed N = 4
SYM, one expects each massive vacuum of the N = 1 quiver theory to be associated with
the condensation of a specific BPS state belonging to the spectrum of the parent conformal
quiver theory. As the generalized duality symmetry of the quiver model described above
must leave the spectrum of the conformal theory invariant, it therefore acts on the BPS
states in the spectrum via permutation. This in turn implies that the action of the duality
group takes the N = 1 quiver theory from one vacuum to another since each of these
vacua is associated with the condensation of a specific BPS particle in the theory.
In addition to the above duality properties, we will argue below that the low-energy
physics, in particular the chiral sector of the N = 1 quiver theory in each vacuum is
invariant under a symmetry which we will refer to as extended
S-duality.
3.3. The quantum vacuum structure of the N = 1 quiver theory
We now turn to the description of the quantum vacuum structure of the N = 1 quiver
theory. As mentioned above, and described in Appendix B from the point of view of brane
rotation, the massive vacua of the theory correspond to the singular points on the Coulomb
branch of the N = 2 quiver theory where the SeibergWitten curve undergoes maximal
degeneration to a surface of genus one.
The locations of these special singular points on the N = 2 moduli space are completely
(i)
specified by the k(N 1) gauge-invariant order parameters ua Tr ia , i = 1, 2, . . . , k,
and a = 2, . . . , N .6 As we have seen above, these order parameters are encoded in the
curves for the theory (modulo ambiguities to be discussed later). The values of these order
parameters at the singular points (as a function of the k couplings i and mass parameters
mi ) yield the condensates of the N = 1 quiver theory and provide complete information
on the chiral sector of that theory.
At the special singular points which descend to massive N = 1 vacua, the Seiberg
Witten curve degenerates maximally to an unbranched (unramified) N -fold cover of the
6 Here, we use the same symbol to denote the chiral superfield and its lowest component.
115
Fig. 5. The N -fold cover of E with = q/p. The s represent the positions of the k NS5-branes.
spacetime torus E .7 This fact and the associated ellipticity properties will allow us to
explicitly construct the functions va (z) at the singular points in the moduli space of the
curve Eq. (3.5) and extract the condensates for the N = 1 theory.
An unbranched N -fold cover of E is also a torus, but with a different complex structure
. As noted in [3] and [12] such covers are classified by 3 integers p, q and l where pq = N
and l = 0, 1, . . . , p 1. They are genus one Riemann surfaces E with the complex
structure = (q + l)/p. E can be defined as usual as the quotient of the complex
plane by the -lattice, namely E C/ . In what follows, we focus primarily on the
cases with l = 0, since the more general covering can be obtained by suitable modular
1 Z
2 Z with
1 = p1 and
2 = q2
transformations. Fig. 5 illustrates the case
and =
2 /
1 . The fundamental parallelogram of the lattice shown in Fig. 5 consists
of N -copies of the fundamental parallelogram of the -lattice with l = 0. The NS5-branes
which extend in the 012345 directions are represented by marked points (the s) on the
-torus. Each -parallelogram represents a single type IIA four-brane that has grown an
extra dimension in the 1 direction (which is roughly speaking, the M-dimension).
The SeibergWitten curve controls the physics of the holomorphic sector of the theory
and as discussed above, at the maximally singular points, the curve degenerates into a
torus with complex structure . As usual, the low-energy physics depends only on the
complex structure and must be invariant under modular transformations acting on .
Thus VEVs of chiral operators in each vacuum of the N = 1 quiver theory are expected
to be modular functions of . This duality, referred to as
S-duality [12] in the N = 1
theory is further extended in the case of the mass-deformed quiver theories. In addition to
modular transformations on , we may also move each NS5-brane independently around
non-trivial cycles of the -torus, inducing corresponding shifts in the individual gaugecouplings. Such -elliptic shifts in the positions of each NS5-brane are also symmetries of
S-duality.
each vacuum of the N = 1 quiver theory and form an extended version of
It is worth noting that there is a simple and direct mapping between the -lattice
sketched in Fig. 5 and the type IIA picture for a massive SU(p)k vacuum. Recall that
va (z) is simply the v-coordinate of the ath type IIA four-brane expressed as a function
of the (x 6 , x 10 ) coordinates of the compact dimensions. In the classical (type IIA) limit
the x 10 dimension is vanishingly small and the -parallelogram describes p coincident
7 Note usually in N = 2 theories as in [10] the curves maximally degenerate into a sphere. In the elliptic
models the curves degenerate at most into a genus one Riemann surface. In the M-theory construction this reflects
the presence of the unbroken and decoupled U(1) factor.
116
four-brane helices. In the classical limit all the NS5s are lined along the x 6 direction
leading to discontinuities in the four-brane helices. Quantum effects (or the growth of the
M-dimension) not only cause the four-branes to grow an extra dimension but also lead to a
separation between the helices in the x 10 -direction.
Vacuum counting in the quantum theory
Let us now see how the quantum theory reproduces the degeneracy factor of pN k1
associated with each classical SU(p)k vacuum. Recall that the duality group of the
Ak1 quiver gauge theory acts on the massive vacua of the N = 1 quiver model via
permutations. Thus one may sweep out all the vacua of the theory by performing the duality
operations on a given vacuum configuration in the IIA/M-theory setup.
Firstly, for a given p that divides N there are p inequivalent N -fold covers of E related
to one another by SL(2, Z) operations on that sweep out the p values of = (q + l)/p
with l = 0, 1, . . . , p 1. Secondly, each NS5-brane may be shifted by integer multiples of
1 and 2 and placed in any one of the N -parallelograms that make up the N -fold
cover. While the first operation constitutes the action of SL(2, Z) on E , the second
represents the motion of an NS5-brane around the cycles of the spacetime torus E . These
operations are precisely the duality symmetries of the parent conformal theory discussed
in the previous section. Thus they act on the vacua of the N = 1 quiver by permutation.
Since only the relative separations (and not the actual positions) of the NS5s are physically
meaningful, and each five-brane may intersect any one of the N four-branes, there are
N k1 such distinct configurations. Thus the total number of vacua from configurations
with a classically unbroken SU(p)k gauge symmetry is pN k1 . We have already argued
that the degeneracy factor of p arises from SL(2, Z) transformations on the coupling of
the diagonal part of the gauge group. Displacing the NS5-branes by integer multiples of
1 leads to a shift in the theta angles i of each gauge group factor in the quiver and
yields pk1 distinct configurations. Moving the five-branes along the other period gives
rise to q k1 ground states. The former is in line with our semiclassical arguments wherein
a multiplicity of pk appears as a consequence of the Witten index for N = 1 SUSY
gauge theories with massive matter and SU(p)k gauge symmetry. The classically obscure
degeneracy factor of q k1 associated with the position of the discontinuities in the fourbrane helices in the type IIA limit, can now be clearly understood to be a manifestation of
the non-trivial duality symmetry of the underlying conformal quiver theory.
The singular curves and condensates
The calculation of the condensates of chiral operators in the vacua with N = 1 SUSY
relies on the key property that at the associated singular points in the N = 2 moduli space
the curve degenerates into the torus E . This translates directly into the requirement
that the functions va (z) describing the individual four-brane positions be invariant under
a -elliptic transformation, i.e., under translations of z by periods of E . At this point the
functions va (z) must be thought of as different branches of a complex function that is
single-valued on E but not on E . Translations by periods of the -torus will smoothly
take us from one branch to the next, i.e., from one four-brane to the next. In particular
the complete set of functions va (z) will still be invariant under periodic shifts on E ,
but the individual va s will get permuted by this action. We can readily write down the
117
explicit form of these functions at the singular points. It will be sufficient to consider a
generic vacuum with N = pq and l = 0. All other massive vacua may be obtained by the
operations generating the duality group of the theory.
In a vacuum with a classically unbroken SU(p)k gauge symmetry, it is convenient
to denote the positions of the four-branes with two subscripts va (z) vsr (z), s =
0, 1, 2, . . . , p 1; r = 0, 1, 2, . . ., q 1. It is well-known that every elliptic function has an
expansion in terms of the Weierstrass zeta function (z) (see Appendix A for details) and
its derivatives [24]. This function has only a simple pole at z = 0 with unit residue. Hence,
knowledge of the singularities of an elliptic function and associated residues completely
determines its expansion in terms of the zeta function and its derivatives. In addition, the
Weierstrass zeta function (z| ) on a torus with complex structure has the following
anomalous transformation property:
(z + 23 | ) = (z| ) + 2 (3 | ),
3 = 1, 2.
(3.7)
We have argued that the four-brane positions vsr (z) must be elliptic on the -torus. Thus
they must have an expansion in terms of the zeta function on the -torus, which we denote
as (z):

(z) z ,
(z + 2p1 ) = (z) + 2 (p1 ),
(z + 2q2 ) = (z) + 2 (q2 ).
(3.8)
Based on the properties of vsr (z) reviewed in the Section 4.1, we also know that these
functions can only have simple poles located at the positions of the NS5-branes with
residues given by the hypermultiplet masses Nmi . Finally, we must also have that at each
NS5 location, one and only one vsr (z) has a simple pole. Following these requirements, we
can deduce the form of the four-brane coordinates vsr (z). In particular, taking the vacuum
with l = 0 where the ith NS5-brane lies on the (si , ri ) sheet of the covering, we have
vsr (z) = N
k

mi z zi + 2(s si )1 + 2(r ri )2
i=1
p1 q1

(z + 2t1 + 2u2 ) + 2qs (p1 ) + 2pr (q2 ) ,
t =0 u=0
s = 0, 1, 2, . . ., p 1, r = 0, 1, 2, . . ., q 1.
(3.9)
Note in particular that this satisfies the requirement that at each five-brane position,
exactly one vsr (z), namely vsi ri (z), has a simple pole and the residue is given by the
hypermultiplet mass: Nmi . All the vsr (z) also have a simple pole at z = 0 with exactly
the same residue m. Our expressions Eq. (3.9) for the four-brane coordinates vsr (z)
are uniquely determined up to additive constants, by the required singularity structure
and the property of -ellipticity. However, these additive constants do not affect the
computation of condensates of gauge-invariant chiral operators. Due to the anomalous
transformation properties of the zeta functions, the functions vsr (z) transform into one
another upon going around one of the cycles of E . For example, vsr (z + 21 ) = vs+1r (z)
118
and vsr (z + 22 ) = vsr+1 (z) where the subscripts are defined modulo (p, q) as usual.
Going around the periodic dimensions takes us smoothly from one four-brane to the
next. This reflects the fact that the type IIA four-branes lifted to M-theory form an
unbranched (unramified) N -fold cover of the torus E . The other vacuum configurations
with l
= 0, may be obtained by using suitable SL(2, Z) modular transformations on the
overall coupling .
(i)
We will now argue that the condensate u2 = Tr i2 , namely the expectation value of
the adjoint scalar in the ith SU(N) factor is encoded in the function
q
p
2
1
vsr (z) vs r (z) .
u2 (z) =
2N

(3.10)
s,s =1 r,r =1
The simplest way to see that this is the relevant object is to first recall the classical
(type IIA) limit. In this limit the vsr (z) corresponding to the four-brane positions become
independent of z at any given point between two adjacent five-branes. The only position
dependence in the four-brane coordinates arises from discontinuities or breaks as we
move across an NS5-brane. In the classical limit, the positions of the D4-branes in a
given gauge group factor correspond precisely to the eigenvalues of the adjoint scalar i
in that gauge group factor. It is then straightforward to show that classically Tr i2 =
q
1 p
2

2N
s,s =1
r,r =1 (vsr vs r ) provided i is defined to be traceless.
When the same quantity is lifted to M-theory it naturally acquires non-trivial
position-dependence (z-dependence) in that the classical discontinuities at the five-brane
locations get smoothed out and the simple geometrical relationship between the classical
condensates u(i)
2 and Eq. (3.10) is lost. Now u2 is a smooth function of z with k double
poles and k simple poles. The strengths of the double poles are simply determined by the
(i)
mass parameters mi , while the values of the condensates u2 are encoded in the residues at
the simple poles and the constant part of u2 (z). The simple pole residues and the constant
part of u2 (z) are moduli of the SeibergWitten curve and are related to the Coulomb branch
moduli of the N = 2 quiver theory.
It is easy to see that the function u2 (z) is an elliptic function on the -torus with modular
weight 2. This follows from the fact that the four-brane positions vsr (z) essentially
transform into one another under z z + 2m1 + 2n2 ; m, n Z, and that the -functions
have modular weight 1. The -ellipticity of u2 (z) provides a natural way to separate out
the singular and constant pieces since every elliptic function has an expansion in terms of
-functions (see Appendix A)
u2 (z) = N(N 1)
k

i=1
m2i (z zi ) +
k
(z zi )Hi + H0 .
(3.11)
i=1
There
are only k 1 independent simple pole residues Hi since we must have
k
i=1 Hi = 0. Thus, including the constant part of u2 (z) there are k independent para(i)
meters which coincides with the number of condensates u2 = Tr i2 . The connection
between u2 (z) and the condensates is also clearly seen via the intimate relationship between the Coulomb branch physics of the N = 2 quiver theory and integrable models.
This aspect will be explored in great detail in Section 4 and via a completely different
119
physical picture, will lead to an extremely non-trivial check of the results for the condensates presented in this section.
At this stage it must be pointed out that the exact relationship between the parameters
(i)
H0 and Hi on the one hand and the physical condensates u2 on the other, is far from
obvious. There is no unambiguous way to map the residues in the above expressions to
the VEVs of the operators Tr i2 in each gauge group factor. In general, one expects
(i)
the physical u2 to be some linear combination of the residues Hi and H0 and the
identity operator. Specifically, we may identify a set of k condensates that are modular
covariant with the correct modular weight, respecting the symmetries of a given vacuum.
However, there is no unambiguous way to determine the exact relation between such
a set of condensates and the VEVs of the operators Tr i2 . This is a generalization of
the ambiguity encountered in the basic N = 1 case [12,19,25]. However although it
appears difficult to obtain the condensates within a given gauge group factor, there is a
natural unambiguous combination of the condensates H0 and Hi that can be identified

(i)
with the average condensate 1k ki=1 u2 = 1k ki=1 Tr i2 . As we will argue below, this
combination will have the correct transformation properties under the duality group and
only suffer from a mild vacuum-independent additive ambiguity, i.e., mixing with the
identity operator.
Using Eqs. (3.9) and (3.10) we may readily calculate the residues of the function u2 (z)
at the simple poles z = zi , corresponding to the location of the NS5-branes. The calculation
of the z-independent part H0 is tedious and can be performed in two different ways. One
method is to use -ellipticity of each term (vsr vs r )2 in the summation in Eq. (3.10),
expand in terms of -functions and subsequently obtain the constant pieces by evaluating
the expressions at certain special points. The other technique is to make use of various
identities for elliptic functions provided in Appendix A, particularly Eq. (A.20). Below,
we simply present our final results for the condensates H0 and Hi for the vacuum with
l = 0 and (si , ri ) = (0, 0). (As we have emphasized the values of the condensates in the
other massive vacua can then be deduced by moving NS5-branes by periods of the torus
and also by modular transformations in .)
It is convenient to introduce the following shorthand notation for certain combinations
of variables that appear in the formulas for the condensates:

sr = 2s1 + 2r2 ,
(z)
= z ,
(z) = z .
zij zi zj ,
(3.12)
We then find that the simple pole residues are given by
Hi |l=0;si =ri =0
= 2Nmi

j
=i

mj (N 1) (zij )

(r,s)
=(0,0)

(zij + sr ) (sr ) .
(3.13)

Note that
Hi = 0 as required. Note also that they are modular functions with weight
1 under the action of SL(2, Z) on . Furthermore, shifting one of the NS5-branes by a
period of the -torus leaves the Hi invariant. As argued earlier, shifting the five-branes by
periods of the -torus has the effect of permuting the vacua of the N = 1 quiver theory.
120
But since the degenerate curve at a given vacuum is simply the -torus, taking an NS5brane around one of the cycles of this torus must be a symmetry of that vacuum and must
leave the order parameters unchanged. Hence the Hi have the requisite properties to be
identified as condensates characterizing a given massive vacuum of the N = 1 quiver
theory. However, it is clear from Eq. (3.11), that H0 cannot share this property. Moving a
five-brane around the -torus will necessarily lead to shifts in our definition of H0 simply
because of the anomalous transformation properties of the -functions characterizing the
simple pole terms. This can be explicitly checked from our expression for H0 :
H0 |l=0;si =ri =0
k

N2 2 1
q
=
mi +
mi mj E2 ( ) E2
12
2
p
i
=j
i=1

N 1 2
1
(zij ) N (z
mi mj
ij ) +
+ N2
(zij )
2
2

i
=j

1
2 (zij + sr s r )
+
2N

(r,s)
=(r ,s )
2(r r )
2(s s )
+
(p1 ) +
(q2 )
p
q
1 q (p1 ) (1 )
zij
N
1

(N 1) (zij )
2

(zij + sr ) (sr )
.
(r,s)
=(0,0)
(3.14)
In writing the above expression we have used the identities Eq. (A.16) and Eq. (A.21).
Eqs. (3.13) and (3.14) are our results for Hi and H0 , respectively, in the vacuum with
= q/p and with all NS5s intersecting the same IIA four-brane. The expressions for
the vacua with l
= 0 may be obtained by the replacement 2 2 + l1 /q in the above
equations. All other massive vacua can be obtained by displacing the five-branes by periods
of the -torus to yield inequivalent brane configurations. As discussed earlier, for every
choice of p, q and l, there are N k1 such vacua.
Under the duality transformation z1 z1 + 2p1 the quantity H0 is not invariant,
S-duality and by
rather H0 H0 + 2p (1 )H1 . This implies that H0 does not respect
itself cannot be identified with the order parameter in a given vacuum. However, there is a
linear combination of the quantities H0 and Hi that respects the duality symmetries of the
N = 1 vacuum; namely
1
H = H0
(3.15)
(zil )Hi .
k
i
=l
The redefined
constant piece H appears to have the properties required of the average
1
condensate k i Tr i2 . It has dimension 2, modular weight 2, and is invariant under
121
permutations of the gauge group factors and importantly is invariant under -elliptic shifts
of each five-brane, i.e.,
S-duality, which was argued to be a symmetry of each massive
vacuum. As we will see below, up to an additive vacuum-independent
ambiguity, in the

classical limit it indeed corresponds to the average condensate k1 i Tr i2 .
A superpotential for the N = 1 quiver theory
For generic N = 1 deformations of the quiver theory, the superpotential in each
vacuum of the low energy theory takes the value
W =
k
!
1
i Tr i2 .
2
g
(3.16)
i=1
Our inability to map the coordinates H0 and Hi to the physical condensates in each gauge
group factor, implies that we cannot determine this superpotential for generic deformations.
However, for a subclass of deformations of the quiver theory where all the adjoint
masses are equal i =
= 0, the effective superpotential in each vacuum is simply given
by the expectation value of the average condensate:
!
1
1
Tr i2 = 2 kH .
2
g
g
k
W =
(3.17)
i=1
This result for the superpotential in the massive vacua of the N = 1 quiver theory along
with Eqs. (3.14) and (3.15) are among the central results of this paper. When k = 1,
Eq. (3.14) reduces to the expression for the elliptic superpotential for N = 1 gauge theory
obtained in [4,12].
A few remarks about the condensates H , Hi and the superpotential in Eq. (3.17) are in
order. Under SL(2, Z) transformations on , these order parameters in a vacuum with given
p, q and l transform with a definite modular weight into the corresponding condensates
in a vacuum with a different value of p, q and l and with different values of the gauge
couplings. For example under S-duality, H (p, q, l = 0) transforms into 2 H (q, p, l = 0)
and with zij zij . It is also worth noting that in the classical theory (type IIA picture),
the relative positions of adjacent five-branes appear to play a special role in that they
correspond to gauge couplings of the individual SU(N) factors in the quiver. However,
the exact expressions for the condensates in a given vacuum exhibit complete democracy
in the relative positions of any two five-branes in the setup.
Finally, the condensates contain a wealth of information on instanton and fractional
instanton contributions in these theories. This can be clearly seen by studying the
expressions in a semiclassical expansion, i.e., in the g 2 , gi2 1 limit. For example, in
this limit, the condensates in the confining vacuum with p = N, q = 1 and l = 0 may be
expanded in powers of e2i/N and ezij /N . There is no obvious semiclassical origin for
such terms in the 4D theory as they appear to be contributions from objects with fractional
topological charge in both the diagonal and the individual gauge group factors. However, as
seen in [4], in the theory on R3 S 1 , such terms can arise from semiclassical configurations
corresponding to 3D monopoles carrying fractional topological charge.
122
The classical limit

We now demonstrate that H indeed reduces to the average condensate in the classical
theory in the appropriate limit. The classical limit of the theory is obtained by taking
i which corresponds to blowing up the radius of the x 6 -circle, simultaneously
scaling the five-brane positions so that i i in each gauge group factor. In this limit
we should expect our general expression for u2 (z) to yield the condensate in the ith SU(N)
factor provided zi < z < zi+1 , z zi and zi+1 z . For simplicity, we have
taken all the {zi } to be real. The -functions then reduce to (see Appendix A):
(z zj )
(z zj ) 1
+ sign(z zj ),
12
2
(3.18)
1
while the Weierstrass function simply tends to a constant (z zj ) 12
. Note the
presence of discontinuities at the positions of the NS5-branes in the classical limit. The
condensate in the ith SU(N) factor in the classical limit is then
(i)
u2 =

N(N 1) 2 zj
Hj +
mj +
Hj + H0 ,
12
12
k
j =1
j =1
j =1
(3.19)
so that the average condensate is

1
1 (i) N(N 1) 2 zj
Hj +
u2 =
mj +
Hj + H0 .
k
12
12
k
k
i=1
j =1
j =1
(3.20)
i=1 j =1
It is straightforward to see that in the classical limit our definition for H in Eq. (3.15)
approaches precisely this value up to a vacuum-independent additive constant:
H
k
k
i

zj
1
Hj +
Hj + H0 .
12
k
j =1
(3.21)
i=1 j =1
The discrepancy is vacuum-independent because we can trace its origin to the double
pole terms in u2 (z) involving the Weierstrass functions. The order parameters of the theory
and therefore the vacuum-dependent information is completely encoded in the simple poles
and constant pieces, Hi and H0 . The quantity H contains precisely the right combination
of these parameters that reproduces not only the correct semiclassical limit of the average
condensate but also has the right transformation properties under the duality symmetries.
4. Compactification to three dimensions

In the previous section, we solved for the vacuum structure of the mass deformed
theory by lifting the type IIA brane configuration to M-theory. As we saw, the massive
vacua have a very simple interpretation as unbranched multiple covers of the basic
underlying torus E . There is another way to solve the theory involving compactification
on a circle to three dimensions [4,17]. On compactification, the 2n = 2(kN k + 1) real
dimensional Coulomb branch of the quiver theory is enlarged. This is because the U(1)n
gauge field can have a non-trivial Wilson line around the compact dimension, which we
123
choose as x 3 . In addition, the three-dimensional gauge field transforming in the unbroken

U(1)n , on the Coulomb branch, can be exchanged, under S-duality, for n real scalar
fields. In four dimensions, the effective couplings of the U(1)n theory are encoded in the
(n n)-dimensional matrix uv which is precisely the period matrix of the SeibergWitten
Riemann surface . The Wilson line and dual photons naturally pair up to form n complex
scalars Xu which are periodic variables
Xu Xu + 2imu + 2iuv nv ,
mu , nu Z.
(4.1)
This means that the Wilson line and dual photon are valued in the Jacobian variety J().
So the dimension of the Coulomb branch, denoted M, of the three-dimensional theory is 4n
and can be thought of as a fibration of the Jacobian variety J() over the four-dimensional
Coulomb branch . It turns out that M is rather special: it is hyper-Khler space.
The low-energy effective three-dimensional theory is described by a supersymmetric
sigma model with M as target. When an N = 1 mass deformation is added to the
four-dimensional theory, the sigma model is perturbed by a superpotential. The phase
structure can then be determined from the superpotential. The key point that makes the
compactification approach attractive even when one is interested in the four-dimensional
theory is that the superpotential is completely independent of the compactification radius.
Therefore the vacuum structure and the values of condensates apply also in the decompactification limit. By solving the model, we mean finding the exact form for the
superpotential.
For the N = 1 deformation of the N = 4 theory, the form of the superpotential
was arrived at from purely physical reasoning: by imposing all the symmetries of
the low-energy three-dimensional theory [4]. In addition, in the limit of small radius,
semiclassical reasoning applies and provides a strong constraint on the functional form
of the superpotential. However, there is a more direct way to construct the superpotential
which relies on the fact that the SeibergWitten fibration, and hence M, is the phase
space of a complexification of a classically integrable dynamical system8 (for a recent
thorough review of this connection see [26] and references therein). In this description, the
Coulomb branch of the four-dimensional theory is parameterized by the set of n (complex)
Hamiltonians and the fibre is parameterized by n (complex) angle variables. The N = 1
deforming operator is then associated with a particular combination of the Hamiltonians
and this gives the superpotential. The problem of finding the explicit superpotential is then
the problem of finding a convenient and unconstrained parameterization of the phase space
M and then finding the appropriate Hamiltonian describing the N = 1 deformation.
4.1. The three-dimensional Coulomb branch
The first problem is to find a convenient description of the hyper-Khler manifold M.
Remarkably, following the approach adopted by Kapustin [16], this can be achieved by
using a version of mirror symmetry. What results is a description of M in terms of a hyperKhler quotient [27] which can be made very explicit.
8 This means that the dynamical variables, Hamiltonians, etc., are taken to be complex.
124
The most convenient way to describe the mirror transform is to return to the type IIA
brane setup. For now let us ignore the NS5-branes and concentrate on the N D4-branes
alone. We also suppose that the D4-branes are all coincident. We now compactify x 3 on a
circle of radius R , to distinguish it from the radius of x 10 defined earlier. The theory on
the N D4-branes is effectively three-dimensional in the infra-red. The low energy degreesof-freedom include the Wilson line around x 3 which is an adjoint-valued scalar field that
generically breaks the U(N) gauge symmetry of the three-dimensional gauge theory to
U(1)N via the Higgs mechanism. We define the Abelian components of the Wilson line
" 2R a 3
as a = 0
A3 dx , a = 1, . . . , N . Large gauge transformations on the circle imply that
the Wilson line is actually a periodic variable
a a + 2na ,
na Z.
(4.2)
U(1)N
In additional to the Wilson line, is the three-dimensional

gauge field. The bosonic
part of the classical effective three-dimensional action is

#
2R
1 a 2 1 a 2
i
a a
F
F
Scl = 2
(4.3)
d 3x
i
ij k i
jk .
g
4 2 R 2
2 ij
8 2
Here, i, j, k = 0, 1, 2, and we assume that the adjoint scalar in the four-dimensional theory
has no VEV. The gauge fields may be exchanged under S-duality for N real scalars a , the
dual photons. To do this one adds a new term to the action
#
i
(4.4)
d 3 x Dij k i a Fjak ,
2
involving a which acts as Lagrange multipliers for the Bianchi identity. The Abelian field
strength Fija can be now integrated out as a Gaussian field to give

$
%2
#
1
1 a 2
g2
3
a
a
.
i
i +
Scl =
(4.5)
+
d x 2 i
2R
2
g
16 2
Due to the quantization of magnetic charge, the dual photons are also periodic variables:
a a + 2ma ,
ma Z.
(4.6)
The classical effective theory has the form of a sigma model in the N complex scalar fields

Xa = i a + a ,
(4.7)
which are valued on the torus that we defined earlier E with periods 21 2i and
22 2i . The pairing of the Wilson line and dual photons is natural when one includes
supersymmetry since they form the scalar component of a superfield. The effective threedimensional theory, including quantum corrections, is a sigma model involving N chiral
superfields whose scalar components are the fields Xa .
This story can be given a brane interpretation in the following way. We start with the
D4-branes in type IIA wrapped around x 3 with radius R and x 6 with radius L. For small
radius R , we can now perform a T-duality in x 3 to yield the type IIB configuration of
D3-branes
spanning x 0 , x 1 , x 2 , x 6 . Under this duality, the string coupling is transformed

to gs = gs /R . We follow this with a S-duality on the four-dimensional theory on
125
the D3-branes. Under this duality, the three-dimensional effective Abelian gauge field is
exchanged with the Wilson line of the dual gauge field around x 6 . This means that the dual
photons are
2L
#
A a6 dx 6 ,
=
a
(4.8)
where A is the dual gauge field of the D3-branes. Finally, we perform, once again,
a T-duality in x 3 to return to a type IIA configuration with D4-branes spanning
x 0 , x 1 , x 2 , x 3 , x 6 . However, due to the intervening S-duality,
the radius of the x 3 is not
returned to its original value. The new radius is R gs = gs . In other words, it is
independent of the radius R .9 The Wilson line a in the original theory is now identified
with the Wilson line of the dual gauge field component A 3 around the dual x 3 circle:
2R
#
a =
2g
#s
A a3 dx 3.
Aa3 dx 3
0
(4.9)
The theory describing the collective dynamics of these D4-branes is the mirror dual, or
magnetic, theory. It is a five-dimensional theory compactified on R3 T 2 . The most
significant fact is that the torus T 2 has complex structure and, up to an overall rescaling,
is therefore identified with E . At low energies the effective three-dimensional theory is
described by the Wilson lines around the two cycles of the torus which are identified with
the Wilson line and dual photons, (4.9) and (4.8), respectively.
The discussion so far has been simplified because we have ignored the fact that there
are NS5-branes in the original type IIA setup on which the D4-branes can fractionate.
The gauge group of the effective low theory should be U(1)n (recall n = kN k + 1)
rather than U(1)N . So there should be n complex fields associated to n Wilson lines and
n dual photons. Furthermore, these complex fields will be valued on tori which are not
simply copies of E but have renormalized complex structures. This is clear when one
looks at the theory for large R . In that limit, one can think of the problem in two stages.
First, there is an effective four-dimensional theory la Seiberg and Witten. This is a U(1)n
gauge theory with coupling constants encoded in the period matrix uv of . The way
that all these extra Wilson lines and dual photons arise in the dual theory will emerge in
due course. For now we must put back the NS5-branes. Under the first T-duality the NS5branes become type IIB NS5-branes. Then under S-duality they becomes D5-branes (to
distinguish them from the other D-branes in the problem). Finally the T-duality around
x 3 changes them into D4-branes spanning x 0 , x 1 , x 2 , x 4 , x 5 , but localized at points on
the x 3 , x 6 torus. In the original theory, the NS5-branes allow the D4-branes to fractionate
and the gauge group is enhanced from U(N) to SU(N)k U(1). In the dual magnetic
theory, the freedom for the branes to fractionate is encoded by presence of VEVs for the
impurities. The impurities somehow must also encode all the extra Wilson lines and dual
9 All memory of R is not lost because the string coupling in the dual theory is R 2 /(g ).
s
126
photons that we expect, as we shall see later. As usual in a mirror transform we have
mapped the Coulomb branch of the original theory, where the D4-branes were prevented
from moving off the NS5-branes, to the Higgs branch of the magnetic theory, where the
impurities gain a VEV Higgsing the dual U(N) gauge group.
The configuration that we are considering preserves eight real supersymmetries. So we
have a realization of M as the Higgs branch of an U(N) impurity gauge theory with eight
real supercharges. This is why the mirror map is a useful devise. The Higgs branch will not
be subject to quantum corrections and in this way we are able to solve the theory. It is
naturally described by a set of D-flatness equations which involve the, suitably normalized,
components of the dual U(N) gauge field A z,z = 12 (A 3 i A 6 ) of the D4-branes along the
torus10 and the adjoint-valued complex scalar field ,11 describing the fluctuations of the
D4-branes in the x 4 , x 5 direction. This field is directly related to the M-theory picture
described in the previous section: the N eigenvalues of are precisely the va . In addition,
)i ) transforming in the (N , N
the D4-brane impurities give rise to k hypermultiplets (Qi Q
representation of U(N), which are localized on the torus at points zi , i = 1, . . . , k. The Dflatness conditions, with some convenient choice of normalization of the hypermultiplets,
read
k

Q

zz , = 2i
2 (z zi ) Qi Qi Q
F
i i g,
(4.10a)
i=1
z = 2i
D
k
i .
2 (z zi )Qi Q
(4.10b)
i=1
z = z +[A z , ].
Here, (4.10a) is a real equation and (4.10b) is a complex equation and D
These equations are a generalization of Hitchins self-duality equations reduced to two
dimensions [28].
The space M described as the solution to (4.10a) and (4.10b) modulo U(N) gauge
transformations, has another interesting interpretation. Consider a configuration of N D2branes lying inside k coincident D6-branes. The D2-branes are embedded as a charge
N YangMills instanton solution in the U(k) gauge theory on the D6-branes in the four
directions orthogonal to the D2-branes inside the D6-branes. In the brane picture, the
moduli space of the instantons is described by the Higgs branch of an U(N) gauge theory,
). This yields
with eight real supercharges and k hypermultiplets transforming in the (N , N
4
nothing but the ADHM construction of instantons in R U(k) gauge theory. Now imagine
compactifying two spatial dimensions in the D6-branes orthogonal to the D2-branes. The
D2-branes now correspond to YangMills instantons in a U(k) gauge theory on R2 T 2
rather than R4 . Performing a T-duality along each compactified direction, we now have the
configuration of N D4-branes and k D4-branes described above. So the Coulomb branch
of the compactified quiver theory is identified with the moduli space of N instantons in
U(k) gauge theory on R2 T 2 .
10 As previously we perform an overall re-scaling of the torus T 2 so that it becomes parameterized by the
holomorphic coordinate z periodic on = 21 Z 22 Z.
11 Not to be confused with the Wilson line a .
127
A hyper-Khler quotient
Lurking within the generalized Hitchin equations (4.10a) and (4.10b) is a complexified
classical integrable dynamical system and uncovering it will be key to our story of
the superpotential of the three-dimensional theory. But this is jumping ahead. The first
thing to say about these Hitchins equations is they are an example of a hyper-Khler
quotient, although of a rather novel infinite-dimensional kind. The quotient construction
[27] provides a way of constructing a new hyper-Khler manifold M in terms of an old one
with some group of isometries G. These isometries must preserve both the metric and
M
and the three complex structures and so are described by tri-holomorphic Killing vectors
Xr , r = 1, . . . , dim G. Each Killing vector defines a triplet of moment maps
! Xr defined
X
r
by d
! = i(Xr ),
! where
! are the triplet of Khler forms. The quotient space is defined
is defined by (
as the coset manifold M = N/G, where the level set N M
! Xr )1 (0),

for all r, i.e., the subspace of M on which the moment maps vanish. It can be shown that
If M
has real dimension 4n then the above
M inherits the hyper-Khler structure of M.
construction leads to M with dimension 4(n dim G).
In the present context, the novelty arises from the fact that the space M, parameterized
Q and Q,
is infinite-dimensional because and the gauge field A are functions
by , A,
on E . However, the quotient constructionat least formallyproceeds in the usual way.
In this case the quotient group G is the group of U(N) local gauge transformations on E .
by the infinite-dimensional
The idea is that quotienting the infinite-dimensional space M
gauge group leads to a finite-dimensional space M. The Eqs. (4.10a) and (4.10b) are really
a triplet of equations since (4.10b) is a complex equation while (4.10a) is real, and they are
nothing but the moment maps for the local U(N) action.
But, as pointed by Kapustin [16], there is a problem. The fields and A must have
simple poles at the punctures zi with residues that are variable, given as they are by the Q
This means that the variations of and A will have simple poles as well and so
and Q.
their norm will be logarithmically divergent. As a consequence, the corresponding tangent
vectors to M will have infinite norm. Physically this is a different manifestation of the fact
that in the type IIA brane setup the relative centres-of-mass of the stacks of D4-branes are
not genuine moduli. These parameters are, of course, the masses mi of the bi-fundamental
hypermultiplets. All this means is that we have not quite described the correct hyper-Khler
in some way to ensure that
quotient. What we need to do is freeze the impurities Q and Q
the non-normalizable modes are discarded. But we must do this in a way which still leads
to a hyper-Khler manifold. The way to do this was described by Kapustin [16], however,
we will present a re-interpretation of his procedure which is more in keeping with the
quotient construction.
We now hypothesize that the correct construction involves a hyper-Khler quotient by
a larger group which involves not only local U(N) gauge transformations on the torus E ,
but also U(1) transformations at each puncture:
Qai Qai ei ,
ia ei Q
ia .
Q
(4.11)
In fact the additional part of the quotient group is U(1)k1 because the transformation
by the same phase is part of the U(N) gauge group. The
that rotates all the Q and Q
128
associated moment maps are simply

Q

Qai Qia Q
ai ia = 0,
(4.12a)
ia = 0.
Qai Q
(4.12b)
The question is where are the mass parameters? In the hyper-Khler quotient construction,
the definition of the moment maps for any Abelian components of G is ambiguous. Any
constant can be added. When the hyper-Khler quotient construction is viewed as the Higgs
branch of a supersymmetric gauge theory with eight supercharges, these parameters are
the FayetIlliopolos (FI) couplings of the Abelian part of the gauge group. In the present
context, the masses mi are the complex FI couplings associated to the Abelian subgroup
U(1)k which acts as in (4.11). The complex components of the moment maps (4.10b) and
(4.12b) are modified to
%
$
k

m
2

Dz = 2i
(4.13a)
(z zi ) Qi Qi 1[N][N] ,
k
i=1

ia = Nmi .
(4.13b)
Qai Q
a

As before m = i mi . The form of (4.13a) is consistent with (4.13b) in the following way.
The quantity = Tr satisfies
%
$
k

m
2
,
(z zi ) mi
z = 2iN
(4.14)
k
i=1
and hence, is a meromorphic function on E with simple poles at z = zi and residues

N(mi m/k). For consistency the sum of the residues must vanish and this is guaranteed
by (2.4).
We can now go on and find a very concrete realization of the quotient. Firstly, as
we have already mentioned, the real and complex moment maps, (4.10a) and (4.12a)
verses (4.13a) and (4.13b), can be viewed as the D- and F -flatness conditions of a Higgs
branch of supersymmetric gauge theory with four real supercharges. As usual, we can
relax the D-flatness condition and simply impose the F -flatness conditions (4.13a) and
(4.13b) at the expense of modding out by the complexified gauge group. Picking out the
complex moment map amounts to picking out a preferred complex structure from the three
independent complex structures of the hyper-Khler manifold. The associated preferred
Khler form will turn out to be the symplectic form of the complexified dynamical system.
It is convenient to introduce the spins Si , N N matrices at each puncture, with
elements
def
ib m ab .
Siab = Qai Q
(4.15)
k
The complex moment map equations (4.13a) and (4.13b) are then
z = 2i
D
k

i=1
Si 2 (z zi ),
(4.16a)

a
$
%
m
Siaa = N mi
.
k
129
(4.16b)
The constraint (4.16b) is equivalent to Kapustins residue condition, generalizing the

residue condition of Donagi and Witten [3], because it implies that the spins lie in a
particular conjugacy class of the complexified quotient group:
Nm m
Si = Ui
mk
..
1
U ,
i
(4.17)
mk
for elements Ui in the complexified quotient group.12
To proceed, it is very convenient to use up (most of) the local part of the quotient group,
GL(N, C), to transform the anti-holomorphic component Az into a constant diagonal
matrix:
i
1
diag(X1 , . . . , XN ) =
diag(X1 , . . . , XN ).
A z =
(4.18)
2(
2 1
1 2 )
16g 2
The only local transformations that remain act by shifting the Xa by periods of E ,
Xa Xa + 2n1 + 2m2 ,
m, n Z.
(4.19)
The Xa are nothing but the complex combination of the Abelian Wilson lines and dual
photons defined in (4.7). The global part of the gauge group is also fixed, up to permutations
of the Xa and the GL(1, C)N diagonal subgroup, as well as the complexified GL(1, C)k1
flavour symmetries (4.11). Putting these symmetries together, we must mod out by the
action
Qai a Qai i ,
ia 1 Q
ia a1 .
Q
i
(4.20)
We can now solve explicitly for to get a very concrete parameterization of the quotient
space M and the associated dynamical system. Roughly speaking, the elements of
must have simple poles at z = zi to account for the -functions. Candidate functions are
(z zi ) and (z zi )1 ,13 however, these must be put together in the right way to ensure
periodicity on E . After some trial and error, one is led to the solution
aa (z) = pa +
k
Siaa (z zi ),
(4.21)
i=1
for the diagonal components, where the pa are new parameters. There are extra constraints
on the spins, arising from the fact that in the gauge we have chosen, the diagonal elements
aa (z) are meromorphic functions on E and so the sum of the residues must vanish:

Siaa = 0.
(4.22)
i
1
12 To see this identify (U ) = Q /N m and (U 1 ) = Q
ia / N mi . The fact that
i a1
ai
i
1a
a (Ui )1a (Ui )a1 =
i
1 follows from (4.13b).

13 A short review of elliptic functions and their properties is provided in Appendix A.
130
The off-diagonal elements are

k

(Xab + z zi ) (zi ,zi )Xab
e
Siab
ab z, z = e(z,z)Xab
(Xab ) (z zi )
(a
= b).
(4.23)
i=1
Here Xab Xa Xb , and we have defined

def
z, z =

1
1 z 1 z (1 )
2 z 2 z .
(2 )
2 1

1 2
(4.24)
One can readily verify that ab (z, z ) is periodic on E . Furthermore, a shift of Xa by

a lattice vector 23 , can be undone by a large gauge transformation on the torus as
anticipated earlier.
We now have an explicit parameterization of M furnished by pa , Siab and Xa . In order to
determine the dimension of M we now count the number parameters. First of all the spins.
subject to 2k and 2N real constraints, (4.22) and
There are 4Nk real quantities Q and Q
(4.16b), respectively. The group action (4.20) further reduces the number of variables by
2(N + k 1) (the 1 arising from the fact that not all the parameters a and i in (4.20) are
independent). Hence the number of real independent spin variables is 4(Nk N k + 1).
The remaining variables are pa and Xa , giving 4N real parameters. Hence, the overall
dimension of the quotient space M is 4n 4(kN k + 1). From the point of view of the
dynamical system, the variables pa are naturally the momenta conjugate to the
Xa . Notice
X
;
hence
that (z, z ) is onlydependent on the differences Xab
a
b
a pa is not

dynamical. In fact a pa , and its conjugate position a Xa , correspond to the decoupled
overall U(1) factor of the gauge group in the quiver theory. We now choose to set

pa =
Xa = 0.
(4.25)
a
Notice that although we have a concrete parameterization of M, the relation with the
physical parameters of the three-dimensional Coulomb branch is not obvious. In the basic
N = 1 case, recovered by taking k = 1, the spins are completely absent. The coordinates
in this case are pa and Xa . Writing Xa = i( a + a ), the real components a and
a are precisely the dual photons and Wilson lines of the effective U(1)N theory. In the
general case, the relation with the dual photons and Wilson lines and the coordinates is less
obvious. Intuitively, the Xa are dual photons and Wilson lines of the diagonally embedded
SU(N).
4.2. A dynamical system and the exact superpotential
As we have already alluded to above, there is also a completely integrable dynamical
system underlying the construction of M, for which M is the phase space with symplectic
form given by the Khler form singled out by the complex moment map. It is the
rather esoteric spin generalization of the elliptic CalogeroMoser system which was first
described in Ref. [6] and further studied in Refs. [7,8]. The integrable system has an
associated spectral curve which is defined by [7]
F (z, v) = det(v1[N][N] ) = 0.
(4.26)
131
This is precisely the SeibergWitten curve , the branched N -fold cover of E which
appeared in the M-theory interpretation, Eq. (3.5). Since the dynamical system is
completely integrable, there are n (complex) Hamiltonians. These are identified with
coordinates on the Coulomb branch of the four-dimensional theory. The conjugate angle
variables (also complex), Xu , u = 1, . . . , n, take values in the Jacobian variety J().
Finally we have identified all the Wilson lines and dual photons of the three-dimensional
effective theory. The explicit maps between the variables {pa , Xa , Si } and the angle
variables was found in [7,8], but we shall not need them here.
Before proceeding, the resulting equations are cleaner if the spin variables are re-defined
by
Siab Siab e(zi ,zi )Xab .
(4.27)
The Hamiltonians of the dynamical system arise as the residues of the gauge invariant
quantities Tr l (z), l = 1, 2, . . . , N . Since the system is integrable, there will be 2(kN
k + 1) independent Hamiltonians which will be identified with coordinates on the Coulomb
branch of the four-dimensional N = 2 theory that we started with. Of particular importance
(i)
will be the k combinations of Hamiltonians that correspond to the condensates u2 =
Tr i2 for each of the SU(N) factors of the gauge group. They must come, on dimensional
grounds, from expressions quadratic in (z). There are two such terms (Tr (z))2 and
Tr (z)2 . We have
already argued that the quantity that relates to the N = 1 deformation
1
is u2 (z) = 2N a
=b (va (z) vb (z))2 , and since va (z) are the eigenvalues of (z), this
identifies
1
(Tr )2 .
(4.28)
N
We now compute this quantity given our solution for (z) in (4.21) and (4.23).
Since (z) has simple poles, the quadratic invariant u2 (z) has double poles at zi . Since
it is manifestly elliptic, the expansion must be of the form (compare (3.11))
u2 (z) = Tr 2
u2 (z) =
k
i (z zi ) +
i=1
k
(z zi )Hi + H0 ,
(4.29)
i=1

where ki=1 Hi = 0. It is tedious but a straightforward exercise to extract the residues and
constant part. Firstly, the residues of the double poles are constants
2

1 i
i
i
i =
(4.30)
Sab Sba
S
= N(N 1)m2i .
N a aa
ab
The residues of the simple poles are non-trivial functions on M, to wit

$
$
%
%

m
m
Hi = 2
pa Siaa 2N mi
mj
(zij )
k
k
a
+2

a
j (
=i)
j (
=i)
j
Siaa Saa (zij ) 2

a
=b j (
=i)
Siab Sba
(Xab + zj i )
,
(Xab ) (zj i )
(4.31)
132
where zij zi zj . One can verify that

constant piece in the expansion is
H0 =
pa2
a
=b

a
=b i
=j

i
Hi = 0. To complete the expansion, the
Siab Siba (Xab )
i
j
Siab Sba

(Xab + zj i )
(Xab + zj i ) (Xab )
(Xab ) (zj i )
&
$
%$
%'

m
m
1 i j
Saa Saa N mi
mj
(zij ) (zij )2 .
2
k
k
a
i
=j
(4.32)
The k independent functions on M, H0 and Hi , are a subset of the Hamiltonians of the
integrable system.
The Hamiltonians are not simply invariant under these shifts (3.6). Rather the shifts can
be undone by an appropriate transformation on M. To find this transformation, under a
shift of zl zl + 23 , we have

j
j
Hi pa , Xa , Sab zj + 2j l 3 = Hi pa , Xa , Sab zj ,
(4.33)
where
pa
$
$
%%
m
l
= pa 2 (3 ) Saa 2 mi
,
k
j
Sab = Sab e2jl Xab (3 ) .
(4.34)
The remaining Hamiltonian transforms similarly, but with an additive anomaly

j
j
j
H0 pa , Xa , Sab zj + 2j l 3 = H0 pa , Xa , Sab zj + 2(3 )Hl pa , Xa , Sab zj .
(4.35)
We have already described, based on the structure of the massive vacua in the M-theory
picture, how to relate the quantities Hi and H0 , now interpreted as Hamiltonians, to the
condensates. The conclusion was that a basis of functions with the right modular properties
is provided by the Hi and H defined in (3.15). But, using the parameterization of the
Coulomb branch provided by the integrable system, we can now see that the basis {Hi , H }
is precisely the one with good modular properties, not just at the massive vacua, but also at
generic points on the Coulomb branch. To see this, notice that under shifts of zi on E , H
transforms covariantly as the Hi (4.33) since the anomaly piece in (4.35) is compensated.
Based on the semiclassical limit, the quantity H was then identified with the diagonal
combination. Explicitly
H =
pa2
a
=b

Siab Siba (Xab )
2
pa Siaa
k
a
i
=l
%
%
$
$
m
m
(zij )
mj
N mi
k
k
j (
=i)

j
i
Saa Saa (zij ) (zil )
+
a
133
j (
=i)

a
=b
j (Xab + zj i )
Siab Sba
(Xab ) (zj i )
i
=j

2
(zil )
(Xab + zj i ) (Xab ) +
k
l(
=i)

%$
%
$

m
m
1 i j
mj
(zij ) (zij )2 .
Saa Saa N mi
2
k
k
a
i
=j
(4.36)
The exact superpotential of the three-dimensional theory corresponding to the diagonal
N = 1 deformation is then simply W = kH /g 2 .
The massive vacua
The superpotential W = kH /g 2 determines the vacuum structure of the N = 1
deformation of the theory. A full analysis of the vacuum structure is beyond the scope of
the present work. Actually, even in the N = 4 case, where the superpotential is a good deal
simpler, there is only a systematic treatment of the massive vacua for N 3 [4]. We shall
achieve as much for the finite N = 2 theories.
The massive vacua have a very beautiful interpretation from the point-of-view of the
dynamical system: they are precisely equilibrium configurations with respect to the space
of flows defined by the n Hamiltonians.14 To see this, recall that the massive vacua
correspond to points of the four-dimensional Coulomb branch for which degenerates
to a torus: cycles pinch off and one is left with an N -fold unbranched cover of E . This
means that the Jacobian variety J() itself degenerates: at these points the period matrix
only has rank 1, with non-zero eigenvalue . The remaining torus is associated with the
overall U(1) factor of the gauge group which we have removed from the integrable system
(4.25). So at a massive vacuum, the remaining angle variables must stay fixed under any
time evolution. Since the Hamiltonians are by definition constants of the motion, this means
that the entire dynamical system must be static at a massive vacuum and the system is at
an equilibrium point. Consequently, a massive vacuum is not only a critical point of the
14 Here, time is an auxiliary concept referring to evolution in the dynamical system and not a spacetime
concept in the field theories under consideration.
134
Hamiltonian describing the N = 1 deformation, but simultaneously of all the other n 1

Hamiltonians. On the other hand, for the massless vacua, this is no longer true.
First of all, it is instructive to recall some details of the N = 4 case described in [4]
which is recovered in our formalism by setting k = 1. This will provide an important clue
for quiver theories. When k = 1, the constraints on the single spin mean that it is not
dynamical:
k = 1:
Sab = m(1 ab ).
(4.37)
There is a single quadratic Hamiltonian,

pa2 m2
(Xab ).
k = 1: H0 =
a
(4.38)
a
=b
At the critical points of H0 the momenta pa conjugate to Xa vanish. The massive vacua are
associated to the finer lattices which contain as a sublattice of index N . This means
that there are N points of contained in a period parallelogram of . The simplest kind,
labelled by two integers p and q with pq = N , are generated by 21 /q and 22 /p. All
the other cases can be generated from these by modular transformations, as we shall see
later. Each a {1, . . . , N} is uniquely associated to the pair (ra , sa ), with 0 ra < q and
0 sa < p. The critical point of H0 associated to (q, p) is then
Xa =
2ra
2sa
1 +
2 ,
q
p
0 r < q, 0 s < p.
(4.39)
One only needs to use the

The proof that this is a critical point of H0 is delightfully simple.

fact that (z) is an odd elliptic function. Terms in the sum b(
=a) (Xab ) either cancel in
pairs or vanish because Xab is a half-lattice point. As we mentioned, the set (4.39) does not
exhaust the set of massive vacua. For a given pair (q, p) we can generate p 1 additional
vacua by performing the suitable modular transformation on to give
$
%
2ra
2sa
l
2 + 1 , 0 l < p.
1 +
Xa =
(4.40)
q
p
q

So the total number of massive vacua is equal to p|N p, as expected on the basis of the
semiclassical analysis.
Returning to the finite N = 2 theories, we hypothesize that the massive vacua are
also associated to the finer lattices with Xa given in (4.40). Once again we can
start with the configurations (4.39) and the additional massive vacua will be obtained by
modular transformations. We will now find a set of critical points common to each of the
Hamiltonians Hi and H0 , and so, by implication, of H . Firstly, by varying H0 with respect
to the pa , we find, as in the N = 4 case described above, that pa = 0. Varying each Hi
with respect to pa , one finds
m
Siaa = mi .
(4.41)
k
ia ) equations
What remains is to impose the Xa and Siab (more properly the Qai and Q
and hence find the spins at the critical points. Rather than write down these equations and
find their solutions, which is necessarily complicated and uninspiring, we can motivate the
135
form of the solution and then verify that the ansatz is a critical point. The critical point
equations are significantly simplified if, for a
= b,
j
Siab Sba
(Xab + zj i )
,
(Xab ) (zj i )
(4.42)
with Xa as in (4.39), are periodic in the indices ra (modq) and sa (modp). This can be
achieved if
Siab ira rb si a sb e
2zi
ra rb
q
(1 )+
sa sb
p
(2 )
(4.43)
Here, i and i are arbitrary qth and pth roots of unity, respectively. In fact, this periodicity
requirement determines the spins completely, up to an overall factor which is easily fixed.
With a little more work one is lead to the ansatz
Siab = mi ira rb si a sb e
2zi
ra rb
q
(1 )+
sa sb
p
(2 )
m
ab .
k
(4.44)
Notice that this is consistent with (4.15) where

2z ra ( )+ saps (2 )
mi ira si a e i q 1
,
ra

s
a
ia = mi ra sa e2zi q (1 )+ p (2 ) .
Q
i
i
Qai =
(4.45)
In particular, the constraints (4.22) and (4.16b) are satisfied. In addition, the solution is
consistent with (4.41). The undetermined roots of unity i and j label inequivalent critical
points. Hence the number of critical points appears to be pk q k = N k . But this over counts.
The residual U(1)k transformations described previously can be used to set i = i = 1
for one particular 1 i k, and so there is an N k1 additional degeneracy of vacua for
each N = pq: precisely the same counting that we found in (2.9) for the massive vacua.
This gives an important clue that the critical points we have found are to be identified
with the massive vacua. But there is more. We saw that the massive vacua were related by
(21 , 22 ) translations of the punctures zi . This should be reflected by the critical points
of the superpotential. Consider the quantity (4.42) with the spins as in (4.44). Under the
shift zi zi + 2m1 + 2n2 , for a
= b,
j
Siab Sba
s s
r r
(Xab + zj i )
j (Xab + zj i )
2im a p b +2in a q b i
e
.
Sab Sba
(Xab ) (zj i )
(Xab ) (zj i )
(4.46)
To show this one has to employ the identity 2 (1 ) 1 (2 ) = i/2. Consequently,

these transformations change the roots of unity labelling the vacua as
zi zi + 21 :
i i e
zi zi + 22 :
i i ,
i
2
p
, i i ,
i i e
2 i
q
(4.47)
Hence, as found earlier, shifts of the punctures by the lattice (21 , 22 ) does indeed
permute the N k1 massive vacua for a given pq = N . Shifts on the larger lattice, generated
by (2
1 , 2
2 ) (2p1 , 2q2 ), leave the vacua invariant. (Notice that the relation of p and
q to 1 and 2 swaps over relative to (4.39).)
136
Proof. We now prove that our ansatz (4.39) and (4.44) is a critical point of Hi and H0 .
First of all, consider the Xa variations. The Xa -derivative of the second term of H0 in
Eq. (4.32) is

2
(4.48)
m2i
(Xab ).
i
b(
=a)
This vanishes for the same reason as in the k = 1 case. For each a and b there exists a
unique b (possibly b = b) such that
Xab = Xb a mod .
(4.49)
b
= b ,
since (Xab ) + (Xab ) = 0, or

Then we can see that terms cancel in pairs when

(Xab ) = 0 when b = b , since Xab is then a half-lattice point and (z) is an odd elliptic
function. Now consider the Xa -derivatives of the third term of H0 , in (4.32), and the fourth
term of Hi , in (4.31). In both cases, the resulting expression involves terms like

j (Xab + zj i )
j (Xba + zj i )
(4.50)
f (Xab ) Siba Sab
f (Xba ) ,
Siab Sba
(Xab ) (zj i )
(Xba ) (zj i )
b(
=a)
for some function f (Xab ) elliptic in Xab . Using the special periodicity properties of the
quantity (4.42), one can show
j
Siba Sab
(Xba + zj i )
j (Xab + zj i )
= Siab Sb a
.
(Xba ) (zj i )
(Xab ) (zj i )
(4.51)
Furthermore, f (Xba , zj i ) = f (Xab , zj i ). After re-labelling the second term in Eq. (4.50)
with b b, we see that the two terms in (4.50) cancel.
Now we turn to the spins. It is convenient to parameterize
ia = yai .
Qai = tai ,
(4.52)
Q
tai
The constraints (4.22) and (4.16b) are then linear in yai :

yai = Nmi ,
yai = m.
a
(4.53)
The symmetries (4.20) can be used to set ta1 = t1i = 1. Now consider
%

$

H0
m
(zij ) (zij )2
yaj
= 2
ybi (Xab )
yai
k
b(
=a)
j (
=i)

2 i j (Xba + zj i )
(Xba + zj i ) (Xba )
+
Sba Sab
yai
(Xba ) (zj i )
b(
=a) j (
=i)
+ Bi + Ca .
(4.54)
Here, Bi and Ca are the Lagrange multipliers for the constraints (4.53). In order to show
that the derivative vanishes, it is sufficient to show that the three terms in (4.54), besides
the Lagrange multipliers, are independent of a, since then one can solve for the Lagrange
multipliers. Recall that our ansatz for the solution has yai = mi /N , independent of a. The
137
second term in (4.54) is then manifestlyindependent of a. Contrary to appearances, the

first term is also independent of a, since b(
=a) (Xab ) is, itself, independent of a due to
the form of Xa (4.39) and the elliptic periodicity of (z). The second term involves

j (Xba + zj i )
(Xba + zj i ) (Xba ) .
(4.55)
Siba Sab
(Xba ) (zj i )
b(
=a)
This is also independent of a. To show this, one uses the special periodicity property that
we established for the quantity (4.42). Hence, there exists Lagrange multipliers such that
H0 /yai = 0. The same kind of reasoning shows that Hj /yai = 0.
Finally, the tai derivatives of Hi and H0 involve quantities like

1
j (Xab + zj i )
f (Xab , zj i )
Siab Sba
tai
(Xab ) (zj i )
b(
=a)

j (Xba + zj i )
(4.56)
f (Xba , zj i ) ,
Siba Sab
(Xba ) (zj i )
where f (Xab , zj i ) is elliptic in Xab . But this is of the form (4.50) which we have already
shown vanishes.
As in the softly broken N = 4 case, there are additional massive vacua that are obtained
by modular transformations in associated to the more general configurations (4.40).
Finally, we can evaluate the Hamiltonians on the vacua. It suffices to pick the vacua with
i = i = 1 and l = 0, since all the others are related either by modular transformations
or shifts of the {zi }. One finds
H |l=0;i =i =1

= N
m2i
i

(r,s)
=(0,0)
2r
2s
1 + 2
q
p
mi mj N 2 (z
ij ) + N (zij )

i
=j

2r
2s
2
+N
(sr ) (q2 ) (p1 )
q
p
(r,s)
=(0,0)

(zij + sr ) (sr )
&
'

+ N (N 1) (zij )
(zij + sr ) (sr )
(r,s)
=(0,0)
&
2
(zil )
(zij )
k
l(
=i)
(N 1)

(r,s)
=(0,0)
'
2r
2s
1 + 2
q
p
%
,
(4.57)
along with
138
Hi |l=0;i =i =1
= 2N
mi mj
&
(N 1) (zij )
j (
=i)
'

(zij + sr ) (sr ) .
(4.58)
(r,s)
=(0,0)
The expression for the residues, Hi , above matches (3.13) precisely. It now remains to
show that the expression for H in (4.57) matches with the M-theory result (3.15) using the
expression for H0 in (3.14). This can be achieved by noting that both the expressions for
H are in fact -elliptic in the variables zij . It is then a straightforward but tedious exercise
(making use of various identities provided in Appendix A) to show that the residues of
the double poles and simple poles and the constant pieces of these expressions treated as
functions of zij , are indeed identical. As the expressions are -elliptic in the zij , this is
sufficient to demonstrate that they are in fact the same.
Recall that in the M-theory picture one has a freedom to choose which of the N branches
of the covering each NS5-brane lies on. This freedom, represented by the parameters
(si , ri ) in (3.9), is encoded in the integrable system as the N choices for the roots of unity
i and i . In particular, the covering where all the NS5-branes lie on the same branch,
(si , ri ) = (0, 0) corresponds to i = i = 1.
5. Conclusions and future directions

In this paper we have provided a classification of the vacuum structure and duality
properties of the N = 1 deformations (mass-deformations) of the N = 2 quiver theories.
We have also obtained exact results for certain chiral condensates in the massive vacua
of these theories, following two completely different approaches, namely: (i) by lifting
the corresponding type IIA brane setups to M-theory and (ii) by studying the theory
on R3 S 1 . For a certain class of mass-deformations, both approaches were used to
independently evaluate the exact superpotential (Eqs. (3.13), (3.14) and (3.17)) in each
massive vacuum and extremely non-trivial agreement was found. In particular, one of the
main results of this paper is an exact superpotential Eq. (4.36) for the theory on R3 S 1 ,
which coincides with a certain linear combination of the quadratic Hamiltonians of the
spin-generalization of the elliptic CalogeroMoser system. This is a generalization of the
corresponding results for the mass-deformed N = 4 theory obtained in [4]. Although we
have only concentrated on the massive vacua, the superpotential (4.36) also determines all
the massless vacua. However, even in the basic k = 1 case the structure of the massless
vacua is not known beyond N = 3. It would be interesting to understand the structure of
these vacua as well.
Some immediate applications of our results include the calculation of physical
quantities such as the gluino condensate and tensions of domain walls interpolating
between the massive vacua [19,30]. In particular, these quantities may be evaluated in
the large g 2 N , large-N limit for a direct comparison with the appropriate deformation of
the type IIB backgrounds on AdS5 S 5 /Zk . In the k = 1 case, i.e., in the N = 1 theory,
the corresponding string backgrounds [11] were asymptotically AdS5 S 5 containing D3branes polarized to 5-branes in the interior, wrapping 2-cycles of the S 5 . From the point
139
of view of the string dual, it would be interesting to understand the characterisation of the
extra vacua which arise in the k > 1 theories. As remarked in [11], they are presumably
associated with the values of twisted sector fields.
The superpotential Eq. (4.36) and the expressions for the condensates Eqs. (3.13)
(3.15) and (4.57) contain a wealth of information regarding instanton and fractional
instanton contributions in the massive vacua. In the three-dimensional picture, the
superpotential receives contributions from semiclassical monopoles carrying fractional
topological charge. The nature of these contributions is visible in the condensates in
the semiclassical limit. For example, in the N k1 confining vacua with p = N , q = 1
and l = 0, the semiclassical expansion reveals contributions from instantons as well as
fractional instantons in each gauge group factor. In the k = 1 theory, i.e., mass-deformed
N = 4 SUSY YangMills, in the large g 2 N limit, the fractional instanton series can be
resummed using
S-duality to obtain a dual expansion. Terms in this dual expansion can
be elegantly described in the IIB string dual of Polchinski and Strassler as arising from
world-sheet instantons wrapping the 2-cycles of the 5-branes polarized from the D3-branes.
While one expects a similar interpretation to arise in the k > 1 theories, the appearance of
different types of fractional instantons (from each gauge group factor) and their associated
actions needs to be understood better from the point of view of the string dual.
Finally, we point out that in the mass-deformed N = 4 theory on R3 S 1 , the
appearance of the elliptic superpotential [4] encoding pairwise interactions between N
particles on a torus can be given a very nice, physical interpretation. Realizing the
compactified N = 4 theory on N D3-branes wrapped on a circle, one may perform
T-duality and lift the resulting setup of N D2-branes to M-theory. We thus obtain N
M2-branes with two transverse compact directions (the second compact direction being
the M-dimension), which may now be thought of as the N particles on a torus. Upon
introducing the N = 1 mass-deformation the M2-branes exert forces on each other which
is described by the Weierstrass superpotential obtained in [4]. It would be extremely
interesting to generalize this picture to the superpotential Eq. (4.36) for the N = 1 quiver
theory and obtain an interpretation for the spinspin interactions in the Hamiltonian in
terms of appropriate interactions between M2-branes.
Appendix A. Some properties of elliptic functions
In this appendix we provide some usefulbut far from completedetails of elliptic
functions and their near cousins. For a more complete treatment we refer the reader to one
of the textbooks, for example [24]. An elliptic function f (z) is a function on the complex
plane, periodic in two periods 21 and 22 . We will define the lattice = 21 Z 22 Z
and define the basic period parallelogram as
D = {z = 21 + 22 , 0 < 1, 0 < 1}.
(A.1)
The archetypal elliptic function is the Weierstrass (z) function. It is an even function
which can be defined via

1
1
1
,
(z) = 2 +
(A.2)
z
(z 2m1 2n2 )2 (2m1 + 2n2 )2
m,n
=(0,0)
140
where the sums is over all integer pairs except m = n = 0. The function (z) is analytic
throughout D, except at z = 0 where it has a double pole:

1
(A.3)
+ O z2 .
z2
The other two important functions for our purposes, are (z) and (z). They are both
odd functions but are not elliptic, since they have anomalous transformations under shifts
by periods:
(z) =
(z + 23 ) = (z) + 2 (3 ),
(z + 23 ) = (z)e2(z+3 )(3 ) .
(A.4)
The three functions are related via

(z) =
(z)
,
(z)
(z) = (z).
(A.5)
In D, (z) has a simple pole and (z) a simple zero at z = 0:

1
+ O z3 ,
(z) = z + O z5 .
z
Some useful identities for (z) and (z) evaluated on half-periods are
(z) =
(A.6)
i
,
2
(1 + 2 ) = (1 ) + (2 ),
(A.7b)
(1 + 2 ) + (1 ) + (2 ) = 0.
(A.7c)
2 (1 ) 1 (2 ) =
(A.7a)
An elliptic function f (z) can always be expressed as

f (z) = c0 +
n

ck,1 (z ak ) + + ck,rk (rk 1) (z ak ) ,
(A.8)
k=1
for constants
c0 and ck,i with the restriction that the sum of the simple pole residues
vanishes, nk=1 c1,k = 0. In particular, an elliptic function which has no poles is a constant.
Of particular importance to us is the behaviour of our basic functions under modular
transformations of the complex structure of the torus defined by . These SL(2, Z)
transformations act as
$
%$ %
$ %
a b
2
2
,
(A.9)
c d
1
1
for a, b, c, d Z subject to ad bc = 1. Since we choose 1 = i and 2 = i ,
after the transformation (A.9), one has to perform a re-scaling by (c + d)1 so that the
transformation on has the usual form:
a + b
.
c + d
A function f (z) has modular weight w if

f z = (c + d)w f z(c + d)| .
=
(A.10)
(A.11)
141
The functions (z), (z) and (z) have modular weights 2, 1 and 1, respectively.
We will need to explore the semiclassical limit of our functions. This is the limit g 2 0,
i or 2 with fixed 1 = i . In this limit, for z D,
(z)
1
1
+
,
12 4 sinh2 2z
z
1
z
+ coth ,
12 2
2
z
2
(z) 2ez /24 sinh .
2
(z)
(A.12)
In particular, for fixed 2 < i < 2,

lim
(i 2 ) =

i
i = 0,
1
sign(i ).
2
(A.13)
Below we collect, without proof, various useful identities. The untilded functions are
defined with respect to the lattice = 21 Z 22 Z while the tilded functions are defined
with respect to the lattice = 2p1 Z 2q2 Z.
(z) =
(z + 2s1 + 2r2 )
(r,s)
=(0,0)
(r,s)
q (p1 ) (1 )
z,
1

(z
+ 2s1 + 2r2 )
(z) =
(r,s)

(r,s)
=(0,0)
(2s1 + 2r2 )
(A.14)
(2s
1 + 2r2 ),
(A.15)
(r,s)
=(0,0)
2r
2s
1 + 2 = pq
q
p
(2s
1 + 2r2 ),
(A.16)
(r,s)
=(0,0)
2
$
%
2s
2s
2r
2r
1 + 2 (1 ) (2 )
q
p
q
p
(r,s)
=(0,0)
$
%
pq 2
2r
2s
1 + 2 ,
=
pq
q
p
(r,s)
=(0,0)
%

$ 2r
2r
2s
2s
2i rrq +2i ssp
1 + 2 (1 ) (2 ) e
q
p
q
p
(r,s)
=(0,0)

2r

2s
= pq 2s 1 + 2r 2
(q2 )
(p1 ) ,
q
p

2
(x + y) (x) (y) = (x + y) + (x) + (y).

For i i = 0
(A.17)
(A.18)
(A.19)
142

i
2
i (zi )
2i (zi ) +
Finally

(r,s)
=(0,0)

1
1
i j 2(zi )(zj zi ) (zi zj ) + (zi zj )2 .
2
2
i
=j
(A.20)
%

2r
2s
q
pq
1 + 2 =
E2 ( ) E2 ,
q
p
12
p
(A.21)
where E2 ( ) is the second Eisenstein series [29] which has the modular transformation
properties E2 ( + 1) = E2 ( ) and E2 ( ) = E2 (1/ )/ 2 6/i .
Appendix B. Rotating the brane configuration
For the non-elliptic models it is well established that the breaking supersymmetry by
adding mass terms for the adjoint chiral multiplets can be realized in the type IIA brane
configurations by relative rotations of the NS5-branes into the w = x 8 + ix 9 direction [20].
This way of realizing soft breaking to N = 1 in the non-elliptic models gives a very simple
picture of why the Riemann surface has to degenerate at a vacuum. The point is that
only very particular surfaces are rotatable. Let us quickly review the argument in the
context of a non-elliptic model with two NS5-branes describing a model with U(N) gauge
symmetry. Breaking to N = 1 is achieved by rotating the first NS5-brane into the (v, w)
plane, along the line w = v, while leaving the second NS5-brane intact. Since v diverges
at the positions of the five-branes, w must diverge at the first five-brane like v, but vanish
at the position of the second five-brane. Since is a free parameter we can construct the
order-by-order in . Working to first order in allows us to find the
rotated surface
constraints on the original in order that it be rotatable. To first order, the problem is
to find a meromorphic function w on the initial surface , which has a simple pole at
the position of the first five-brane.15 But a function with a single simple pole cannot exist
on surfaces of genus > 0, and so in order that be rotatable it must have completely
degenerated to a surface of genus 0.
The goal of the present section will be to generalize the brane rotation story to the
elliptic models. We shall see that the situation is rather more involved but the essence of
the problem is the same. The condition that a surface be rotatable will boil down to
the existence, or otherwise, a certain meromorphic function on with a prescribed pole
structure. To start with we consider the original N = 1 models where there is only a single
NS5-brane. The obvious problem is that there is no immediate sense rotating a single NS5brane. Thinking about the periodicity in terms of an infinite string of images, what we
need to do is rotate each image relative to the last. The situation is rather similar to the
15 Technically is non-compact at the positions of the five-brane, so we must compactify it by adding these
points to get a compact surface.
143
introduction of the global N = 2 mass m. Recall that in order to introduce this mass we
had to modify the spacetime itself introducing a non-trivial bundle over E . We need to do
an analogous thing in order to incorporate the N = 1 rotation. The twist acts as a complex
rotation on the complex combination g = v + iw:
(
1 + i
i
i
g e g, e
(B.1)
,
1 i
where is the N = 1 mass. Comparing with the discussion of how the global mass m was
introduced, we can trivialize the bundle at the expense introducing a suitable singularity at
an arbitrary point of E . Recalling that the shift v v + m required that v(z) to a simple
pole at an arbitrary point (chosen to be z = 0) on each sheet with residue m/N , we see
that g(z) must have an essential singularity of the form
$
%
i/N
g(z) exp
(B.2)
,
z z0
at a new arbitrary point z0 on each sheet. The function g(z) must also have simple poles at
z = 0, as before to incorporate the global mass m, on each sheet and also a simple pole of
residue m at z = z1 on the single sheet where the NS5-brane is located.
We now follow the logic of the non-elliptic case and work to first order in in order
to derive the condition that is rotatable. To leading order we think of g(z) as a function
on . When we turn off the rotation g(z) v (0) (z), where v (0) (z) describes the unrotated
surface . Consider the function h(z) = g(z) v (0) (z). The simple pole at the NS5-brane
is now cancelled and so h(z) has simple poles at z = 0 on each sheet, of the same residue,
and essential singularities
at z = z0 of the form (B.2) on each sheet. It is instructive to
consider the function a ha (z), a sum over the images of h(z) on each sheet. This is a
bona fide meromorphic function on the torus E itself with a simple pole at z = 0 and an
essential singularity of the form (B.2) at z = z0 . The unique function with these properties
is
(z + i/N) i (zz0 )/N
e
.
(z)
(B.3)
We can find a necessary condition that be rotatable by working to first order in (or ).
To this order, we expand (B.2) to find simple poles at z = z0 , with residue i/N on each
sheet. So the surface will be rotatable provided there exists a meromorphic function h(z)
on , more properly its compactification, with simple poles on each sheet at z = 0 and z0
of equal and opposite residue i/N , respectively.16
We can translate the condition into something more convenient by the following chain of
arguments. Since z0 is arbitrary, we can take z0 0. In this limit the simple poles merge
together in pairs on each sheet yielding double poles. So will be rotatable provided
there exists a meromorphic function on with double poles at z = 0 on each sheet with
16 This last requirement follows from the fact that h (z) is a meromorphic function on the torus E with
a a
two simple poles the sum of whose residues must vanish, as can be verified by expanding (B.3) to linear order
in .
144
the same coefficient and vanishing residue. Finally, by taking a linear combination of this
function, v (0) (z)2 and v (0) (z) we can find a function which is now regular at z = 0, but
now has a double pole at z = z1 on the single sheet where the NS5-brane is located. So a
necessary condition that be rotatable is that there exists a meromorphic function f (z)
on its compactification which has a single double pole at the position of the NS5-brane.
The RiemannRoch theorem implies that, generically, f (z) can only exist on surfaces
with genus 0 or 1. So generically in our example, in order for to be rotatable it must
completely degenerate to an unbranched (unramified) N -fold cover of the torus. The
function with a single double pole is then the Weierstrass function. In this case, as we
shall see, the theory has a mass gap (ignoring the overall U(1) factor). Exceptionally,
however, a suitable function f (z) can exist on a higher genus surface, in which case the
deformed theory is massless since there is an unbroken U(1)g1 symmetry (in addition
to the overall U(1) factor) where g is the genus. In fact, we can say more about these
exceptional surfaces. In order that there exists a meromorphic function with a single double
pole at the position of the NS5-brane must necessarily be hyper-elliptic and moreover
the NS5-brane must necessarily be positioned at one of the 2g + 2 Weierstrass points of
the surface.
The generalization to the case with more NS5-branes is now clear. As well as an
overall rotation, there are k 1
relative rotations of the NS5-branes parameterized by
i = i /k, i = 1, . . . , k and i i = . The surface will be rotatable if h(z) has the
pole structure established in the k = 1 case above but, in addition, has a simple pole at the
position of the ith NS5-brane with residue i for i = 1, . . . , k. For example, if = 0, then
h(z) only has simple poles at the NS5-branes with residues i . As before the Riemann
Roch theorem dictates that such a function will generically only exist when undergoes
complete degeneration to the torus. However, exceptionally there will exist higher genus
surfaces which admit such a function. Notice that these exceptional g > 1 surfaces will
depend on the i : in other words, as we vary the N = 1 deformation the rotatable surface
will change accordingly.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Phys. Rep. 323 (2000) 183, hep-th/9905111.
M.R. Douglas, G. Moore, hep-th/9603167.
R. Donagi, E. Witten, Nucl. Phys. B 460 (1996) 299, hep-th/951010.
N. Dorey, JHEP 9907 (1999) 021, hep-th/9906011.
E. Witten, Nucl. Phys. B 500 (1997) 3, hep-th/9703166.
J. Gibbons, T. Hermsen, Physica 11D (1984) 337.
I. Krichever, O. Babelon, E. Billey, M. Talon, hep-th/9411160.
N. Nekrasov, Commun. Math. Phys. 180 (1996) 587, hep-th/9503157.
S. Kachru, E. Silverstein, Phys. Rev. Lett. 80 (1998) 4855, hep-th/9802183.
N. Seiberg, E. Witten, Nucl. Phys. B 426 (1994) 19, hep-th/9407087.
J. Polchinski, M. Strassler, hep-th/0003136.
N. Dorey, O. Aharony, S.P. Kumar, JHEP 0006 (2000) 026, hep-th/0006008.
A. Gorsky, I. Krichever, A. Marshakov, A. Mironov, A. Morozov, Phys. Lett. B 355 (1995) 466, hepth/9505035;
E. Martinec, N. Warner, Nucl. Phys. B 459 (1996) 97, hep-th/9509161.
[14] E.J. Martinec, Phys. Lett. B 367 (1996) 91, hep-th/9510204.
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
M.A. Olshanetsky, A.M. Perelomov, Phys. Rep. 71 (1981) 313.

A. Kapustin, Nucl. Phys. B 534 (1998) 531, hep-th/9804069.
N. Seiberg, E. Witten, hep-th/9607163.
K. Intriligator, N. Seiberg, Phys. Lett. B 387 (1996) 513, hep-th/9607207.
N. Dorey, S.P. Kumar, JHEP 0002 (2000) 006, hep-th/0001103.
K. Hori, H. Ooguri, Y. Oz, Adv. Theor. Math. Phys. 1 (1998) 1, hep-th/9706082.
C. Vafa, E. Witten, Nucl. Phys. B 431 (1994) 3, hep-th/9408074.
J. Polchinski, Int. J. Mod. Phys. A 16 (2001) 707, hep-th/0011193.
S. Katz, P. Mayr, C. Vafa, Adv. Theor. Math. Phys. 1 (1998) 53, hep-th/9706110.
E.T. Whittaker, G.N. Watson, A Course of Modern Analysis, 4th edn., Cambridge Univ. Press, 1927.
N. Dorey, V.V. Khoze, M.P. Mattis, Phys. Lett. B 396 (1997) 141, hep-th/9612231.
A. Gorsky, A. Mironov, hep-th/0011197.
N.J. Hitchin, A. Karlhede, U. Lindstrom, M. Rocek, Commun. Math. Phys. 108 (1987) 535.
N. Hitchin, Duke Math. J. 54 (1987) 91.
N. Koblitz, Introduction to Elliptic Curves and Modular Forms, Springer, 1984.
N. Dorey, T.J. Hollowood, S.P. Kumar, in preparation.
145

String equation for string theory on a circle

Ivan K. Kostov
Service de Physique Thorique, CNRSURA 2306, C.E.A. Saclay, F-91191 Gif-Sur-Yvette, France
Abstract
We derive a constraint (string equation) which together with the Toda Lattice hierarchy determines
completely the integrable structure of the compactified 2D string theory. The form of the constraint
depends on a continuous parameter, the compactification radius R. We show how to use the string
equation to calculate the free energy and the correlation functions in the dispersionless limit. We
sketch the phase diagram and the flow structure and point out that there are two UV critical
points, one of which (the sine-Liouville string theory) describes infinitely strong vortex or tachyon
perturbation. 2002 Published by Elsevier Science B.V.
PACS: 04.20.Jb; 04.60.-m; 04.60.Kz; 05.70.Jk
1. Introduction
Bosonic string theory with a two-dimensional target space, or c = 1 string theory (see,
e.g., the review [1]), is commonly considered as completely solved. In fact, this is not quite
true, and some of the most intriguing questions here are still waiting to be answered. One
of these questions is whether strong perturbations of the world sheet action can change the
geometry of the target space. According to the FZZ conjecture [2], the compactified c = 1
string theory in presence of a sufficiently strong vortex or tachyon source is related by
S-duality to a 2d string theory in a Euclidean black hole background [3,4]. The string theory
in a strong vortex background forms a new phase, which we call the sine-Liouville phase,
because the role of the Liouville potential here is played by a sine-Liouville potential. 1
A vortex perturbation of the 2d string theory is equivalent to a time-dependent tachyon
perturbation in the T-dual theory. The simplest nontrivial example of such a perturbation is
the sine-Gordon model coupled to quantum gravity, which was first studied by Moore [6].
E-mail address: kostov@spht.saclay.cea.fr (I.K. Kostov).
1 The relation between the topological version of the 2d Euclidean black hole and the c = 1 string compactified
at the self-dual radius has been previously pointed out by Mukhi and Vafa [5].
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 5 0 - 2
I.K. Kostov / Nuclear Physics B 624 (2002) 146162
147
Some interesting results followed for the general case, in particular, the discovery of the
integrable structure of the Toda Lattice hierarchy [7].
Recently, this problem has been reconsidered in [8], where the fact that the flows
generated by allowing vortices on the world sheet are described by the Toda Lattice
hierarchy was used. In this paper we complete the approach of [8] by deriving a constraint
(a string equation), which, together with the Toda Lattice hierarchy, completely
determines the theory. We consider in more detail the dispersionless (genus zero) limit
and show how to use the string equation to perform calculations. We then check that the
string equation correctly reproduces the free energy in the presence of a pair of vortex
operators obtained in [8], as well as the one- and two-point correlators in the Liouville
phase, recently calculated in [9]. Finally, we discuss the phase structure of the theory and
give a qualitative description of the sine-Liouville phase.
2. Tachyon and vortex sources in the compactified c = 1 string theory
Euclidean 2D string theory in a flat background is described on the string worldsheet
by the conformal field theory of a massless scalar x coupled to a c = 25 Liouville field ,
with worldsheet Lagrangian
L=

1
+ e2 + Lghost.
(x)2 + ()2 2R
4
(1)
If the c = 1 coordinate is compactified at radius 2 R, then the theory has a discrete spectrum
and the excitations carrying the momentum and the winding modes are represented by the
tachyon operators Tn and the vortex operators Vn (n = 0, 1, 2, . . .)

Tn
(2)
einx/R e(|n|/R2) ,
world sheet
Vn
einxR
e(|n|R2) ,
(3)
world sheet
where the field x is T-dual to x. The vortex operators can be created as topological defects
describing localized winding modes: a vortex of charge nR is located at the endpoint of a
line along which the time coordinate has discontinuity 2nR.
3. Matrix model formulation and Toda Lattice symmetry

The c = 1 string theory compactified at length 2R can be constructed as a large
N matrix model (Matrix Quantum Mechanics), which can be viewed as a dimensional
reduction of an N -color 2d YangMills theory (for details see the review [1]). After being
dimensionally reduced to the circle x + 2R x, the YM theory is described in terms
2 We will use units in which the T-duality acts as R 1/R and thus R
KT = 2.
148

j
of two Hermitian N N matrix fields, the Higgs field M = Mi (x) and the gauge field
j
A = Ai (x). It is more convenient to consider the grand canonical partition function, in
which the chemical potential plays the role of a cosmological constant:
Z(, R, h)
1 2RN
=
e h
N=0
DMDAe h Tr
2 R
0
1
1
( 2 [ix +A,M]2 + 2 M 2 ) dx
(4)
A(x+2 R)=A(x)
M(x+2 R)=M(x)
We have introduced explicitly the Planck constant h,

which is also the string interaction
constant. The dominant contribution of the sum over N comes from N 1/h . In order to
make sense of the path integral, the measure DM should be stabilized by an appropriate
cutoff, for example, an infinite potential wall placed far from the origin.
The tachyon operators are represented as the discrete momentum modes of the Higgs
field M(x) 3
2R
|n|
n
dx ei R x M(x) R
Tn
(5)
while the vortex operators can be constructed as the moments of the holonomy factor of
the gauge field A(x) around the spacetime circle [8]
Vn Tr ,
n
= T exp
2R
dx A(x).
(6)
Note that, in spite of the asymmetry in the definition of the tachyon and vortex operators,
their expectation values and correlation functions are related by the T-duality (R 1/R). 4
The integral over the gauge field A can be done explicitly and the resulting integral with
respect to the eigenvalues of the Higgs field M can be written as the partition function of
free nonrelativistic fermions defined on the spacetime circle. This remains true also if a

purely tachyonic source n n Tn is added to the exponent in (4). Using the properties of
the amplitudes for tachyon scattering [10], Dijkgraaf, Moore and Plesser showed in [7] that
the partition function (4) with purely tachyonic source added is a -function of the Toda
Lattice hierarchy [11].

On the other hand, if the theory is perturbed by a purely vortex source n=0 tn Vn ,
then the integral with respect to the Higgs field M can be done exactly 5 and one obtains
an effective one-plaquette gauge theory whose only variable is the unitary matrix , the
3 This expression for the tachyon operators in the scaling limit is due to V. Kazakov.
4 The T-duality can be made explicit if the matrix model on a circle (4) is considered as the limit of YangMills
theory defined on a two-dimensional torus, when one of the periods tends to zero. Then the tachyon and vortex
operators (5), (6) are constructed as Polyakov loops winding n times around one of the two periods of the torus.
5 This observation was first made by by Boulatov and Kazakov [12], who used it to study the non-singlet states
of the c = 1 string theory.
149
holonomy factor around the spacetime circle [8]. 6 The partition function of this theory can
be expressed in terms of free fermions defined on the unit circle and is shown to be again
a -function of the Toda Lattice hierarchy.
In this way the c = 1 theory perturbed by a purely vortex or purely tachyonic source
is completely integrable and its integrable structure is described by the Toda Lattice
hierarchy. 7 The Toda Lattice hierarchy is formulated in terms of the time variables
t1 , t2 , . . . and a discrete spatial variable s hZ.
It is natural to split the time variables
where t t . The correspondence with 2D
and
t
=
{
t
}
into two sets t = {tn }
n
n
n
n=1
n=1
string theory requires to consider a complex variable s, which is related to the cosmological
constant as s = 12 i. The Planck constant h , which appears as the spacing unit of the
difference operators in the Lax formalism, plays the role of string interaction constant. The
exact statement, made in [7] on the basis of the analysis of the tachyon S-matrix in [10],
is 8 that the free energy of the string theory

F (, t, t ) =
(7)
h 2g e n=0 tn Vn ,
g0
where g means the connected expectation value on a genus-g worldsheet, is related to

the -function of the Toda Lattice hierarchy as
= eF /h .
2
(8)
Thus the free energy of the string theory satisfies an infinite chain of PDE, the first of which
is the Toda equation

1

F () + exp 2 F ( + i h ) + F ( i h ) 2F () = 0.
(9)
t1 t1
h
This equation determines the flow in the compact c = 1 string theory perturbed by the
lowest vortex operators V1 . In [8], the free energy on the sphere and on the torus was
obtained as the solution of (9) satisfying at t = t = 0 the boundary condition

R + R 1
1
log + O h 2 .
F () = R2 log h 2
2
24
(10)
The string susceptibility = 2 F at genus zero was obtained as a solution of a simple

algebraic equation,
e/R + (R 1)t1 t1 e(2R)/R = 1,
(11)
which resums the perturbative expansion found in [6]. The same approach was subsequently applied to calculate the vacuum expectation values of the vortex operators and the
two-point correlators [9].
The algebraic equation (11) suggests strongly that an operator constraint (a string
equation) similar to those found for the c < 1 matrix models [18] might exist. Below
6 This matrix model belongs to a class of integrable models [1315] representing various generalizations of the
Gaudin gas [16].
7 The integrability is however lost if the theory is perturbed by both tachyon and vortex sources.
8 The derivation of [7] was critically reconsidered in [17].
150
we derive the string equation, which leads to (11), but before doing that we will briefly
recall the Lax formulation of the Toda hierarchy [11].
4. Lax formalism
The Lax formalism is based on finite-difference operators where the Planck constant h
(which in our interpretation plays the role of string interaction constant) emerges as a
spacing unit [19]. The Lax operators L and L are defined as series expansions in the shift
operator
= ei h /
(12)
with coefficients depending on and the couplings t and t

L = r +
uk k ,
L = 1 r +
k=0
k u k .
(13)
k=0
The commuting flows along the times tn and tn are generated by the operators

1
1
n = L n
H
+ L n 0 , n = 1, 2, . . . ,
Hn = Ln >0 + Ln 0 ,
(14)
<0
2
2
where the symbol ( ) means the positive (negative) parts of the series in the shift operator
and ( )0 means the constant part. 9 If we define the covariant derivatives
n = h
Hn
tn
(n = 1, 2, . . .),
(15)
then the Lax equations

= 0,
[n , L] = [n , L]
(16)
are equivalent to the zero-curvature conditions

[m , n ] = 0.
(17)
The Lax operators L, L can be thought as the result of dressing transformations

L = W W
1 ,
1 ,
1 W
L = W
have series expansions

where the dressing operators W and W

=
W=
wn n ,
W
w n n .
n0
(18)
(19)
n0
Lax-equations can be converted into Sato equations for the dressing operators
= 0,
n W = n W
n = 1, 2, . . . .
(20)
9 We use slightly different notations than in [11] because we would like to preserve the symmetry between L
which is appropriate for the double scaling limit of the matrix model.
and L,
151
The dressing operators are related to the -function (8) by

zi W zi = 1 eh Dt (z) 2 i h ,
1
z i = 1 eh Dt(z)+ 2 i h ,
z i W
1
(21)
where we introduced the spectral parameters z and z and the symbols

1
1
,
D
(
z
)
=
.
Dt (z) =
t
nzn tn
nzn tn
n
(22)
The -function contains all the data of the Toda system. For example, the first coefficients
in the expansion of Lax operators (13) and the dressing operators are related to the
-function as
( + i h ) ( i h )
.
r r () =
(23)
2 ()
The dressing operators are determined by (18) up to a factor that commutes with the
shift operator. In particular, Eq. (18) is satisfied by the so-called wave operators

n
n

W = W exp
tn ,
W = W exp
(24)
tn
n1
n1
which have the following important property. The operator

A = W 1 W
(25)
does not depend on t and t (see Theorem 1.5 in Ref. [11] and Proposition 1.2 in Ref. [20]).
The operator (25) characterizes the particular solution of the Toda Lattice hierarchy.
5. OrlovShulman operators
The Lax system can be extended by adding OrlovShulman operators [21]

n
ntn W 1 ,
M =W +i
n1

=W
i
1 .
M
ntn n W
(26)
n1
The Lax and OrlovShulman operators are expressed in terms of the wave operators (24)
as
L = W W
1 ,
1 W
1 ,
L = W
M = WW 1 ,
W
1 .
= W
M
(27)
M
satisfy the same commutation relations as the pair
Therefore, the pairs L, M and L,
(,
):
[,
] = i h ,
[L, M] = i h L,
M]
= i h L.
[L,
(28)
152
Since the spatial parameter of the Toda system s = 12 i is continuous in our

consideration, one can also consider arbitrary powers of the shift operator and
generalize (28) to

= i h L .
L , M = i hL
L , M
, = i h ,
(29)
,
6. Gaussian field representation
The OrlovShulman operators can be represented as the currents of the gaussian field
z) whose left and right chiral components are defined by
(z, z ) = (z) + (

zn
1
zn tn i
,
(z) = log z i
+i
2
n tn
1
z) = log z + i
+i
(
2
n=1
z n tn i
n=1
z n
.
n tn
(30)
Indeed, introducing the spectral parameters z and z , we can write (21) in the form

z)
z i = e(
zi Wzi = e(z) ,
z i W
,
where for any differential operator O in t, t and we denote O =
and (27) we find

z i = z z (
z) .
z i M
zi Mzi = zz (z) ,
1 O .
(31)
From (31)
(32)
Eq. (32) yields, together with (14), the following expansions for the OrlovShulman
operators as Laurent series in L and L

ntn Ln + vn Ln ,
M =+i
n=1

=i
M
ntn L n + vn L n
(33)
n=1
with coefficients vn and vn equal to the expectation values of the vortex operators
vn =
F
= Vn ,
tn
vn =
F
n .
= V
tn
(34)
z) can interpreted as creating and annihilating world-sheet

The operators (z) and (
boundaries and the spectral parameters z and z play the role of boundary cosmological
create and annihilate boundaries with marked
constants. Similarly, the operators M and M
points because of the derivatives in z and z . In this way OrlovShulman operators allow
the spectrum of local scaling operators on the world sheet to be studied.
For sufficiently weak perturbations, these are the vortex and anti-vortex operators
For strong
associated with the expansion (33) in integer powers of z = L and z = L.
can be given
perturbations, where the series (33) do not converge, the operators M and M
meaning via analytic continuation. We will return to this point at the end of the paper.
153
7. The string equation

Now we are ready to proceed with the derivation of the constraint which, together with
the Toda Lattice hierarchy, defines the integrable structure of the compactified string theory.
Such constraints are commonly called string equations.
Let us first consider the case of the unperturbed theory (t = t = 0). In this case the
all-genus free energy of the 2d string theory compactified at radius R is given by the
integral [1]
h 2
Im
F () = h log () =
4
e2iRy/h
dy
y sinh(yR) sinh(y)
(35)
and therefore satisfies the functional equation

4 sin(h /2R) sin(h /2) log = log(1/).
(36)
Now we would like to rewrite (36) as an algebraic relation for the Lax operators. In the
case of zero potential, only the first term in the expansions (13) survives
L = r = W W
1 ,
1 W
1 ,
L = 1 r = W
(37)
() are ordinary functions. By (21) we find for their ratio

where W () and W
W () ( i h /2)
,
=
() ( + i h /2)
W
(38)
and therefore the functional equation (36) is equivalent to

( + i h /2R)W ( i h /2R)
W
= ,
( i h /2R)
W ( + i h /2R)W
(39)
satisfies the following identities

which means that the operator A = W 1 W
A 1/R A1 1/R = i h /2R,
A1 1/R A 1/R = + i h /2R.
(40)
A third identity follows from the fact that for t = t = 0 the dressing operators do not contain
shifts and therefore commute with :
AA1 = .
(41)
coincide with the wave operators W

Further, at t = t = 0 the dressing operators W and W

and W defined by (24) and we can replace the operator A in (40), (41) by A = W 1 W.
Therefore, the identities (40), (41) actually hold for arbitrary couplings t and t.
M and M,

Eqs. (40) and (41) can be formulated as algebraic relations between L, L,
namely,
i h ,
L 1/R L1/R = M
2R

M = M.
L1/R L 1/R = M +
i h
,
2R
(42)
(43)
154
Thus we arrive at the constraint, which allows to express the OrlovShulman operators as
bilinears of Lax operators

= 1 L1/R L 1/R + L 1/R L1/R .
M =M
(44)
2
Given the string equation (44), the canonical commutation relations (29) are equivalent
to the following constraint for the Lax operators
1/R
1/R
1/R
= i hL
= i h L 1/R .
,
L , L
L , L
(45)
8. The dispersionless limit h

0
In the sequel we will concentrate on the genus zero case and consider the dispersionless
limit h 0 of Toda hierarchy, which is a particular case of the universal Whitham
hierarchy introduced by Krichever [22]. The solutions of the dispersionless hierarchy can
be parametrized by canonical transformations in a 2D phase space so that any solution is
determined by the choice of a canonical pair of variables [19].
In the limit h 0, the operator can be considered as a classical variable , conjugate
to
{, } = ,
(46)
where the Poisson bracket is defined by

{f, g} = i( f g g f ),
(47)
and the Lax operators L, L are considered as c-functions of the phase space coordinates
and . The Hamiltonians as functions of and give the expectation values of the vortex
operators, Hn = tn F = Vn .
The Lax equations are equivalent to the exterior differential relations

= d log d + i
d log L dM = d log L d M
(48)
dHn dtn
n=0
, t, t ) whose differentials
which imply the existence of functions S(L, , t, t ) and
S(L,
are given by

dS = Md log L + log d + i
Hn dtn ,
n=0
log L + log d + i
d
S = Md
Hn dtn .
(49)
n=0
The functions S and

S can be thought of as the generating functions for the canonical
M).
The new coordinates z =
transformations (, ) (L, M) and (, ) (L,
L(, ) and z = L(, ) are analytic functions of , given by the h 0 limit of the
expansions (13). The classical Hamiltonians Hn (, ) associated with the times tn are
related z = L() and z = L()

by (14).
155
Conversely, the coordinate can be considered as a meromorphic function of either z

or z

= e S(z) = e S(z) .
(50)
M).

Eq. (50) defines the canonical transformation between (L, M) and (L,
Further, the functions S(z) and S(z) are equal to the expectation values of the chiral
components of the gaussian field (30)

z)
S(z) = (z) ,
S(z) = (
(51)
and the functions representing OrlovShulman operators are given by the derivatives
M(z) = zz S(z),
z) = z z
S(z).
M(
(52)
In the dispersionless limit we can write (44) and (45) as

= L1/R L 1/R
M =M
and
(53)

L1/R , L 1/R = 1/R.
(54)
The second constraint is a consequence of the first one and the canonical commutation
relations

M
= L.
{L, M} = L,
L,
(55)
The dispersionless string equation (53) generalizes the constraint found for the selfdual radius (R = 1) by Eguchi and Kanno [23], and for all integer values of = 1/R by
Nakatsu [24]. In this last case one can rewrite the string equation as a linear constraint
(W1+ constraint) on the -function [25].
9. Two-point correlators in the dispersionless limit

Knowing the functions z() and z () or their inverse functions (z) and (z), we can
evaluate, using (50) and (51), the two-point functions n F :

1
1
n F =
(56)
zn d(z),
n F =
z n d(z) (n 1).
2i
2i
The generating functions for the two-point correlators m n F and m n F (m, n > 0) are
given by
1 2
(z1 ) (z2 )
F + Dt (z1 )Dt (z2 )F = log
,
2
z1 z2

1
,
Dt (z1 )Dt(z2 )F = log 1
(z1 )(z2 )
(57)
(58)
where the operators Dt (z) and Dt(z) are defined by (22) and the functions (z) and (z)
are defined by (50). Eqs. (57) and (58) are obtained directly from the dispersionless Hirota
equations for Toda hierarchy [19] and hold independently of the constraint (54).
156
10. LandauGinsburg description

The topological description of the c = 1 string theory proposed for the self-dual radius
in [26] and [27] can be readily generalized for any value of the compactification radius.
It was shown in [28] that the WDVV equations are a consequence of the dispersionless
Hirota equations.
11. Relation to conformal maps

The solutions of the dispersionless Toda hierarchy can be given a geometrical
interpretation as conformal maps from simply connected domains in the complex plane
to the unit disk. This correspondence was first established for a particular solution of the
hierarchy [29,30] and then for any generic solution in [31].
The general form of a conformal map from the exterior of the unit disc to a simply
connected domain D containing the point z = is given by the first series expansion in
(13). The parameter r, which we will assume to be real and positive, is called the conformal
radius of the domain D. Let us further assume that the couplings tn and tn are complex
conjugate. Then, along the boundary of the unit disc = 1/, z = L()
and z = L()
are complex conjugates and define a smooth curve in the complex z-plane, which is the
boundary of the domain D. If t = t = 0, then is a circle of radius r = R/2 ; in the general
case the form of the contour depends on the values of the couplings tn . The couplings tn
(n = 0) and can be thought of as a set coordinates in the space of closed curves. The
coordinates of the curve = D are given by the moments of the domain D with respect
to the measure = (z, z ) dz d z

1
1
zn (z, z ) d 2 z,
tn =
z n (z, z ) d 2 z,
tn =
n
n
D
D

1
=
(59)
(z, z ) d 2 z
R
D
and the free energy is

2
1
2
F=
d
z
d 2 (z, z ) logz1 1 (, ).
2
2
D
In [29] it is assumed that the density is homogeneous, which in our case is so only at the
selfdual point R = 1. The case of generic density (z, z ) was worked out in [31].
The density function is fixed by the constraint imposed on the Toda system. In our case,
the string equation (54) means that if we parametrize the phase space by z = L and z = L,
1/R
1/R
then the volume form is (z, z ) dz d z = R d(z ) d(z ). This means that the density
function is
(z, z ) =
1R
1
(zz) R .
R
(60)
157
(z, z ), the couplings t, t and

If we introduce the potential U (z, z ) such that (z, z ) = U
the cosmological constant are given by contour integrals along the boundary = D

1
1
n
tn =
z z U dz,
tn =
z n z U d z ,
2in
2in
1
2iR

z U dz.
(61)
In our case U (z, z ) = R(zz)1/R . The derivative of the potential can be considered as a
meromorphic function in the vicinity of the curve and is related to the OrlovShulman
operator by zz U (z, z ) = M(z). Similarly, z z U (z, z ) = M(z). The dispersionless string
= R(LL)
1R
R = 1 (L, L).
equation (54) is equivalent to {L, L}
The formulation of the dispersionless limit of the theory as a conformal map problem is
analogous to the now standard geometrical description of the tachyon excitations in the 2D
string theory within MQM. The tachyons there appear as small fluctuations of the profile
of the Fermi surface [32]. In our case, similarly, a small change of the couplings tn leads
to a small deformation of the boundary of the domain D. Typically the deformations of
the domain satisfy a W symmetry, but a local realization of this symmetry exists only
for R = 1. In order to have the standard realization of the W symmetry, we have to go
1
1
to coordinates y = z R , y = z R . In these coordinates the density is homogenuous, but the
conformal map has a conical singularity in the origin.
12. How to use the string equation

It is convenient to rewrite the expansions (13) as

1
L1/R = (r) R 1 + a1 / + a2 /2 + ,

1
L 1/R = (r /) R 1 + a 1 + a 2 2 + ,
(62)
where ak , a k are some functions of the couplings tk , tk (k = 1, 2, . . .) and . The constraint

(53) then gives

1
(r r ) R 1 + a1 1 + a2 2 + 1 + a 1 + a 2 2 + = M().
(63)
In order to use (63) we need to express the coefficients of the Laurent series M() in
terms of the couplings t1 , . . . , tn and . This can be done using the expansions (33)
as follows. It is easy to see that if we consider a perturbation by t1 , . . . , tn , then all
the coefficients but a1 , . . . , an and a 1 , . . . , a n in (62) are zero. Then the coefficients of
the positive (negative) powers of are obtained by plugging M() = M(L()) [M() =
L())]
M(
into (33). Comparing the coefficients of 1 , . . . , n on both sides of (63), we
obtain 2n conditions, which determine a 1 , . . . , a n .
158
13. Example: sine-Gordon model coupled to gravity (tk = k,1 tn + k,1 tn )

Let us solve, as an example, the string equation for the case of only one pair of nonzero
vortex couplings, t = t1 and t = t1 , which corresponds to a sine-Liouville term
(R2)
xR
e
te + tei xR
(64)
added to the world-sheet action (1). In this case one can retain only a = a1 and a = a 1 in
(62)
L = r 1 (1 + a)
R
L = r(1 + a/)R ,
(65)
and Eq. (63) yields

=
(r r ) R (1 + a/)(1 + a)
rt ( + Ra) + + ,
for large ,
+ + , for small .
r t(1 + R a)
(66)
Comparing the coefficients in front of 0,1 we get three identities

1
(r r ) R a = rt,
(r r ) R a = r t,
(R 1)a a + (r r ) R = 1.
(67)
Taking the limit h 0 limit of (23), we can express the product r r through the
susceptibility 2 F by
r r = e .
(68)
From Eqs. (67) and (68) we get an algebraic equation for the susceptibility
1
e R + t t(R 1)e
2R
R
=1
(69)
and, choosing the gauge r = r , the explicit form of the functions z = L(), z = L():
R

1
2R
z = e 2 1 + te 2R 1 ,

R
1
2R
z = e 2 1 1 + te 2R .
(70)
The two-point correlators can now be obtained by plugging (z) and (z) in (57) and (58)
and expanding in z and z . To compute the one-point correlators, one should integrate (56)
with respect to . We will not do this here, since the calculation is presented in detail in [9].
In a similar way, if all couplings but tn and tn are zero, the string equation is solved by
1
e R + (nR 1)tn tn e
2nR
R
=1
(71)
R

1
2nR
z = e 2 1 + tn e 2R n ,

R
1
2nR
z = e 2 1 1 + tn e 2R n .
(72)
and
As expected, the theory compactified at radius R and perturbed by t1 and t1 is equivalent

to the theory compactified at radius R/n and perturbed by tn and tn .
159
14. UV IR flows in the compactified the c = 1 string theory in a vortex

background
The critical points in the space of couplings describe conformal matter coupled to
gravity. The typical size of the world sheet diverges when one approaches the critical point
and is inversely proportional to the cosmological constant, defined as the distance to this
point. At a critical point the third derivative of the free energy in the cosmological constant
diverges.
Let us analyze the solution of the string equation in the example considered above, where
only the three couplings, , t = t1 and t = t1 are nonzero. In this case the theory contains
a single dimensionless parameter
t tR2
(73)
which measures the strength of the sine-Liouville perturbation. The algebraic equation for
the susceptibility (69) has three singular points
= 0,
= ,
where we denoted =
free energy behaves as
3 F 1 ,
= ,
(1R)1R
(2R)2R
(74)
. In the vicinity of these points the third derivative of the
1/2

3 F
,
1

31 F 1 .
(75)
The critical point = 0 corresponds to the unperturbed string theory (c = 1 compactified

boson coupled to gravity). It was argued in [8] that if 1 < R < 2, the critical point =
describes another c = 1 string theory, in which the role of the cosmological constant is
played by the sine-Liouville coupling. (The latter theory, which can be called sine-Liouville
or SL string theory is expected to be dual, according to the FZZ conjecture, 10 to the 2D
string theory in Euclidean black hole background.) Finally, = describes the trivial
c = 0 string theory in which the matter field is the vortex-antivortex plasma with finite
correlation length.
In this way the string theory with a vortex source exists in two phases, the ordinary
Liouville phase (0 < < ) and the sine-Liouville phase ( < < ). In both phases
the theory flows to the same c = 0 IR fixed point = .
Each of the two phases is characterized by its spectrum of local operators (the operators
creating microscopic loops in the world sheet). In the Liouville phase these are the
vortex operators of all possible vorticities, which appear in the Laurent expansion of the
loop operator M(z). The corresponding coupling constants tn are the times of the Toda
system.
If we are interested in the sine-Liouville phase, we should be more careful, because the
loop operator has two branches for large z. Indeed, let us examine the function log (z) =
S(z), which can be thought of as the partition function of a string having a punctured
10 The FZZ conjecture [2] actually concerns only the point R = 3/2, which can be considered as a c = 1 string
theory compactified at R = 3/2 and perturbed by t1 V1 + t1 V1 .
160
Fig. 1. The Riemann surface of (z).
disk as world sheet, with boundary cosmological constant z. The Riemann surfaces of
this function is shown in Fig. 1. When 1 < R < 2, it follows from (65) that the functions
log (z) has two branches at large z, which are associated with the two UV fixed points
= 0 and =

z,
if 0 (branch I),
1
(z)
(76)
z 1R , if (branch II).
In the SL phase ( < < ) it is the second branch which is the relevant one. In this
branch the partition function on a punctured disk is expanded as
log (z) = log z +
ck z k ,
z = z 1R .
(77)
k0
If we want to calculate the correlation functions of the microscopic loop operators in the SL
phase (which are no vortex operators), we can still use the general formulas (57) and (58),
but we should expand (z) in fractional powers of z. Near the SL critical point the theory
can be again described in terms of a gaussian field (z) with local expansion parameter
1
parameter z = z R1 .
Let us see how the phase transition at = occurs within the description using
conformal maps. The ground state in the dispersionless limit is described by a closed
contour

= z(); || = 1
in the z-plane, or more correctly in the Riemann surface of the function log (z). In terms
of the variable = log z, this Riemann surface is made by gluing together three semiinfinite cylinders of widths 1, R 1 and R. If we are in the Liouville phase, then the
contour goes around the cylinder representing the branch I. As grows from 0 to ,
the contour moves from infinity towards the branch point. As we will show in a moment,
the IR critical point occurs when the contour hits the branch point. If we are in the SL phase,
then the contour wraps the cylinder II and hits the branch point when approaches
from above.
We have to show that the contour hits the branch point when the coupling reaches
the critical value . At the branch point dz/d = 0 we have = (R 1)a. On the other
hand, the contour corresponds to the unit circle = 1. The condition (R 1)a = 1 is
satisfied exactly at the critical coupling = defined by (74).
161
15. A description of the IR critical point in terms of the dual sine-Gordon theory
An intuitive physical picture of the flows in the perturbed theory can be made in terms
of the T-dual field x.
Below we sketch the scenario proposed by Kutasov in [33] (see
2
< R < m2 , then only the
also [34]). If the compactification radius is in the interval m+1
vortex operators V1 , . . . , Vm are relevant. At large distances on the world sheet the
effective couplings grow, the barriers between the minima of the effective potential V (x)
=
+ + tm emx/R
become infinite and the original c = 1 theory splits up into to
tm emx/R
m theories with c = 0.
In general, the IR matter fields need not be massive. By tuning the coefficients
of the potential appropriately one can achieve multicritical behavior V (x)
V (x0 ) =
C(x x 0 )2k + , which gives, as was argued in [33], a string theory with central charge
6
. Therefore one can expect that there can be multicritical points that
c = 1 (k+1)(k+2)
correspond to a direct product of rational c < 1 conformal theories coupled to gravity. The
general form of the solution (62) indeed confirms this physical picture. For example, if we
switch on all 2m couplings t1 , . . . , tm , for which the perturbation is relevant, then we can
tune the coefficients in (62) and the cosmological constant so that the first m derivatives
of z() vanish at some point zc such that |(zc )| = 1. Then the IR central charge of the
6
. In this way all unitary c < 1 string theories can
matter field will be cIR = 1 (m+1)(m+2)
be obtained by turning on various perturbations of the compactified c = 1 string theory, as
it was suggested in [33]. It would be interesting to exhibit the flow of the constraint (53) in
Toda hierarchy to the Douglas equations [18] for generalized KdV hierarchies.
Acknowledgements
The author thanks S. Alexandrov, P. Dorey, T. Eguchi, M. Fukuma, V. Kazakov,
D. Kutasov, A. Marshakov, D. Serban,
P. Wiegmann and A. Zabrodin for stimulating

discussions and useful comments. This research is supported in part by European network
EUROGRID HPRN-CT-1999-00161. Part of the work is done during the stay of the author
at Yukawa Institute, Kyoto.
References
[1] I. Klebanov, Lectures delivered at the ICTP spring school on string theory and quantum gravity,
Trieste, April 1991, hep-th/9108019.
[2] V. Fateev, A. Zamolodchikov, Al. Zamolodchikov, unpublished.
[3] E. Witten, String theory and black holes, Phys. Rev. D 44 (1991) 314.
[4] R. Dijkgraaf, E. Verlinde, H. Verlinde, String propagation in a black hole geometry, Nucl.
Phys. B 371 (1992) 269.
[5] S. Mukhi, C. Vafa, Two-dimensional black hole as a topological coset model of c = 1 string
theory, hep-th/9301083.
[6] G. Moore, Gravitational phase transitions and the sine-Gordon model, hep-th/9203061.
[7] R. Dijkgraaf, G. Moore, M.R. Plesser, Nucl. Phys. B 394 (1993) 356.
162
[8] V. Kazakov, I. Kostov, D. Kutasov, A matrix model for the two-dimensional black hole, hep-th/
0101011.
[9] S. Alexandrov, V. Kazakov, Correlators in 2D string theory with vortex condensation, hep-th/
0104094.
[10] G. Moore, M. Plesser, S. Ramgoolam, Exact S-matrix for 2D string theory, hep-th/9111035.
[11] K. Ueno, K. Takasaki, Toda lattice hierarchy in: Group representations and systems of
differential equations, Adv. Stud. Pure Math. 4 (1984) 1.
[12] D. Boulatov, V. Kazakov, Int. J. Mod. Phys. 8 (1993) 809.
[13] V. Kazakov, I. Kostov, N. Nekrasov, D-particles, matrix integrals and KP hierachy, Nucl.
Phys. B 557 (1999) 413, hep-th/9810035.
[14] J. Hoppe, V. Kazakov, I. Kostov, Dimensionally reduced SYM4 as solvable matrix quantum
mechanics, Nucl. Phys. B 571 (2000) 479, hep-th/9907058.
[15] I.K. Kostov, Exact solution of the six-vertex model on a random lattice, Nucl. Phys. B (2000)
513, hep-th/9911023.
[16] M. Gaudin, Une famille un paramtre densembles unitaires, Nucl. Phys. 85 (1966) 545;
Travaux de Michel Gaudin, Modles exactement rsolus, Les Editions de Physique, 1995, p. 25.
[17] C. Imbimbo, S. Mukhi, Nucl. Phys. B 449 (1995) 553.
[18] M. Douglas, Strings in less than one dimension and the generalized KdV hierarchies, Phys.
Lett. B 238 (1990) 279.
[19] K. Takasaki, T. Takebe, Quasi-classical limit of Toda hierarchy and W -infinity symmetries,
Lett. Math. Phys. 28 (1993) 165, hep-th/9301070.
[20] T. Takebe, Toda lattice hierarchy and conservation laws, Commun. Math. Phys. 129 (1990) 129.
[21] A. Orlov, E. Shulman, Lett. Math. Phys. 12 (1986) 171.
[22] I. Krichever, Func. Anal. i ego pril. 22 (3) (1988) 37, (English translation: Funct. Anal. Appl.
22 (1989) 200);
I. Krichever, The -function of the universal Witham hierarchy, matrix models and topological
field theories, Commun. Pure Appl. Math. 47 (1992), hep-th/9205110.
[23] T. Eguchi, H. Kanno, Toda lattice hierarchy and the topological description of the c = 1 string
theory, hep-th/9404056.
[24] T. Nakatsu, On the string equation at c = 1, Mod. Phys. Lett. A 9 (1994) 3313, hep-th/9407096.
[25] K. Takasaki, Toda lattice hierarchy and generalized string equations, Commun. Math. Phys. 181
(1996) 131.
[26] D. Ghoshal, S. Mukhi, Topological LandauGinsburg model of two-dimensional string theory,
hep-th/9312189.
[27] A. Hanani, Y. Oz, R. Plesser, Topological LandauGinsburg formulation and integrable
structure of the 2d string theory, Commun. Math. Phys. 173 (1995) 559.
[28] A. Boyarsky, A. Marshakov, O. Ruchhayskiy, P. Wiegmann, A. Zabrodin, On associativity
equations in dispersionless integrable hierarchies, hep-th/0105260.
[29] P. Wiegmann, A. Zabrodin, Conformal maps and dispersionless integrable hierarchies,
Commun. Math. Phys. 213 (2000) 523, hep-th/9909147.
[30] I. Kostov, I. Krichever, M. Mineev-Veinstein, P. Wiegmann, A. Zabrodin, -function for
analytic curves, hep-th/0005259.
[31] A. Zabrodin, Dispersionless limit of Hirota equations in some problems of complex analysis,
math.CV/0104169.
[32] J. Polchinski, What is string theory, Lectures presented at the 1994 Les Houches Summer
School, Fluctuating Geometries in Statistical Mechanics and Field Theory, hep-th/9411028.
[33] D. Kutasov, Irreversibility of the renormalization group flow in two-dimensional quantum
gravity, hep-th/9207064.
[34] E. Hsu, D. Kutasov, The gravitational sine-Gordon model, Nucl. Phys. B 396 (1993) 693, hepth/9212023.

String inspired Z model with stable proton

and light neutrino masses
Alon E. Faraggi a,b , Marc Thormeier a
a Theoretical Physics Department, University of Oxford, Oxford, OX1 3NP, United Kingdom
b Theory Division, CERN, CH-1211 Geneva, Switzerland
Abstract
Grand unification and its incarnation in the form of heterotic-string unification, are the only
extensions of the standard model that are rooted in the structure of the standard model itself. In
this context it was proposed that the exclusiveness of proton stability and suppression of neutrino
masses necessitates the existence of an additional U (1)Z symmetry, which is of non-GUT origin
and remains unbroken down to intermediate or low energies. Realistic string models frequently give
rise to non-GUT U (1) symmetries, which arise from the flavor symmetries in the models. In this
paper we demonstrate in a string-inspired toy model that such a stringy Z can indeed guarantee
proton longevity and viable phenomenology in the neutrino sector as well as in the quark and charged
lepton sectors. 2002 Published by Elsevier Science B.V.
PACS: 11.25.M; 12.60.-i; 14.70.Pw; 14.60.Pq
1. Introduction
Following the astounding experimental achievements of the previous decade, the
standard model of particle physics is firmly established as the correct effective description
of nature in the accessible energy range. The experimentally observed multiplet structure of
the standard model, as well as other qualitative indications, strongly favor the embedding
of the standard model in larger unifying groups. This embedding can be realized either at
the level of an effective field theory below the Planck scale, or directly in a heterotic-string
realization, formulated as a model of perturbative quantum gravity at the Planck scale.
E-mail addresses: faraggi@thphys.ox.ac.uk (A.E. Faraggi), thor@thphys.ox.ac.uk (M. Thormeier).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 4 6 - 0
164
A.E. Faraggi, M. Thormeier / Nuclear Physics B 624 (2002) 163180
A general consequence of the standard model multiplet unification is that the unification
scale is separated from the electroweak scale by fifteen orders of magnitude. This arises
from the fact that the multiplet unification of the standard model spectrum results in proton
decay mediating operators. Compatibility of this hypothesis with the measured gauge parameters at the electroweak scale and the observed suppression of the left-handed neutrino
masses then provide further strong support for the validity of the grand unification scenario.
The unification hypothesis substantially reduces the number of free parameters of the standard model and is hence also well motivated from the perspective of reducing some of the
arbitrariness found in the standard model. String theory then provides a consistent framework for quantum gravity unification, in the context of which all the standard model parameters should, in principle, be calculable from a minimal set of fundamental parameters.
An important augmentation to the unification program is that of spacetime supersymmetry. While not yet substantiated experimentally, supersymmetry is supported by several
observations, including the consistency of a large top quark mass with the supersymmetric
radiative electroweak symmetry breaking. However, proton longevity is more problematic
in supersymmetric extensions of the standard model [1], which admit dimension four and
five baryon and lepton number violating operators [2]. In the MSSM one imposes a global
symmetry, R-parity (Rp ), which forbids the dangerous dimension four operators, while
the difficulty with the dimension five operators can only be circumvented if one further assumes that the relevant Yukawa couplings are sufficiently suppressed. The proton lifetime
problem becomes especially acute in the context of heterotic-string unification, in which all
operators that are compatible with the local gauge symmetries are expected to arise from
non-renormalizable terms. Indeed this issue has been examined in the past by a number of
authors [37]. The avenues explored range from the existence of matter parity at special
points in the moduli space of particular models to the emergence of non-Abelian custodial
symmetries in specific compactifications [5].
The most realistic heterotic-string models constructed to date are the models constructed
in the free fermionic formulation [8]. This has given rise to a large set of semirealistic models [5,915], which differ in their detailed phenomenological characteristics
and share an underlying Z2 Z2 orbifold structure [16]. Past investigations examined
several possibilities to explain the proton longevity. For example, Ref. [17] stipulated
the possibility that the U (1)Z , which is embedded in SO(10), remains unbroken down
to the TeV scale and suppresses the problematic dimension 4 operators. In Ref. [5] it
was shown that the free fermionic string models occasionally give rise to non-Abelian
custodial symmetries, which forbid proton decay mediating operators to all orders of nonrenormalizable terms. These proposals, however, fall short of providing a satisfactory
solution, as they are, in general, exclusive to the generation of light neutrino masses
through a see-saw mechanism. For example, the absence of the SO(10) 126 representation
in string models [18] necessitates that the SO(10) U (1)Z be broken at a high scale,
rather than at a low scale. Similarly, to date, the existence of the custodial non-Abelian
symmetries seems to be exclusive to the generation of a see-saw mass matrix.
Refs. [6,7] proposed that heterotic-string unification necessitates the existence of an
additional U (1) symmetry, beyond the standard model, which remains unbroken down
to low or intermediate energy. Invariance under the extra U (1) forbids the proton
decay mediating operators, which can be generated only after U (1)Z breaking. The
165
magnitude of the proton decay mediating operators is therefore proportional to the U (1)Z
breaking scale, Z , which is in turn constrained by the proton lifetime limit, and other
phenomenological constraints. On the other hand the additional U (1) should not forbid
quark, lepton and see-saw mass terms. It was argued [6,7] that the required U (1) symmetry
is not of the type that arises in SO(10) or E6 GUTs. Rather, it should arise from the U (1)
symmetries in the string models that are external to the GUT symmetries.
By studying the spectrum and symmetries of the string model of Ref. [12] Pati
showed [6] that U (1) symmetries with the required properties do indeed exist in the string
models, and arise from combinations of the flavor symmetries. In Ref. [7] it was shown that
the required symmetries can in fact remain unbroken by the choices of supersymmetric flat
directions in the string vacuum. In the present paper we continue to explore these ideas.
The basic approach that we pursue is to develop field theory models that are inspired from
the string models. By this we aim to confront the string features with the experimental
data. Our main goal in the present paper is to examine whether the string symmetries that
are used to guarantee the proton longevity can indeed at the same time allow for adequate
suppression of the neutrino masses. That this is a non-trivial requirement can be seen from
the fact that if we insisted on gauged B L as the proton protector, it would imply a
neutrino in the MeV range [20].
We note that the possibility of utilizing U (1) gauge symmetry to insure proton longevity
has also been discussed in the context of purely field theory models [21]. The novelty
here is the consideration of U (1) symmetries that arise in the context the realistic free
fermionic string models. From the string models we extract the charges of the standard
model fields under the various U (1) symmetries. We examine which combinations of
the U (1) symmetries forbid the dangerous baryon and lepton number violating operators
while still allowing for the see-saw mass terms. We then assume that all but two of the
additional U (1)s are broken in the string vacuum by the choices of supersymmetric flat
directions. The first one being the U (1)Z which is embedded in SO(10), and the second
being the additional unbroken U (1) that arises from the flavor symmetries of the string
model. We proceed to construct the see-saw mass matrix by postulating additional Higgs
spectrum and comment how it may arise from the string models. We then demonstrate that
such additional stringy U (1) symmetries can indeed guarantee proton longevity and viable
phenomenology in the neutrino sector, as well as in the quark and charged lepton sectors.
2. Gauge symmetries in free fermionic models

In this section we discuss the general structure of the realistic free fermionic models, and
of the additional U (1) symmetries that arise in these models. The free fermionic heteroticstring formulation yields a large number of three generation models, which possess an
underlying Z2 Z2 orbifold structure [16], and differ in their detailed phenomenological
characteristics. We emphasize the features of the models that are common to this large
class of realistic models.
The free fermionic models are constructed by specifying a set of boundary conditions
basis vectors and the one-loop GSO projection coefficients [8]. The physical massless
states are obtained by acting on the vacuum with bosonic and fermionic operators and
166
by applying the generalized GSO projections. The U (1) charges, Q(f ), with respect to the
unbroken Cartan generators of the four-dimensional gauge group, which are in one-to-one
correspondence with the U (1) currents f f for each complex fermion f , are given by:
1
Q(f ) = (f ) + F (f ),
2
(2.1)
where (f ) is the boundary condition of the world-sheet fermion f in the sector , and
F (f ) is a fermion number operator.
The four-dimensional gauge group in the three generation free fermionic models arises
as follows. The models can in general be regarded as constructed in two stages. The first
stage consists of the NAHE set, {1, S, b1 , b2 , b3 } [13]. The gauge group after imposing
the GSO projections induced by the NAHE set basis vectors is SO(10) SO(6)3 E8
with N = 1 supersymmetry. The spacetime vector bosons that generate the gauge group
arise from the NeveuSchwarz sector and from the sector 1 + b1 + b2 + b3 . The Neveu
Schwarz sector produces the generators of SO(10) SO(6)3 SO(16). The sector
1 + b1 + b2 + b3 produces the spinorial 128 of SO(16) and completes the hidden gauge
group to E8 . At the level of the NAHE set the sectors b1 , b2 and b3 produce 48 multiplets,
16 from each, in the 16 representation of SO(10). The states from the sectors bj are singlets
of the hidden E8 gauge group and transform under the horizontal SO(6)j (j = 1, 2, 3)
symmetries. This structure is common to all the realistic free fermionic models. At this
stage we anticipate that the SO(10) group gives rise to the standard model group factors,
whereas the SO(6)3 groups may produce additional symmetries that can play a role in
safeguarding the proton lifetime.
The second stage of the free fermionic basis construction consists of adding to the
NAHE set three (or four) additional boundary condition basis vectors. These additional
basis vectors reduce the number of generations to three chiral generations, one from each
of the sectors b1 , b2 and b3 , and simultaneously break the four-dimensional gauge group.
The SO(10) symmetry is broken to one of its subgroups SU(5) U (1) [9], SO(6) SO(4)
[11], SU(3) SU(2)2 U (1) [15] or SU(3) SU(2) U (1)2 [10,12,14]. Similarly, the
hidden E8 symmetry is broken to one of its subgroups by the basis vectors which extend the
NAHE set. This hidden E8 subgroup may, or may not, contain U (1) factors which are not
enhanced to a non-Abelian symmetry. As the standard model states are not charged with
respect to these U (1) symmetries, they cannot play a role in suppressing the proton decay
mediating operators, and are therefore not discussed further here. On the other hand, the flavor SO(6)3 symmetries in the NAHE-based models are always broken to flavor U (1) symmetries, as the breaking of these symmetries is correlated with the number of chiral generations. Three such U (1)j symmetries are always obtained in the NAHE based free fermionic
models, from the subgroup of the observable E8 , which is orthogonal to SO(10). These are
produced by the world-sheet currents (j = 1, 2, 3), which are part of the Cartan subalgebra of the observable E8 . Additional unbroken U (1) symmetries, denoted typically by
U (1)j (j = 4, 5, . . .), arise by pairing two real fermions from the sets {y 3,...,6 }, {y 1,2 , 5,6 }
and { 1,...,4 }. The final observable gauge group depends on the number of such pairings.
Subsequent to constructing the basis vectors and extracting the massless spectrum the
analysis of the free fermionic models proceeds by calculating the superpotential. The cubic
167
and higher-order terms in the superpotential are obtained by evaluating the correlators

f f
AN V1 V2 V3b VNb ,
(2.2)
where Vi (Vib ) are the fermionic (scalar) components of the vertex operators, using
the rules given in [19]. Typically, one of the U (1) factors in the free-fermion models
is anomalous, and generates a FayetIlioupolos term which breaks supersymmetry at
the Planck scale. The anomalous U (1) is broken, and supersymmetry is restored, by a
non-trivial VEV for some scalar field that is charged under the anomalous U (1). Since
this field is in general also charged with respect to the other anomaly-free U (1) factors,
some non-trivial set of other fields must also get non-vanishing VEVs V to ensure a
supersymmetric vacuum. Some of these fields will appear in the non-renormalizable terms
(2.2), leading to effective operators of lower dimension. Their coefficients contain factors
of order V/M 1/10. Typically the solution of the D- and F -flatness constraints break
most or all of the horizontal U (1) symmetries.
The question is whether there exist string symmetries, which are beyond the GUT
symmetries and can provide an appealing explanation for the proton lifetime. In a
beautifully insightful paper [6] Pati studied this question in the model of Ref. [12], and
showed that such symmetries indeed exist in the string models. In Ref. [7] it was shown
that the required symmetries can in fact remain unbroken below the string scale, and hence
provide the needed suppression. However, as emphasized above, an important question is
whether the additional symmetry that suppresses the baryon and lepton number violating
operators, still allows at the same time the generation of a neutrino see-saw mass matrix
and hence an acceptable neutrino mass spectrum. In this paper we address this question.
3. Proton decay and superstring Z

For concreteness we focus on the string model of Ref. [12], giving rise to the observable
gauge symmetry
SU(3) SU(2) U (1)C U (1)L U (1)1 U (1)6 .
(3.1)
U (1)C and U (1)L will, apart from the SMs hypercharge, produce one of the additional
gauge bosons which appears in the string models, and constrain the proton decay mediating
operators. The charges of the standard model states under U (1)L and U (1)C are fixed in
the string model, as are the charges of U (1)1,...,6 . We further assume that the low energy
spectrum contains two light Higgs representations, which are the h 1 and h45 of [12], with

1)
H U = h 1 = (1, 0); (2,
,
0,1,0,0,0,0

H D = h45 = (1, 0); (2, 1) 1 , 1 ,0,0,0,0.
2
(3.2)
Prior to rotating the anomaly into a single U (1)6 , all of the U (1)1,...,6 are anomalous:
Tr U1 = Tr U2 = Tr U3 = 24, Tr U4 = Tr U5 = Tr U6 = 12. These can be expressed by
168
one anomalous combination which is unique and five non-anomalous ones:1
U (1)1
1 1 0
0
0
0
U (1)1
0
0 U (1)2
U (1)2 1 1 2 0
0
1 1 0 U (1)3
U (1)3 0 0
=
.
0
1
1 2 U (1)4
U (1)4 0 0
1 1
1
2
2
2
U (1)5
U (1)5
2 2
2 1 1 1
U (1)6
U (1)6
(3.3)
The standard model weak hypercharge U (1)Y and the orthogonal U (1)Z combinations are
given by

1

1
U (1)Y
U (1)C
2
(3.4)
= 3
.
U (1)Z
1 1
U (1)L
The complete massless spectrum and the charges under the four-dimensional gauge group
are given in Ref. [12]. After the rotation (3.4) one gets the charges
Field

E

U
U (1)Y
23
U (1)Z
1
2
1
2

N

D
1
3
32
1
6
1
2
5
2
HU
HD
12
1
2
12
32
(3.5)
in agreement with the hypercharges of the standard model. Furthermore (3.3) leads to the
charges
Field
U (1)1 U (1)2 U (1)3 U (1)4 U (1)5 U (1)6
1
3
1
1 , N
1
Q1 , D
12
12
12
2
2
2
1
1
1
3
1
1
1 , U
1
L1 , E
2
12
2
1
2
1
2
2
1
2
12
0
1
0
1
1
1
0
0
0
1
0
0
2 , N
2
Q2 , D

L2 , E2 , U2
12
3 , N
3
Q3 , D
3 , U
3
L3 , E
HU
HD
12
12
1
2
3
2
12
3
2
2
3
2
1
2
3
2
1
2
1
1
(3.6)
2
2
We note that U (1)1 , . . . , U (1)4 are generation-dependent.

The superpotential of the MSSM + N + R
/ p is given by
) i
U
ab xy (D) i
D
W = ab xy G(U
ij Qax Hb Uy + Gij Qax Hb Dy
j
i
D j
ab (N) i
U j
+ ab G(E)
ij La Hb E + Gij La Hb N
i N
j
+ ab HaD HbU + ij N
1 The normalization of the different U (1) combinations is fixed by the requirement that the conformal
dimension of the massless states gives h 1 = 1 in the new basis. Here we neglect the normalization factors as
we are only interested in the invariant superpotential terms.
169
1
j k
+ ab ij k Lia Lb E
2
j k
i
+ ab xy ij k Qiax Lb D
y + i N
1
xi D
yj D
zk + ab i N
i HaD HbU
+ xyz ij k U
2
i j k
+ ab Ki Lia HbU +
ij k N N N .
(3.7)
Here the particle content is given by the MSSM-superfields plus three generations of righthanded neutrinos. Q, L, U, D, N, E represent the two left-handed SU(2)-doublets of the
quarks and leptons, and the right-handed up-quarks, down-quarks, neutrinos and electrons,
respectively; H D , H U are the two left-handed SU(2) Higgs doublets; an overbar denotes
charge conjugation. The gauge group is SU(3) SU(2) U (1)Y . a, b are SU(2)-indices
(taking on the values 1, 2), x, y, z are SU(3)-indices (taking on the values 1, 2, 3), i, j, k
are generational indices, taking on the values 1, 2, 3; the heaviest generation is labeled
/ p ; concerning ij k ,
by 1, which is not the same convention as, e.g., in some papers on R
rather than LQD.
Summation over repeated
note that we work with the operator QLD
...
...
indices is implied; is the Kronecker symbol, symbolizes any tensor that is totally
antisymmetric with respect to the exchange of two indices, with 12... = 1. All other
symbols are coupling constants.
The unbroken new symmetries U (1)1 ,...,6 strongly reduce the number of renormal/ p : Imposing
izable operators in the low-energy superpotential of the MSSM + N + R
U (1)Z U (1)... on (3.7) gives that
(U )
(N)
(D)
(D)
(D)
(E)
(E)
(E)
U (1)Z U (1)1
allows G22 , G22 , G12 , G21 , G33 , G12 , G21 , G33 ,
U (1)Z U (1)2
)
(N)
(D)
(E)
allows G(U
ij , Gij , Gij , Gij
U (1)Z U (1)3
allows
U (1)Z U (1)4
allows
U (1)Z U (1)5
allows
U (1)Z U (1)6
allows
for i, j = 1, 2,
)
(N)
G(U
for i = 1, 2, 3;
ii , Gii
(D)
(D)
(E)
(E)
(E)
G12 , G21 , G(D)
33 , G12 , G21 , G33 , ,
(U )
(N)
Gii , Gii
for i, j = 1, 2, 3;
(U )
(U )
(N)
(N)
G21 , G12 , G21 , G12 ; ,
)
(N)
G(U
for i, j = 1, 2, 3,
ij , Gij
)
(N)
G(U
for i, j = 1, 2, 3.
ij , Gij
(3.8)
2 and L2 H U N
2 in (3.7)
If U (1)Z and all U (1)... were exact, only the two terms Q2 H U U
would be allowed. Thus we assume that all but one, labeled U (1)? , get broken near the
heterotic-string scale MS 1018 GeV. Hence the gauge group at high energies below the
Planck-scale is given by
SU(3) SU(2) U (1)Y U (1)Z U (1)? .
i
N
(3.9)
We assume that the

do not get a VEV. Thus to break the U (1)Z U (1)? gauge
symmetries we add the left-handed SM-singlet fields A, B and R, S, T to the fields of
the MSSM + N . The underlying SO(10) structure of the model demands A, B to be a
vector-like couple with respect to U (1)Z , with A having the charge of the right-handed
170
. We take A, B to be uncharged with respect to U (1)? , as we require that the

neutrino N
VEVs of A, B do not break U (1)? because we assume the breaking scale of U (1)Z to
be much higher than the breaking scale of U (1)? . The phenomenological constraints that
we discuss below imply that ? has to be considerably above the electroweak scale. The
stringy background of the model requires R, S, T to be uncharged under U (1)Z . Fixing
their U (1)? -charges basically is the only freedom in our string-inspired model. The choices
we make here aim at generating a simple see-saw matrix. In the string models the additional
Higgs spectrum may arise from fundamental states in the massless string spectrum, or from
product of fields that produce these effective quantum numbers [22].
From (3.8) we see that U (1)4 , U (1)5 , U (1)6 are too restrictive, and U (1)1 still allows
too few entries in G(U ) , G(N) to be U (1)? . U (1)2 and U (1)3 look more promising, and
we study both cases. U (1)2 leads to a model with no mass-scale, as it forbids the term. If U (1)2 suppresses, or forbids, the proton decay mediating operators, while still
allowing the see-saw mass terms, and setting the order of the -parameter by its breaking
scale, then U (1)2 yields an example of a U (1) symmetry that, provided that it remains
unbroken down to a sufficiently low scale, can address the proton lifetime and neutrino
mass issues, as well as provide a solution to the -problem. However, U (1)2 also forbids
the mass terms for one generation, which are therefore suppressed by 2 /MS . One could
contemplate the possibility that this deficit is cured by radiative corrections. U (1)3 allows
both masses for all generations and also a -term. In this case therefore we have to
ascertain that the additional Higgs VEVs do not generate a large -term. However, unlike
U (1)2 , U (1)3 does not forbid the proton-destabilizing dimension five operators, and in
particular, the operators Q3 Q3 Q3 L3 and U 3 U 3 D 3 E 3 . Therefore in the case of U (1)3
additional suppression of the dimension five operators is required. In the context of the
string models such suppression is anticipated due to the additional U (1) symmetries that
are broken by the choices of supersymmetric flat directions at the string scale.
3.1. The case of U (1)3
We first focus on the case of U (1)Z U (1)3 . We work with the following charges:
Field
U (1)Z
5
2
52
1
2
12
U (1)3
(3.10)
As T is completely uncharged it cannot break any symmetry and we therefore impose

T = 0. Instead of (3.7) we now obtain
y + ab xy G Qiax HbD D
y + ab G Lia HbD E
j
W = ab xy Gij Qiax HbU U
ij
ij
(U )
(D)
(E)
i
U j
ab
D U
ab
D U
ab
3 U
+ ab G(N)
ij La Hb N + Ha Hb + P T Ha Hb + ALa Hb
3 + MR RS + MA AB + MT T T + BR N
1
+ T + MN B N
2 + BT N
3 + P ABT + P RST + P T T T ,
+ BS N
(3.11)
171
()
with the Gij having only three entries (see (3.8)). Guided by the structure of the string
models we impose = 0. From the superpotential above we obtain the following seesaw mass matrix (n3L being the fermionic component of L3 = NL3 , a being the fermionic
component of A, etc.),
( n1L
n2L
n3L
nR 1
nR 2
nR 3
H U
H U
H U
H U
H U
B
R
r
H U
B
S
b)
n1L
n2L
3
H U
H U
nL
1
B
R
n
R
B
S nR 2
3
B
MN nR
MR S
r
s
R
MR
B
A t
B
S R MT
MN
B
A
MA
MA
a
b
(3.12)
n3L a-entry
where in (3.12) coefficients of order one are not displayed explicitly, and the
is
discussed further below. The entries in (3.12) are constrained by several phenomenological
considerations. In order to avoid D-term SUSY-breaking contributions that are larger than
1 TeV we impose |R2 S2 | 1 (TeV)2 and |A2 B2 | 1 (TeV)2 . As the U (1)Z
and U (1)3 breaking scales are assumed to be above the SUSY-breaking scale we thus have
A B,
R S.
(3.13)
After U (1)Z breaking the term AL3 H U may generate a very large effective K3 ,
see (3.7). One might argue that it is possible to rotate this term away by a unitary fieldredefinition of L3 and H D , so that

D

cH D
H
(3.14)
( 0 ) 3 ,
( K3 )
3
L
L
but this would induce a very large ,
namely, ||
2 = ||2 + | A|2 , which is not desired,
as A. We therefore have to demand that

H U
.
A
(3.15)
Thus, in the see-saw mass matrix M in (3.12) we can neglect the n3L a-entries, as
H U 2 /A H U .
We impose the following relation on the parameters appearing in (3.12) which will yield
appealing neutrino mass and mixing spectrum
MN
A3/2
,
R1/2
(3.16)
with MN much larger than A. Eq. (3.16) is constrained by requiring no F -term
SUSY-breaking contributions that are bigger than 1 TeV, i.e., MN A 1 TeV MS .
172
Similar constraints on products like AR are automatically fulfilled with the condition
guaranteeing no large contributions to the -term, that we discuss further below.
We must also insure that the dangerous proton decay mediating operators are sufficiently
suppressed. Such operators may arise from non-renormalizable terms of the form of 2
AR 3 3 k
Q L D
MS2
AR 3 3 k
U D D .
MS2
and
(3.17)
After U (1)Z and U (1)3 symmetry breaking one gets

33k
and 33k
AR
.
MS2
Proton decay limits impose that (see, for example, [23])

mk 2
dR
33k 33k 2 1027
,
100 GeV
(3.18)
(3.19)
taking the masses of the SUSY-particles to be of O(1 TeV), one gets that the relation above
is fulfilled if
33k 5 1013
and 33k 5 1013.
(3.20)
Comparing this with Eq. (3.18) gives the condition

AR 5 1023 (GeV)2 .
(3.21)
Similarly, potential terms contributing to the electroweak Higgs mixing parameter

should be adequately suppressed or absent altogether. As the VEV of T vanishes, the
superpotential term T H D H U does not generate a large -term. To guarantee that the two
gauge-invariant non-renormalizable terms
AB D U
H H ,
MS
RS D U
H H
MS
(3.22)
do not contribute substantially to the -term after U (1)Z and U (1)3 gauge symmetry
breaking, we impose
A 1010 GeV.
(3.23)
Since we have A R Eq. (3.23) implies

AR 1020 (GeV)2 ,
(3.24)
so that the operators (3.17) are rendered harmless, and also F -term SUSY-breaking is
avoided.
As stated earlier, the U (1)2 /U (1)3 -charges of the three generations are not family
universal. Consequently, the breaking scale of U (1)2 /U (1)3 has to be sufficiently high to
2 k = 1, 2, because k = 3 does not have to be considered due to the antisymmetry of under the exchange
ij k
of j, k.
173
avoid contradiction with Flavor Changing Neutral Currents (FCNC):

R 30 103 GeV.
(3.25)
For simplicity we assume3

MR MT R,
MA A.
(3.26)
We plug (3.13), (3.15), (3.16) and (3.26) into (3.12) and determine its eigenvalues. The
eight large ones are approximately given by
MN ,
A,
A,
A,
(3.27)
all large enough that the corresponding particles cannot have been observed. However, we
are of course most interested in the three small eigenvalues. Writing
U

H
0
0
0 0 0 0 0
U
H=
(3.28)
0
H
0
0 0 0 0 0
0
0
H U 0 0 0 0 0
and
A
R
A
R
3/2
A
A
R1/2
A
R
R
,
J=
(3.29)
A
R
R
A R R R A A
A
A
R
R
A3/2
R1/2
A
A
we can express the see-saw mass matrix as

0 H
.
M=
HT J
(3.30)
Thus the three lightest eigenvalues and their corresponding eigenvectors are approximately,
and to lowest order in R/A, those of the 3 3 matrix

H U 2 R 0 1 0
T 1
H J H
(3.31)
1 0 0 ,
A2
0 0 2
namely,4
H U 2 R
,
A2
2H U 2 R
.
A2
(3.32)
3 Other choices either did not lead to substantially different results or produced contradiction with experiment.
4 This result can be found by an alternative method which for simplicity we demonstrate only for the simple
conventional 2 2 see-saw matrix, which gives the characteristic polynomial x 2 Mx m2 = 0. Assuming that
2
|x 2 | |Mx|, m2 one has Mx m2 0 so that x m

M , reproducing the well-known result. Plugging it into
4
2
2
the initial assumption above gives m 2 M m
M , m , justifying the assumption in hindsight.
174
The mass-degeneracy of the first two eigenvalues can be lifted by including higher orders,
and/or non-universal coefficients, to obtain (3.36). All expressions go to zero for vanishing
Higgs VEV, as they should. The diagonalizing matrix of (3.31) is given by
1 1 0 1
nL
1
2
2
1
1
(3.33)
n2L = 2 .
0
2
2
3
n
3
0
0
1
L
Here symbolizes the light mass-eigenstates. This result can be interpreted as the
(currently disfavored by SNO data [25]) maximal atmospheric neutrino mixing [24]
(23 45 ), and small mixing angle MSW solar neutrino oscillation (13 12 0 ),
i.e., one has to fix the values of R and A with
2

m m2 = 3.5 103 (eV)2 ,
(3.34)
2
3

2
m m2 = 3.5 103 (eV)2 ,
(3.35)
3
1
2

2
6
2
m m = 6 10 (eV) .
(3.36)
1
2
From the eigenvalues (3.32) we know that m3 2m2 , 2m1 . Therefore,
3 H U 2 R
3.5 103 (eV)2 ,
A2
(3.37)
and thus
R 3 1015 (GeV)1 A2 .
From (3.25) we hence obtain A2
we have the narrow range
(3.38)
31015
GeV
30 103 GeV, so that together with (3.23)
3 109 GeV A 1010 GeV.
(3.39)
Hence
30 TeV R 300 TeV,
(3.40)
which leads to
MN 1012 GeV,
(3.41)
A = 3 109
hence fulfilling the constraints of F -term SUSY breaking with

GeV.
When A acquires a VEV several renormalizable R
/ p couplings are generated from the
terms
A

LLE,
MS
A

QLD,
MS
A
U D D.
MS
(3.42)
Assuming their coupling constants to be O(1) one gets

231 ,
132 ,
123 ,
231 ,
333 ,
113 ,
223 ,
312 ,
132 ,
113 ,
311 ,
322,
223 ,
(3.43)
175
all being suppressed by a factor of A/MS 108 , which is below the current upper
bounds at high energies.5 We also note that the lepton number violating operators in (3.42)
can contribute to the left-handed neutrino masses through fermionsfermion loops, with
2
m 3 /(16 2 )m2b /MSUSY [27]. However, with the constraint (3.39), this contribution
is at most m 5 1011 eV and therefore negligible.
3.2. The case of U (1)2
Turning to U (1)Z U (1)2 we work with the following charge assignment:6
Field
U (1)Z
U (1)2
5
2
52
0
0
12
0
12
0
1
(3.44)
It allows the non-renormalizable term

T T HUHD
,
MS
(3.45)
so that we can generate the -term with

T 2 102 GeV MS
T 1010 GeV.
(3.46)
Thus if N , N , N (see (3.50)) are 10 the superpotential terms RST , RRT , SST
cannot contribute substantially to F -term SUSY-breaking (see (3.51) and (3.52)). Since
the U (1)2 -charge of (H UH D ) is 2, and neither a three-field combination of R, S, T or
one of these fields alone has a compensating U (1)2 -charge of +2, the U (1)2 -charges
forbid
(R or S or T )(R or S or T )(R or S or T )H UH D
MS2
and
AB(R or S or T )H UH D
.
MS2
(3.47)
Furthermore, since
3
QR , QS , QT = ,
2
the terms
1,2
A(R or S or T )Q3 L3 D
2
MS
(3.48)
and
3 D
1,2
3 D
A(R or S or T )U
,
2
MS
(3.49)
5 See Table III in [26]. Note that their convention is to label the heaviest generation 3, and that they work
rather than QLD.

with LQD
6 Being far from obvious we have not found any other U (1) -charge assignment for R, S, T that produced a
2
stable proton, a sensible see-saw matrix and the generation of a -term. Just as for U (1)3 , the U (1)2 -charges of
R, S, T are opposite to the ones of the neutrinos.
176
are forbidden (unlike the case with U (1)3 ) and thus the proton is protected. The
dimension five operator M1S Q3 Q3 Q3 L3 is also forbidden (again unlike U (1)3 ). It can
be generated from T T T5 T Q3 Q3 Q3 L3 , which however is highly suppressed. Similarly,
MS
all other dimension five operators are suppressed by at least T /MS and are therefore
adequately suppressed. The renormalizable superpotential is given by
y + ab xy G Qiax HbD D
y + ab G Lia HbD E
j
W = ab xy Gij Qiax HbU U
ij
ij
j
(U )
(D)
(E)
(N)
j + MA AB + i BR N
i + i BS N
i + BT N
3
+ ab Gij Lia HbU N
+ N RST + N RRT + N SST ,
(3.50)
with i, j = 1, 2. Looking at the charges we see that we can avoid D-term SUSY breaking
with
R = S = T ,
A = B.
(3.51)
Thus (3.25) is fulfilled.

As in the case of U (1)3 we have R A. MA and A have to be chosen such that
the SUSY breaking F -terms are adequately suppressed. Thus, we impose
MA A 1 TeV MS ,

i and i and RA 1 TeV MS A
1011 GeV
.
(i and i and )
(3.52)
The see-saw matrix is again of the form (3.30), with

U

H H U 0 0 0 0 0 0
U
U
H = H H 0 0 0 0 0 0
0
0
0 0 0 0 0 0
and
A
J=
A
A
A
R
R
A
A
A
A
A
R
R
R
R
R
R
R
(3.53)
R
R
R
MA
A
R
R
(3.54)
MA
where we have not displayed explicitly the coupling constants Gij , i , i , , N ,N ,
N .7 The eight large eigenvalues corresponding to heavy particles are approximately
(N)
A,
A,
A,
MA .
(3.55)
7 These couplings however cannot be completely random, otherwise the matrix could be singular, e.g., all
couplings exactly equal to unity does not work, unlike the case for U (1)3 .
177
It turns out that to a good approximation the small eigenvalues of the see-saw matrix,
corresponding to the very light neutrinos, are independent of MA , one has

U 2 R 0
H
HT J1 H
(3.56)
0 ,
A2
0 0 0

, , being complicated functions of G(N)
ij , i , i , , N , N , N . From the null-row
in the matrix above it follows that one of the neutrinos is massless. It is possible to assign
(N)
values between 0.1 and 10 to Gij , i , i , , N , N , N such that one gets nearly
degenerate masses for the other two neutrinos, and maximal mixing between them, obeying
(3.58). In order for this to reproduce the experimental data (3.34) with m3 0 one thus
needs
m1,2 =
H U 2 R
= 6 1011 GeV A 1012 GeV.
A2
(3.57)
In order to avoid substantial contributions to SUSY breaking F -terms we thus have to

impose (see (3.52))
i , i , 101 ,
(3.58)
and MA 109 GeV.8 From the terms (3.42) one gets

131 ,
231 ,
132 ,
232 ,
123 ,
131 ,
231 ,
132,
232 ,
123 ,
113 ,
311 ,
321 ,
312 ,
322 ,
213 ,
223,
131 ,
123 ,
213 ,
223 ,
312 ,
(3.59)
O(106 ),
this time all of them are suppressed by A/MS =

again not contradicting the
bounds of Table III in [26].9 Similarly, we note that the fermionsfermion contribution to
the left-handed neutrino masses [27] are adequately suppressed.
4. Discussion and conclusions

The structure of the standard model spectrum indicates the realization of grand
unification structures in nature. On the other hand the proton longevity severely constrains
the possible extensions of the standard model and serves as a useful guide in attempts
to understand the origin of the standard model gauge and matter spectrum. The realistic
free fermionic heterotic-string models reproduce the grand unification structures that
are suggested by the standard model and represent the most realistic string models
constructed to date. As such the realistic free fermionic models serve as a useful
probe to the fundamental characteristics of the possibly true string vacuum, as well
8 However, M should be sufficiently large so that the two particles with mass M would not have been
A
A
detected.
9 We note again that in our convention 1 labels the heaviest generation; we also work with QLD
instead of

LQD.
178
as to various properties that the string vacuum should possess in order to satisfy
various phenomenological constraints. It was proposed previously that proton stability
necessitates the existence of an additional U (1) symmetry, which remains unbroken down
to intermediate or low energies. Furthermore, the required symmetry is not of the type
that arises in Grand Unified Theories, but is of intrinsic string origin. The realistic free
fermionic models do indeed give rise to U (1) symmetries, which are external to the GUT
symmetries, and forbid the proton decay mediating operators.
One of the vital issues in heterotic-string Grand Unification is whether it is possible
to adequately suppress the proton decay mediating operators, and at the same time insure
that the left-handed neutrino masses are sufficiently small. That this is a non-trivial task
is seen, for example, from the fact that a VEV of the neutral component in the 16 and 16
of SO(10), which is used in the string models to generate the required large see-saw mass
scale, at the same time can induce the dangerous proton decay mediating operators from
non-renormalizable terms. The solution pursued here therefore advocates the existence
of an additional U (1) symmetry, which forbids the proton decay mediating operators,
while it permits the neutrino see-saw-matrix mass terms, as well as those of the quarks
and charged-leptons. Thus, provided that the additional U (1) is broken at a sufficiently
low scale, the proton decay mediating operators will be adequately suppressed. In this
paper we demonstrated that indeed the symmetries that appear in the string models can
guarantee proton longevity and simultaneously suppress the left-handed neutrino masses.
We achieved this by assuming additional Higgs spectrum that is needed to break the
U (1)Z U (1)? and by detailed construction of the neutrino see-saw mass matrix. Such
additional Higgs spectrum arises in the string models from fundamental states or from
product of states that produce the effective quantum number of the states that we assumed
here. We then analyzed the neutrino see-saw mass matrix and showed that the resulting
neutrino mass spectrum is in qualitative agreement with the experimentally observed
values.
It is interesting to note that among the semi-realistic orbifold models constructed to date
the free fermionic models are unique in the sense that they are the only ones that have been
shown to admit the SO(10) embedding of the standard model spectrum. This is true of the
Z3 heterotic string models [28] as well as the type I string models, which do not admit the
chiral 16 of SO(10) in the perturbative massless spectrum. This, in particular, means that
these models do not admit the canonical SO(10) embedding of the weak-hypercharge.
In these models the weak-hypercharge arises from a non-minimal combination of the
world-sheet U (1) currents, which correctly reproduces the standard model hypercharge
assignments. Therefore, the models do contain a number of U (1) symmetries that may
suppress the proton decay mediating operators, and the suppression depends on the U (1)Z
symmetry breaking scale. However, the non-canonical embedding of the weak hypercharge
typically results in exotic states which cannot be decoupled from the massless spectrum. As
regard the issue of neutrino masses, the non-standard embedding of the weak-hypercharge
implies that this issue can only be studied on a case by case basis, and general patterns are
more difficult to decipher. The case of type I constructions presents additional novelties. In
this case the standard model gauge group typically arises from a product of U (n) groups.
Therefore, embedding of the standard model gauge group in an enlarged gauge structure is
not obtained in type I string models. Consequently, one then often finds a gauged U (1)B
179
in these models, which guarantees proton longevity. The suppression of neutrino masses in
this context still remains an open issue that can only be addressed on a case by case basis.
However, due to the non-standard embedding of the weak-hypercharge, these models have
typically been considered in the context of low scale gravity models [29]. In which case
the suppression of of left-handed neutrino masses may arise from the large volume of the
extra dimensions [30].
To conclude, the standard model multiplet structure, augmented with the right-handed
neutrinos, strongly indicates the realization of an underlying SO(10) structure in nature.
Preservations of the SO(10) embedding in the context of string unification has proven to
be a highly non-trivial task. In fact, among the semi-realistic orbifold models, only the free
fermionic models are known to reproduce this desired structure. However, the problems of
proton longevity and the simultaneous suppression of the left-handed neutrino masses, still
pose an extremely severe challenge to Grand Unified, and heterotic-string, models. It has
been proposed that these challenges necessitate the existence of an extra non-GUT U (1)
symmetry, broken at low energies, and that the required symmetries do arise in the free
fermionic heterotic-string models. In this paper by constructing detailed string inspired
models, we showed that the additional stringy U (1) symmetries can indeed perform their
designated task. The relative simplicity of the additional Higgs spectrum that we assumed
here, suggests that further more detailed models might be able to fully reproduce the
observed quark and lepton mass spectrum, as well as safeguarding the proton lifetime.
Acknowledgements
We would like to thank Sacha Davidson and Athanasios Dedes for useful discussions
and the CERN theory group for hospitality. This work was supported in part by
PPARC. M.T. would like to thank Bonn University for hospitality and the Evangelisches
Studienwerk and Worcester College for financial support.
References
[1] For a review see, e.g., H.P. Nilles, Phys. Rep. 110 (1984) 1.
[2] S. Weinberg, Phys. Rev. D 26 (1982) 287;
N. Sakai, T. Yanagida, Nucl. Phys. B 197 (1982) 533;
R. Arnowitt, A.H. Chamsdine, P. Nath, Phys. Rev. D 32 (1985) 2348.
[3] R. Arnowitt, P. Nath, Phys. Rev. Lett. 62 (1989) 2225;
A. Font, L.E. Ibanez, F. Quevedo, Phys. Lett. B 228 (1989) 79;
J. Ellis, J.L. Lopez, D.V. Nanopoulos, Phys. Lett. B 252 (1990) 53.
[4] A.E. Faraggi, Nucl. Phys. B 428 (1994) 111, hep-ph/0107094.
[5] A.E. Faraggi, Phys. Lett. B 339 (1994) 223;
A.E. Faraggi, Phys. Lett. B 398 (1997) 95;
J. Ellis, A.E. Faraggi, D.V. Nanopoulos, Phys. Lett. B 419 (1998) 123.
[6] J.C. Pati, Phys. Lett. B 388 (1996) 532.
[7] A.E. Faraggi, Phys. Lett. B 499 (2001) 147.
[8] H. Kawai, D.C. Lewellen, S.H.-H. Tye, Nucl. Phys. B 288 (1987) 1;
I. Antoniadis, C. Bachas, C. Kounnas, Nucl. Phys. B 289 (1987) 87.
180
[9] I. Antoniadis, J. Ellis, J. Hagelin, D.V. Nanopoulos, Phys. Lett. B 231 (1989) 65;
J.L. Lopez, D.V. Nanopoulos, K. Yuan, Nucl. Phys. B 399 (1993) 654.
[10] A.E. Faraggi, D.V. Nanopoulos, K. Yuan, Nucl. Phys. B 335 (1990) 347;
A.E. Faraggi, Phys. Rev. D 46 (1992) 3204.
[11] I. Antoniadis, G.K. Leontaris, J. Rizos, Phys. Lett. B 245 (1990) 161;
G.K. Leontaris, J. Rizos, Nucl. Phys. B 554 (1999) 3.
[12] A.E. Faraggi, Phys. Lett. B 278 (1992) 131;
A.E. Faraggi, Nucl. Phys. B 403 (1993) 101.
[13] A.E. Faraggi, D.V. Nanopoulos, Phys. Rev. D 48 (1993) 3288;
A.E. Faraggi, Nucl. Phys. B 387 (1992) 239, hep-th/9708112.
[14] G.B. Cleaver et al., Phys. Lett. B 455 (1999) 135;
G.B. Cleaver et al., hep-ph/9904301;
G.B. Cleaver et al., hep-ph/0104091.
[15] G.B. Cleaver, A.E. Faraggi, C. Savage, Phys. Rev. D 63 (2001) 066001;
G.B. Cleaver, D.J. Clements, A.E. Faraggi, hep-ph/0106060.
[17] A.E. Faraggi, D.V. Nanopoulos, Mod. Phys. Lett. A 6 (1991) 61.
[18] K.R. Dienes, Nucl. Phys. B 488 (1997) 141.
[19] S. Kalara, J.L. Lopez, D.V. Nanopoulos, Nucl. Phys. B 353 (1991) 651.
[21] See, for example: M. Aoki, N. Oshimo, Phys. Rev. Lett. 84 (2000) 5269;
A. Aranda, C.D. Carone, Phys. Rev. D 63 (2001) 075012.
[22] A.E. Faraggi, E. Halyo, Phys. Lett. B 307 (1993) 311;
A.E. Faraggi, J. Pati, Phys. Lett. B 400 (1997) 314.
[23] H.K. Dreiner, hep-ph/9707435.
[24] SuperKamiokande Collaboration, Y. Fukuda et al., Phys. Lett. B 436 (1998) 33.
[25] SNO Collaboration, Q.R. Ahmad et al., nucl-ex/0106015.
[26] B.C. Allanach, H.K. Dreiner, A. Dedes, Phys. Rev. D 60 (1999) 075014, hep-ph/9906209.
[27] See, for example, S. Davidson, M. Losada, hep-ph/0010325, and references therein.
[28] L. Ibez, J.E. Kim, H.P. Nilles, F. Quevedo, Phys. Lett. B 191 (1987) 282;
J.A. Casas, C. Muoz, Phys. Lett. B 209 (1988) 214;
J.A. Casas, C. Muoz, Phys. Lett. B 214 (1988) 157;
A. Font, L. Ibez, H.P. Nilles, F. Quevedo, Phys. Lett. B 210 (1988) 101;
A. Font, L. Ibez, F. Quevedo, A. Sierra, Nucl. Phys. B 331 (1990) 421;
J. Giedt, hep-th/0108244.
[29] G. Aldazabal, L. Ibez, F. Quevedo, JHEP 0001 (2000) 031;
L.E. Ibanez, F. Marchesano, R. Rabadan, hep-th/0105155.
[30] N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, J. March-Russell, hep-ph/9811448;
A.E. Faraggi, M. Pospelov, Phys. Lett. B 458 (1999) 237.

On SUSY breaking and SB from string duals

Jaume Gomis
Department of Physics, California Institute of Technology, Pasadena, CA 91125, USA
Received 17 December 2001; accepted 21 December 2001
Abstract
We find regular string duals of three-dimensional N = 1 SYM with a ChernSimons interaction
at level k for SO and Sp gauge groups. Using the string dual we exactly reproduce the conjectured
pattern of supersymmetry breaking proposed by Witten by showing that there is dynamical
supersymmetry breaking for k < h/2 while supersymmetry remains unbroken for k h/2, where h
is the dual Coxeter number of the gauge group. We also find regular string duals of four-dimensional
N = 1 SYM for SO and Sp gauge groups and exactly reproduce the expected pattern of chiral
symmetry breaking Z2h Z2 by analyzing the symmetries of the string solution. 2002 Published
by Elsevier Science B.V.
PACS: 11.25.Mj; 11.15.-q
1. Introduction
The description of asymptotically free, confining gauge theories via gravity has shed
new light on familiar questions regarding the non-trivial infrared dynamics of gauge theory.
Despite the lack of proper string techniques to study the strong coupling dynamics of
an asymptotically free dual gauge theory, many expectations from gauge theory have
been realized in gravity duals. Understanding how to deal with string theory in highly
curved backgrounds in the presence of RamondRamond fields will eventually yield a
calculationally reliable solution of the large N limit of gauge theory.
In this paper we find the string duals of minimally supersymmetric gauge theories in
three and four dimensions for SO and Sp gauge groups and exactly reproduce some of the
expected non-trivial infrared dynamics of these gauge theories.
In the three-dimensional case we generalize the construction of Maldacena and
Nastase [1] and construct regular string duals of N = 1 SYM with a ChernSimons
E-mail address: gomis@theory.caltech.edu (J. Gomis).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 5 8 - 7
182
J. Gomis / Nuclear Physics B 624 (2002) 181199
interaction at level k for SO and Sp gauge groups. Witten has conjectured that
supersymmetry is spontaneously broken in this theory when k < h/2, where h is the dual
Coxeter number of the gauge group, and has shown that supersymmetry is unbroken for
k h/2. Generalizing the results of [1] we show that the string dual we describe exactly
reproduces the conjectured pattern of supersymmetry breaking.
We also construct a regular supergravity solution describing four-dimensional N = 1
SYM for SO and Sp gauge groups. Here we generalize the solution of Maldacena and
Nuez [2] to find solutions in which the expected pattern of chiral symmetry breaking
Z2h Z2 in gauge theory is realized in the string construction by showing that an
asymptotic symmetry of the solution that exists in the UV region is broken in the IR region.
An important tool in finding the string duals of these gauge theories is the proper
treatment of orientifold projections1 when branes wrap supersymmetric cycles and
understanding how to extend the orientifold action in the gravity dual. Precise agreement
with field theory expectations is obtained by taking into account an interesting subtlety
on the proper description of the WessZumino coupling on a D-brane in models with
unoriented open strings which we describe in Section 3.
The plan of the rest of the paper is as follows. In Section 2 we briefly summarize the field
theory results in [4] and the solution found in [1] which is dual to D = 3 SYM for SU(N)
gauge group. In Section 4 we make a suitable orientifold construction which generalizes
the gauge groups to SO and Sp and identify the effect of the orientifold construction in the
gravity dual. Following [1] we realize Wittens conjecture about supersymmetry breaking.
In Section 4 we recall the field theory results on chiral symmetry breaking in D = 4 N = 1
SYM and the dual supergravity solution in [2] when the gauge group is SU(N). In Section 5
we find the string dual for SO and Sp gauge groups and reproduce using the string dual
the expected pattern of chiral symmetry breaking. Finally, Appendix A rederives the BPS
equations in [1] from an effective superpotential method.
2. SUGRA dual to D = 3 N = 1 SYM with a ChernSimons interaction

A supersymmetric ChernSimons interaction at level2 k can be added to N = 1 SYM
with gauge group G
k
LCS =
4

2
Tr A dA + A A A + .
3
(2.1)
Classically, the supersymmetric gauge theory has a mass gap when k = 0 [5]. The mass
of the gluon and gluino fields is of order g 2 k and for k 1 one can integrate out
the massive gluinos while retaining a local purely bosonic low energy effective action.
Integrating out a massive fermion at one loop in a three-dimensional gauge theory induces
a ChernSimons interaction whose level is determined by the sign of the fermion mass
1 In a similar context to the one studied here, Ref. [3] considered orientifolds of wrapped branes.
2 In this paper we will take k > 0.
183
and the representation of G under which the fermion transforms [6,7].3 Therefore, in the
supersymmetric example we are considering, integrating out the massive gluino shifts the
effective value of k appearing in the bosonic ChernSimons term to keff = k h/2 [8],
where h is the dual Coxeter number of G. At very low energies, where the YangMills term
can be neglected in comparison with the ChernSimons term, the effective description of
the model isfor k 1given by bosonic ChernSimons theory with gauge group G
at level keff . This result was extended to all k in [4] where it was shown that the low
energy effective description of N = 1 SYM with gauge group G with a ChernSimons
interaction at level k is given by bosonic ChernSimons theory with gauge group G at
level keff = k h/2.
Using this effective description, Witten [4] computed the supersymmetry index [9]
Tr(1)F of three-dimensional N = 1 SYM with gauge group G with a ChernSimons
interaction of level k. The number of supersymmetric vacua equal the number of zero
energy states of the effective bosonic ChernSimons theory at level keff , which in turn
equal the number of conformal blocks of the corresponding WZW model.4 The result is

= 0, k h/2,
Tr(1)F =
(2.2)
= 0, k < h/2,
which shows that supersymmetry is unbroken for k h/2. The supersymmetry index vanishes for k < h/2, which is insufficient to decide whether supersymmetry is broken in the
infinite volume limit. However, in [4] Witten conjectured that dynamical supersymmetry
breaking does occur for k < h/2.
In Section 3 we generalize the work of Maldacena and Nastase [1] and exactly reproduce
the conjectured pattern of supersymmetry breaking for SO and Sp gauge groups using a
string dual description. Here we summarize the value of the dual Coxeter number of the
gauge groups for which we find a supergravity dual solution.
2.1. Supergravity dual for G = SU(N)
A simple way5 of realizing D = 3, N = 1 SU(N) SYM is to study Type II string theory
on R1,2 X, where X is a seven-dimensional manifold of G2 holonomy and to consider
the low energy effective field theory living on N suitably wrapped branes.6 There are two
choices that lead to the desired gauge theory. One can either wrap N Type IIB five-branes
on the supersymmetric S3 cycle [1,12] of X = S(S3 )the spin bundle over S3 or wrap
3 The induced level is given by k = 1 C sign(m), where C is the index of the representation R under
R
2 R
a t b ) = 2C ab . For the adjoint representation
which the fermion transforms and m is the fermion mass, i.e., tr(tR
R
R
CR = h.
4 To properly define the index one considers the theory on R T2 and the number vacua equals the number of
conformal blocks of the WZW model on a Rieman surface with T2 topology. On T2 , the conformal blocks are in
at level keff . Moreover,

one to one correspondence with the integrable representations of the affine Lie algebra G
all vacua are bosonic or fermionic so that Tr(1)F = number of conformal blocks.
5 For an alternative realization of SU(N ) SYM with a ChernSimons term see, for example, [10,11], where a
description of supersymmetry breaking for SU(N ) was given.
6 X has to be a non-compact space in order for the flux coming from the wrapped branes to be able to spread
to infinity.
184
N Type IIA D6-branes on the supersymmetric four cycle [1315] Z of X = (Z)the

bundle of anti-self-dual two-forms over Zwhere Z is S4 or P2 .7 Since the cycle which
the branes wrap is rigid, the familiar massless scalars describing the motion of the branes
in the transverse directions are absent in the low energy field theory and one just gets pure
SU(N) gauge theory in three dimensions. In this paper we consider the Type IIB realization
with wrapped five-branes.
In the Type IIB realization a ChernSimons interaction at level k can be incorporated
by turning on three-form flux through S3 S(S3 ) which induces the following bosonic low
energy coupling on the five-branes

1
1
2 3
LWZ =
B Tr(F F ) = 2
H Tr A dA + A
8 2
8
3
R 1,2 S 3
k0
=
4

2
Tr A dA + A3 ,
3
R 1,2 S 3
(2.3)
R 1,2
which together with its superpartner generates the N = 1 supersymmetric ChernSimons

interaction (2.1).
We have been careful in differentiating between the levels k0 and k. The reason for this
is that the low energy effective field theory on the branes is SU(N) SYM with a Chern
Simons interaction with bare level k0 . However, massive N = 1 KaluzaKlein multiplets
arise by reducing the six-dimensional massless vector multiplet on the five-branes over S3
and integrating out the massive KaluzaKlein adjoint fermions may shift the bare Chern
Simons level k0 if there is a net number of multiplets with a given sign for their mass.8 It
has been shown9 in [1] that there is a single KaluzaKlein adjoint massive chiral multiplet
which does not have a partner with opposite mass so that the effective theory on the branes
after integrating out all KaluzaKlein modes is three-dimensional N = 1 SU(N) SYM
with a ChernSimons interaction at level k = k0 h/2 = k0 N/2.
Recently, Maldacena and Nastase [1] have interpreted a solution10 found by Chamseddine and Volkov [17] as describing a collection of N five-branes wrapping S3 S(S3 )
when k0 = N . This regular supersymmetric solution is, therefore, dual to D = 3, N = 1
SU(N) SYM with a ChernSimons term at level k = N/2.
The Type IIB supergravity solution [1] obtained by lifting [1,18] the gauged supergravity solution in [17] to ten dimensions is in the class of supersymmetric supergravity solu7 Since P2 is not a Spin manifold, one must turn on non-zero magnetic flux through P1 P2 [16].
8 Since the shift of the level depends only on the sign of the fermion mass term, fermions with opposite mass
leave k0 invariant and the net shift of the level depends only on the net number of fermions with a specific mass
term sign.
9 One can start with the topological twist corresponding to D = 3 N = 2 SYM, so that the KaluzaKlein
modes come in N = 2 chiral multiplets which have two fermions one with each sign for the mass and, therefore,
produce no shift of the level. One can continuously twist the theory to N = 1 by smoothly changing the
gauge field on the normal bundle. To determine the number of unpaired fermions one computes the number of
eigenvalues of the Dirac operator which change sign under a gauge transformation of the normal bundle, which
is just given by the winding number of the gauge transformation [4], which is 1.
10 In Appendix A we rederive this solution from an effective superpotential for the ansatz.
185
tions considered in [19].11 It is a compactification12 of Type IIB string theory on R1,2 M7

with a non-trivial three-form field strength H and dilaton profile over M7 , where M7 denotes the internal space. The metric of the solution is given by

1,2
1
1 2
2
2
2
wa Aa )2 ,
+ N dr + R (r)wa wa + N (
ds = ds R
4
4
1 + b(r)
wa ,
Aa =
(2.4)
2
where wa , a = 1, 2, 3 parametrize S3 and w
a , a = 1, 2, 3 are another set of SU(2) left

invariant one forms parametrizing
S3 . The radial function b(r) is monotonically increasing

and goes from 0 at r to 1 at r = 0. On the other hand, R 2 (r) behaves linearly in r
near r while R 2 (r) r 2 near r = 0, so that the metric inside the square brackets
in (2.4) is the flat R4 near the origin. Topologically, M7 is asymptotically conical with
base S3
S3 and in the interior the
S3 stays of finite size while S3 becomes contractible.

The supersymmetry left unbroken by the solution, which is N = 1 in three dimensions,
guarantees that M7 is a G2 holonomy manifold with respect to the connection with torsion
= + H , where is the usual Levi-Civita connection and H is the three-form
piercing M7 .
The solution has non-trivial H -flux (see Eqs. (A.5) and (A.15) in Appendix A). The
S3 .
number of five-branes wrapping S3 S(S3 ) is encoded as magnetic flux of H through
3
The solution also has flux through S , and its asymptotic value at the boundary r , as
explained previously, determines the ChernSimons level of the holographic field theory
dual. The fluxes of the solution are13

H
H
=
= N.
(2.5)
2
2
S3
S3
Since k0 = N , this supergravity solution is dual to SU(N) SYM with k = N/2. An external
quark is represented by a string stretching from the boundary and since the solution is
completely non-singular and without a horizon it exhibits confinement as well as a mass
gap as expected from the field theory analysis [4,21].
We note that this duality realizes the general philosophy advocated in [22] in which
the infrared description of the gauge theory on a collection of wrapped branes is given
by a topologically distinct background in which branes have been replaced by fluxes. In
the original description the branes wrap S3 S(S3 ) which is asymptotically a cone over
S3
S3 , where
S3 is contractible in the full geometry while S3 is topologically non-trivial.

The dual description of the gauge theory is given by a geometry M7 in which the roles of
S3 and
S3 are reversed, so that
S3 is now topologically non-trivial and supports the flux

left behind by the branes once they disappear. In effect the infrared dynamics of the gauge
theory is captured by the flopped geometry which the branes wrapped [23].
11 Ref. [19] considered the case D = 6. The analysis was extended when D = 7 in [20].
12 Strictly speaking, it is not a compactification since the internal geometry is non-compact.
13 The flux of H through S3 vanishes at r = 0 as required for a smooth solution since S3 vanishes in the
interior.
186
Since the supergravity solution is supersymmetric, this shows that SU(N) SYM with
a ChernSimons term at level k = N/2 has unbroken supersymmetry. Maldacena and
Nastase showed that one can use this solution to explore what happens when k = N/2.
Changing to k = N/2 + nor equivalently changing to k0 = N + nis accomplished by
wrapping n = k N/2 five-branes14 on
S3 M7 , which adds n units of H -flux through

3
S M7 so that the level is k = N/2 + n. This can be done reliably for n N , when the
backreaction of the n wrapped branes can be safely ignored. Adding n > 0 five-branes does
not break supersymmetry,15 since the wrapped branes preserve the same supersymmetry
as the background [1]. On the other hand, adding n < 0 five-branesthat is |n| antibranesbreaks supersymmetry completely. Therefore, using supergravity one realizes the
expected pattern of supersymmetry breaking for G = SU(N). For k N/2 supersymmetry
is unbroken and for k < N/2 supersymmetry is broken.
For k > N/2 the low energy effective description of the physics is given by the field
theory on the n five-branes wrapped over
S3 . At very low energies, where the YangMills

term can be neglected, one obtains N = 1 SU(n) ChernSimons theory with level k = N .
Integrating out the SU(n) adjoint gluinos and KaluzaKlein modes result in a negligible
shift of the ChernSimons level (of O(n)) in the regime where backreaction can be ignored
so that the effective description is given by SU(n) ChernSimons theory with level k = N
which is precisely the level-rank [24] dual of what was found in [4] which is SU(N) Chern
Simons theory with keff = n . In the next section we generalize this duality to SO and Sp
gauge groups.
3. Supergravity dual for G = SO and G = Sp

The SU(N) theory is realized by wrapping N D5-branes16 on S3 S(S3 ). A simple way
to generalize the construction to other gauge groups is to superimpose an orientifold fiveplanean O5-planeon top of the five-branes. Since the branes are curved this requires
performing an orientifold projection which leaves an O5-plane along R1,2 S3 , which is
the worldvolume of the five-branes. This can be accomplished by considering the following
projection

Type IIB on R1,2 S S3 /{1, },
(3.1)
where is the usual worldsheet parity operator and is the generator of a Z2 symmetry
of S(S3 ) which we will now describe.
14 Since the external quark is a string which can now end on the five-branes, this shows that for k = N/2 that
there is no confinement, in agreement with the field theory analysis of [4].
15 The G structure of M , which is a three-form, is such that when pulled back to
S3 is proportional to the
7
2
volume form on
S3 .
16 In the previous section we wrote the solution corresponding to wrapped NS5-branes. One can trivially
act with S-duality and write the corresponding solution for wrapped D5-branes, which is given by ds 2 (D5) =
eNS ds 2 (NS5), C2R = B2R and D5 = NS .
187
The topology of S(S3 ) where the five-branes are supersymmetrically embedded can be
characterized by the following hypersurface in R8
4

ai2 bi2 = V ,
(3.2)
i=1
S3 .
with ai , bi R. The space is asymptotically conical and the base of the cone is S3
These spheres are parametrized by the coordinates ai and bi , respectively. For V > 0,
S3 is topologically non-trivial while
S3 shrinks smoothly in the interior and is, therefore,

topologically trivial. For V < 0 the roles of S3 and
S3 are reversed and now it is
S3 which
is topologically non-trivial. The transformation V V can then be interpreted as a flop
transition [23].
In this description of S(S3 ) the relevant acts by
a a ,
i
i
:
(3.3)
bi bi .
If one considers the geometry near infinity after quotienting by one gets a cone with base
S3 /Z2 . Since
S3 is contractible in the full geometry the Z2 action has fixed points.

S3
Therefore, an O5-plane sits at the fixed locus of that is bi = 0which is R1,2 S3 ,

which is what we were after. Therefore, we have identified a suitable orientifold projection
in string theory that results, at energy scales smaller than the scale of the S3 to a threedimensional N = 1 gauge theory with SO or Sp gauge groups. For future reference we
note that when V < 0, that is in the flopped geometry, that is a freely acting involution
since the fixed point set bi = 0 is not part of the manifold and thus there is no orientifold
plane.
Three different gauge groups can be realized by placing D5-branes on top of an O5plane. The three types of orientifold planes17 differ in the choice of discrete fluxes in the
3 =
geometry transverse to the orientifold plane, which in this case is RP

S3 /Z2 .
We can now construct SUGRA duals to these gauge theories by implementing the
orientifold action we have just described on the SUGRA solution presented in the previous
section. That background was given by Type IIB string theory on R1,2 M7 where M7
is a G2 holonomy manifold with respect to a connection with torsion. As discussed in
the previous section, topologically M7 is asymptotic to a cone over S3
S3 just like
S(S3 ). However, M7 is topologically the flopped version of S(S3 ). That is, in M7 S3 is
contractible while
S3 is topologically non-trivial. Therefore, acts freely on M7 since

3 never shrinks. Thus, the string dual to D = 3 N = 1 SYM theories with SO and Sp
RP
gauge groups is described by unoriented closed string theory on the smooth background
R1,2 M7 /Z2 , without any open strings nor orientifold planes.
In the SUGRA dual solution the branes wrapped on S3 disappear and are replaced by
flux over
S3 , which is now non-contractible. This still occurs when an O5-plane is added in

the gauge theory realization. The only two differences are that one must take into account
17 There is actually a fourth type [25] of O5-plane, which yields Sp(N ) gauge symmetry. It differs from the
familiar O5-plane yielding the same gauge group by a discrete RamondRamond 0-form -angle. These two
theories differ only in their non-perturbative spectrum.
188
the total five-brane chargethat of the five-branes plus that of the O5-planewhich is then
3 due to the action of . The
converted into flux in the dual theory and replace
S3 by RP
total five-brane charge can be easily computed by remembering that the gauge groups that
we are considering appear by projecting the massless open strings on N branes together
with their imagesso that there are 2N branes in totalapproaching an orientifold plane

and that the charge of an Op plane is 32 2p9 . Finally, one must recall that an Op
plane has an extra Dp-brane which cannot be moved from the orientifold plane since it
has no image under the orientifold action. Therefore, taking into account the charge of the
orientifold we see that the total five-brane charge is given by Table 3.
As described above the corresponding SUGRA solution is found from that in [1] by
3 . Therefore, the fluxes of the
replacing everywhere in the solution N Q and
S3 RP
supergravity solution are given by

H
H
=
= Q.
(3.4)
2
2
S3
S3
Since there is non-zero flux of H through S3 this solution describes D = 3 N = 1 SYM

with a non-vanishing ChernSimons interaction. In order to identify the correct value of the
level of the ChernSimons interaction, one must take into account the following subtlety.18
The WessZumino term

1
LWZ =
(3.5)
B Tr F F
8 2
is the familiar coupling in the worldvolume theory of N D5-branes with gauge group
SU(N). This coupling implies that a D1-brane dissolved inside N D5-branes is equivalent
to a single instanton in the SU(N) gauge theory on the five-branes, so that the induced
D1-string charge is the same as the instanton number in the SU(N) gauge theory. Is
this coupling modified when one considers the worldvolume theory of D-branes with
unoriented open strings? First lets consider the case where the gauge theory on the brane is
SO(2N).19 When the five-branes approach the O5 -plane the gauge symmetry is enhanced
from SU(N) to SO(2N). Since a single instanton in SU(N) corresponds to a unit of D1string charge, in order to determine the correct normalization of the WessZumino coupling
for an SO(2N) gauge theory, one must determine the SO(2N) instanton number of a
SU(N) gauge field configuration when it is embedded in SO(2N). It is straightforward20 to
show that an instanton number one gauge field in SU(N) has instanton number one when
embedded in SO(2N), which in turn corresponds a single unit of induced D1-string charge.
Therefore, the correct coupling is given by
LSO
WZ = LWZ .
(3.6)
18 A similar discussion appeared recently in [26].

19 The discussion that follows trivially extends when the gauge group is SO(2N + 1).
20 One can embed a SU(N ) instanton in SO(2N ) by turning on a SU(2) instanton in SU(N ) via the embedding
given by SU(N 2) SU (2) SO(2N 4) SO(4). The minimal SO(2N ) instanton is then obtained by
embedding the SU(2) SU (N ) instanton in an SU(2) subgroup of SO(4), which clearly has instanton number
one in SO(2N ).
189
Table 1
Value of dual Coxeter number for several gauge groups
SU(N )
SO(2N )
Sp(N )
SO(2N + 1)
2N 2
N +1
2N 1
On the other hand, when the five-branes approach the O5+ -plane the gauge symmetry
is enhanced from SU(N) to Sp(N). Here an instanton number one gauge field in SU(N)
has instanton number two when embedded in Sp(N). This follows by embedding an SU(2)
instanton inside SU(N) into Sp(N) via the splitting SU(N 2) SU(2) Sp(N 2)
Sp(2). If we denote by t an SU(2) gauge field configuration, then the minimal embedding
of SU(2) inside Sp(2) is given by the following Sp(2) Lie algebra valued matrix

t
0
TSp =
(3.7)
.
0 t T
It then follows that the instanton number of TSp is twice that of t. Therefore, the induced
D1-string charge on the D5-brane worldvolume equals twice the instanton number of
Sp(N) so that the correct WessZumino coupling is given by
1
Sp
LWZ = LWZ .
2
(3.8)
Taking into account this subtlety for Sp(N) one finds that the flux of H over S3 in
Table 3 induces a bare ChernSimons interaction at level k0 = h (see Table 1) on the
worldvolume of the wrapped branes for all gauge groups. As explained in [1], k0 is shifted
by the presence of an unpaired massive KaluzaKlein fermion, which is not projected
out by the presence of the O5-plane but transforms now in the adjoint representation
of G. Integrating this fermion out shifts k0 to k = k0 h/2 = h/2. Therefore, we have
generalized the results of of [1] to find the string dual of three-dimensional N = 1 SYM
with gauge group G and a ChernSimons coupling at level k = h/2 and have seen that
supersymmetry is unbroken. Regularity of the solution and the lack of a horizon show
that the dual gauge theory confines and has a mass gap as expected from the field theory
analysis of [4,21].
One can change the value of the level to k = h/2 + n by wrapping n five-branes21
3 M7 , which in turn adds n units of flux through S3 . When n > 0, adding
over RP
five-branes does not break any further supersymmetry and, therefore, supersymmetry
is unbroken. On the other hand, when n < 0, anti-five-branes must be added to the
background, which break supersymmetry completely. Therefore, using SUGRA we have
exactly reproduced the conjecture by Witten regarding supersymmetry breaking as a
function of k for these theories, that is for k h/2 supersymmetry is unbroken while
supersymmetry is spontaneously broken for k < h/2.
The low energy effective theory when n > 0 or k > h/2 is properly described by the
3 . The statement that the five-branes wrap
field theory on the n wrapped branes over RP
21 Just as in the case discussed in [1], this analysis is self-consistent when n h.
190
3 , where Y
3 can be formalized by saying that there is a map between Y and RP
RP
is a closed three-manifold. Since the background we are considering is orientifolded, this
means that in going around a non-contractible loop in the target space, that the orientation
of the worldsheet must be reversed, which means that must satisfy [25] (x) = w1 (Y ),
3 , Z2 ) and w1 (Y ) is the obstruction to the orientability of Y . Following
where x H 1 (RP
[25], one can show that must a map of even degree so we take, for example, Y =
S3 . By
integrating the appropriate WessZumino coupling on the worldvolume of the n wrapped
3 . Therefore, the
five-branes, we find that the flux over Y is twice as large as that over RP
effective field theory on the branes in the limit where n h in which backreaction can be
ignored is given by
G = SO(N) : SO(n) ChernSimons at level k = N ;
G = Sp(N) : Sp(n) ChernSimons at level k = N .
This is to be compared with the expected low energy description found by Witten and
described in Section 2:
G = SO(N) : SO(N) ChernSimons at level keff = n;
G = Sp(N) : Sp(N) ChernSimons at level keff = n.
Remarkably, the effective description predicted by the string dual is precisely the levelrank dual [24] of the expected field theory result, which has the same number of ground
states.
4. SUGRA dual to D = 4 N = 1 SYM
Four-dimensional N = 1 SYM with gauge group G has at the classical level a chiral
U (1)R symmetry that acts on the gluino fields by
ei ,
ei .
(4.1)
Likewise, U (1)R acts with the same charges on the supersymmetry generators of positive
, respectively. Quantum mechanically, the presence of
and negative chirality Q and Q
zero modes of in an instanton background induces a t Hooft vertex which explicitly
breaks the U (1)R symmetry down to an anomaly free chiral Z2h subgroup, where h is the
dual Coxeter number of G. The Z2h symmetry is then spontaneously broken to Z2 , which
by the presence of a gluino condensate
acts by 1 on and ,
h = 43h
QCD ,
(4.2)
which gives rise to h supersymmetric vacua with a mass gap and confinement.
4.1. Supergravity dual for G = SU(N)
A simple way of realizing D = 4 N = 1 SU(N) SYM is to consider the low energy
effective field theory on N five-branes wrapping the S2 of the resolved conifold geometry,
191
which is topologically the X = O(1) O(1) bundle over S2 and admits a metric with
SU(3) holonomy.22
Recently, Maldacena and Nuez [2] have interpreted a gauged supergravity solution
found by Chamseddine and Volkov [28,29] as the supergravity description of the above
wrapped five-brane configuration. This smooth supergravity solution is thus dual to D = 4
N = 1 SU(N) SYM.
The solution they found is a compactification23 of Type IIB string theory on R1,3 M6
with a non-trivial three-form field strength H and dilaton over M6 . The metric of the
solution is given by

1
wa Aa )2 ,
ds 2 = ds 2 R1,3 + N dr 2 + e2h(r) d 2 + sin2 d 2 + N (
4
A2 = a(r) sin d,
A3 = cos d,
A1 = a(r) d,
(4.3)
a are SU(2) left

where and parametrize the S2 which the branes wrapped and w
invariant one-forms parametrizing the transverse
S3 to the five-branes
sin
w1 = cos
+ sin
d
+ cos
.
w3 = d
sin
w2 = sin
+ cos
d
,
(4.4)
The explicit form of the radial functions h(r), a(r) can be found in [2]. Here we will only
need its asymptotics which near r are e2h r and a(r) 0 while near r 0 they
behave as e2h r 2 and a(r) 1 so that the metric in the square brackets in (4.3) is the flat
R3 metric near the origin. Topologically, M6 is asymptotically conical with base S2
S3
3
2
and in the interior S stays of finite size while S becomes contractible. The solution has
N = 1 supersymmetry in four dimensions so that M6 is a SU(3) holonomy manifold with
= + H , where is the usual Levi-Civita
respect to the connection with torsion
24
connection while H is the three-form flux through M6 .
The solution also has the following non-trivial three-form field strength

N
1
1
H=
(4.5)
wa Aa )
(w1 A1 ) (w2 A2 ) (w3 A3 ) + Fa (
2
4
4
whose flux over
S3 encodes number of five-branes wrapping S2 in the X = O(1)

O(1) bundle over S2

H
= N.
(4.6)
2
S3
22 There is a mirror realization, in which pure SYM appears by wrapping N D6-branes on the deformed
conifold. It is in this Type IIA picture where one can make a direct connection with M-theory on G2 holonomy
spaces [23,27].
23 The internal geometry is nevertheless non-compact.
24 The complex geometry of M and the derivation of the BPS equations via an effective superpotential for the
6
ansatz can be found in [30].
192
We note that this solution also realizes25 the general philosophy of [22]. Here the
original geometry of X is such that it is asymptotically a cone over S2
S3 and fiveS3 is contractible in the full geometry.

branes wrap the topologically non-trivial S2 while
In the dual supergravity solution, the roles of S2 and
S3 are reversed since in M6 S2 is

contractible while
S3 supports the flux left behind by the branes once they disappear. The
topology of M6 is that of the deformed conifold.
One can use this supergravity solution, which manifestly exhibits a mass gap and
confinement, to explore the gravity realization [2] of the expected chiral symmetry
breaking Z2N Z2 . From the implementation of the twist required to obtain a
supersymmetric gauge theory on the wrapped brane one can identify the classical U (1)R
, which acts by
+ c.
symmetry of the gauge theory with shift symmetry of
Clearly, from the form of metric (4.3) in the interior (IR) of M6 we see that the metric
by 2
+ 2 , which due to the
periodicity
is only invariant under shifts of
+ 4 generates a Z2 symmetry, which is identified with the Z2 symmetry of N = 1
SYM preserved by the gluino condensate.

In the UV region of the geometry the full U (1)R symmetry acting by arbitrary shifts
is restored. On the other hand the field theory expectation is that this symmetry is
on
broken down to Z2N . In field theory, this explicit breaking is due to YangMills instantons
and in [2] this expectation was realized by taking into account the effect of a YangMills
instanton in the brane realization. By taking into account the WessZumino term (3.5) one
can identify a SYM instanton with a string worldsheet wrapping an S2 in M6 . As explained
in [2] the appropriate description of a SYM instanton is given by a string worldsheet
= and
= . Then the
wrapping an S2 in M6 parametrized by and and with
amplitude for a string instanton wrapping this cycle is given by

v = exp i B .
(4.7)
S2
By using the asymptotic expression for H in (4.5) one finds that

B = 2N .
(4.8)
S2
+
Therefore, the instanton corrected UV string solution is only invariant under
2k/N , where n = 1, . . . , 2N and, therefore, reproduce the expected anomaly free Z2N
R-symmetry.
In this way the expected chiral symmetry breaking pattern Z2N Z2 is realized in
the string dual as the symmetry group of the string solution being broken from the UV to
the IR. The N vacua can then be easily constructed [2] by performing a discrete set of N
rotations of the SU(2) gauge fields Aa in (4.3) described in [2] such that the UV solution is
unchanged while it is rotated but still smooth in the interior. We now extend this realization
to SO and Sp gauge groups.
25 We note that a very interesting description of a confining gauge theory in which branes are replaced by fluxes
also appeared in [31] which is dual to a cascading gauge theory.
193
5. Supergravity dual for G = SO and G = Sp

One can generalize the construction in the previous section to SO and Sp gauge groups
by introducing an O5-plane on top of the five-branes.26 In this case the O5-plane must
along R1,3 S2 , where S2 is the supersymmetric cycle of the resolved conifold geometry
X = O(1) O(1) bundle over S2 . This can be accomplished by considering

Type IIB on R1,2 X /{1, },
(5.1)
where is the usual worldsheet parity operator and is Z2 symmetry of X whose fixed
point set is R1,3 S2 .
We recall that the conifold singularity
z1 z2 z3 z4 = 0
(5.2)
can be resolved by replacing (5.2) by the following pair of equations

z1 z3
1
= 0,
z4 z2
2
(5.3)
where (1 , 2 ) (1 , 2 ) dont vanish simultaneously and therefore parametrize an S2 .

The system (5.3) is the familiar small resolution of the conifold singularity and is
topologically X = O(1) O(1) bundle over S2 . When zi = 0 i (5.3) defines an entire
S2 and therefore resolves the conical singularity at the apex while when any zi = 0 (5.3)
S3 but
S3 is
determines a unique point in S2 . X is asymptotically conical with base S2
contractible. With this description of X it is easy to identify the Z2 involution we were

after, it is given by
: zi zi ,
i = 1, . . . , 4,
(5.4)
so that the fixed point set is zi = 0 where the S2 sits. Therefore, the orientifold (5.1) results
in an O5-plane along R1,3 S2 .
We note for future reference that the singularity (5.2) has an alternative desingularization, in which the singularity is deformed as
z1 z2 z3 z4 = V .
(5.5)
The action of on this geometry is now free since the fixed point set zi = 0 is not part of
the geometry, so no orientifold plane appears in this case.
The Types of gauge groups that one can obtain by performing this orientifold are given
in Table 2. Likewise, the total five-brane chargethat of the five-branes plus that of the
O5-planeis given in Table 3.
Just like in the discussion in Section 3 the corresponding string duals for these
other gauge groups are obtained by implementing the orientifold action above on the
supergravity solution described in the last section. Therefore, the string dual is given by an
orientifold by of Type IIB on R1,3 M6 . M6 is topologically like the deformed conifold
26 In the previous section we described the solution with NS5-branes but can obtain the corresponding one for
D5-branes by performing an S-duality.
194
Table 2
Gauge group on 2N five-branes coincident with an O5-plane
O5
O5+

O5
SO(2N )
Sp(N )
SO(2N + 1)
Table 3
Five-brane charge of a configuration of 2N branes on top of an O5-plane
G
Q
O5
O5+

O5
SO(2N )
2N 2
Sp(N )
2N + 2
SO(2N + 1)
2N 1
geometry of (5.5) so that in this case the action of is free and one obtains a dual given by
unoriented closed string theory on the regular geometry R1,3 M6 /Z2 , without any open
strings nor orientifold planes. In this case the action of is such that the non-contractible
3.
S3 inside M7 is replaced27 by RP
3
3 in M6 one must take into account the total five-brane
Apart from replacing S RP

charge. Just like in Section 3 the total charge due to the branes and the O5-plane that we
3 of M6 so that we can replace
have on X is replaced by flux over the non-contractible RP
in the solution found in [2] N Q, where Q is the total five-brane charge given in Table 3.
We can understand the expected pattern of chiral symmetry breaking for these gauge
theories by studying the symmetries of the string solution in the UV and in the IR. In
3 which makes the
the IR, the metric looks the same modulo the replacement
S3 RP
+ 2 . Just as in the previous section, the symmetry

coordinate 2 periodic, that is
+ 2 . This shift acts as a trivial rotation in

in the IR is just that given by shifts by 2
space-time but since the Killing spinors of the supergravity solution are unmodified by the
generates a Z2 transformation on the Killing

orientifold projection, a rotation by 2 on
spinors which is the expected unbroken symmetry group after gluino condensation.
What is the symmetry in the UV? The symmetry of the metric is again U (1)R acting
. We must analyze the effect on the U (1)R symmetry of a SYM

by arbitrary shifts on
instanton in the string theory realization. This can be accomplished by taking into account
the way a SYM instanton is represented in terms of branes in string theory. Care must be
taken in performing this identification since as we discussed in Section 3 there are subtleties
in the relation between SYM instantons and D-branes for unoriented open strings. We
showed that for SO(N) gauge groups, via WessZumino coupling (3.5), that a D1-string
on the D5-brane worldvolume corresponds to a single SO(N) SYM instanton. Therefore,
the SYM instanton is represented by an Euclidean D1-string wrapping the S2 M6 /Z2
that we described in the previous section. The amplitude for such a wrapped D1-string is
27 This follows from the following observation. The real section of Eq. (5.5) describes
S3 so that the effect of

3.
3 . Topologically, the space is T RP
the Z2 transformation zi zi is to replace it by RP
195
given by

v = exp i B .
(5.6)
S2
By using the asymptotic expression for H in (4.5) and the required replacement of N Q
one finds that
(5.7)
B = 2Q.
S2
that act non-trivially on the Killing spinors are given

Therefore, the discrete rotations on
+ 2k/Q with k = 1, . . . , 2Q. By using Tables 3 and 1 we see that from the
by
string dual that the UV symmetry of the solution is precisely Z2h as required by the field
theory analysis.
For the Sp(N) case there is an interesting subtlety. As we discussed in Section 3 the
relation between brane charge and instanton number is modified for Sp(N) gauge theories.
For an Sp(N) gauge theory on a collection of D5-branes we established that a D1-string
dissolved on the D5-branes actually corresponds to two instantons in the Sp(N) gauge
theory. Therefore, the amplitude of a single SYM instanton is given by v 1/2 where v is
that leave v 1/2 invariant and act non-trivially

given by (5.6). The discrete rotations on
+ 2k/Q with k = 1, . . . , Q. By taking the

on the Killing spinors are given by
expression for Q in Table 3 corresponding to the Sp(N) case we see that we exactly
reproduce the expected Z2h symmetry as can be seen from Table 1.
Therefore, we have realized the expected pattern of chiral symmetry breaking for SO
gauge groups Z2h Z2 as the breaking of the UV symmetry of the string solution to that
of the IR. The h vacua, just like in the SU(N) case, are obtained by performing the h
allowed discrete rotations on the SU(2) gauge fields Aa of the solution such that the new
solutions are non-singular in the interior but are equivalent in the UV region.
Acknowledgements
We would like to thank A. Brandhuber, E. Diaconescu, A. Kapustin, J. Maldacena and
E. Witten for discussions. The work of J.G. is supported in part by the National Science
Foundation under grant No. PHY99-07949 and by the DOE under grant No. DE-FG0392ER40701.
Appendix A. NS5 branes with SU(2)L gauge fields for 2 + 1 SYM
Here we rederive the solution found by Maldacena and Nastase by obtaining the
first order BPS equations from an effective quantum mechanical Lagrangian derived
from gauged supergravity. The relevant gauged supergravity is seven-dimensional SO(4)
gauged supergravity, which appears by reducing Type IIB supergravity on the near horizon
geometry of a NS5-brane. Since we are interested in finding a supergravity dual to D = 3
196
N = 1 SYM, the supersymmetry of the field theory on the wrapped branes requires
considering an SU(2)L truncation of SO(4) gauged supergravity.28
The ansatz for the solution is determined by the symmetries of the background, which
is given by five-branes wrapping S3 S(S3 ). This requires considering the most general
ansatz invariant under the SO(4) isometries of S3 . In the string frame, the ansatz is therefore

e2h(r)
1 + b(r)
wa wa ,
wa ,
ds 2 = ds 2 R1,2 + dr 2 +
Aa =
4
2
where wa , a = 1, 2, 3 are a set of SU(2) left invariant one-forms satisfying
(A.1)
1
dwa = :abc wb wc .
(A.2)
2
The three-form field strength h of SU(2)L gauged supergravity satisfies the following
Bianchi identify29
1
Fa Fa .
8
The SO(4) symmetry together with
dh =
(A.3)

b
1
Fa = dr wa + b 2 1 :abc wb wc
2
8
(A.4)
determine the three-form from (A.3)30

1 3
b 3b + 2 w1 w2 w3 .
h=
32
Plugging the ansatz into the bosonic SU(2)L gauged supergravity Lagrangian

1 a 2
2
h h
+4
R + 4 F Fa
L = ge
8
3
(A.5)
(A.6)
one obtains the following effective quantum-mechanical system

Leff = e2 (T V ),
(A.7)
where
3
T = 4 2 3h 2 e2h b 2,
4
2 3
2

1
V = e6h b 3 3b + 2 + e4h b2 1 6e2h 4,
8
4
(A.8)
with
3
= h ,
2
d
,....
dr
(A.9)
28 One can also have an SU(2) = (SU(2) SU(2) )

D
L
R diag truncation. Then, the solution has double the
amount of supersymmetry. This truncation was used in [20,32] to find the supergravity dual to D = 3 N = 2
SYM.
29 We are writing the action such that the flux of h/2 is quantized in integral units.
30 The constant piece is only fixed by requiring the solution we will find to be regular.
197
The SU(2)L gauged supergravity equations of motion follow from (A.7) together with
the Hamiltonian constraint which follows from having fixed the radial reparametrization
invariance of the ansatz (A.1)
H = T + V = 0.
(A.10)
One can find a set of first order BPS equations whose solutions also solve the equations
of motion if the potential of the effective quantum mechanics can be written in terms of a
superpotential as follows
1
1
1
(h W )2 + e2h (b W )2 W 2 .
12
3
4
The required superpotential is given by
W = 6eh M,
V=
where
M=

1 2h 3
e
b 3b + 2 b
12
2
(A.11)
(A.12)

2 2
4
1
+ e2h b2 1 b 2 1 + e2h .
4
3
9
(A.13)
The BPS equations following from this superpotential are given
by31
3
1
= W = eh M,
4
2

1
2
2
b = e2h b W = eh e4h b3 3b + 2 1 b2
3
8
3 M

+ 1 b3 e2h 2b ,

2
1
(b3 3b + 2)2 e2h 2
eh
+
h = h W =
e4h
3 b 1
6
16
2
3 M

4

2 b 3b2 + 2b + b2 + 2 .
(A.14)
The asymptotics of the solution to these equations can be found in [1].

This gauged supergravity solution can then be lifted to a solution of Type IIB
supergravity using the formulae in [33]. In the string frame the Type IIB solution is

1,2
1
1 2h
2
2
2
ds = ds R
+ N dr + e wa wa + N (w a Aa )2 ,
4
4

1
N
1
(
w1 A1 ) (
H =h+
w2 A2 ) (
w3 A3 ) + Fa (
wa Aa ) ,
2
4
4
(A.15)
31 There is a typo in Appendix of [1], the second equation in (4.4) should an extra multiplicative factor of 1/2
in the second term.
198
where w
a , a = 1, 2, 3 are another set of SU(2) left invariant one forms parameterizing
S3 .
1,2
This metric is defined on R M7 where M7 is parametrized by r, wa and w
a . wa
and w
a parameterize, respectively, S3 and
S3 . As described in the text M7 is asymptotically

topologically a cone over S3
S3 but S3 is contractible in the full geometry while
S3 is
topologically non-trivial.
References
[1] J. Maldacena, H. Nastase, The supergravity dual of a theory with dynamical supersymmetry breaking, hepth/0105049.
[2] J.M. Maldacena, C. Nunez, Towards the large N limit of pure N = 1 super YangMills, Phys. Rev. Lett. 86
(2001) 588, hep-th/0008001.
[3] S. Sinha, C. Vafa, SO and Sp ChernSimons at large N , hep-th/0012136.
[4] E. Witten, Supersymmetric index of three-dimensional gauge theory, hep-th/9903005.
[5] S. Deser, R. Jackiw, S. Templeton, Topologically massive gauge theories, Ann. Phys. 140 (1982) 372;
S. Deser, R. Jackiw, S. Templeton, Ann. Phys. 185 (1982) 406, Erratum;
S. Deser, R. Jackiw, S. Templeton, Ann. Phys. 281 (1988) 409.
[6] L. Alvarez-Gaume, E. Witten, Gravitational anomalies, Nucl. Phys. B 234 (1984) 269.
[7] A.N. Redlich, Parity violation and gauge noninvariance of the effective gauge field action in three
dimensions, Phys. Rev. D 29 (1984) 2366.
[8] H.C. Kao, K.Y. Lee, T. Lee, The ChernSimons coefficient in supersymmetric YangMills ChernSimons
theories, Phys. Lett. B 373 (1996) 94, hep-th/9506170.
[9] E. Witten, Constraints on supersymmetry breaking, Nucl. Phys. B 202 (1982) 253.
[10] T. Kitao, K. Ohta, N. Ohta, Three-dimensional gauge dynamics from brane configurations with (p, q)fivebranes, Nucl. Phys. B 539 (1999) 79106, hep-th/9808111;
K. Ohta, Moduli space of vacua of supersymmetric ChernSimons theories and type IIB branes, JHEP 9906
(1999) 025, hep-th/9904118;
B. Lee, H. Lee, N. Ohta, H.S. Yang, MaxwellChernSimons solitons from type IIB string theory, Phys.
Rev. D 60 (1999) 106003, hep-th/9904181;
T. Kitao, N. Ohta, Spectrum of MaxwellChernSimons theory realized on type IIB brane configurations,
Nucl. Phys. B 578 (2000) 215, hep-th/9908006;
K. Ohta, Supersymmetric index and s-rule for type IIB branes, JHEP 9910 (1999) 006, hep-th/9908120.
[11] O. Bergman, A. Hanany, A. Karch, B. Kol, Branes and supersymmetry breaking in 3D gauge theories,
JHEP 9910 (1999) 036, hep-th/9908075.
[12] B.S. Acharya, J.P. Gauntlett, N. Kim, Fivebranes wrapped on associative three-cycles, Phys. Rev. D 63
(2001) 106003, hep-th/0011190.
[13] J. Gomis, D-branes, holonomy and M-theory, Nucl. Phys. B 606 (2001) 3, hep-th/0103115.
[14] R. Hernandez, Branes wrapped on coassociative cycles, hep-th/0106055.
[15] S. Gukov, J. Sparks, M-theory on spin(7) manifolds. I, hep-th/0109025.
[16] D.S. Freed, E. Witten, Anomalies in string theory with D-branes, hep-th/9907189.
[17] A.H. Chamseddine, M.S. Volkov, Non-Abelian vacua in D = 5, N = 4 gauged supergravity, JHEP 0104
(2001) 023, hep-th/0101202.
[18] M. Schvellinger, T.A. Tran, Supergravity duals of gauge field theories from SU(2) U (1) gauged
supergravity in five dimensions, JHEP 0106 (2001) 025, hep-th/0105019.
[19] A. Strominger, Superstrings with torsion, Nucl. Phys. B 274 (1986) 253.
[20] J.P. Gauntlett, N.W. Kim, D. Martelli, D. Waldram, Fivebranes wrapped on SLAG three-cycles and related
geometry, hep-th/0110034.
[21] E. Witten, Supersymmetric index in four-dimensional gauge theories, hep-th/0006010.
[22] C. Vafa, Superstrings and topological strings at large N , hep-th/0008142.
[23] M. Atiyah, J. Maldacena, C. Vafa, An M-theory flop as a large N duality, hep-th/0011256.
199
[24] S.G. Naculich, H.A. Riggs, H.J. Schnitzer, Group-level duality in WZW models and ChernSimons theory,
Phys. Lett. B 246 (1990) 417;
S.G. Naculich, H.J. Schnitzer, Duality relations between SU(N )k and SU(k)N WZW models and their braid
matrices, Phys. Lett. B 244 (1990) 235;
I.B. Frenkel, in: D. Winter (Ed.), Lie Algebras and Related Topics, Lecture Notes in Mathematics, Vol. 93,
Springer, Berlin, 1982, p. 71;
J. Fuchs, P. van Diel, Some symmetries of quantum dimensions, J. Math. Phys. 31 (1990) 1220;
A. Kuniba, T.N. Nakanishi, Level-rank duality in fusion RSOS models, Bombay Quant. Field Theory (1990)
344;
M. Camperi, F. Levstein, G. Zemba, The large N limit of ChernSimons gauge theory, Phys. Lett. B 247
(1990) 549.
[25] E. Witten, Baryons and branes in anti-de-Sitter space, JHEP 9807 (1998) 006, hep-th/9805112.
[26] M. Atiyah, E. Witten, M-theory dynamics on a manifold of G(2) holonomy, hep-th/0107177.
[27] B.S. Acharya, On realising N = 1 super YangMills in M-theory, hep-th/0011089.
[28] A.H. Chamseddine, M.S. Volkov, Non-Abelian BPS monopoles in N = 4 gauged supergravity, Phys. Rev.
Lett. 79 (1997) 3343, hep-th/9707176.
[29] A.H. Chamseddine, M.S. Volkov, Non-Abelian solitons in N = 4 gauged supergravity and leading order
string theory, Phys. Rev. D 57 (1998) 6242, hep-th/9711181.
[30] G. Papadopoulos, A.A. Tseytlin, Complex geometry of conifolds and 5-brane wrapped on 2-sphere, Class.
Quantum Grav. 18 (2001) 1333, hep-th/0012034.
[31] I.R. Klebanov, M.J. Strassler, Supergravity and a confining gauge theory: duality cascades and chiSBresolution of naked singularities, JHEP 0008 (2000) 052, hep-th/0007191.
[32] J. Gomis, J.G. Russo, D = 2 + 1 N = 2 YangMills theory from wrapped branes, hep-th/0109177.
[33] M. Cvetic, H. Lu, C.N. Pope, Consistent KaluzaKlein sphere reductions, Phys. Rev. D 62 (2000) 064028,
hep-th/0003286.

The enhanon mechanism for fractional branes

Paolo Merlatti
Dipartimento di Fisica Teorica, Universit di Torino, and I.N.F.N., sezione di Torino, via P. Giuria 1,
I-10125 Torino, Italy
Received 30 November 2001; accepted 17 December 2001
Abstract
We study the enhanon mechanism for fractional D-branes in conifold and orbifold backgrounds
and show how it can resolve the repulson singularity of these geometries. In particular, we show
that the consistency of the excision process requires that the interior space be not empty. In the
orbifold case, we exploit the boundary state formalism to obtain an explicit conformal description
and emphasize the non-trivial role of the volume of the internal manifold. 2002 Elsevier Science
B.V. All rights reserved.
PACS: 11.25.-w; 12.10.-g; 11.25.Sq
1. Introduction
In the context of the gauge/gravity duality and to better understand the non-perturbative
dualities between different string theories, a lot of effort has been recently devoted to the
study of fractional D-branes on conifold and orbifold singularities. Indeed the introduction
of fractional branes breaks conformal invariance and introduces RG flow [13], thus
describing more realistic gauge theories [57,14,15].
It is well known that we lack of an explicit description, in terms of two-dimensional
conformal field theory, of ordinary and fractional Dp-branes on a conifold background;
therefore many techniques such as the boundary state formalism or the brane-probe
computations, cannot be applied in this case. Obviously it is still possible to construct
classical supergravity solutions, describing space-filling branes in such a background.
However those solutions are singular [2] at a certain small value of the radial coordinate
of the cone (in the IR of the dual gauge theory).
The appearance of naked singularities in the gravitational dual of non-conformal gauge
theories with reduced supersymmetry is a rather common phenomenon. In the past year
Research partially supported by the EC RTN programme HPRN-CT-2000-00131.

E-mail address: paolo.merlatti@to.infn.it (P. Merlatti).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 4 8 - 4
P. Merlatti / Nuclear Physics B 624 (2002) 200218
201
we learned that these naked singularities are potentially related to interesting IR dynamics
of their dual gauge theories. Thus, understanding the physics of these singularities is an
important problem because it could teach us about non-perturbative effects in the gauge
theory.
From a supergravity point of view, we can think that, at the typically short distances
where these singularities occur, some stringy mechanism comes into play, so that
supergravity solutions cease to make sense. However, in this paper we show that, rather
surprisingly, pure supergravity tools let us to construct non-singular solutions also for these
conifold spaces, with the same techniques used in [3] for the case of a K3 background.
Well find that its possible to have a well defined solution if the fractional branes, instead
of being placed at the origin, form a composite shell, extending along some directions of
the base of the cone, at a radius r r , where r is a small fixed radius we will determine in
each case we will discuss. This is very similar to what happens in the enhanon mechanism
[1,3]. Indeed, we think that also in the case of the conifold background an enhanon
mechanism takes place. In this case the base of the cone at r = r defines the enhanon
locus and we argue that the fractional branes become massless there.
We will show these results in the case of fractional Dp-branes with p 3. We will
particularly emphasize the p = 3 case because of its relevance on the gauge theory side
[57,911]. This case has already been extensively studied from different perspectives:
in [6], Klebanov and Tseytlin found a solution that exhibits a singularity in the IR. In
a subsequent paper [7] Klebanov and Strassler, getting insights from the strong dynamic
of the infrared gauge theory, argued that the naked singularity of the KlebanovTseytlin
(KT) geometry should be resolved via the deformation of the conifold. In [9] instead
this procedure has been reversed, trying to make predictions on the field theory using
string theory computations. It was suggested that a non-extremal generalization of the KT
solution may have a regular Schwarzschild horizon cloaking the naked singularity. Nonextremal supergravity solutions translate in field theories at finite temperature [12]. In this
case, the dual field theory interpretation of this has been then supposed to be the restoration
of chiral symmetry at a finite temperature TC [9].
From the supergravity point of view, we are then in a particular situation: at zero
temperature the theory exhibits chiral symmetry breaking [7], while for temperature above
TC we have chiral symmetry restoration. So, the regular Schwarzschild horizon should
appear only for some finite Hawking temperature. As noticed in [911], one implication
is that, at temperatures below TC , there should be non-extremal generalizations of the
KlebanovStrassler solution which are free of horizons, just like the extremal solutions.
Particularly, the analysis of [911] suggests that gravitational backgrounds which exhibit
regular Schwarzschild horizon only above some critical non-extremality, should be quite
generic for backgrounds dual to gauge theories which undergo finite temperature symmetry
breaking phase transition. Nevertheless, they are unusual and, to our knowledge, new from
the supergravity point of view.
We argue that supergravity backgrounds of that type have to involve the enhanon
mechanism in non-extremal backgrounds, so that a combination of the enhanon shell
effect and of the horizon of the black hole, determines the appearance of regular
202
Schwarzschild horizons only in particular range of the parameters of the solutions. We

leave anyway this issue for future investigation.
We expect similar infrared phenomena to take place in the fractional D0-, D1-, and D2brane cases. Easily generalizing the observation and the computation of [7,8] we can say
that there are two natural ways of smoothing out the singularity at the apex of the conifold:
one can substitute the apex by a (6 p) non-trivial cycle (deformation) or by a 2-cycle
(resolution). We expect anyway that the resolution still displays a naked singularity (as it
happens for the p = 3 case [8]), that can be excised by the enhanon mechanism. Indeed,
as pointed out in [7] for the p = 3 case, the source of the singularity can be traced back
to the infinite energy of the F6p (magnetic) field. At all radii there are M units of flux of
2
F6p along the (6 p) cycle, and when that cycle shrinks to zero size, this causes F6p
to diverge. Since resolution does not affect the (6 p)-cycle, we can think naively that, if
we resolve the conifold, the solution remains singular. However, the enhanon mechanism
2
from diverging, so its possible, as we will show, to have a non-singular
prevents F6p
solution also in the case of resolution. Instead, deforming the conifold does mean exactly
that the (6 p) cycle does not shrink, so the singularity can be avoided (as it happens in
the deformed conifold solution of [7]).
In this paper we will analyze the enhanon mechanism on the conifold with the aim of
checking if its consistent with the physical expectations discussed above. We will perform
this analysis using exclusively supergravity tools. However, the enhanon mechanism in
the presence of fractional branes occurs also in the case of orbifold backgrounds [1,3,
14,16]. These are more tractable examples since we have an exact stringy description of
these geometries. We will then concentrate on those backgrounds, where we can compare
the supergravity solutions with some new insights we can have from the boundary-state
formalism. Thus, we will construct, in Appendix A, the boundary state appropriate to
describe fractional branes.
The analysis of how supergravity results perfectly match with stringy computations
in the orbifold case, corroborates, in our opinion, the interpretation we propose also
for conifold backgrounds. In particular, in the orbifold case, we will see that the
NSNS twisted sector contributes to the definition of what has to be interpreted as
the effective mass of a fractional Dp-brane. This interpretation will be supported by
different calculations, involving the boundary-state techniques as well as the supergravity
description of branes. From both these types of computations, we shall find that, as in
the case of branes wrapped around K3 [1,3], the volume of the internal manifold plays
a fundamental role. In particular, we will then show that when the volume reaches a critical
value, the fractional branes become massless.
Finally, we shall find that, in both conifold and orbifold backgrounds, in order to have a
consistent enhanon mechanism, the shell of branes defining the enhanon locus is made
up by fractional branes only and that the presence of some regular Dp-brane in the origin
is necessary.
203
2. The enhanon on the conifold

Type IIA and type IIB fractional Dp-brane solutions in conifold backgrounds have been
extensively studied in [2]. They are warped product of R1,p flat space-time directions and
a Ricci-flat, (9 p)-dimensional cone C9p . Since the brane are space filling, the warpfactor depends only on the radial coordinate of the cone. Because the cone is Ricci-flat, the
base of the cone is an (8 p)-dimensional Einstein manifold X8p with metric hij . Its
assumed this Einstein manifold has a harmonic 2-form 2 , so that wrapping a D(p + 2)brane around the 2-cycle corresponding to 2 , and letting the remaining p + 1 dimensions
fill R1,p , gives origin to a fractional Dp-brane.
We concentrate now on the fractional Dp-branes, with p < 3.
2.1. Fractional Dp-branes, p < 3
As a generic example and to see what it is meant by warped product we reproduce the D0
fractional brane solution found in [2], but our considerations will be easily generalizable
for every p:
e = H (r)3/4,

1/2
dsE2 = gs H 7/8 dt 2 + H 1/8 dr 2 + r 2 hij dx i dx j .
(1)
The non-zero RR field strengths are:

F2 = dt dH 1 ,
1
4 = H Q dt dr 2 ,
F
r4
(2)
4 = F4 C1 H3 , Q gs M, M is the number of fractional Dp-branes and gs is

where F
the string coupling constant.
We want now to express, for more generality, the NSNS two form field (B2 ) and the
warp factor H for p generic (p < 3):
B2 =
Q
r 3p
2
,
p3
H (r) =
r 7p
Q2
,
(3 p)(10 2p)r 102p
(3)
where as usual gs N and N is the number of ordinary Dp-branes. The warp factor has
to satisfy the following equation of motion (it comes from that for the RR field strength
F2 ):

Q2
d
d
r 8p H (r) = 4p .
(4)
dr
dr
r
This is the Poisson equation for the H -potential. It has been integrated in (3) with the
boundary condition that H approaches zero as r , so this solution is valid for small r
(near-horizon limit) and consequently we are studying the gauge theory (according to the
gauge/gravity dual principle) in the infrared.
Note also that the r-dependence of the source term seems to indicate a spatial extension
of the source in the transverse directions. Moreover, from the precise form of the
204
dependence, its possible to argue that it spreads over (4 p) directions of the base of
the cone.
Moreover, as already pointed out by the authors of [2], such solutions possess a naked
singularity at r = r0 , in the IR. This singularity is of repulson type because before
reaching it, the derivative of G00 changes sign at r > r0 (where H (r ) = 0) and
consequently the gravitational force changes its sign.
We want now study the possibility that an excision process takes place, as it has been
shown to happen in [3] in a different context. There, it has been studied this mechanism for
branes wrapped around the K3 surface, but we think that also in the present case it happens
something very similar, giving rise to the enhanon.
Then, in order to avoid the singularity, we can imagine that the Dp-branes are not in the
origin, but they distribute on a surface at some r = ri r . For generic p < 3 we can try to
calculate the stress-energy tensor of these source branes, interpreted as the discontinuity in
the extrinsic curvature (we are making the calculation in the Einstein frame). Following [3]
we define unit normal vectors
1
n =
.
Grr r
The extrinsic curvature of the junction surface, for the two regions (r > ri and r < ri ) is
1
1 GAB
.
K = nc c GAB =
2
2 Grr r
+
The discontinuity in the extrinsic curvature is then defined as AB = KAB

+ KAB
. Finally,
1
the stress-energy tensor supported at the junction is simply:

1
SAB = 2 AB GAB CC ,
(5)
where is the gravitational coupling constant.

Thus, if we think that in the interior region (r < ri ) the space is flat, with easy
calculations we obtain:

1
H
1
Sij = 0.
S = 2
(6)
G ,
2
Grr H
According to the second line, we see that there is no stress in the directions transverse to
the branes, as it has to be since the branes in consideration are BPS.
From the first line we can argue that for ri = r the constituent branes become massless.
To get more information about this configuration we can now expand the result for
large values of the incision radius ri . The coefficients of the metric components in the
longitudinal directions (, ) can be interpreted as proportional to the effective tension of
the brane. So we see that:
(p 7) +
1
1
= 2
(ri ) = 2
2 H
2 1
H
Q2
3p
(3p)ri
Q2
3p
(3p)(102p)ri
1
ri
1 Throughout this paper, the indices A, B run from 0 to 9, and denote the world-volume directions of the
brane, while the indices i, j denote the transverse ones.

1 p7
Q2
1
2
+
.
2
ri
(10 2p) r 4p
205
(7)
We can notice that this expression seems to indicate a configuration that does not reproduce
the behaviour of the source term in the equation of motion (4).
Thus we now imagine a slightly different configuration: we think that all the N ordinary
branes are in the origin and the fractional branes alone form a shell at r = ri . We can now
repeat the computation of the stress-energy tensor for this configuration and we get always
a vanishing contribution in the transverse directions but in the longitudinal ones we have
the following result:
(ri )
Q
1
1 M
.
2 2 N (10 2p) r 4p
(8)
This result is now compatible with the source term which appear in (4). This expression
for the effective tension is due to the presence of M fractional branes and it contains the
small parameter M/N that is typical for these expansions.
We get then the following solution:

Hp (r) = hs (r) + (r ri ) hf (r) hf (ri ) ,
Hp (r) = h s (r) + (r ri )h f (r),
Hp (r) = h s (r) + (r ri )h f (r) + (r ri )h f (r),
(9)
where
hs (r) =
r 7p
and hf (r) =
1
Q2
.
(p 3)(10 2p) r 102p
For the sake of clarity, we report here the modified ansatz and the equations of motion for
the case of a D0 fractional brane, but everything continues to be easily generalizable to
other p:
Q (r ri )
2 ,
p 3 r 3p

4 = Q(r ri ),
4 = H 1 Q(r ri ) dt dr 2 ,
d e/2 F
F
4p
r
2
2

3
4 = Q (r ri ) Q (r ri ) .
d e 2 F2 = gs e/2 H3 F
r 4p
p 3 r 3p
F2 = dt dH 1 ,
B2 =
(10)
With the new solutions all the singular terms (see in particular the last equation) fit
perfectly. They are those which give rise to the junction conditions. One could also study
the dilaton equation of motion to see it has a discontinuity at r = ri (see [3]) but it is not
relevant here.
Thus the scenario seems to be the following one. We have D(p + 2)-branes at r = ri
that are wrapped around the 2-cycle which is Poincar dual to 2 . According to the form
of their effective tension, it seems these objects extend in (4 p) of the directions of
the base of the cone. Moreover, these have a non trivial coupling with the B2 field, that
206
becomes a source for fractional Dp-branes potential for all r ri (notice the form of the
source term in (10)).
This is anyway suggested also by the running of the fluxes:

3
1 Q2
2
d e F2 = +
(r r ),
=
(11)
3 p r 3p
C9p
where =
1 Q2
3p r 3p .
This behaviour seems indeed to indicate that the fractional branes
are present for all r ri and that, at fixed r, they extend in the (4 p) directions of the
base of the cone where the shell at r = ri spreads.
In our discussion we have never fixed ri , but we found a lower bound for it:
r3 =
Q2
,
(7 p)(3 p)
where we argued the branes become tensionless. This is very reminiscent of the enhanon
locus [1], where we know the branes become effectively massless. We think, indeed, it is
now working the same mechanism, that seems typical for fractional D-branes in various
background [1,14,16]. In this case it would be natural to identify the enhanon radius re
with r . It would be nice to find some interpretation of this with the boundary states or other
conformal techniques, but, as already emphasized in the introduction, we do not have these
tools at our disposal in this case. So, in the following section we apply the same techniques
used in the present one to other backgrounds which do have a conformal description and
we will see how the supergravity computations match with the stringy ones. Well get
therefore convinced of the right interpretation of the supergravity results.
Before that, we consider now the p = 3 case. This case is indeed particularly relevant in
the context of the gauge-gravity duality [5] and the formulae are slightly different from the
other cases.
2.2. Fractional D3-brane
In the context of gauge/gravity duality the basic picture common to all RG problems
is that the radial coordinate defines the RG scale of the field theory, hence the scale
dependence of the couplings may be read off from the radial dependence of the
corresponding supergravity fields. Since the RG flows of couplings in physically relevant
gauge theories are logarithmic, an important problem is to find gravity duals of logarithmic
flows. Thus the importance of the case of the fractional D3-branes. Indeed, the solution
describing a collection of N D3-branes and M fractional D3-branes on the conifold was
constructed in [6] (where the base of the cone is the T 1,1 manifold), finding a logarithmic
radial dependence of certain supergravity fields.
The gauge dual of this background is an N = 1 supersymmetric non-conformal gauge
theory. Besides the conical case we are considering now, other supergravity solutions
representing such configurations have been constructed [7,8] also for the deformed
conifold [7] and the resolved conifold [8]. The two solutions, for large values of the radial
207
coordinate, (in the UV from the dual gauge theory perspective), coincide with the conifold
one we examine. However, the small-distance (IR) behaviour is different in each case. In
particular, the conifold and the resolved conifold have naked singularities at finite values
of the radial coordinate, while the deformed conifold solution is regular.
Let us now perform the excision calculation for the conifold solution. We want to notice
that also in the case of the resolved conifold [8] the metric can be written, in term of
a different radial variable [8], in a way which is very similar to the standard conifold case
(at least in the IR, where the solution exhibits its singularity). The excision mechanism can,
therefore, be used to avoid the singularity also in that case.
Now we illustrate it explicitly.
The metric is a warped product

ds 2 = H (r)1/2 dx dx + H (r)1/2 dr 2 + r 2 hij dx i dx j
(12)
5 reads
and the equation of motion (or the Bianchi identity) for the self-dual field-strength F
5 = H3 F3 ,
dF
5 = dC4 + B2 F3 .
where F
If we make the usual ansatz for fractional D3-branes

5 = d 4 x dH 1 + d 4 x dH 1 ,
F3 = Q3 ,
F
Q
H3 = dr 2 ,
r
Eq. (13) becomes
(13)
(14)
Q2
.
r 5H =
(15)
r
We can integrate it with the appropriate boundary condition to study the near-horizon limit
(IR dynamic of the dual gauge theory), and obtain:
r

1
2 ln R
H = 4 +Q
(16)
,
+
r
4r 4
16r 4
where R is a scale of the same order of r . If we repeat for this case the excision calculation,
with the usual procedure, we get a stress-energy tensor at the incision radius ri :

1 Q2 /4r 5
1
G ,
S = 2
(17)
Sij = 0,
2 2 Grr H (r)
where we have already considered the ordinary branes in the interior region of the shell
and the fractional ones forming the shell.
All the considerations we made for the other p continue to be valid. In particular, notice
that the right expansion for (17) seems to be for r R (where R is a scale such that
R r ). 2 We obtain (for M N ):
S
1 MQ1
.
2 2 N 4 ri
2 Also in this case r is defined as H (r ) = 0.
(18)
208
This is again consistent with the equations of motion and it seems to suggest that the shell is
located precisely at r , the radius we argued has to be identified with the enhanon radius.
We have already mentioned that the deformed conifold solution is regular. In that
case the mechanism that removes the singularity is related to the breaking of the chiral
symmetry in the dual SU(N) SU(N + M) gauge theory. The Z2M chiral symmetry,
which may be approximated by U (1) for large M, is broken to Z2 by the deformation of
the conifold [7].
In [9] a different mechanism for resolving the naked singularity was proposed. It
was suggested that a non-extremal generalization of the solution found in [6] may have
a regular Schwarzschild horizon cloaking the naked singularity. The dual field theory
interpretation of this would be the restoration of chiral symmetry at a finite temperature
TC [9]. In [10] the right ansatz and the system of second order radial differential equations
necessary to study such non-extremal solution have been constructed. Finally, in [11],
smooth solutions to this system have been found in a perturbation theory that is valid when
the Hawking temperature of the horizon is very high.
Clearly, it would be very interesting to study this theory as a function of the temperature
and to identify the phase transition at T = TC . We argue that its possible that the excision
mechanism is valid for studying the physics of the enhanon at finite temperature, as it
seems to happen in orbifold backgrounds [1,3,4]. Notice indeed that this solution does not
break chiral symmetry. Particularly, it would be interesting to investigate the possibility
this mechanism is related to the non-extremal solution at T = TC , where chiral symmetry
is restored, but we leave the study of this possibility for future work.
Let us now concentrate on the orbifold background.
3. The enhanon on the orbifold

3.1. Comparison with the conifold
Now we want to examine the supergravity solutions for fractional Dp-branes in an
orbifold background
R1,5 T 4 /Z2 ,
(19)
where T 4 is the four-dimensional torus. For this purpose we mainly refer to [16], where the
solutions have been written down explicitly for p = 0, 2 in the context of the IIA theory.
We report here the solutions, integrating the equation for the warp factor with the boundary
condition that H approaches zero as r , as we did in the previous paragraph:

ds 2 = H (3p)/4 dx dx + H (p+1)/4 dr 2 + r 2 d 4p ,
e = H (1p)/4,
H=
ea = H 1/4,
Q2p
1 Qp
1
V
,
2 r 3p 2 (2)4 2 r 62p
1 V 1/2 Qp
b =
,
2 4 2 r 3p
(20)
209
where
2 2 MTp orb
Qp =
,
(3 p)4p V 1/2
M is the number of fractional branes, orb = 2 /V 1/2 , 4p is the area of a unit (4 p)dimensional sphere, r is the radial coordinate in the (5 p)-dimensional space transverse
to the brane, is the six-dimensional dilaton, b is the twisted NSNS scalar and a are the
four untwisted NSNS scalars:
Gaa = e2a ,
a = 6, . . . , 9,
(21)
and finally G is the ten-dimensional metric in the string-frame. We do not report here
explicitly the form of the RR twisted and untwisted potentials because they are not
relevant in our discussion.
As already pointed out in [16], this solution presents a naked singularity, that is of
repulson type because its located at r0 < r (r is defined so that H (r ) = 0).
From what we know from the enhanon and from what we have learned in the conifold
case, we are now naturally led to imagine that also in this case the fractional branes form
a shell at r = ri r . Analogously to the computation made in the previous paragraph, we
can then calculate the stress-energy tensor for this configuration at the junction surface. We
get:

1
1
H
S (ri ) = 2
(22)
Sij = 0.
G ,
2orb Grr H
The vanishing of the stress energy in the transverse directions again indicates that the
system is BPS and because the stress-energy tensor in the longitudinal directions is
proportional to H , its again natural to argue that at r the constituent branes become
massless. The advantage we have in this case, its that this interpretation is now strongly
supported by a direct probe calculation (made in [16]). Indeed in the orbifold background
we are considering, we have a stringy description that enables us to construct the worldvolume action describing the interaction of the D-brane with the fields of the spectrum of
the theory. For the background (19), this has been written in [16] and the part of interest
for us (obtained expanding the DBI part of the world-volume action in the velocities and
keeping only the lowest order terms), is:

Tp
x i x j
2V
p+1
ij
d
b .
1+

2 2
2V korb
We see that the interaction term of the brane with the NSNS twisted field seems to add
a contribution to the definition of the tension of the brane.
We can interpret it as if both the NSNS sector (with a constant contribution) and the
NSNS twisted one (with an r-dependent term) contribute to the definition of the running
tension of the brane. Then, it is easily shown [16] that the enhanon radius re , defined as
the radius where the running tension of the brane goes to zero, coincides with r .
210
Comforted by these results, we consider now the stress-energy tensor (22) and, as in the
conifold case, we expand it for large ri . We find again a configuration not consistent with
the equations of motion.
So, we can now imagine of making analogous considerations for a configuration in
which we add N ordinary Dp-branes in the origin (with N M). We assume furthermore
that the M fractional Dp-branes alone form the composite shell. In this case, performing
an excision calculation,we get:
S
1 M (3 p)Q0 V 1
2 2 N (2)4 r 4p
i
(23)
and this is consistent with the equations of motions.

Then we notice that, in order to have a consistent enanhon mechanism, excision
computations tell us we have to insert ordinary branes in the interior region, both in the
conifold and in the orbifold case.
Moreover, it seems again the branes smear in (4 p) directions. Being the space-time
(19) six-dimensional, these directions are naturally interpreted as the angular space-time
dimensions transverse to the brane. However its interesting to note this is the same number
(4 p) that appears in the conifold case, where we have no this interpretation.
Finally, we want to emphasize that this solution shares a particular feature with the
enhanon solution found in [1], that is the r-dependent compactification volume.
Indeed, from (20) and (21) we see that the volume of the orbifold, as measured in the
string-frame, is:

V (r)
=V
ea = VH (r),
2
a
(24)
where the factor 2 is typical for orbifold compactifications. We see it depends on r and that
at r = re it takes the value:

4
V (re ) = 2 .
(25)
3.2. Effective mass and the twisted sectors
We want now to use the description of fractional branes by means of the boundary state
(see Appendix A). In this paragraph, for the sake of clarity, we will consider branes that
do not extend in the compact dimensions. The aim of this analysis is to extract information
about the vanishing of the tension that seems to occur at a particular value of the space-time
radial coordinate (r = re ) and for the value of the compactification volume (25).
To get information about the mass of a brane, it is sufficient to study the long-distance
behavior of the p-brane solution. This can be obtained directly from the boundary state.
Indeed, as shown in [18], it is possible to derive the most singular term in k (the transverse
impulse) of the classical solution by projecting the boundary state onto the level-one states
with the closed string propagator inserted in between:
J (k) = 0; 0|V NSNS (k; z, z )P |BNS ,
(26)
211
where P is the closed string propagator, V NSNS (k; z, z ) is the vertex operator producing
the level-one states and |B is the boundary state (see Appendix A). The symmetric
irreducible component of the current gives the long-distance behavior of the graviton, from
which we can easily extract the (density of) mass M of the brane. Indeed, when saturated
with the graviton polarization (h ), we expect it has the following form:
1
Vp+1 Tr(h L),
k2
where the longitudinal matrix L is the projector on the world-volume directions of the
brane.
We can follow the same procedure for the fractional branes. The difference is that the
normalization constant of the boundary state is now different in the NSNS untwisted
sector. We then obtain:
Tp
Tp 1
Vp+1 Tr(h L), and then M = .
Tr(hJ ) =
(27)
2
2V k
2V
Tr(hJ ) = M
But in the case of fractional branes we think this is not the end of the story. Indeed we
have that, even if the linear coupling between the NSNS twisted sector and the graviton is
vanishing, this sector provides a two-point amplitude between the graviton and the NSNS
twisted scalar, which is different from zero (see Fig. 1). This amplitude is given by
0; 0|V NSNS T (k; z2, z 2 )V NSNS (k; z1, z 1 )P |BNS .
(28)
The main point is that the NSNS twisted sector, even if not linearly, couples to the
graviton. This suggests it can participate in some way at the definition of mass, in
agreement with the probe calculation. In [17] the amplitude (28) has been calculated,
obtaining:
= NT ,p Vp+1 Tr(h L)b,

A(h, b)
(29)
where NT ,p is defined in Appendix A to be:
2 Tp
.
NT ,p =
(4 2 )
We can now extract from (29) the contribution (we call it M) of the NSNS twisted sector
to the (density of) mass of the brane. We argue indeed that (29) can be brought to the form:
Fig. 1. The two-point diagram.
212
1
Vp+1 Tr(h L).
(30)
k2
We have then to identify g1 and g2 (see Fig. 1). At this purpose its useful to note that this
amplitude is different from zero because of the presence of the NSNS twisted scalar in the
closed string propagator. We can then identify g1 with its linear coupling with the brane
and g2 with the strength of the coupling of the graviton with two twisted NSNS scalars.
Thus we have [17]:
g1 = 2 NT ,p and g2 = k6 .
(31)
= g1 g2 M
A(h, b)
in the impulse-space, has the following

Moreover, its easy to see from (20) that the field b,
form:

1
b(p)
= 2 NT ,p 6 2 .
(32)
k
Finally, from (29)(32), we argue that:
M = NT ,p .
(33)
If we then consider both the contributes (27)(33) we get a total density of effective mass:
Tp
M=
NT ,p .
2V
(34)
Its now easy to see, from the definition of Tp and NT ,p (see Appendix A), it vanishes
precisely at the value (25) of the internal volume.
We can then conclude that the vanishing of the (effective) mass of a Dp-brane is
strictly linked to the presence of twisted sectors and, particularly, to a precise value of
the compactification volume. These ingredients are all encoded in the boundary state. In
principle one could then expect to obtain this result by direct computations involving only
branes and their description via the boundary state. The straightforward calculus we can
make its the tree-level closed string amplitude:
B|P |B.
(35)
We make this computation and comment it in the Appendix A, where we also construct the
generic boundary state describing fractional wrapped branes.
4. Conclusions
In this paper we have shown how the enhanon mechanism can resolve the repulson
singularities both in conifold and orbifold backgrounds.
In the case of the conifold there are no conformal techniques at our disposal but,
uniquely by means of supergravity tools, we have shown how the enhanon mechanism
seems to take place and resolve the singularities. We have then studied the enhanon
mechanism by means of the same supergravity tools in the orbifold background, where
we have in hand also conformal techniques. We have found a perfect matching between
213
the two computations, getting therefore persuaded of the correctness of the supergravity
computations and of their interpretation in the case of the conifold.
In the conifold background we have examined the solutions for Dp-branes with p 3.
We have shown how, for all those cases, the excision computation is consistent with the
enhanon mechanism. We have paid particular attention to the case p = 3, because of his
relevance for the dual gauge theory. In that case, this mechanism is good at avoiding the
singularity also in the case of the resolved conifold [8] and, particularly, we expect it to
be good at describing the non-extremal solutions at temperatures above chiral symmetry
breaking temperature (TC ) in the gauge theory. We leave this issue for future investigation.
In the orbifold background, we have emphasized the role of the twisted NSNS sector
in contributing to the definition of an effective mass for a fractional brane. We have shown
how the vanishing of this effective mass is related to the volume of the internal manifold.
This relation is supported in Appendix A by the computation of the three level closed string
amplitude between two Dp-branes. This calculation is based on the open/closed string
duality. We show that a convenient choice of the sign of the orbifold projection on the
ground state of the open string leads to determine the value of the internal volume at which
the brane becomes effectively massless. The opposite choice leads instead to suppose the
existence of non-BPS truncated Dp-branes that are stable for particular values of their
moduli. We make that calculation for generic wrapped branes.
Finally, we have shown that, in both the conifold and orbifold backgrounds, consistency
of the excision process requires that the interior space is non-empty.
Acknowledgements
I would like to thank A. Lerda and G. Sabella for very useful discussions. I am also
grateful to M. Bertolini and M. Bill for enlightening comments on a preliminary version
of the paper.
Appendix A. Construction of the boundary state for wrapped fractional branes

We want to derive the structure and the normalizations of the boundary state in the
orbifold background
R1,5 T 4 /Z2 ,
(A.1)
where Z2 is the reflection parity that changes the sign to the four coordinates of the torus
T 4 . The Dp-brane is intended to have r + 1 longitudinal space-time directions and s that
extend in the compact manifold (p = r + s). In an orbifold background like (A.1) the
boundary state has four different components which correspond to the (usual) NSNS and
RR untwisted sectors and to the NSNS and RR twisted sectors [19,21,22,24].
In each sector of the theory we can construct the boundary state:

1

i
|B, k, = e exp
S l + i
m S m |B, k, (0) , (A.2)
l l
l>0
m>0
214
where l and m are integer or half-integer depending on the sector, k denotes the momentum
of the ground state and is a phase equal to in the untwisted RR and twisted NSNS
sectors, to 3/2 in the untwisted NSNS sector and to 0 in the twisted RR sectors.
The parameter = 1 describes the two different spin structures [25,26], and the matrix
S encodes the boundary conditions of the Dp-brane which we shall always take to be
diagonal. For an exhaustive review on this subject see [20].
We omit always the ghost and superghost part of the boundary states, which are as in the
usual case [23,27]. Finally, we want to stress that in the two untwisted sectors, the ground
state is in addition characterized by a winding number ni for each compact direction that
its tangential to the world-volume of the brane.
In order to obtain a localized Dp-brane (say, in y = 0), we have to take the Fourier
transform of the above boundary state, where we integrate over the directions transverse to
the brane. For the untwisted sectors we obtain:

9s
i ni q
1
Ri
|B, y = 0, =
(A.3)
e
dk 5r eik q |B, k, ,
(2)5r
n
i=6
where with Rl we indicate the value of the internal radii of the dimensions where the brane
extends, while with Ri we refer to the transverse ones and, according to our discussion,
this localized bound state also contains a sum over the winding states.
For the twisted sectors we get:

1
dk 5r eik q |B, k, .
|B, y = 0, =
(A.4)
(2)5r
The invariance of the boundary state under the GSO and the orbifold projection
always requires that the physical boundary state is a linear combination of the two states
corresponding to = . In the conventions of [23,27] these linear combinations are of the
form:

1
|B, +NS,U |B, NS,U ,
2

1
|BR,U = |B, +R,U + |B, R,U ,
2

1
|BNS,T = |B, +NS,T + |B, NS,T ,
2

1
|BR,T = |B, +R,T + |B, R,T .
2
|BNS,U =
(A.5)
(A.6)
(A.7)
(A.8)
A fractional Dp-brane state can be written as:

|Dp = NU,p |BNS,U |BR,U NT ,p |BNS,T |BR,T .
(A.9)
Now we want to determine the constants NT ,U solving the openclosed consistency

condition. We have to compare the closed string cylinder diagram with the open string
one-loop diagram. We report the operator structure of this comparison in (A.12)(A.15),
215
while for the normalizations we obtain:

s
1
1
l=1 (2Rl )
NU,p = Tp ns
,
n
(2R
)
2 2
i
i=1

1r
NT ,p = 2s/2 2 4 2 2 ,
(A.10)
(A.11)
where
Tp =
2 3p
4 2
and is the self-dual volume.

NS NS + |elHc |NS NS + = NS NS |elHc |NS NS
8

f3
=
= TrNS e2t Ho ,
f1
NS NS + |elHc |NS NS = NS NS |elHc |NS NS + =
RR + |e
lHc
|RR + = RR |e
lHc
f4
|RR =
f1
8
(A.12)

f2
f1
8

= TrR e2t Ho ,

= TrNS e2t Ho ()F ,
RR + |elHc |RR = RR |elHc |RR + = 0,

NS NS; T + |elHc |NS NS; T + = NS NS; T |elHc |NS NS; T

f4 f3 4
=
= TrNS e2t Ho g ,
f1 f2
(A.13)
NS NS; T + |elHc |NS NS; T = NS NS; T |elHc |NS NS; T +

= 0 = TrR e2t Ho g ,
RR; T + |elHc |RR; T + = RR; T |elHc |RR; T

f3 f4 4
=
= TrNS e2t Ho g ()F ,
f1 f2
RR; T + |elHc |RR; T = RR; T |elHc |RR; T +

= 0 = TrR e2t Ho g ()F .
(A.14)
(A.15)
The definition of the function fi appearing here is as in [20]. Notice that the identification
of the twisted sectors is due to the relations (A.13) and (A.14). Examining closely these
relation, we see that this identification is very sensitive to the sign of the g-orbifold
projector on the ground state of the open-string and, moreover, this sign can always be
absorbed in the global normalization of the twisted sectors of the boundary state. This
ambiguity is characteristic only of the Z2 orbifold.
Now, we assume to make the opposite choice for the identification of the twisted sectors
(on the closed side of the openclosed string duality we exchange the RT sector with the
NST one). Then the tree level closed string amplitude
B|P |B
(A.16)
216
continues to be zero, as it has to be. However it happens something particular at the value
(25) of the internal volume: the contributions from the twisted and untwisted NSNS
sectors cancel between themselves. So, if we interpret the NSNS T-sector as contributing
to the definition of an effective mass for the brane, as argued in Section 3.2, we can say that
at that particular value of the internal volume (25) the fractional branes become massless,
and this has been shown to be consistent with the enhanon mechanism.
The other choice of the sign, made frequently in the literature [28], led instead to suppose
the existence of a non-BPS D(r, s)-truncated branes (for s = r = 0 you have the famous
non-BPS D-particle [29]), that are stable for the value (25) of the internal volume.
We report here, for generic r and s, the calculation that determines the critical volume:

s

s

l=1 (2Rl )
2
2

exp R tnl /2
B|P |B = const 4s
i=1 (2Ri ) l=1 nl
ns

f3 (q)8 f4 (q)8

2
2
exp tnl /2R

f1 (q)8
l=1 nl

s2 f2 (q)4 f3 (q)4
,
24 4 2
f1 (q)4 f4 (q)4
where q = et . Using the relations

2
et l = f1 et f32 et ,

f4 f3 f2 et = 2,
we get that this amplitude is zero for the following values of the radii:

,
Ri = 2 .
Rl =
(A.17)
2
Note that this result is consistent with T-duality.
We see that for s = 0 we get precisely the value (25) for the critical volume and that it
slightly changes for s = 0.
References
[1] C.V. Johnson, A.W. Peet, J. Polchinski, Gauge theory and the excision of repulson singularities,
Phys. Rev. D 61 (2000) 086001, hep-th/9911161;
J. Polchinski, N = 2 gauge-gravity duals, Int. J. Mod. Phys. A 16 (2001) 707, hep-th/0011193.
[2] C.P. Herzog, I.R. Klebanov, Gravity duals of fractional branes in various dimensions, Phys.
Rev. D 63 (2001) 126005, hep-th/0101020.
[3] C.V. Johnson, R.C. Myers, A.W. Peet, S.F. Ross, The enhancon and the consistency of excision,
hep-th/0105077.
[4] C.V. Johnson, R.C. Myers, The enhancon, black holes, and the second law, hep-th/0105159.
[5] I.R. Klebanov, N.A. Nekrasov, Gravity duals of fractional branes and logarithmic RG flow,
Nucl. Phys. B 574 (2000) 263, hep-th/9911096;
L.A. Pando Zayas, A.A. Tseytlin, 3-branes on spaces with R S 2 S 3 topology, Phys.
Rev. D 63 (2001) 086006, hep-th/0101043;
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
217
J.M. Maldacena, C. Nunez, Towards the large N limit of pure N = 1 super-YangMills, Phys.
Rev. Lett. 86 (2001) 588, hep-th/0008001.
I.R. Klebanov, A.A. Tseytlin, Gravity duals of supersymmetric SU(N) SU(N + M) gauge
theories, Nucl. Phys. B 578 (2000) 123, hep-th/0002159.
I.R. Klebanov, M.J. Strassler, Supergravity and a confining gauge theory: duality cascades and
chiSB-resolution of naked singularities, JHEP 0008 (2000) 052, hep-th/0007191.
L.A. Pando Zayas, A.A. Tseytlin, 3-branes on resolved conifold, JHEP 0011 (2000) 028, hepth/0010088.
A. Buchel, Finite temperature resolution of the KlebanovTseytlin singularity, Nucl. Phys.
B 600 (2001) 219, hep-th/0011146.
A. Buchel, C.P. Herzog, I.R. Klebanov, L. Pando Zayas, A.A. Tseytlin, Non-extremal gravity
duals for fractional D3-branes on the conifold, JHEP 0104 (2001) 033, hep-th/0102105.
S.S. Gubser, C.P. Herzog, I.R. Klebanov, A.A. Tseytlin, Restoration of chiral symmetry:
a supergravity perspective, JHEP 0105 (2001) 028, hep-th/0102172.
S.S. Gubser, I.R. Klebanov, A.W. Peet, Entropy and temperature of black 3-branes, Phys. Rev.
D 54 (1996) 3915, hep-th/9602135;
E. Witten, Anti-de-Sitter space, thermal phase transition, and confinement in gauge theories,
Adv. Theor. Math. Phys. 2 (1998) 505, hep-th/9803131.
O. Aharony, A note on the holographic interpretation of string theory backgrounds with varying
flux, JHEP 0103 (2001) 012, hep-th/0101013;
N. Evans, C.V. Johnson, M. Petrini, The enhancon and N = 2 gauge theory/gravity RG flows,
JHEP 0010 (2000) 022, hep-th/0008081.
M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, R. Marotta, I. Pesando, Fractional D-branes and
their gauge duals, JHEP 0102 (2001) 014, hep-th/0011077.
M. Billo, L. Gallot, A. Liccardo, Classical geometry and gauge duals for fractional branes on
ALE orbifolds, hep-th/0105258;
M. Petrini, R. Russo, A. Zaffaroni, N = 2 gauge theories and systems with fractional branes,
hep-th/0104026;
M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, R. Marotta, N = 2 gauge theories on systems
of fractional D3/D7 branes, hep-th/0107057;
M. Grana, J. Polchinski, Gauge/gravity duals with holomorphic dilaton, hep-th/0106014.
M. Frau, A. Liccardo, R. Musto, The geometry of fractional branes, Nucl. Phys. B 602 (2001)
39, hep-th/0012035.
P. Merlatti, G. Sabella, World-volume action for fractional branes, Nucl. Phys. B 602 (2001)
453, hep-th/0012193.
P. Di Vecchia, M. Frau, I. Pesando, S. Sciuto, A. Lerda, R. Russo, Classical p-branes from
boundary state, Nucl. Phys. B 507 (1997) 259, hep-th/9707068.
M. Bill, B. Craps, F. Roose, Orbifold boundary states from Cardys condition, hep-th/0011060.
P. Di Vecchia, A. Liccardo, D branes in string theory I, hep-th/9912161;
P. Di Vecchia, A. Liccardo, D-branes in string theory II, hep-th/9912275.
M.R. Gaberdiel, J.B. Stefanski, Nucl. Phys. B 578 (2000) 58, hep-th/9910109.
T. Takayanagi, String creation and monodromy from fractional D-branes on ALE spaces,
JHEP 0002 (2000) 040, hep-th/9912157.
P. Di Vecchia, M. Frau, A. Lerda, I. Pesando, R. Russo, S. Sciuto, Nucl. Phys. B 507 (1997)
259276, hep-th/9707068;
P. Di Vecchia, M. Frau, A. Lerda, A. Liccardo, Nucl. Phys. B 565 (2000) 397, hep-th/9906214.
D.E. Diaconescu, J. Gomis, JHEP 0010 (2000) 001, hep-th/9906242.
J. Polchinski, Y. Cai, Nucl. Phys. B 296 (1988) 91.
C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Nucl. Phys. B 308 (1988) 221.
M. Bill, P. Di Vecchia, M. Frau, A. Lerda, I. Pesando, R. Russo, S. Sciuto, Nucl. Phys. B 526
(1998) 199, hep-th/9802088.
218
[28] A. Sen, Stable non-BPS bound states of BPS D-branes, JHEP 9808 (1998) 010, hep-th/9805019;
M.R. Gaberdiel, A. Sen, Non-supersymmetric D-brane configurations with BoseFermi
degenerate open string spectrum, JHEP 9911 (1999) 008, hep-th/9908060;
M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, R. Marotta, R. Russo, Is a classical description
of stable non-BPS D-branes possible?, Nucl. Phys. B 590 (2000) 471, hep-th/0007097;
M.R. Gaberdiel, B.J. Stefanski, Dirichlet branes on orbifolds, Nucl. Phys. B 578 (2000) 58,
hep-th/9910109.
[29] E. Eyras, S. Panda, The spacetime life of a non-BPS D-particle, Nucl. Phys. B 584 (2000) 251,
hep-th/0003033;
O. Bergman, M.R. Gaberdiel, Stable non-BPS D-particles, Phys. Lett. B 441 (1998) 133, hepth/9806155.

Geometry of WZW orientifolds

L.R. Huiszoon a , K. Schalm a,b , A.N. Schellekens a
a NIKHEF Theory Group, P.O. Box 41882, 1009 DB Amsterdam, The Netherlands
b Department of Physics, Columbia University, New York, NY 10027, USA 1
Received 29 November 2001; accepted 3 January 2002
Abstract
We analyze unoriented WessZuminoWitten models from a geometrical point of view. We show
that the geometric interpretation of simple current crosscap states is as centre orientifold planes
localized on conjugacy classes of the group manifold. We determine the locations and dimensions
of these planes for arbitrary simply-connected groups and orbifolds thereof. The dimensions of
the O-planes turn out to be given by the dimensions of symmetric coset manifolds based on
regular embeddings. Furthermore, we give a geometrical interpretation of boundary conjugation in
open unoriented WZW models; it yields D-branes together with their images under the orientifold
projection. To find the agreement between O-planes and crosscap states, we find explicit answers for
lattice extensions of Gaussian sums. These results allow us to express the modular P -matrix, which
is directly related to the crosscap coefficient, in terms of characters of the horizontal subgroup of the
affine Lie algebra. A corollary of this relation is that there exists a formal linear relation between the
modular P - and the modular S-matrix. 2002 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Hf
1. Introduction
There has been much progress in the study of interacting conformal field theories
on open unoriented surfaces [14]. Compared with the closed case, this requires the
introduction of two new objects, boundary and crosscap states, which act as sources
for closed strings. Starting with [5], there has been much interest in the geometrical
interpretation of the first of these two: the boundary states. By extension of the toroidal
results, they are believed to give a stringy description of D-branes [6] in curved
E-mail address: t58@nikhef.nl (A.N. Schellekens).
1 Address since September 1st.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 0 0 5 - 6
220
L.R. Huiszoon et al. / Nuclear Physics B 624 (2002) 219252
backgrounds. Owing to the presence of localized YangMills fields on their surface, their
natural geometric interpretation is as vector-bundles wrapping closed submanifolds or
more precisely coherent sheaves. And owing to the geometric interpretation the abstract
CFT results may be extrapolated away from the special rational point to other areas of the
moduli space of the theory.
In many realistic string compactifications however, the presence of boundary states
leads to tadpoles, and we are forced to include crosscap states in CFT. In geometric
language one needs to absorb the excess charges carried by the D-branes by the
introduction of (some superposition of) orientifold planes [7]. Outside of toroidal or
orbifold compactifications, little is known about the geometry of orientifolds (see [8] for a
first attempt to study O-planes in a curved geometry). It is not clear whether orientifolds
also should be interpreted as independent objects wrapping closed submanifolds. From a
CFT point of view it is a priori not even clear that crosscap states yield properly localized
charge distributions which may be interpreted as (lower-dimensional) orientifold planes.
Known examples of toroidal or orbifold models are limited to special points in the stringy
region of the geometric moduli-space where the target space is flat except for some singular
points. Indeed from a perturbative string-theory point of view orientifolds seem little
more than a bookkeeping device for excess curvature or charge, because they have no
inherent moduli and are strictly rigid objects. On the other hand, non-perturbative studies
do indicate that they may possess some properties which allow them to be seen as bound
states of independent higher-dimensional objects in string theory. An example hereof is
the splitting of O7-planes into D7-branes under splitting of the singularities of F-theory
compactifications on K3 [9].
Here we wish to make a first step towards determining the geometrical characteristics
of crosscap states. Understanding these would allow us to extend unoriented type I
compactifications to regions away from rational points in the moduli space, and make
contact with large volume geometry. We have to be careful, for it has not yet been
established whether this is a truly sensible question to ask. The orientation reversal
projection, characteristic of the crosscap state, may project out some blow-up modes which
are necessary to reach the large volume point of the moduli space. Hence the targetspace geometry, or more precisely its moduli space, of unoriented models is not directly
equivalent to that of its oriented parent. In turn this means that we should also reconsider
the geometry of boundary states in unoriented models, as it may need reinterpretation.
In this article, we will start to address these questions by studying orientifold planes
on group manifolds. Strings on group manifolds are described by WessZuminoWitten
(WZW) models, which are a well-understood class of interacting CFTs. In particular,
the crosscap states which may arise in WZW-models are known. On the other hand, to
consider the group manifold of a WZW-model as target spaces does not always agree with
the conventional interpretation of the target space of string theories. Most importantly, in
standard string scenarios we think of the string moving in a target space with dimension
equal to the central charge of the bosonic CFT. With WZW models, on the other hand, one
thinks of the string living on the group manifold with dimension that of the group itself.
The word geometry depends on the setting in which it is discussed.
We will proceed with this caveat in mind. In Section 2 we will use simple geometrical
(group theoretical) techniques in the Lagrangian description of WZW models to classify
221
the possible WZW orientifold planes by their locations and dimensions. For every element
of the centre of the group, we find a configuration of O-planes that are localized on
conjugacy classes. In Section 3, we review the construction of abstract unoriented CFTs
in general and WZW-models, in particular. We show that in order to determine the
characteristics of WZW crosscap states, we need an expression for the pseudo-modular
P -matrix in terms of characters of the horizontal subalgebra. This specific mathematical
problem we answer in Section 4, whereafter in Section 5 we reach the objective of this
article. We demonstrate that there is indeed a one-to-one correspondence between the
so-called simple current crosscap states [10] of WessZuminoWitten models and the
geometrical O-planes. We also show that the crosscap states of [11] describe configurations
of O-planes on non-simply connected group manifolds (orbifolds). In the final section,
Section 6, we describe the geometry of branes in unoriented WZW models, thereby
completing the earlier study of branes on group manifolds [1215].
The important calculation of Section 4 that expresses the P -matrix of a WZW-model
in terms of characters of the horizontal subalgebra is the heart of this paper. There are at
least two reasons for us to surmise that it may have consequences beyond the immediate
question posed here. In deriving the relation between the P -matrix and characters, one has
to perform mathematically interesting quadratic Gaussian sums on the weight lattices
of the horizontal algebra. These are extensions of one-dimensional Gaussian sums and
d-Cartesian dimensional Kloosterman sums, which play a role in analytic number theory.
Physically interesting is the fact that the relation with characters of G implies a linear
relation between the modular P - and the modular S-matrix for WZW-models, albeit an
analytic continuation of the latter, which we define in Section 4.2. The construction of
open-string theories relies heavily on the properties of both, and if the existence of such a
linear relation were to hold for all abstract CFTs, this could have important consequences,
for instance for tadpole cancellation.
Concerning the notation: a tilde is used for the rightmoving sector, but is often omitted
for the sake of clarity. The affine algebra will be denoted by g and the horizontal
subalgebra by g . By extension all affine quantities are unbarred, for instance weights
and characters , and all horizontal quantities are barred. However, when the formulas are
unambiguous, we will often drop the bars. Lie group characters are denoted by X .
Two final remarks: one, to prevent an overcluttered notation we have normalized the
length squared of the highest root for every group to two: (, ) = 2. Two, in trying
to keep the presentation as simple as possible, we will assume that the group centres are
cyclic. At the end of some sections, we will comment on the generalization to non-cyclic
centres, that occur for Dr , r even, or tensor products of WZW models.
While we were applying the final touches, a paper [34] appeared which also finds some
of the results discussed here.
2. Semiclassical analysis
WZW models, as do free CFTs, have the advantage that there also exists a Lagrangian
description of the theory. String theory on a group manifold G can either be described
by the abstract CFT built on an affine extension g of the algebra g of G or by the Wess
222
ZuminoWitten action [16,17]

2

3
k
k
d 2 x Tr g 1 g +
d 3 y Tr g 1 dg .
S =
16
24
(2.1)
Here g is a field that takes values in the group G and lives on the string worldsheet . The
second term is topological; B is a three-manifold whose boundary is , and the level k has
to be integer. The Lagrangian description emphasizes the geometry and it is, therefore, the
natural place to explore the geometrical consequences of orientation reversal.
The WZW theory possesses the chiral currents [16]
W = g 1 g,
1 ,
= gg
W
(2.2)
whose modes span two commuting copies of the affine Lie algebra g. To construct an
unoriented theory, consider the involutions
n := Rn .
(2.3)
The worldsheet parity transformation interchanges the chiral coordinates and Rn acts
on the coordinates as
Rn : g n g 1 ,
n = 0, . . . , N 1,
(2.4)
where generates the ZN centre of G. n interchanges the chiral currents and is,
therefore, a global invariance of this theory. When we gauge this symmetry, we obtain
orientifold fixed planes at those points of the target space G that are fixed under n [7].
Note that by itself is not a symmetry. This tells us right away that WZW models do not
possess spacetime filling orientifold-planes, analogous to O9-planes in superstring theory.
Instead, we only get lower-dimensional planes at the points g G for which
g = n g 1 .
(2.5)
Note that, if g solves (2.5), so does every point in the set

C(g) = {hgh1 |h G},
(2.6)
so our planes are located at conjugacy classes of G.

To solve (2.5) explicitly, recall that every conjugacy class contains at least one element
of a maximal torus T of G. So choose a Chevalley basis of the rank r algebra g and let H i ,
i = 1, . . . , r, be a basis for the Cartan subalgebra. In this basis their action on the weights
returns the integral Dynkin labels. Then an element of the corresponding maximal torus
can be written as

r

i
ti H .
gt = exp 2i(t, H ) := exp 2i
i=1
In this basis, the centre elements

are given by

n
n = exp 2i
ci H i ,
N
i
(2.7)
223
where C := i ci H i is tabulated in Table 1. The eigenvalue of C modulo N is known as

the conjugacy class of a representation2 of the algebra g . It is constant modulo N within
a representation, which means that C commutes modulo N with all other generators in
that representation. This shows that n indeed belongs to the centre (and, by turning the
argument around, that the elements of the centre of G are in one-to-one correspondence
with the conjugacy classes of g ).
Table 1
Currents, conjugacy classes and the number of O-planes. Column two gives the Dynkin labels of the current
(at level k) that corresponds to the conjugacy class generator in column three according to (3.4). An asterisk
indicates orientifold configurations related to a previous one by an even current, in which case they are just centre
translations of each other
Current J
g
Ar , r even
Ar , r even
Ar , r odd
Ar , r odd
Br , r even
Br , r even
Br , r odd
Br , r odd
Cr
(k, 0, . . . , 0)
(k, 0, . . . , 0)
(k, 0, . . . , 0)
(k, 0, . . . , 0)
(0, . . . , 0, k)
Cr
Dr , r even
# O-planes
1 + 2r
i
1 + 2r
1 + (r1)
2
i=1 iH
i=1 iH
()
1 + (r+1)
2
2 + 2r
Hr
r
2
1 + (r+1)
2
Hr
(r+1)
2
[(r+1)/2]
i=1
1+r
H 2i1
1
3 + 2r
r/21
(k, 0, . . . , 0)
2i1 + (2 r)H r1 rH r
i=1 H
r1
2H
+ 2H r
(0, . . . , 0, k, 0)
H 2i1 rH r1 + (2 r)H r
Dr , r even
(0, . . . , 0, k)
Dr , r even
Dr , r even
Dr , r odd
C
0
r/21
i=1
r
2
1 + r+1
2
(r1)/2
(k, 0, . . . , 0)
H 2i1 + (2 r)H r1 rH r
i=1
r1
2H
+ 2H r
(0, . . . , 0, k, 0)
H 2i1 rH r1 + (2 r)H r
Dr , r odd
(0, . . . , 0, k)
Dr , r odd
Dr , r odd
(r1)/2
i=1
1 + r+1
2 ()
G2
F4
H1 H2 + H4 H5
H4 + H6 + H7
E6
E6
(0, 0, 0, 0, k, 0)
E7
E7
(0, 0, 0, 0, 0, k, 0)
E8
2 Not to be confused with the conjugacy classes for groups used above.
()
()
224
It is now easy to find the solutions to (2.5). An element gt is a fixed point of Rn when
for all i
ui
nci
nci
mod 1 ti =
mod 1.
2ti =
(2.8)
N
2N
2
Here ui are vectors in a basis dual to the Dynkin basis, whose entries are either zero or one
modulo even integers. We can phrase this in a basis-independent way, using that the dual
of the weight space Lw is the coroot lattice L . Eq. (2.8) is a vector equation on a basis of
r simple coroots i , i = 1, . . . , r;
u
nc
nc
mod L t =
with u L mod 2L ,
(2.9)
N
2N
2
with t = i ti i , etc. The minus sign in front of u is chosen for later purposes (to make
contact with (5.5)). So we have found 2r solutions gn,u to (2.5). In general, however, not
all of them lie on different conjugacy classes. A quite non-trivial theorem states that when
two elements of the maximal torus are conjugate, they are related by the action of the Weyl
group W [18]. Now we can count how many and what type of O-planes we get.
Consider first the projection with n = 0 and define
2t =
gu := g0,u = ei(u,H ).
(2.10)
Then, according to this theorem, gu and gu lie on the same conjugacy class when u and u
belong to the same Weyl orbit, i.e., when there is a w W such that
u = w(u) mod 2L .
(2.11)
So the number of O-planes equals the number of Weyl orbits of g . We determined these
numbers empirically and displayed them in Table 1.
Next, consider the orientifold projection by n where n is even, i.e., n = 2m. Note that,
when the order of the centre N is odd, we can always bring n to this form. It is easy to see
that (2.5) is solved by
g2m,u = m g0,u ,
(2.12)
i.e., the locations of the n = 2m planes are simply translations of the canonical (n = 0) ones
by centre elements m . It is not hard to see that translation by an element of the centre is
consistent with conjugation, i.e., that the conjugacy classes of g0,u map one-to-one to the
conjugacy classes of g2m,u . Hence the n = 2m configuration of O-planes directly follows
from the n = 0 result.
For N even, we have the additional possibility that n = 2m + 1 odd. Then
g2m+1,u = m g1,u ,
(2.13)
solves (2.5) where g1,u is a solution for n = 1. So the locations of the n = 2m + 1 planes
are translations of the n = 1 configuration by centre elements m . The n = 1 and n = 0
configuration appear to be related by translations
gn=1,u = gn=0,u ,
(2.14)
225
where is one of the elements of the maximal torus that squares to .3 Of course
itself is not an element of the centre. We must check anew how the 2r solutions gn=1,u split
into conjugacy classes. The total number of O-planes is now given by the number of Weyl
orbits containing the 2r solutions g1,u . Again we determined these numbers empirically
and they are also given in Table 1.
Note that for N even, a given configuration (both n = 0, 1) is invariant under translations
by the Z2 subgroup of the centre. Indeed, from (2.5) it is easy to see that when g is a
solution, so is g.
2.1. Dimensions of O-planes
To determine the dimension of the O-planes we need to know which elements of the
group commute with gn,u , the element of the maximal torus that defines the conjugacy
class. This analysis can be done in the tangent space at the point gn,u of the group manifold,
in other words we have to determine which elements of the Lie algebra commute with gn,u .
For a generic element of the maximal torus, this set (called the commutant Gc henceforth)
consists of only the Cartan sub-algebra, and then the dimension of the O-plane is d r,
where d is the dimension of the group and r its rank. For a centre element the commutant
is of course the entire group, and therefore the O-plane has dimension 0 (i.e., it is an
O0-plane) if and only if gn,u is an element of the centre. In all other case we have to
examine the commutator of the root generators E with gn,u .
This can be done as follows. Note that

m

m1

E (t, H ) = (t, H ) E (t, H ), E (t, H )
m1

= (t, H ) (, t) E (t, H )
.
(2.15)
Iterating this we get

m
m
E (t, H ) = (t, H ) (, t) E ,
(2.16)
and hence
E e2i(t,H ) = e2i[(t,H )(,t )]E .
E
(2.17)
e2i(t,H )
Therefore,
commutes with
if and only if (, t) is an integer. Obviously linear
combinations of root generators are in the commutant if and only if every term separately
is in the commutant. Hence the commutant is in fact a regular subalgebra. The dimension
of the plane is drM, where M is the number of roots with (, t) integer. One may also
write this as d dc , where dc is the dimension of the commutant Gc . In other words, the
dimension of the O-plane is equal to the dimension of the coset space G/Gc .
We can say a little more about these coset spaces. Since [e2i(t,H )]2 is an element of the
centre, it must commute with all E , and therefore 2(, t) must be an integer. Hence the set
of root generators splits into two sets, one with (, t) even that commutes with e2i(t,H ) ,
and one with (, t) odd that anti-commutes with it. This means that the Lie-algebra g of G
3 We may restrict our attention to elements of the maximal torus, since we are solving g = n g 1 only modulo
conjugation. In fact Eq. (2.14) would not be a solution otherwise.
226
where g c is the Lie algebra of the commutant

has a direct sum decomposition g = g c + k,
and k is the odd part. Note that the commutator of two elements of k commutes with
[e2i(t,H )]2 , and hence is in g c . This implies that the coset G/Gc is a symmetric space.
The converse is also true: let Gc G be a regular embedding, such that G/Gc is a
symmetric space. Then the Lie algebra of G can be decomposed as a direct sum g = g c + k,
and there exists an automorphism that fixes g c and sends k to minus itself. If the embedding
is regular, g c contains the entire Cartan subalgebra of g , and hence the Cartan subalgebra
of g is left unchanged by the automorphism. This implies that the automorphism is inner,
i.e., there exists an element x G such that xhx 1 = h; xkx 1 = k for h g c and k k.
This element is unique up to multiplication by an element of the centre. Furthermore x 2
commutes with both h and k, and hence it is an element of the centre. Therefore, x is a
solution to (2.5).
Hence the classification of O-planes is now a matter of inspecting the table of symmetric
spaces [19], and selecting the ones where the embedding is regular. The results are
summarized in Table 2. In the table the trivial case, Gc = G, corresponding to O0-planes
is omitted. This is always a solution for n = 0, and corresponds to g = 1. This table was
actually obtained by counting the number of odd root generators in order to identify which
conjugacy class belongs to which symmetric space. The relation between conjugacy classes
and regular symmetric spaces is not one-to-one because of the possibility of multiplying
Table 2
Dimensions of O-planes. The first column indicates the group manifold, the second the power n of the current,
and the third the subgroup Gc that commutes with the orientifold map. The dimension of the O-plane is equal to
the dimension of the tangent space of G/Gc given in column 4. In all cases p + q = N . For SO(N ), N even, we
indicate in column 2 the conjugacy class under consideration
G
Gc
Dimension
SU(N )
SU(N )
SO(N ), N
SO(N ), N
SO(N ), N
SO(N ), N
SO(N ), N
SO(N ), N
SO(N ), N
Sp(N )
Sp(N )
G2
F4
F4
E6
E6
E7
E7
E7
E8
E8
0
1
0
1
0
1 (s)
0
1 (s)
1 (v)
0
1
0
0
0
0
0
0
1
1
0
0
S(U (p) U (q)), q even

S(U (p) U (q)), q odd
SO(p) SO(q), q = 0 mod 4
U (N/2)
U (N/2)
Sp(p) Sp(q)
U (N/2)
SU(2) SU(2)
SO(9)
Sp(6) SU(2)
SO(10) U (1)
SU(6) SU(2)
SO(12) SU(2)
E6 U (1)
SU(8)
SO(16)
E7 SU(2)
2pq
2pq
pq
pq
pq
1 N (N 2)
4
pq
1 N (N 2)
4
pq
pq
1 N (N + 2)
4
8
16
28
32
40
64
54
70
128
112
odd
odd
= 2 mod 4
= 2 mod 4
= 0 mod 4
= 0 mod 4
= 0 mod 4
227
x with a center element. This may or may not change the conjugacy class to which x
belongs. For example, in the trivial case G = Gc (yielding O0-planes) valid choices for x
are 1 and 1 (if 1 is an element of G), but 1 and 1 are obviously not conjugate. Up to
this freedom of multiplying x by elements of the centre, the number of different choices of
Gc for given G and n corresponds to the number of O-planes in the last column of Table 1.
In most cases where there are more conjugacy classes per symmetric space this degeneracy
is due to multiplication by 1, but for Dr , r even, there are additional degeneracies.
This can be illustrated rather nicely for G = SU(N), because in that case one can solve
(2.4) explicity in the group. For n = 0 the condition g 2 = 1 combined with det(g) = 1
allows diagonal matrices with p eigenvalues 1 and q eigenvalues 1, with N = p + q and
q even. This breaks SU(N) to S(U (p) U (q)). Therefore, the corresponding O-planes
have dimension 2pq.
For n = 1 we may take a square root of the center element diag(exp(2i/N)). If we
take all N entries of the square root equal to exp(i/N) we would get determinant 1.
Hence we have to flip an odd number of signs. The corresponding group element g breaks
SU(N) to S(U (p) U (q)), but this time with q odd. Note that if N is odd either p is odd
or q is odd, so that there is no real difference between n = 0 and n = 1, as expected.
Note that O-planes have smaller dimensions than D-branes except if Gc is precisely the
Cartan sub-algebra. This happens only for SU(2), n = 1.
To summarize, simple geometric semiclassical analysis of the WZW-action tells us that
the O-plane characteristics depend on the order of the centre. When the order of the centre
N is odd, there is only one configuration of O-planes modulo centre translations. This
configuration has one O0-plane at the origin; the remaining O-planes are localized on
conjugacy classes of various dimensions; their number is given in Table 1. For even N ,
there are two inequivalent configurations of O-planes modulo centre translations. The
even configuration has two O0-planes at g = 1, and a remainder (see Table 1) of planes
localized on conjugacy classes, whereas the odd configuration consists completely of
O-planes localized on certain conjugacy classes. This pattern generalizes to the non-cyclic
centre of Dr , r even, in a straightforward manner. When there are more cyclic factors in
the centre, the number of inequivalent configurations is 2s , where s is the number of even
cyclic factors.
2.2. Planes in orbifolds
As for free CFTs, the WZW action (2.1) may also be used to describe orbifolds. For
every subgroup H of the centre, we can construct the orbifold G/H . Let H = ZM , with
M a divisor of N , be generated by := N/M . Then the orbifold consists of equivalence
classes (m = 0, . . . , M 1)
g m g.
(2.18)
The n = 0 O-planes in orbifolds are located at those points of G/H for which
g = g 1 .
(2.19)
By (2.18), this relation only has to hold modulo translations by m . In terms of the covering
group G, every O-plane configuration corresponding to m appears in the orbifold. We will
228
see in Section 5.1 that the planes form H -translation invariant combinations. Obviously
this generalizes directly to n = 0 planes.
2.3. An example: SU(2)
Let us illustrate these results with SU(2).4 The group manifold is a three-sphere;
the conjugacy classes are two-spheres of constant latitude and the centre elements are at the
north and south poles. The standard orientifold symmetry g g 1 is a reflection through
the axis of rotation with fixed points at the poles. In terms of the solutions (2.10) we
simply have gu=0 = 1 and gu=1 = 1.
The action of the centre is to identify antipodal points on the sphere. So the combined
action g g 1 leaves the equator fixed. The solution (2.14) indeed describes one
O-plane at the equator, since = diag(i, i).

So the orientifold group (1, 0 ) gives two O0-planes at the north and south pole and
the orientifold group (1, 1 ) gives one O2-plane at the equator.
When we identify g g, we obtain the group manifold of SO(3). The orientifold
projection g g 1 gives the following configuration. The two SU(2) O0-planes at the
poles are identified by the orbifold symmetry, leaving one O0-plane at the pole for SO(3).
There is also one O2-plane at the equator, since this submanifold is now fixed under the
combined action of the orientifold and orbifold symmetry.
Recalling the caveat regarding differing views of target space between conventional
string compactifications and WZW models, it is interesting to briefly compare the SU(2)
results to the well known results for U (1); for, the U (1) theory at the self-dual radius is
equal to the SU(2) level one model. In terms of U (1) the standard orientifold symmetry
g g 1 is just the reflection x x modulo 2R, with fixed points x = 0 and x = R
corresponding to g = 1. The n = 1 orientifold symmetry g g 1 corresponds to the
identification x x + R modulo 2R, with fixed points at x = R/2. Within SU(2)
these solutions are conjugate, as are all solutions to the conditions g 2 = , U (1):
the center is the entire group. If we only consider the locations there is therefore just one
solution in U (1), up to equivalence. However, if we also consider the tensions there are
two orientifold configurations for any radius R, one with O0-planes with opposite tensions
and one with O0-planes with the same tension [21]. In the case of U (1) at the self-dual
radius or SU(2) level 1, the underlying CFT interpretation of these two configurations is
the same, and the difference corresponds in both cases to a simple current modification of
the crosscap, as discussed in the next section, Eq. (3.8). In Section 5 we will see that the two
O0-planes in SU(2) actually have opposite tension. Hence this configuration corresponds
to the analogous one in the U (1) picture. The other configuration is, in SU(2) language, an
O2-plane with non-zero tension. This corresponds in U (1) language to two O0-planes with
the same tension. In both cases the latter configuration is the one that one obtains from the
simple current modified crosscap (3.8) with n = 1.
4 This example was studied prior to our work by C. Bachas et al. [20].
229
3. Stringy analysis
The above geometric analysis is only valid in the large volume limit, i.e., at infinite
level. At finite k, we expect our results to acquire stringy corrections. Conformal field
theory, being exact in the string tension, should provide these corrections.
We therefore consider the chiral algebra A of a WZW model. It contains both the
Virasoro algebra and an affine Lie algebra g spanned by the modes of a chiral current
W given by (n, m Z and a, b = 1, . . . , d)
a

c
Wn , Wmb = fcab Wn+m
(3.1)
+ m ab n,m K.
Here K is the central element whose eigenvalue is the level k and fcab the structure
constants of the horizontal Lie algebra g . At finite level, WZW models are rational (i.e.,
the number of unitary highest weight representations is finite). The unitary highest weight
k) where is a highest weight representation
representations of A are labeled by = (,
) k with the highest root of g . The set
of the horizontal subalgebra g such that (,
of highest weights which obey this constraint, form the dominant affine Weyl chamber
Wk+ .
For closed oriented theories the chiral and antichiral algebras have to be combined in
such a way that the torus partition function

Z ,M M
Z=
(3.2)
,M
is modular invariant. Here are the affine g-characters of a representation . For

modular invariance, it is necessary and sufficient to require the matrix Z to commute
with the modular matrices S and T ; explicit expressions for these matrices are known
in the case of WZW models and are given in Appendix A. The invariant Z corresponds
to the geometry of the space on which the strings live. When we take for Z the charge
conjugation invariant, this space is the compact, simply-connected Lie group G whose Lie
algebra is the compact real form of g . With the use of simple currents we can construct a
large class of inequivalent invariants. Simple current invariants correspond geometrically
to orbifolds of G by (subgroups of ) the centre [22,23]. For instance, for A1 level k, the
modular invariant choices for Z come in an ADE classification. The A-series correspond
to the charge conjugation invariant (which is just the diagonal invariant in this case); it
describes an SU(2) manifold. The D-series are simple current invariants that describe
SO(3). Finally, the E-series, that only exist for three values of k, do not correspond to
a simple current.
Recall that simple currents [24] are primary fields whose fusion rules are such that
fusion with any other primary field yields exactly one other field and not several. The
fusion rules of simple currents among themselves form an Abelian group. This group is
isomorphic to the centre of G.5 We will denote the field that generates the simple current
group H by J , the elements by J n , n = 0, . . . , N 1, and the identity by 0, i.e., the
vacuum representation. An important quantity is the monodromy charge, defined by (h
5 Except for E level 2 [26].
8
230
is the conformal weight of and J is the unique fusion product of J and ).

QJ () := hJ + h hJ mod 1.
(3.3)
The charges of the other currents in H are related to that of the generator by QJ n () =
nQJ () mod 1. The monodromy charge is related to the conjugacy class C of the g representation as
QJ () =
C()
mod 1.
N
(3.4)
To construct an unoriented theory, we project the spectrum described by (3.2) to states

that are invariant under the interchange of left- and right-movers by adding the Klein bottle
partition function [3]

K=
(3.5)
K .
Wk+
To ensure that the total closed unoriented spectrum, encoded in (T + K)/2, produces nonnegative state multiplicities, K must be an integer that has the same parity and is bounded
by Z, . The perturbative worldsheet expansion in this theory contains unoriented surfaces, i.e., surfaces with crosscaps. The worldsheet may also have boundaries. Their properties are encoded in the annulus and Mbius strip partition function:

A
Ma =
Ma .
Aab =
(3.6)
ab ,
Wk+
Wk+
The ChanPaton labels a, b are used to distinguish the possible boundaries of the open
strings. Open-closed string duality means that we may equivalently describe the three
partition functions K, A and M as the tree-level exchange of closed strings, propagating
between boundary-states |Ba and crosscap states | :

Ba, |B ,
| =
| .
|Ba =
(3.7)
Wk+
Wk+
We have expanded each onto a basis of Ishibashi boundary and crosscap states,
respectively, in order to make the conformal properties of these states manifest; Ba,L
and are known as the boundary and crosscap coefficients. These are not arbitrary
but must obey various integrality constraints. For all simple current modifications of the
charge conjugation invariants solutions to these constraints are known [11]. In case of the
charge conjugation invariant, we have N solutions; one for every simple current J n :
Sa,
(n)
,
Ba, =
SJ n ,
PJ n ,
(n)
=
,
SJ n ,
n = 0, . . . , N 1,
(3.8)
whereSa,b is the
modular S-matrix and Pa,b the pseudo-modular matrix equaling
P := T ST 2 S T . This matrix will form an important part of our story, and we will
discuss it in more detail below. Historically, the boundary state with the n = 0 coefficient is
the famous Cardy state. The crosscap coefficient with n = 0 first appeared in [27]. In [10],
231
it is shown that the above set of boundary and crosscap coefficients gives rise to a consistent
unoriented spectrum for every simple current J n . Not all currents give rise to a different

spectrum, however. In [28], it is shown that two Klein bottle currents J n and J n produce
the same spectrum if n + n even modulo N .
We claim that, just as the boundary states describe stringy D-branes, these crosscap
states describe the orientifolds found in Section 2 in the stringy regime. As a check,
we would like to recover the semiclassical properties and positions in the large volume
limit. In order to extract the position of the defects described by the crosscap state, we
follow the method of [13,29], by calculating the coupling of a graviton to the crosscap in
momentum space. Extrapolating from the U (1) free CFT and its chiral extension U (1)k ,
a graviton is created by the vertex
a b
b a
W1
M.
W1 + W1
W1 |,
(3.9)
The weights , are the left- and right-momentum, and required to be weights of M
and M c , respectively (since we are discussing the charge conjugation modular invariant).
Due to the multi-dimensionality of the ground-state of the Verma-module (by acting with
the horizontal subalgebra), the range of is not just the set of affine highest weight
states Wk+ = { Lw |(, ) < k}, but includes all weights found by acting with the
horizontal subalgebra on Wk+ . To find the location of the orientifold we compute the onepoint function of a graviton, up to a normalization of the crosscap states which will be
discussed in Section 5.
Gab
=
M (, )
b
1a W
M , |W
1
a | .
+ W1b W
1
(3.10)
We will need the explicit form and reflection properties of the Ishibashi crosscap state

a
m
| ,
| =
|, m U (1)m | = , m,
Wma | = (1)m W
levels m
(3.11)
where U is an anti-unitary operator that maps weights M to their duals c M c . Here
|, m is a condensed notation for the complete set of weights of the horizontal algebra g
that appear at level m in the Verma module based on the highest weight . With the help
of (3.11) one finds that
b
a
Gab
= M , |W
1a W1
+ W1b W1
|
M (, )

ab
ab
M , |
= 2k
= 2k ,
c M .
(3.12)
It is obvious from the computation that 2k ab just signifies the tensorial nature of the
overlap, and that one may use any arbitrary closed string state to derive the momentumspace distribution (e.g., as in [15]). We will therefore drop this factor from now on, and
call the remainder GM (, ).
To find the spatial-distribution we need to Fourier transform. The analogue of Fourier

modes for group-manifolds are the matrix elements |R (g)| , where R (g) stands for
the group element g in the representation and , are weights and c . The
232
Fourier transformation to find the distribution amounts to computing

=
G(g)
G (, )|R (g)| =
X (g) .
,
(3.13)
Wk+
The last step follows from the fact that highest weights
/ Wk+ do not appear in the
theory.
The above is simply repeating what has been done for the boundary states in [13].
There on the r.h.s. of (3.13) the boundary coefficient, and thus the modular S-matrix
appears instead of the crosscap coefficient which depends on the modular P -matrix.
Mathematically it was known that the modular S-matrix is expressible in terms of a
character of the horizontal Lie group. This meant that the boundary state analogue of (3.13)
was easily evaluated by using completeness of the group characters, and the geometric
analysis could proceed straightaway. Our aim in the next section will be to derive an
expression for the P -matrix in terms of group characters of G as well.
4. The P -matrix, characters and Gaussian sums

4.1. Properties of the P -matrix
Before we turn to this calculation of the P -matrix in WZW models, we briefly discuss
some of its properties that hold for arbitrary CFTs. As we indicated, the P -matrix plays
an important role in the construction of open unoriented string theories [3]. The P -matrix
encodes the channel transformation for the Mbius strip, i.e., it relates the open and closed
string channels of this diagram [3].The action
P may be represented in
terms of the
modular T - and S-matrix as P = T ST 2 S T , with T diagonal and T defined as
exp[i(ha c/24)].
When the CFT has a ZN simple current group generated by J , the matrix elements of S
are related as follows (n = 0, . . . , N 1) [24,25]:
SJ n i,j = e2inQJ (j ) Si,j ,
(4.1)
where Q is the monodromy charge (3.3). A similar relation can be derived for the matrix
elements of the P -matrix [11]
PJ 2m i,j = (2m, i)e2imQJ (j ) Pi,j ,
(4.2)
where

(2m, i) := exp i hi hJ 2m i 2QJ m J m i .
(4.3)
In the application in the main text, we have i = 0 or i = J and J integer spin, in which
case the phases (2m) := (2m, 0), (2m + 1) := (2m, J ) are just signs.
Note that there is no simple relation for matrix elements that are related by odd currents.
This paper basically explains why: the corresponding O-plane configurations are not
related by a centre translation.
233
4.2. The P -matrix in WZW models and G-characters

The objective of this section, to which we now return, is to find an expression for the
P -matrix of WZW models in terms of group characters, since this enables us to find the
location of the O-planes by using orthogonality of these characters.
For WZW models we know the explicit expressions for both the modular S- and
the modular T -matrix. We have given them in Appendix A. The P -matrix is therefore
explicitly known. Some properties of Lie algebra characters and the different forms in
which they may be expressed are also given in Appendix A. Inspection shows that to find P
in terms of characters we must perform the matrix multiplication inherent in the definition
of P . After substituting the formulas for the S and T matrix we therefore have to compute

i
kd
P,N = NS2 exp i
exp
(, + 2) + (N, N + 2)
4h
2h

2i
w( + ), R + (R, R + 2)
sign(wv) exp
h
+
RWk
w,vW

+ v(N + ), R + .
(4.4)
Here Wk+ is the set of affine dominant integral weights, i.e., the set of weights R of G with
non-negative Dynkin labels such that (R, ) k, with the highest root. The vector is
the Weyl vector, h := k + g the height, g the dual Coxeter number, W the Weyl group
of G and sign(w) equals (1)# generators of wW . NS is the normalization of the S-matrix
given in Eq. (A.12).
We can simplify this expression by shifting the sum over R to a sum over M = R + .
When (R, ) k then (R + , ) k + (, ) = h 1. Furthermore, as the Dynkin
labels Ri 0, it follows from the definition of the Weyl vector that Ri + i = Ri + 1 1.
So this shifted weight M lives in the interior of the strictly dominant affine Weyl chamber
Wh++ . However, since the S-matrix vanishes at the boundaries of this chamber, we can
extend the sum back to the dominant Weyl chamber Wh+ .

i
kd
2
exp
P,N = NS exp i
(, + 2) + (N, N + 2) 4(, )
4h
2h

2i
w( + ), M (M, M)
sign(wv) exp
h
+
MWh
w,vW

+ v(N + ), M .
(4.5)
Next, for every v we reorder the sum over w to vw. Then, since Weyl reflections preserve
the inner product, we can combine the sums over M and v to a sum over weights M that
live in all Weyl reflections of the dominant Weyl chamber, i.e., the set Wh . It may appear
that the fixed points of the Weyl group at the boundaries are now undercounted. However,
234
as we have seen before, they do not contribute to the P -matrix, so we can add or subtract
them at will.
There is, however, a more natural range for M. Note that the summand is invariant under
a translation M M + h with an arbitrary simple coroot. Recall that for every element
of the weight lattice Lw that does not lie on the boundary of some Weyl chamber, there is
a unique element of the affine Weyl group (at level h) that maps it to the dominant Weyl
chamber Wh+ . The affine Weyl group is the (semi-direct) product of the horizontal Weyl
group W and the group of translations by hL , with L the coroot lattice. So every weight
in Wh = W (Wh+ ) that does not lie at the boundary of some chamber, represents one of the
elements in L w / hL , where L w is the weight lattice where all boundaries are removed.
So instead of summing over all elements in Wh , we may sum over cosets M Lw / hL ,
where we included the boundaries again since they do not contribute to P anyway. So

i
kd
2
exp
P,N = NS exp i
(, + 2) + (N, N + 2) 4(, )
4h
2h

2i
w( + ), M (M, M)
sign(w) exp
MLw / hL
wW

+ (N + , M) .
(4.6)
We can now complete squares in the second line:

i
2
P,N = NS exp
kd + 12(, )
4h

i
w( + ), N +
exp
MLw / hL
wW

2i
w( + ) + N + 2
exp
.
M
h
2
(4.7)
Upon using the strange formula for the length of (recall that we are using the
normalization (, ) = 2)
1
dg ,
12
we can write this as

id
i
2
w( + ), N +
P,N = NS exp
sign(w) exp
4
h
wW

Gh C 1 , w( + ) + N + ,
(, ) =
where we introduced the lattice Gaussian sum on Lw with metric

1
X 2
2i
M .
exp
Gh C , X :=
h
2
MLw / hL
(4.8)
(4.9)
C 1 .
(4.10)
In the next section, see Eq. (4.31), we prove that this sum equals

1 Lw 1/2 hi r/2
h

Gh C , X =
(1) 2 (u,u)+(u,X),
L
2
235
(4.11)
uL /2L
where u is the
set of all coroots modulo even coroots. In other words, u is the set of
vectors u
= i ui i on the coroot lattice L whose entries ui are
either 0 or 1. Hence
(u, X) = i ui Xi where Xi are the Dynkin labels, and (u, u) = rj =1 ui C ij uj where C
is the inner product matrix of the coroots C ij = ( i , j ).
Taking everything together, we find

h
P,N = NP
(1) 2 (u,u)+(u,N+)
uL /2L

wW

i
w( + ), N + + hu
sign(w) exp
h
(4.12)
with

1/2 1
NP := 2r/2NS = i(dr)/2Lw /L
.
(2h)r
(4.13)
Two results immediately follow from Eq. (4.12). Using the Weyl character formula
(A.3) P can be seen to equal the sum of characters X of the horizontal subgroup,

Nu(N) X guN
P,N =
(4.14)
u
evaluated at the group element

i
N
(N + + hu, H ) ,
gu = exp
h
(N)
where Nu
(4.15)
equals
h
Nu(N) : = NP (1) 2 (u,u)+(u,N+)

i
w(), N + + hu .
sign(w) exp
h
(4.16)
wW
And, due to the formal equivalence between characters X and the modular S matrix, one
derives a linear relation between S and P ,

h
(1) 2 (u,u)+(u,N+) S, N+hu .
P,N = 2r/2
(4.17)
u
This relation exists only in the formal sense, as the S-matrix for non-integer weights is
only defined by analytic continuation through its equivalence with characters. Non-etheless
the very existence of these relations is remarkable. Both, however, hinge on the proof of
Eq. (4.11) with which we now proceed.
236
4.3. Gaussian sums for lattices

The Gaussian sum gk (a, 0) is the discrete analog of the Gaussian integral:
gk (a, 0) =
k1

2 ian2
k
(4.18)
n=0
Gaussian sums play a role in analytic number theory (for a review, see [30]). Their main
application is to count the number of solutions to the congruence x 2 a(mod p); the
number a is called a quadratic residue mod p.
This sum has been evaluated to
gmk (a, 0) = mgk (a, 0),
gcd(ma, mk) = m Z,

a
gk (a, 0) =
g(1, k),
gcd(a, k) = 1,
k

1
gk (1, 0) = (1 + i) 1 + i k k,
(4.19)
2

where (a/k) is the Legendre symbol. Decomposing k into primes k = pi , it equals the
product of Jacobi symbols,

a
a
,
=
k
pi

i;k= pi

a
if x 2 = a(mod p) has a solution for x Z,
= 1
(4.20)
p
1 if not.
For purposes of comparison, we also give the Gaussian sum over half integers
1
k

2 2 ian2
1
1
=
e k = g(a, 4k) g(a, k).
gk a,
2
2
1
(4.21)
n= 2
For a = 1 this is simply

1
1
gk 1,
= (1 + i) 1 i k k.
2
2
(4.22)
For SU(2) the weight lattice is one-dimensional and the sum in (4.9) equals a regular
Gaussian sum. For higher rank groups we will need a lattice generalization of this sum.
Consider an even lattice L with volume |L | and dual lattice Lw . In the application
we have in mind, L is the coroot lattice of a Lie algebra g and Lw the weight lattice, but
the following discussion applies to arbitrary even lattices. The restriction to even lattices is
important, since only then is L (and every integer multiple hL ) a sublattice of Lw . The
quadratic Gaussian sum on Lw / hL is defined as

1
X 2
2i

exp
Gh C , X :=
h
2
Lw / hL

mi

2i
Xi
Xj
exp
mi
Cij1 mj
,
h
2
2
237
(4.23)
where X is an integral weight and |y|2 is the length squared of y, computed with the lattice
metric C 1 of Lw . The precise summation range of mi follows from the first form of the
expression. In order to evaluate this sum, it is convenient to define (E > 0)

iE
1 1
= + 2 + O E2 .
h h
Consider then the sum over the full weight lattice

X 2
2i
,
exp
( ) :=

2
:= h iE
(4.24)
(4.25)
Lw
which due to the small regulating parameter introduced in (4.24) is now finite and well
defined (it is a theta function). We can write in two different ways. In the first, we split
the sum over in a E-dependent sum over h, L and a E-independent sum over6
Lw / hL :

X 2
2i

( ) =
exp
h
2
Lw / hL

2
X/2 2

exp 2E +
(4.26)
+O E .
h
The last sum can be approximated by a Gaussian integral that to leading order in E does
not depend on X/2, and we recognize the Gaussian sum in the prefactor:

1 1
2x 2
( ) = Gh C , X
(4.27)
+ O(1) .
dx e
E
On the other hand, we can first use Poissons resummation formula (which is allowed since
E > 0) to write the sum (4.25) as

r/2

i 2
i
X
|| exp 2i
, ,
exp
( ) = |Lw |1
(4.28)
2
2
2
and then
the sum over in a E-dependent sum over 2, L and a E-independent
sum over u L /2L

r/2

u 2
1 i
(u,X)+ 2h (u,u)

(1)
exp 2E .
( ) = |Lw |
2
2
split7
uL /2L
(4.29)
6 We have to be careful here, since the summand is not independent of the representative of the coset
Lw / hL . We must therefore choose a particular set of representatives and this is implicit in the notation. In
the final answer, Eq. (4.31), this ambiguity disappears.
7 Again, at this stage we must choose a representative for u in order for this expression to be well-defined.
238
As before, we can approximate the sum over by an integral that does not depend on u/2
to leading order in E:
r/2

h
hi
1
2
( ) = |Lw |1
(1)(u,X)+ 2 (u,u)
dx e2X + O(1) .
2
E
uL /2L
(4.30)
We can now equate the two results (4.27) and (4.30). Comparing the O(1/ E ) terms, we
find for the quadratic Gauss sum

Lw 1/2 ih r/2
h
Gh C 1 , X =
(4.31)
(1)(u,X)+ 2 (u,u) ,
L
2
u
where we used |Lw | = |L |1 in obtaining the final equation. For rank one SU(2) this
indeed reduces to the known results (4.20) and (4.22).
To appreciate how remarkable this result is note that we started with a summation of
Gaussians involving the integrable weights at a certain level. Yet the result depends on the
level in a very simple way, namely only through the volume of the affine Weyl chamber.
Although not shown here explicitly, the dependence on X is also very simple. A priori it
is clear that the result only depends on the classes Lw /2Lw , and is constant on the Weyl
orbits of these classes (the number of distinct classes is in fact precisely given by the entries
in the last column of Table 1). If h is even, the summation over u yields 2r for X = 0, and
zero if X belongs to a non-trivial class. For odd h the result is slightly more complicated.
The summation can take three values on the classes, either 0 or 2l , with l r. There
are many more interesting features we could mention, and undoubtedly more still to be
discovered, but let us return now to the main subject of this paper.
5. The semiclassical limit

We are now in a position to show that there is a one-to-one correspondence between the
O-planes constructed in Section 2 and the crosscap states of Section 3. We will do this by
demonstrating that the distribution (3.13), that gives the location of the defects described by
the crosscap states in the large volume limit, is peaked at the solutions gn,u of (Eq. 2.5). Let
us first consider the charge conjugation invariant. There is a normalization issue we have
suppressed up till now. In the more extensively studied case of D-branes, the graviton
crosscap coupling is not directly given by the boundary coefficients, but by a reflection
coefficient R,a that differs from it by a factor [31,32]. By boundary coefficient we
concretely
here the coefficients B that appear in the expression of the annulus as
mean
i
Ba Bb , and that are determined by integrality. These are the coefficients
Aiab = S
presented in (3.8). The additional factors come from the definition of the annulus amplitude
and the normalization of the identity near a boundary. In the Cardy
case, the reflection
coefficients are Sa /S0a , and the boundary coefficients Sa / S0 . We should expect

similar factors to appear between the crosscap states (3.8) and the O-plane reflection
coefficients. The latter we will denote as U . An additional complication is the fact that we
have to know these factors for non-zero values of J . Our proposal is to use the following
239
definitions
(n)
Ra, =
Sa,
,
S0a
(n)
U =
PJ n ,
,
S00
n = 0, . . . , N 1.
(5.1)
It may appear counter-intuitive that for n = 0 the phases from the denominator in (3.8)
were dropped. The rationale behind this is that the D-brane reflection coefficients should
not change if we modify the O-plane reflection coefficients, by going from P0 to PJ n , .
On the other hand, at the one-loop level, the Mbius strip amplitude does change, and this
in its turn enforces (through open sector integrality) a change in the annulus amplitude.
The latter change correponds always to a different choice of the boundary conjugation
matrix A0ab . This can be taken into account by extra phases in the boundary coefficients.
In other words, the phases in the denominators in (3.8) belong naturally in the boundary
and crosscap coefficients that enter in the definition of the one-loop partition functions, but
not in the reflection coefficients. Obviously this is a conjecture, and not a proof. The fact
that this definition will turn out to make sense geometrically gives additional support for
this conjecture, and we hope that it may help resolve a long-standing debate on the proper
description and the origin of these phases. In fact, we will see that the geometric picture
gives a nice way of understanding boundary conjugation.
The precise distribution that we therefore wish to evaluate is
(n)
(n) (g) =
G
(5.2)
U X (g)
Wk+
rather than Eq. (3.13). In the infinite volume limit, k , the sum is over all highest
weight representations of g . We can then use completeness of the group characters,
provided we have an expression for PJ n , as a sum of G-characters. This is the very fact
we accomplished in (4.14),

PJ n , =
(5.3)
Nu(n) X gu(n) ,
u
where
h
Nu(n) : = NP (1) 2 (u,u)+(u,J +)

i
w(), J n + + hu ,
sign(w) exp
h
n
(5.4)
wW
and the character in the P -matrix is evaluated at the group element

i n
J + , H exp i(u, H ) .
gu(n) := exp
h
(5.5)
We will start with n = 0. In the infinite volume limit, the argument of the characters in (5.3)
becomes (we ignore / h)
h
gu(0) ei(u,H ) = gu ,
(5.6)
240
where gu is the solution to (2.5) for n = 0. To work out the normalizations we make use of
the denominator identity,

(, )
(w(),)
,
sign(w)e
=
2 sinh
(5.7)
2
wW
positive roots
applied to the ratio

i

wW sign(w) exp h (w(), + hu)

2i

wW sign(w) exp h (w(), ) .

The result is

positive roots
sin
2h(, + hu)

sin h (, )
(5.8)
(5.9)
If (, u) is even, the u term may be dropped, and in the limit h the corresponding
factor in the product contributes a factor 1/2. This is obviously the correct answer if u = 0:
(0)
N0
r/2 (dr)/2
Q(0)
2
= 2d/2 .
0 := 2
S0,0
(5.10)
If on the other hand (, u) is odd (since u is a co-root it is integer) the numerator contributes
sin( 12 (, u)), which remains finite in the limit h . We can write the limit of the ratio
as follows

1
sin 2h
(, + hu)
for (, u) even,
1

e 2 i(,u) 2 h
(5.11)
sin h (, )
i(,) for (, u) odd.
The result is then
D/2

(0)
h
Nu
h
1
(0)
D/2
2 (u,u)
Q(0)
:=
2
(1)
Q0 ,
u
S0,0
i
(,
)
, odd
(5.12)
where the product is over all positive roots with (, u) odd, and where D is the dimension
of the O-plane. This is equal to D = d r M, where M is the number of roots with
(, u) even. The number of odd positive roots is equal to 12 (d r) 12 M = 12 D. Since
we started from a real expression, the result is of course real. This implies that D must
be a multiple of four, as indeed it is according to Table 2 (with n = 0!). Note that the
result diverges for large h as hD/2 . Obviously this is a volume factor. At finite h, previous
results on D-branes suggests that the branes are fuzzy. Hence even an O0-plane, a point,
has then a finite volume. But as h tends to infinity, the ratio of the volume of a higherdimensional object and that of a point goes to infinity. This phenomenon has not been
observed for WZW D-branes for the simple reason that they all occupy regular conjugacy
classes of the same dimension d r [13]. Although we believe we correctly understand
the relative normalization, the overall normalization is another matter, as is the relative
normalization between D-branes and O-planes. Overall normalizations of amplitudes are
tricky in general, and even more so in our case where one does not consider a genuine string
theory, so that graviton scattering is only called that way by analogy. We will not try to
241
address this issue here and turn instead to the main goal, the localization of the planes, for
which the precise normalization is of little relevance.
The distribution (5.2) can now easily be computed using completeness of the group
characters:

(0) (g) =
G
Q(0)
X (gu )X (g)
u
u
Q(0)
u (gu
g).
(5.13)
This shows that the crosscap state | (0) corresponds to a localized charge distribution
in curved geometries and that it provides a conformal field theory description of the
n = 0 orientifold planes of Section 2. Note that, since the Weyl group preserves inner
products, the contributions to the distribution from characters X evaluated at gu and gw(u)
(0)
have the same coefficient Qu . This is what one should expect, since this factor can be
interpreted as a charge (in fact a tension) and gu and gw(u) describe the same plane, as we
discussed in Section 2. The delta function is actually a -function on conjugacy classes,
not on individual group elements. The sum on u usually contains several members of the
same conjugacy class, thus producing a multiplicity factor which one may absorb in the
(0)
definition of Qu . Note that the distribution we get is peaked at all locations of O-planes
corresponding to n = 0, in contrast to the situation for D-branes, which live on a single
conjugacy class. This reflect the well known fact that D-branes come with ChanPaton
multiplicities which can be switched on or off at will, whereas there is no such freedom for
O-planes: one can either choose to have one or not to have one, but one cannot distribute
them freely over the available locations.
In a similar way, the crosscap states | (n=0) describe O-planes that correspond to a
non-trivial element of the centre. First take n even, n = 2m. Now (note that (2m) is a
sign, see the discussion after Eq. (4.3))
PJ 2m , = (2m)e2iQJ m () P0,

mC()
= (2m)
Nu(0) e2i N X gu(0)
u
= (2m)

Nu(0) X m gu(0) ,
(5.14)
where we used (3.4) and (A.5). In the large volume limit, gu(0) gu and the distribution
becomes

m

(2m) (g) = (2m)
G
(5.15)
Q(0)
u gu g .
u
So the corresponding O-plane configurations are translations of the canonical ones by m ,

up to an overall sign. When we compare this with the discussion in Section 2, we conclude
that the crosscap state | (2m) describes the O-plane configuration with centre element
2m in the defining orientifold group.
242
Now the computation for n = 1, N even, remains. Following the same steps in
Eq. (5.14), we can show that for n = 2m + 1 the crosscap reflection coefficient is

PJ 2m+1 , = (2m + 1)
(5.16)
Nu(1) X m gu(1) .
u
Note that in the large h limit, J n / h is finite. Hence in the large volume limit, the group
element (5.5) approaches, for n = 1,

r

1
(1)
i
gu exp i
(5.17)
ci H exp i(u, H ) = gu ,
N
i=1
where is a group element that squares to the generator of the centre . In obtaining this
i
result, we have written (J, H ) in Dynkin basis as j i Jj Cj1
i H . The Dynkin labels of a
simple current are always of the form Jj = kqj for some q. Now we observe the following
property of the quadratic form matrix C 1 : for all simple Lie algebras the column vector
1
Ciq
equals
ci
mod 1,
(5.18)
N
as can be shown by simple inspection of the tables in for instance [18]. From this relation
between the quadratic form matrix and the conjugacy classes, it follows that indeed
squares to and that gu(1) approaches the solutions (2.14) in the large volume limit. To
compute the tension in the large volume limit we need to consider
1
=
Ciq
Nu(1)
Nu(1)
= lim
h S0,0
h S0,J
Q(1)
u := lim
= lim 2
h
r/2
(1)
h
2 (u,u)+(u,J +)
sin
>0
2h(, + J + hu)

sin h (, + J )
(5.19)
The J in the denominator is introduced just for convenience. At level k the simple currents
can be written as kJ1 = (h g )J1 , where J1 is a fundamental weight. The ratio of the
sines can then be written as follows

sin 2h
(, + J + hu)

sin h (, + J )

i
i
exp 12 i(, J1 + u) exp 2h
y exp(i(, J1 + u)) exp 2h
y
=
(5.20)
,
i
i
exp(i(, J1 )) exp h y exp h y
where y = (, g J1 ). Using (5.18) the inner product (, J1 ) equals

i Cij1 (J1 )j =
i ci /N mod 1.
(, J1 ) =
i,j
(5.21)
Shifting the values of ci by N does not affect the corresponding centre element. Therefore
it is possible to define ci in such a way that the last relation holds exactly, not just modulo
243
integers. Then we get, using (2.8), (, J1 + u) = 2(, t). For the ratio we find then

1
for 2(, t) even,
1
i(,J
+u)
i(,J
)
2
1
1
e2
(5.22)
e
h
i(,y) for 2(, t) odd.
Putting everything together and using = 12 >0 we find

D/2
h
1
(1)
D/2 d/2
(u,u)+(u,J ) i(,J1 ) h
2
, (5.23)
(1)
e
Qu = 2 2
i
(, g J1 )
,odd
where , odd means all positive roots with 2(, t) odd. As in the case n = 0 the tension
diverges as hD/2 . In this case D is not always a multiple of 4, but the factor ei(,J1 ) is
purely imaginary precisely if D = 2 mod 4, so that the final result is real.
The distribution becomes

m

(2m+1) = (2m + 1)
Q(1)
gu g .
G(g)
(5.24)
u
u
| (2m+1)
describes the O-planes with a non-trivial centre element 2m+1

This shows that
in the defining orientifold group. We have thus succeeded in showing that the geometric
interpretation of simple current crosscap states for the charge conjugation invariant of a
WZW model are the centre O-plane configurations on a compact, simply-connected Lie
group manifold. The exact CFT computation has also given us the (relative) charges of
each.
5.1. Planes in orbifolds
We will now consider a WZW model with a simple current modular invariant. To be
more precise, we will only consider invariants that are extensions of the chiral algebra [24].
Such an invariant can be built with the use of an integer spin (conformal weight) simple
current group H , that is a subgroup of the centre. We take H = ZM , where M a divisor
of N and denote the generator by L := J N/M . Only fields that have monodromy charge
zero with respect to all currents in H contribute to the torus partition function. The
geometric interpretation of this invariant is that of strings living on an orbifold G/H .
Only representations of G that are not-faithful with respect to the conjugacy classes
corresponding to H appear as representations of this coset. Given the relation between the
simple current charges and conjugacy classes, Eq. (3.4), it follows that all representations
of G/H appear in the torus partition function.
For simple current modifications of the charge conjugation invariant, the crosscap
coefficients are [11]
=
M1
1
PLn ,
(n)
.
|H | n=0
S0,
(5.25)
The signs (n) are constrained by the requirement that vanishes for charged fields:

(n) = (n + 2m) exp i(hLn hLn+2m ) ,
(5.26)
244
for all m = 0, . . . , M.
We now want to give a geometrical interpretation of the crosscap states for this
theory, based on the crosscap coefficients in Eq. (5.25) and the solutions of the signrule (5.26). First note that the distribution corresponding to this crosscap state is still
given by (5.2), simply because the crosscap coefficient vanishes for those that are not
representations of the orbifold when the satisfy the sign-rule. Again, we must make a
distinction between the crosscap coefficient and the graviton-crosscap coupling. Following
our previous argument, we presume the last is given by:
M1
1
PLn ,
U =
(n)
.
S00
|H | n=0
(5.27)
Let us first consider M odd. We can rewrite the sum over P -matrix elements as follows:
M1

(n)PLn , =
n=0
M1

(2m)PL2m ,
m=0
= (0)
M1

e2iQLm () P0,
m=0
= (0)
M1

Nu(0) X m gu(0) .
(5.28)
m=0 u
In the first step we used that all currents in Zodd groups are even. In the second step, we
combined the signrule (5.26) and the identity (4.2). In the last step, we used our character
formula (4.12) for the P -matrix and used Eq. (A.5) from Appendix A. In the limit h ,
becomes
the group element gu(0) gu , so the distribution G
M1

(0) (0) m
=
G(g)
Qu gu g .
|H | m=0 u
(5.29)
Hence this crosscap state describes an H -translation invariant combination that survives
the orbifold projection. The overall sign (0) can be interpreted as the overall sign of the
charges (tensions) of the O-planes.
For M even, we have to make a distinction between even and odd currents:
M1

(n)PLn ,
n=0
M/21

(2m)PL2m , +
m=0
= (0)
M/21

(2m + 1)PL2m+1 ,
m=0
M/21

m=0
M/21

Nu(0) X m gu(0) + (1)
Nu(1) X m gu(1) .
m=0
(5.30)
245
In the large volume limit, this crosscap state describes a configuration of O-planes given
by
M/2
M/2

(1) (1) m
(0) (0) m

Qu gu g +
Qu
gu g .
G(g) =
|H | m=0 u
|H | m=0 u
(5.31)
This describes two H -invariant combinations, if we take into account that every m in the
sum describes a Z2 translation-invariant configuration. The relative sign is actually the
relative sign of the tensions of the two H -invariant configurations, and is therefore relevant.
Similar remarks apply for general simple current invariants. The crosscap state is always
of the form (3.8). The number of free signs equals the number of even cyclic factors p in
the simple current group, so we have 2p different crosscaps. In geometric terms, this is
the number of relative tensions between the 1 + p translation invariant combinations of
O-plane configurations.
6. Branes and planes

The geometry of D-branes in open oriented WZW models has been discussed
in [1215]. What we would like to stress here, is that this geometry differs in the unoriented
open string. As reviewed in Section 3, boundaries in CFT are described by a boundary
state |Ba . The boundary and crosscap states must be such that the open unoriented
string partition function (A + M)/2 produces non-negative integers. Due to the Mbius
projection, the set of boundary labels a is reduced; some boundaries are identified and
some boundaries are projected. More precisely, define the boundary conjugate a of a by

1 when b = a ,
0
Aab =
(6.1)
0 when b = a ,
then complex boundaries a = a are identified and real boundaries are projected. In terms
of ChanPaton gauge groups, on a complex pair of boundaries lives a unitary group.
Real boundaries carry a symplectic or orthogonal gauge group, depending on whether the
Mbius coefficient Ma0 is plus or minus one.
In [11], the boundary and crosscap coefficients (and therefore the annulus and Mbius
strip coefficients) for arbitrary simple current invariants are presented (see [28,33] for
proofs). We will discuss the charge conjugation invariant first. The boundary labels a are
in one-to-one correspondence with the primary fields (so we label them by A). Recalling
the boundary coefficients corresponding to the crosscap coefficients (3.8)
S,A
(n)
,
B,A =
S,J n
(6.2)
we can calculate the boundary conjugation matrix

A = J n Ac .
(6.3)
The geometric interpretation of these branes can be derived in a similar way as was
done in [13]: calculate the branegraviton coupling and Fourier transform this to a
246
distribution on group space. Again the coupling of a graviton to a brane is given by the
boundary reflection coefficient [13] R,A , which differs from the boundary coefficient by
a normalization. The reflection coefficient is exactly equal to a character of the Lie group G
R,A =
S,A
= X (gA )
S0,A
evaluated at

r

2i
gA = exp
(A + ), H .
h
(6.4)
(6.5)
i=1
For a boundary A the distribution that gives the localization of the defect in the limit of
infinite level is given by

A (g) =
G
R,A X (g)
X (gA )X (g)
= (gA g),
where gA is the group element in the h limit:

r

(, H ) ,
gA exp 2i
(6.6)
(6.7)
i=1
where we keep the relative weight i = Ai / h R fixed in taking the limit. To obtain the
location of the conjugate boundary, note that
1

n
1
S,J n Ac = e2iQJ n () S,A

(6.8)
= S0A X n gA
,
= e2i N C () S0A X gA
where we used two well-known properties of the S-matrix, Sij = Sij c and Eq. (4.1) in the
first step, the correspondence between monodromy charges and conjugacy classes (3.4) in
the second and the property (A.5) in the last. The location of the conjugate boundary is
therefore

A (g) = n g 1 g .
G
(6.9)
A
So the boundary conjugation (6.3) has a nice geometric interpretation: given a boundary
1
A at gA , its charge conjugate is located at Rn (gA ) = n gA
. Real boundaries, therefore,
correspond to branes that are left fixed (not necessarily pointwise) by the orientifold group,
whereas complex boundaries form an Rn invariant pair.
Take again SU(2). In the oriented theory, the boundaries are located at circles of constant
latitude. When we perform the standard orientifold projection, i.e., reflection through the
axis of rotation, all these circles are left fixed. The branes either have an SO or Sp gauge
group, depending on the FrobeniusSchur indicator of the boundary label A. When A is a
vector representation of SU(2), the projection is orthogonal and branes labelled by a spinor
247
representation are symplectic.8 When we gauge the orientifold group with a non-trivial
element of the centre, a brane is identified with its image obtained by reflection through
the equator. These branes carry unitary gauge theories on their worldvolumes. Only for
even level we have a real boundary with an orthogonal gauge group, labelled by A = k/2,
that lies on top of the orientifold plane.
6.1. Branes in orbifolds
Boundary states for a (symmetric) simple current invariant are presented in [11]
(see [28] for a proof of integrality of the annulus coefficients). We will again consider
an extension invariant by a group H = ZM and generator L = J N/M . We will assume that
the simple currents do not have fixed points (for a definition, see [24]). Then the boundary
states are labelled by H -orbits [A] := (A, LA, L2 A, . . . , LM1 A) and the Ishibashi states
only exist for chargeless fields . The reflection coefficients are given by

S,A
.
R,[A] = |H |
(6.10)
S0,A
The distribution to be calculated is

[A] = |H |
X (gA )X (g).
G
(6.11)
,CH ()=0
We restricted the sum over to non-faithful G representations with respect to the

subgroup H , since the boundary coefficient is not automatically zero for chargeless fields.
We can extend the sum to all by inserting a projector
CM(),0 =
M1
1 2imC ()/M
e
,
|H |
m=0
where the superscript M on the delta-function means modulo M. Then

[A] = 1
G
|H |
1
=
|H |
M1

X m gA X (g)
m=0
M1

m gA g .
(6.12)
m=0
We used again the equivalence between characters of the Lie group and the S-matrix,
the relation between conjugacy classes and monodromy charges and, in the last step, we
inserted a projector on the trivial class and used (A.5). From this result we immediately see
that the boundary states in the orbifold theory describe H -invariant combinations.
When the simple currents that define the modular invariant partition function do have
fixed points, some boundary and Ishibashi labels may have multiplicities. In the presence of
O-planes, i.e., the crosscap given by (5.25), the boundary coefficients have an extra phase
8 Here, and in the following, we made the choice 1 for the overall sign of the Mbius strip.
248
that depends on the multiplicity label and the relative O-plane charge (the signs ). (In [11]
this phase is J .) Although it would be interesting to give a geometrical interpretation of
the multiplicities and this phase, we will leave this for future research.
7. Conclusion
We have seen that crosscap states in WZW models and orbifolds thereof correspond to
localized orientifolds, as they do in free CFT. The centre of the horizontal subalgebra plays
an important classifying role in the possible orientifold projections; this is the geometrical
counterpart of the role of simple currents in the exact CFT description. To compute the
various numbers, locations, dimensions and charges of the possible geometric O-planes,
we needed to prove mathematically important lattice generalization of Gaussian sums. The
consequence of this is a highly remarkable relation between the pseudo-modular P -matrix
and characters of the horizontal subalgebra.
This in turn implies a linear relation between the P - and the S-matrix. It is intriguing
that linear relations between the two matrices is precisely what is needed in order to cancel
tadpoles in open string theories. It would be very interesting to know if such a relation also
exists in other CFTs. Although on the one hand there is no obvious generalization of the
analytic continuation of the labels of S, on the other hand most CFTs are WZW-based. It
is, therefore, conceivable that such a relation might be derivable at least for coset CFTs.
Whether relations of this kind actually do help with tadpole cancellation is an interesting
question which we hope to address in the future.
Acknowledgements
A.N. Schellekens wishes to thank C. Bachas for a discussion of his work on the
SU(2) case. We thank B. Pioline for pointing out [35] in which lattice Gaussian sums are
considered. Although the method used to derive the sum is essentially the same as ours,
the Gaussian sum considered in [35] is different; there the range is over Lw / hLw instead
of Lw / hL .
The work of K. Schalm was supported in part by DOE grant DE-FG02-92ER40699.
Appendix A. Characters
In this appendix, we list some formulas that are used in the main text. For an extensive
review on Lie algebras and groups, we refer to [18].
A character maps the information of a highest weight representation R of the Lie
algebra g to the complex numbers:

X (ht ) = Tr exp 2iR (ht ) ,
(A.1)
where ht = ti H i (t, H ) is an element of the Cartan subalgebra, whose generators

we denote as H i . The trace is over all weight vectors in the representation with highest
249
weight and R (ht ) is the representation of ht . As weights are eigenstates of the

Cartan subalgebra, R (H i )| = i |, we have

mult () exp 2i(, t) ,
X (ht ) =
(A.2)
where mult () denotes the multiplicity of the weight in the representation R .

A second, more powerful, expression for the character is given by the Weyl character
formula,
wEW sign(w) exp[2i(w( + ), t)]

X (ht ) =
(A.3)
.
wEW sign(w) exp[2i(w(), t)]
The sum is over the Weyl group W weighted by the sign of each element defined as
sign(w) = (1)# of generators of W . The vector is the Weyl vector and equals
half the
sum of positive roots, or, equivalently, the sum of the fundamental weights = ri=1 i .
These expressions are basis-independent. It is often convenient to choose the Dynkin
basis in weight space and to choose a basis in the Lie-algebra (Chevalley basis) so that
the eigenvalues
of H i are the Dynkin labels. The inner product in weight space is then
(, ) = i,j i Cij1 j , where C 1 is the inverse of the symmetrized Cartan matrix.

Group characters
Eq. (A.1) may also be read as a character of the Lie Group G,
X (g) = Tr (g),
(A.4)
evaluated at the group element gt = exp{2i i ti H i }. In the main text we use two
properties. One is that characters shift by a phase under transformations g g, where
is the generator of the centre of G:

n
n
X g = exp 2i C() (g).
(A.5)
N
What is really important to us, however, is that characters serve as the analogue of
Fourier modes for group manifolds. They are orthogonal and complete

G (g)X (g)XM (g) = ,M ,
(A.6)
X (g)X (h) = (g h).
The sum is over dominant weights

only and the delta-function (g h) is defined with

respect to the Haar measure dG (a) F (a)(a b) = F (b). The -function is defined on
conjugacy classes, i.e., it vanishes unless g and h belong to the same conjugacy class.
A.1. The modular S- and T -matrix
The characters that appear in the partition functions, i.e., the characters of the chiral
algebra A, form a representation of the modular group, generated by T : + 1 and
250
S : 1/ :
( + 1) =
T M M ( ),
(A.7)
S M M ( ).
(A.8)
(1/ ) =

M
The T matrix is diagonal and equals

c
,M
T,M = exp 2i h
24

1 kd
(, + 2)
= exp 2i
,M .
2h
24 h
The modular S-matrix is unitary and symmetric and equals

2i
w( + ), M + ,
S,M = NS
sign(w) exp
h
(A.9)
(A.10)
(A.11)
wW
with normalization

NS = e
2i
dr
8

Lw 1/2 1

.
L
hr
(A.12)
Here |Lw /L | is the number of weights inside the unit cell of the coroot lattice.
Comparison with the Weyl character formula Eq. (A.3) shows that the modular S-matrix
is proportional to a character of the horizontal sub-algebra

2i
(M + , H )
w(), M +
X
S,M = NS
sign(w) exp
h
h
wW

(M + , H )
.
= S0,M+ X
(A.13)
h
Note that
if we express M + in Dynkin basis and H in Chevalley basis, then (M +
, H ) i (M + )i Cij1 H j .
References
[1] J.L. Cardy, Boundary conditions, fusion rules and the verlinde formula, Nucl. Phys. B 324 (1989) 581.
[2] N. Ishibashi, The boundary and crosscap states in conformal field theories, Mod. Phys. Lett. A 4 (1989) 251.
[3] A. Sagnotti, in: Non-Perturbative Quantum Field Theory, G. Mack et al. (Eds.), Pergamon Press, 1988,
p. 521;
M. Bianchi, A. Sagnotti, On the systematics of open string theories, Phys. Lett. B 247 (1990) 517;
M. Bianchi, A. Sagnotti, Twist symmetry and open string Wilson lines, Nucl. Phys. B 361 (1991) 519;
G. Pradisi, Open superstrings, Nuovo Cimento B 112 (1997) 467, hep-th/9603104;
A. Sagnotti, Y.S. Stanev, Open descendants in conformal field theory, Fortsch. Phys. 44 (1996) 585;
A. Sagnotti, Y.S. Stanev, Nucl. Phys. Proc. Suppl. B 55 (1996) 200, hep-th/9605042.
251
[4] J. Polchinski, Y. Cai, Consistency of open superstring theories, Nucl. Phys. B 296 (1988) 91;
P. Horava, Strings on worldsheet orbifolds, Nucl. Phys. B 327 (1989) 461;
J. Dai, R.G. Leigh, J. Polchinski, New connections between string theories, Mod. Phys. Lett. A 4 (1989)
2073.
[5] I. Brunner, M.R. Douglas, A.E. Lawrence, C. Romelsberger, D-branes on the quintic, JHEP 0008 (2000)
015, hep-th/9906200.
[6] J. Polchinski, Dirichlet branes and RamondRamond charges, Phys. Rev. Lett. 75 (1995) 4724, hepth/9510017.
[7] For a review, see: A. Dabholkar, Lectures on orientifolds and duality, hep-th/9804208.
[8] S. Forste, D. Ghoshal, S. Panda, Phys. Lett. B 411 (1997) 46, hep-th/9706057.
[9] A. Sen, Orientifold limit of F-theory vacua, Phys. Rev. D 55 (1997) 7345, hep-th/9702165;
A. Sen, F-theory and orientifolds, Nucl. Phys. B 475 (1996) 562, hep-th/9605150.
[10] L.R. Huiszoon, A.N. Schellekens, N. Sousa, Klein bottles and simple currents, Phys. Lett. B 470 (1999) 95,
hep-th/9909114.
[11] J. Fuchs, L.R. Huiszoon, A.N. Schellekens, C. Schweigert, J. Walcher, Boundaries, crosscaps and simple
currents, Phys. Lett. B 495 (2000) 427, hep-th/0007174.
[12] A.Y. Alekseev, V. Schomerus, D-branes in the WZW model, Phys. Rev. D 60 (1999) 061901, hepth/9812193.
[13] G. Felder, J. Frohlich, J. Fuchs, C. Schweigert, The geometry of WZW branes, J. Geom. Phys. 34 (2000)
162, hep-th/9909030.
[14] S. Stanciu, D-branes in group manifolds, JHEP 0001 (2000) 025, hep-th/9909163.
[15] J. Maldacena, G. Moore, N. Seiberg, Geometrical interpretation of D-branes in gauged WZW models,
JHEP 0107 (2001) 046, hep-th/0105038.
[16] E. Witten, Nonabelian bosonization in two dimensions, Commun. Math. Phys. 92 (1984) 455.
[17] P. Goddard, D.I. Olive, KacMoody and Virasoro algebras in relation to quantum physics, Int. J. Mod. Phys.
A 1 (1986) 303.
[18] J. Fuchs, Affine Lie Algebras and Quantum Groups, Cambridge Univ. Press, 1992;
J. Fuchs, C. Schweigert, Symmetries, Lie Algebras and Representations, Cambridge Univ. Press, 1997.
[19] S. Helgason, Differential Geometry Symmetric Spaces, Academic Press, 1962.
[20] C. Bachas, private communication, April, 2001;
C. Bachas, N. Couchoud, P. Windey, hep-th/0111002.
[21] A. Sagnotti, private communication.
[22] D. Gepner, E. Witten, String theory on group manifolds, Nucl. Phys. B 278 (1986) 493.
[23] D. Bernard, String characters from KacMoody automorphisms, Nucl. Phys. B 288 (1987) 628;
D. Altschuler, J. Lacki, P. Zaugg, The affine Weyl group and modular invariant partition functions, Phys.
Lett. B 205 (1988) 281;
G. Felder, K. Gawedzki, A. Kupiainen, Spectra of WessZuminoWitten models with arbitrary simple
groups, Commun. Math. Phys. 117 (1988) 127;
G. Felder, K. Gawedzki, A. Kupiainen, The spectrum of WessZuminoWitten models, Nucl. Phys. B 299
(1988) 355.
[24] A.N. Schellekens, S. Yankielowicz, Simple currents, modular invariants and fixed points, Int. J. Mod. Phys.
A 5 (1990) 2903.
[25] K. Intriligator, Bonus symmetry in conformal field theory, Nucl. Phys. B 332 (1990) 541.
[26] J. Fuchs, Simple WZW currents, Commun. Math. Phys. 136 (1991) 345.
[27] G. Pradisi, A. Sagnotti, Ya.S. Stanev, Planar duality in SU(2) WZW models, hep-th/9503207;
G. Pradisi, A. Sagnotti, Ya.S. Stanev, Phys. Lett. B 354 (1995) 279.
[28] L.R. Huiszoon, dissertation, to appear.
[29] P. Di Vecchia, M. Frau, I. Pesando, S. Sciuto, A. Lerda, R. Russo, Classical p-branes from boundary state,
Nucl. Phys. B 507 (1997) 258, hep-th/9707068.
[30] B.C. Berndt, R.J. Evans, K.S. Williams, Gauss and Jacobi Sums, Wiley, 1998.
[31] J. Fuchs, C. Schweigert, Branes: from free fields to general backgrounds, Nucl. Phys. B 530 (1998) 99,
hep-th/9712257.
[32] J. Cardy, D. Lewellen, Bulk and boundary operators in conformal field theory, Phys. Lett. B 259 (1991) 274.
[33] J. Walcher, Worldsheet boundaries, supersymmetry and quantum geometry, dissertation. ETH No. 14225.
252
[34] I. Brunner, On qrientifolds of WZW models and their relation to geometry, hep-th/0110219.
[35] L. Jeffrey, Commun. Math. Phys. 147 (1992) 563.

Mesoscopic QCD and the -vacua

Jonathan T. Lenaghan , Thomas Wilke
The Niels Bohr Institute, Blegdamsvej 17, DK-2100 Copenhagen , Denmark
Received 24 August 2001; accepted 12 December 2001
Abstract
The partition function of QCD is analyzed for an arbitrary number of flavors, Nf , and arbitrary
quark masses including the contributions from all topological sectors in the LeutwylerSmilga
regime. For given Nf and arbitrary vacuum angle, , the partition function can be reduced to Nf 2
angular integrations of single Bessel functions. For two and three flavors, the dependence of the
QCD vacuum is studied in detail. For Nf = 2 and 3, the chiral condensate decreases monotonically
as increases from zero to and the chiral condensate develops a cusp at = for degenerate quark
masses in the macroscopic limit. We find a discontinuity at = in the first derivative of the energy
density with respect to for degenerate quark masses. This corresponds to the first-order phase
transition in which CP is spontaneously broken, known as Dashens phenomena. 2002 Published
by Elsevier Science B.V.
PACS: 12.38.Lg
1. Introduction
For Nf flavors of massless quarks, the Lagrangian of quantum chromodynamics (QCD)
is invariant under the global symmetry group SU(Nf )R SU(Nf )L U (1)A at the
classical level. At the quantum level, however, the SU(Nf )R SU(Nf )L symmetry is
spontaneously broken to the diagonal subgroup of vector transformations, SU(Nf )R+L =
SU(Nf )V , by an expectation value for the quarkantiquark condensate. The U (1)A
symmetry is explicitly broken to Z(Nf )A by a nonvanishing topological susceptibility [1].
For scales well below , the typical hadronic mass scale, the effects of the explicit breaking
of the U (1)A symmetry can be essentially ignored and the dynamics of the theory are
dominated by the Nf2 1 Goldstone bosons that arise from the spontaneous breaking of
* Corresponding author.
E-mail addresses: lenaghan@nbi.dk (J.T. Lenaghan), wilke@nbi.dk (T. Wilke).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 3 5 - 6
254
J.T. Lenaghan, T. Wilke / Nuclear Physics B 624 (2002) 253282
the chiral symmetry. For small, nonzero quark masses, these excitations become pseudoGoldstone bosons.
In addition to the quark masses, the QCD Lagrangian depends on another parameter,
, the so-called vacuum angle. Though this parameter explicitly violates the discrete CP
symmetry for all noninteger multiples of , experimental results for the value of the
neutron dipole moment constrain to be zero within a deviation of less than 109 [2,
3]. The reason why = 0, however, is still poorly understood, and there is considerable
theoretical interest in the physics of QCD at = 0. The most well-known example is the
VenezianoWitten formula which in the large-Nc limit relates the mass of the meson
to the second derivative of the pure YangMills vacuum energy with respect to [4,
5]. Another phenomenologically motivated example is the work of Refs. [69] in which
it is argued that metastable states in which is effectively nonzero may be created in
ultrarelativistic heavy ion collisions.
Physics at = 0 is inherently nonperturbative and one must rely either on effective theories or perform lattice gauge theory simulations [10]. The latter approach is prohibitively
difficult at present because at nonzero values of the action is complex and standard importance sampling methods that are usually employed are no longer applicable. This difficulty
is similar to that for nonzero values of the baryonic chemical potential, but there has been
recent progress in solving the complex action problem in simpler models [11,12]. While
many observables associated with the quark masses have been studied extensively in lattice
gauge theory computations, properties of the QCD vacuum associated with such as the
topological susceptibility have only recently been given extensive attention by the lattice
community [1320]. So, most of our knowledge of the physics of QCD at = 0 has been
gleaned from effective theories. One of the most striking examples of novel physics is the
emergence of two CP violating degenerate vacua separated by a potential energy barrier at
= known as Dashens phenomenon [21]. While QCD is seemingly invariant under the
discrete CP group for equal to integer values of , this invariance is spontaneously broken in a first-order phase transition at = . Chiral effective Lagrangians in the large-Nc
limit were used to investigate this and other properties of the -vacua in Refs. [22] and [23].
Subsequent elaborations and refinements using this approach were made in Refs. [2429].
Additionally, random matrix models [30,31] as well as numerical simulations of CP N1
models [3235] have also been employed to investigate the physics of nonzero values of .
In this work, we investigate the properties of QCD at = 0 in the LeutwylerSmilga
finite volume regime [36], also known as the mesoscopic regime. Here, chiral perturbation
theory is valid but the volume is taken such that the Goldstone modes are constant and their
kinetic term can be ignored. This approach differs from those of Refs. [2229], in which
the large-Nc is taken and the Lagrangian is studied to lowest order in the chiral fields.
The LeutwylerSmilga regime, on the other hand, enables exact, analytical predictions to
be made. Indeed, this limit is realized in many lattice simulations [37]. Many of the lowenergy aspects of QCD have been studied using effective chiral Lagrangians in a finite
volume [38]. In Ref. [36], Leutwyler and Smilga demonstrated that such theories contain
information beyond pion dynamics and the spontaneous breaking of chiral symmetry,
including the low-lying spectrum of the QCD Dirac operator and the topology of the gauge
fields. The purpose of this work is twofold. First, we reduce the QCD partition function
for Nf 3 in the LeutwylerSmilga scaling regime including the contributions from all
255
topological sectors to Nf 2 angular integrations over single Bessel functions. The latter
half of this work is devoted to examining the vacuum properties of QCD in the Leutwyler
Smilga scaling regime at nonzero values of .
The partition functions for one and two flavors of quarks including contributions from
all topological sectors were computed in Refs. [39] and [40]. As emphasized in Ref. [41],
the contributions to the partition function from sectors with nonzero topological charge are
crucial in the finite volume regime. For example, if one considers the partition function
at any fixed topological charge (with the exception of = 0), then chiral condensate
diverges as one of the quark masses approaches zero. This completely unphysical behavior
is remedied after summing over all topological charges, a procedure which becomes
significantly more complicated for three or more quark flavors. We show in the following
that the summation over all topological sectors can be performed for arbitrary Nf and
with the reduction of the partition function to Nf 2 angular integrations over single
Bessel functions.
This paper is organized as follows. In Section 2, we briefly review the reduction of
the QCD partition function to a finite dimensional group integral in the LeutwylerSmilga
scaling regime. In Section 3, we review the summation over all topological charges for the
partition function for Nf =1 and 2. We then calculate the summation for arbitrary Nf and
quark masses. While the full partition function for Nf = 2 has been known for some time
[36], there has not been a detailed study of the vacuum properties of this theory for = 0.
In Section 4, we examine the dependence of the chiral condensate, chiral susceptibility,
topological density and topological susceptibility for two flavors. In Section 5, we extend
this discussion to three flavors. Finally, we give concluding remarks in Section 6.
2. Partition functions in the LeutwylerSmilga regime

Consider a SU(Nc ) gauge theory with Nf flavors of fermions in the fundamental
representation on a four-dimensional torus of volume V = L4 . The Lagrangian is
Nf

1 a a
a a
L = 2 F
F
/ mf )f i F
f (iD
F ,
4g
(1)
f =1
a
a = 1

where F
2 F . The last term in Eq. (1) is a total derivative and so does not affect
the field equations or any of the perturbative aspects of the theory. The integral of this term
over the four-volume is quantized and is given by

1
a
a
(x)F
(x) Z.
d 4 x F
=
(2)
32 2
This quantity, known as the topological charge, contributes a phase factor to the path
integral and is associated with transitions between topologically nontrivial gauge field
configurations. The full partition function is given by a weighted sum over the partition
function for each topological sector:

(N )
ei Z f {mk } .
Z (Nf ) , {mk } =
=
(3)
256
The individual contributions are

(Nf )

{mk } =

[dA]
Nf

det iD[A]
/ mf exp d 4 x LY M [A] ,
f =1
(4)
where the integration is taken only over gauge field configurations with topological
charge .
In principle, all of the observables of QCD can be calculated by evaluating Eq. (3)
and its various functional derivatives. Typically, however, this can only be done in some
approximation since the integration is taken over an infinite dimensional functional space
and there are ultraviolet and infrared divergences. Formulating the theory on a discrete,
Euclidean lattice has the advantage of eliminating the ultraviolet divergences on account of
a nonzero lattice spacing, a, and the infrared divergences on account of a necessarily finite
volume. A second approach is to analyze the partition function for the effective theory
which describes the low-energy behavior. For QCD, this is chiral perturbation theory.
Restricting the Euclidean four-volume, V = L4 , to the range
1
1
L
,
(5)
where is the typical hadronic scale and m is the mass of the Goldstone excitations,
results in tremendous simplifications and exact analytical predictions are possible [36,38].
This is possible since the lower limit ensures that the partition function is dominated by
Goldstone modes and the upper limit ensures that these modes are constant, i.e., that the
kinetic term in the partition function factorizes from the mass dependent term.
There have been a number of advances in recent years towards evaluating the QCD finite
volume partition function. With the four-volume taken according to Eq. (5), the partition
function was shown in Refs. [36,38] to reduce to the finite dimensional group integration

(Nf )
Z
(6)
, {mi }, V =
dU exp V Re ei/Nf Tr MU ,
SU(Nf )
where M = diag(m1 , . . . , mNf ) is the mass matrix for the quark fields and is the chiral
condensate in the chiral and infinite volume limit. We shall henceforth refer to as the
macroscopic chiral condensate. The integration is taken over the group manifold of the
Goldstone modes, SU(Nf ). Note that the dependence on the volume, the macroscopic
chiral condensate and the quark masses is only through the dimensionless scaling variable
i = V mi .
(7)
The Fourier coefficients conjugate to Z (Nf ) (, {i }) are obtained by taking the Fourier
transformation of Eq. (3),

(N )
Z f {i } =
1
2
2
0

d Z (Nf ) , {i } ei .
(8)
257
They have been computed for arbitrary Nf and Nc 3 [39,42], and can be expressed as
det A ({i })
,
{i }, V =
({2i })
(Nf )
(9)
where
I (1 )

..
A {i } =
.
I (Nf )
1 I+1 (1 )
..
.
Nf I+1 (Nf )
..
.
I+Nf 1 (1 )
..
.
Nf 1
Nf I+Nf 1 (Nf )
N 1
1 f
is a Nf Nf matrix and In is a modified Bessel function of order n. The denominator,

2

2i =
(10)
i 2j ,
i>j
is the Vandermonde determinant. For equal masses, i = , Z () can be further

simplified using properties of Bessel functions and determinants to give
I ()

I ()
1

(N )
Z f () =
..

.

INf +1 ()
I+1 ()
I ()
..
.
INf +2 ()
..
.
I+Nf 1 ()
I+Nf 2 ()
.
..

.

I ()
(11)
Combinatorial formulas for generating the character expansions of the U (N) group were
recently given in Ref. [43] and used to very efficiently derive Eq. (9). The close connection
to random matrix theory was firmly established in Refs. [4446].
The vacuum properties of the theory are defined by Eq. (6) and its derivatives. Recall
that the first derivative of the logarithm of the partition function with respect to a parameter
determines the mean value of the conjugate variable to the parameter, while the second
derivative serves as a measure for the fluctuations around the average value. We focus on
the derivatives with respect to the quark masses, mi , and the vacuum angle, . The chiral
condensate for the ith quark species is

(N )
(i) f , {mk }, V =

1
log Z (Nf ) , {mk }, V ,
V Nf mi
(12)
with the average value

N
(Nf )
f

(N )
, {mk }, V =
(i) f , {mk }, V .
(13)
i=1
This quantity should approach macroscopic chiral condensate, , at = 0 and for large
values of the scaling variable, . The chiral (scalar) susceptibility is defined as
(Nf )
ij

1

2
, {mk }, V =
log Z (Nf ) , {mk }, V .
V mi mj
(14)
258
Vacuum properties associated with the vacuum angle, , can be analogously defined.
The topological density is

1
log Z (Nf ) , {mk }, V ,
, {mk }, V =
(15)
V
which at = 0 has the interpretation of the mean topological charge. From Eqs. (3) and
(9), one can see that the topological density vanishes if is an integer multiple of . The
topological susceptibility is defined by

1 2
log Z (Nf ) , {mi }, V ,
top , {mi }, V =
(16)
V 2
which at = 0 is the mean square deviation of the topological charge and is in general
nonzero.
3. Summation over topological charges

When studying the vacuum properties, e.g., the chiral condensate, in the Leutwyler
Smilga regime, the necessity of including the contributions to the partition function from
every topological sector was established in Ref. [41]. For one and two quark flavors,
the summation can be performed using standard Bessel function identities. For Nf 3,
however, the summation is more complicated and to the best of our knowledge has not
been performed previously in the literature. It is instructive to review the derivation of the
results for one and two flavors before proceeding to the derivation for Nf 3.
3.1. Nf = 1
For Nf = 1, the summation over all topological sectors is straightforward to calculate.
Using the definition of the generating function for modified Bessel functions [47],

x
t + t 1 =
exp
(17)
t I (x),
2
=
the summation can be performed:
Z (Nf =1) (, ) =

i
ei I () = exp
e + ei = exp[ cos ].
2
=
(18)
This result was first found in Ref. [36].

3.2. Nf = 2
For two flavors, the calculation is more involved but the method generalizes very
naturally to Nf 3. From Eq. (9), the partition function is
Z (Nf =2) (, 1 , 2 )

1
ei 2 I (1 )I+1 (2 ) 1 I (2 )I+1 (1 ) .
= 2
2
2 1 =
(19)
259
Since the partition function is symmetric in the two scaling variables, we consider only the
first term between the brackets in Eq. (19). Following Ref. [47], we use the contour integral
representation for I (x) which follows from Eq. (17),

ds 1
x
I (x) =
(20)
s
exp
s + s 1 ,
2i
2
where the contour is the standard Bessel function contour surrounding the negative real axis
and the origin in a counterclockwise fashion. Using this representation, the summation can
be calculated:
ei I (1 )I+1 (2 )
ei I (1 )
=

=

ds 2
2
s
exp
s + s 1
2i
2

i 1
2
s 2 exp
I (1 )
e s
s + s 1
2

1 i
2
s 2 exp
s + s 1 exp
se
+ s 1 ei .
2
2

ds
2i
ds
2i
(21a)
(21b)
(21c)
By making the change of variables,

=
1 =
(2 + 1 ei )
s
12 ( )
(22a)
(2 + 1 ei ) 1
s ,
12 ( )
(22b)
where
12 ( )
21 + 22 + 21 2 cos ,
(22c)
Eq. (21a) can then be simplified to yield

=
2 + 1 ei
I (1 )I+1 (2 ) =
12 ( )
=

12 ( )
d 2
1
exp
+
2i
2

2 + 1 ei
I1 12 ( ) .
12 ( )
(23)
Combining the two contributions, the partition function for Nf = 2 is

Z (Nf =2) (, 1 , 2 ) =
I1 (12 ( ))
.
12 ( )
(24)
Eq. (22c) defines a reduced mass determined by a triangle law. The triangle has sides 1 ,
2 and 21 and the angle subtended by 1 and 2 is .
260
3.3. Nf 3
For Nf 3, we consider cases of degenerate and nondegenerate quark masses
separately. In the case of nondegenerate quark masses, the summation over topological
charge is a generalization of the summation for Nf = 2. In the process of the derivation,
we find interesting relationships between sums of products of arbitrarily many Bessel
functions and angular integrations over single Bessel functions.
3.3.1. Nondegenerate quark masses
From Eq. (9), the quantity that needs to be calculated is a summation of products of Nf
modified Bessel functions weighted by a phase, since
Nf

j 1
ij I+j 1 (ij ).
det A {i } = i1 ,...,iNf
(25)
j =1
Each term in the summation is of the form
B=
ei I+m1 (x1 )I+m2 (x2 ) I+mNf (xNf )
= eim1
ei I (x1 )I+n2 (x2 ) I+nNf (xNf )
im1
C,
(26)
where nj = mj m1 . For the calculation of the partition function, B is actually more

general than necessary and the particular choices of xj = ij and mj = j 1 specialize to
the appropriate terms in the partition function.
By expressing all but the first of the modified Bessel functions by their contour integral
representations, C can be written as
C=

=
ei I (x1 )

Nf

xj
dsj nj 1
sj
sj + sj1
exp
.
2i
2
(27)
j =2
The definition of the generating function for In (x), Eq. (17), can be used to perform the
summation over :
Nf

s2 sNf
xj
dsj nj 1
ei
x1
1
s
C=
.
exp
+
exp
sj + sj
2i j
2
2 s2 sNf
ei
j =2
(28)
We now make a change of variables which is suggested by Eqs. (22a) and (22b):

s3 sNf s2
= x2 + x1
(29)
,
ei
1
s2
ei
1
= x2 + x1
(30)
,
s3 sNf
261
where

= x12
+ x22
+ x1 x2
s3 sNf
ei
+
s3 sNf
ei
(31)
The contour integration over s2 is rewritten as

s3 sNf
1
ei
s2
ds2 n2 1
s2
x2 + x1
+
x
exp
+
x
2
1
2i
2
ei
2s2
s3 sNf

=
x2 + x1 s3 sNf ei
n2
x2 + x1 s3 sNf ei

d n2 1

1
exp
+
2i
2
n2
In2 ().
(32)
The expression for C is then

C=
Nf

x2 + x1 s3 sNf ei n2
dsj nj 1
xj
sj
sj + sj1
exp
In2 ().
2i
2
j =3
(33)
Making the second change of variables, sk = eik , for k = 3, . . . , Nf , and deforming the
contours to the unit circle, the final expression for C is obtained:
C=
Nf
2
0
j =3

dj
exp xj cos(j ) inj j
2
In2
x2 + x1 e
n2
i(3 ++Nf )
x22 + x12 + 2x2 x1 cos(3 + + Nf )

x22 + x12 + 2x2 x1 cos(3 + + Nf ) .
Collecting the various contributions, the partition function for Nf quark flavors is

Z (Nf ) , {i } =
1
i1 ,...,iNf
({2i })
Nf

i2 S(i1 , i2 ; i3 , . . . , iNf ), (34)
k1
ik
k=1
where
S(i1 , i2 ; i3 , . . . , iNf )
=
Nf
2
0 k=3

I1 (i1 i2 ( ; 3, . . . , Nf ))
dk
exp i(k 1)k + ik cos(k )
, (35)
2
i1 i2 ( ; 3, . . . , Nf )
262
and
i1 i2 (, 3 , . . . , Nf ) =
2i1 + 2i2 + 2i1 i2 cos(3 + + Nf ).
(36)
Note that S(i1 , i2 ; i3 , . . . , iNf ) is symmetric under the interchange of i1 and i2 .

In the case of complete nondegeneracy, i = j for all i = j , the partition function and
its derivatives are analytic functions to all orders. Any potential nonanalyticities in Eq. (35)
and its derivatives occur when i1 i2 ( ; 3 , . . . , Nf ) vanishes which is only possible if
at least two quark masses become degenerate.1 This becomes apparent when Eq. (36) is
rewritten in the form

lim i1 i2 (, 3 , . . . , Nf ) = 2 + 2 cos(3 + + Nf )
i1 ,i2

3 + + Nf

= 2cos
(37)

2
which has a cusp at 3 + + Nf = . This behavior is a necessary condition for the
derivatives of Eq. (34) to have discontinuities. On account of the integration, this is not a
sufficient condition.
3.3.2. Degenerate quark masses
The limit of equal quark masses, i , can be calculated either by summing over
all topological sectors for nondegenerate quark masses and then taking the limit i ,
or starting from the equal mass partition function in a given topological sector, Eq. (11)
and then summing over all topological charges. Using the first approach, one can derive
equal mass limit from Eqs. (34)(36) and the resulting expression for the partition function
involves derivatives up to order Nf 1 of modified Bessel functions with respect to
the masses. This approach is quite cumbersome and, moreover, the derivatives are by
construction taken at the cusps given by Eq. (37). The second approach is more tractable
and is the one used here. After expanding the determinant in Eq. (11),
(N )
Z f () = i1 ,...,iNf
Nf

I+j ij (),
(38)
j =1
and using the contour integral representation for each Bessel function, the expression for
the equal mass partition function in a sector of given topological charge becomes

Nf

j 1
iN
dsj

(N )
exp
sj + sj1 sj
Z f () =
i1 ,...,iNf s1i1 sNff
2i
2
j =1
Nf
2
0 j =1

iNf iN
dj
f .
exp cos(j ) ij ij j i1 ,...,iNf eii1 e
2
(39)
1 This is true for finite . As shown for N = 2 in Ref. [48], there can be a nonanalyticity if the two scaling
i
f
(1 2 )2
variables are taken to infinity such that lim1 ,2
0. We expect that this limit can be generalized
1 2
to arbitrary Nf in a straightforward manner.
263
The last equality is obtained by change of variables, sj = eij , deforming the integration
contours to the unit circle, and making use of the identity
k1 ,...,kNf eik1 e
iNf kN
= k1 ,...,kNf eik1 1 e
ikNf Nf
(40)
The authors of Ref. [36] showed using a result of Weyl that

(N )
Z f () =
1
Nf !
Nf
2
0 j =1

i
dj
e k eil 2 .
exp cos(j ) + ij
2
(41)
k<l
Eqs. (39) and (41) can be transformed into one another by observing that
i

iNf iN
f = e ij
i1 ,...,iNf eii1 e
=
e k eil
(42)
k<l
is a Vandermonde determinant, and so

eik eil 2 = eij eij .
(43)
k<l
By expanding ({eij }), one can transform both expressions into one another by an
appropriate relabeling of the integration variables.
The summation over the topological charges can be performed using
2
0

d1
i(1 Nf )
f (1 , . . . , Nf , )
e
2
=
= f (2 + + Nf , 2 , . . . , Nf , ),
(44)
provided f is 2 -periodic in 1 . For the subsequent integration over 2 , an integral

representation of the Bessel functions is used. The details of the calculation are given in
the appendix. The result is
Z (Nf ) (, )

Nf 2

dm cos(m ) Ia1 a2 ((3 , . . . , Nf , ))
e
= a1 ,...,aNf (a2 a1 )
2
(3 , . . . , Nf , )
m=3 0

cos (3 a1 a2 )( + 3 + + Nf )

/2 (3 a3 )3 (Nf aNf )Nf ,
(45)
where

(3 , . . . , Nf , ) = 2 + 2 cos (3 + + Nf ).
(46)
The summation over the completely antisymmetric tensor can be simplified to give a
(Nf 2)-fold integration over a sum of Nf 1 Bessel functions multiplied by some
264
phases,
Z
(Nf )
(, ) = 2e
i/2
Nf
2
0 m=3
Nf 1
r(1)r
Ir ((3 , . . . , Nf , ))
(3 , . . . , Nf , )
r=1
(Nf )
The phases r
dm cos(m )i(m3/2)m
e
2
(Nf )
(3 , . . . , Nf , ).
(47)
(3 , . . . , Nf , ) are given by
Nf r
(N )
r f (3 , . . . , Nf , ) =

j =1
i(j +r/2)(3 ++Nf ) (Nf )

j,j +r (3 , . . . , Nf ),
(48)
with
(N )
k,l f (3 , . . . , Nf )

1 ei(k2)3
.
..
= ..
.
1 ei(k2)Nf

ei(Nf 1)3

..
.
.

ikNf
i(l2)Nf
ilNf
i(Nf 1)Nf
e
e
e
e
(49)
Despite appearances, this last expression is real. As mentioned above, the possible
emergence of a cusp in the integrand occurs in the ratios of the Bessel functions and
(3 , . . . , Nf , ). At this stage, it is not a priori obvious that there are nonanalyticities
after the integrations are performed. Thus, one must consider them on a case by case basis.
This is done for Nf = 2 and 3 in the following sections.
eik3
..
.
ei(l2)3
..
.
eil3
..
.
4. Vacuum properties for Nf = 2

Since the partition function for Nf = 2, Eq. (24), can be expressed in closed form
even after summing over all topological charges, calculating the vacuum observables is
straightforward. Most of the work in the literature has focused upon the behavior of the
chiral condensate and the topological susceptibility as functions of the scaling variable for
either = 0 or in a sector of fixed topological charge. In this section, we demonstrate that
QCD in the LeutwylerSmilga scaling regime exhibits a great deal of nontrivial behavior
at nonzero values of for Nf = 2.
4.1. Partition function
The partition function for two quark flavors is

I1 ( 21 + 22 + 21 2 cos )
Z (Nf =2) (, 1 , 2 ) =
.
21 + 22 + 21 2 cos
(50)
265
At = (2n + 1) , where n is an integer, it reduces to

I1 (|1 2 |)
,
Z (Nf =2) = (2n + 1), 1 , 2 =
|1 2 |
(51)
i.e., the theory at = is equivalent to the theory at = 0 but taking one quark masses to
be negative. This is in accordance with the standard lore. Taking the two quark masses to
be equal, however, leads to the surprising result that the partition function is independent
of the scaling variable,

1
Z (Nf =2) = (2n + 1), = .
2
This can be seen also by starting from the equal mass partition function:
I1 ( 2 + 2 cos )
(Nf =2)
(, ) =
Z
.
2 + 2 cos
(52)
(53)
Since Z (Nf =2) (, 1 , 2 ) is a smooth function of 1 , 2 and , it does not matter in

which order the limits of degenerate masses and = (2n + 1) are taken. This result may
be understood in terms of the microscopic spectral density. As shown in Ref. [41], after
summing over all topological charges, the full microscopic spectral density is related to a
particular quotient of the partition function for Nf flavors at and the partition function
for Nf + 2 flavors at + . Thus, the partition function for two flavors at = is related
to the quenched (Nf = 0) theory at = 0. This fact also demonstrates at a qualitative level
that the partition function becomes independent of the quark masses if and only if Nf = 2
and = [49]. It is apparent already at the level of the partition function that, at least in
the LeutwylerSmilga scaling regime, QCD is very different qualitatively at = n than
at all other values of .
This can also be understood directly from chiral perturbation theory. For Nf = 2, the
leading order term in chiral perturbation theory, Re[Tr{Mei/Nf U }], vanishes at =
when M = m1 since the trace of an SU(2) matrix is always real. This term, however, is
the only relevant term in the LeutwylerSmilga scaling regime as can be seen from Eq. (6).
From this observation, it is clear that the partition function at = is simply a constant in
this case. The resolution to this problem was given in Ref. [24] where higher order terms
quadratic in masses were included in the effective chiral Lagrangian. A more extensive
analysis in terms of chiral perturbation theory was given in Refs. [28,29]. By including
higher order terms in the chiral expansion, the energy density then depends once again
explicitly on the quark masses. These terms, however, are suppressed in the Leutwyler
Smilga regime and so we do not include them.
The most fundamental quantity that can be derived from the partition function is the
energy density which is defined by
1
log Z (Nf =2) (, 1 , 2 ).
V
An expansion for large 12 ( ) gives
E(, 1 , 2 ) =
E(, 1 , 2 ) =

12 ( )
+ O log 12 ( ) .
V
(54)
(55)
266
Fig. 1. The energy density for Nf = 2 in the macroscopic limit. The lower curve corresponds to degenerate
quark masses, while the upper curve corresponds to an infinite mass splitting. The region between the two curves
contains all finite mass splittings. The global minimum of the energy density for all cases occurs at = 0.
Note that the volume dependence drops out in the leading order term. In the macroscopic
limit, 12 ( ) 1,

m1 m2 cos + O(m2 /m1 ), m1 m2 ,
lim E(, m1 , m2 , V ) =
(56)
2m| cos(/2)|,
m1 = m2 = m.
V
There is a constant shift, E0 = m1 , in the energy density which we subtract in the
following. The energy density in both limits is plotted in Fig. 1. It is minimized at = 0
and maximized at = . For degenerate quark masses, there is cusp at = .
4.2. Chiral condensate
The chiral condensate is calculated from Eq. (12) and for the first quark flavor is
(N =2)
(1)f
(, 1 , 2 ) =
(1 + 2 cos ) I2 (12 ( ))
.
2
12 ( )
I1 (12 ( ))
(57)
The condensate for the second quark flavor is given by the same expression with 1 and
2 interchanged. The total chiral condensate given by the sum is
(Nf =2) (, 1 , 2 ) =
(1 + 2 )(1 + cos ) I2 (12 ( ))

.
2
12 ( )
I1 (12 ( ))
(58)
For = 0, the individual contributions are equal. For = , however, they are equal in
magnitude but opposite in sign:
(N =2)
(1)f
(, 1 , 2 ) = sign(1 2 )
I2 (12 ())
(N =2)
= (2)f (, 1 , 2 ), (59)
2 I1 (12 ())
267
with their sum, the total chiral condensate, vanishing for any value of the quark masses.
Taking of the limit of degenerate quark masses, the chiral condensate is
2 + 2 cos I2 ( 2 + 2 cos ) (Nf =2) (, )

(Nf =2)
.
=
(1,2) (, ) =
(60)
4
2
I1 ( 2 + 2 cos )
Three observations may be made based upon Eq. (60) about the physics of QCD in the
LeutwylerSmilga scaling regime at nonzero values of :
For fixed , the chiral condensate decreases monotonically in the interval [0, )
and increases monotonically for (, 2).
As = mV , the chiral condensate develops a cusp at = .
For any value of , the chiral condensate vanishes identically at = .
This behavior is shown in Figs. 2 and 3. The last observation follows of course from the
constancy of the partition function for Nf = 2 at = and is not expected to be true
outside the LeutwylerSmilga regime.
For small values of the scaling variable, the chiral condensate has the expansion

cos4 (/2) 3 cos6 (/2) 5
(Nf =2) (, ) cos2 (/2)
=
+
+ O 7 .
(61)
2
12
48
As expected, the chiral condensate is linear to lowest order in the quark mass and vanishes
term-by-term at = (2n + 1) . The expression for the chiral condensate simplifies in the
limit of degenerate quark masses and in the limit of large quark mass splitting:
(Nf =2) (, m1 , m2 , V )

m1 )
O(m2 /m1 ),
cos2 (/2) II21 (V
(V m1 ) +
=
I2 (mV 2+2 cos )
,
2 2 + 2 cos
I1 (mV
2+2 cos )
m1 m2 ,
m1 = m2 = m.
(62)
The volume dependence of the chiral condensate in these two limits is easily calculated.
In the limit of small volume, the chiral condensate vanishes in both limits of degenerate
quark masses and large mass splittings. In the macroscopic limit, V 1/m, the chiral
condensate is
lim (Nf =2) (, m1 , m2 , V )

cos2 (/2) + O(m2 /m1 ), m1 m2 ,
=
| cos(/2)|,
m1 = m2 = m;
(63)
and is independent of the quark masses in both cases. Any nonzero quark mass splitting,
m1 = m2 , interpolates between these two cases as shown in Fig. 2. The value of the chiral
condensate is remarkably insensitive to the magnitude of the quark mass splitting.
In Fig. 3, the chiral condensate as a function of the scaling variable is plotted for =
0, /2, and . As increases from zero to , the condensate decreases monotonically
and vanishes identically at = . The monotonic decrease in the chiral condensate can be
understood from the underlying QCD Lagrangian [49]. If one performs a chiral rotation
5 . For
by an angle , then and the chiral condensate,
rotates into
268
Fig. 2. The chiral condensate in the macroscopic limit as a function of normalized to the infinite volume value,
. The lower curve corresponds to asymptotically large quark mass splitting and the upper curve is limit of
degenerate quark masses. The area between the two curves contains all finite mass splittings between the quark
masses. There is a cusp at = for equal quark masses.
Fig. 3. The chiral condensate for Nf = 2 as a function of . The upper, middle and lower curves correspond to
= 0, /2 and , respectively. As increases from zero to , the condensate decreases and vanishes identically
at = .
269
Nf = 2, the (scalar) chiral condensate is completely rotated into the pseudoscalar chiral
condensate at = . This can be demonstrated analytically by introducing a complex mass
matrix as a source for both the scalar and pseudoscalar chiral condensates.
4.3. Chiral susceptibility
The chiral susceptibility has been previously studied in Ref. [50] but the analysis was
restricted to fixed values of . As shown in Ref. [41], however, the contributions from all
topological sectors can be important in the mesoscopic regime. Defining

I3 (12 ( )) I2 (12 ( ))2
S 12 ( ) =
(64a)
,
I1 (12 ( )) I1 (12 ( ))2

1 I2 (12 ( ))
,
T 12 ( ) =
(64b)
12 ( ) I1 (12 ( ))
the diagonal terms in the chiral susceptibility are

(1 + 2 cos )2
2
11 = V
S 12 ( ) + T 12 ( ) ,
12 ( )2

(2 + 1 cos )2
22 = 2 V
S
(
)
+
T
(
)
.
12
12
12 ( )2
(65a)
(65b)
The off-diagonal terms are

(1 + 2 cos )(2 + 1 cos )
2
S 12 ( ) + cos T 12 ( ) .
12 = 21 = V
12 ( )2
(66)
In the limit of degenerate quark masses, m1 = m2 = m, chiral susceptibilities become

cos2 (/2)S ( ) + T ( ) ,
m

12 = 21 =
cos2 (/2)S ( ) + cos T ( ) .
m
At = , the diagonal and off-diagonal elements are maximally different,
11 = 22 =
2 V
=
,
4m
4
2V
12 ( = ) = 21 ( = ) =
=
.
4m
4
An expansion in small yields

1

cos2 (/2)(2 + cos )3 + O 5 ,

11 = 22 =
m 4 24

1
cos
cos2 (/2)(1 + 2 cos )3 + O 5 .

12 = 21 =
m
4
24
11 ( = ) = 22 ( = ) =
(67a)
(67b)
(68a)
(68b)
(69a)
(69b)
270
On the other hand, in the limit 1 for = ,

3 cos
sin2 (/2)
2
+
,
+
O
1/
m 2| cos(/2)| 8 cos(/2)2

3
sin2 (/2)
2
+
12 = 21 =
+ O 1/ ,
m 2| cos(/2)| 8 cos(/2)2
11 = 22 =
(70a)
(70b)
therefore, to leading order in 1/, 11 = 22 = 12 = 21 . The chiral susceptibility

was shown to be proportional to 3/8m2V at = 0 in Ref. [50]. The importance of summing
over all topological charges is clear in this expression, since the leading order terms in
Eqs. (70a) and (70b) vanish at = 0.
4.4. Topological density: first-order phase transition at =
The topological density, defined by Eq. (15), is
(, 1 , 2 ) =
1 1 2 sin I2 (12 ( ))
.
V 12 ( ) I1 (12 ( ))
(71)
At = 0, (, 1 , 2 ) has the strict interpretation as the mean topological charge. Since in

a large enough ensemble of gauge fields the average number of instantons should be equal
to average number of anti-instantons, the topological density should be zero at = 0, and,
indeed, the right-hand side of Eq. (71) vanishes for equal to any integer multiple of .
In the limit of degenerate quark masses, the topological density becomes
(, ) =
1
I2 ( 2 + 2 cos )
sin
.
V 2 + 2 cos I1 ( 2 + 2 cos )
(72)
The most interesting property of this relation is that in the limit of very large scaling
variable, 1, (, ) develops a discontinuity at = . This is the first-order phase
transition proposed by Dashen [21].
m1 )
m2 sin II2 (V
+ O(m2 /m1 ), m1 m2 ,
1 (V m1 )
(, m1 , m2 , V ) =
(73)
m sin(/2) I2 (2V m cos(/2)) ,
m1 = m2 = m.
I1 (2V m cos(/2))
Then, in the infinite volume limit,

m2 sin + O(m2 /m1 ),
m1 m2 ,
lim (, m1 , m2 , V ) =
m sin(/2) sign(cos(/2)), m1 = m2 = m.
V
(74)
The limit m1 m2 coincides with the one-flavor case. For two flavors of degenerate
quarks, there is a first-order phase transition at = , while for two flavors of
nondegenerate quarks the phase transition disappears as long as the quark masses are kept
finite. As mentioned above, it was shown in Ref. [48] there exists a particular scaling limit
in which there may be a phase transition for nondegenerate quark masses. See Fig. 4.
271
Fig. 4. The topological density of the vacua. The solid line corresponds to the limit where one quark mass is
very heavy, m1 m2 , the long dashed line to mass splitting of m1 /m2 = 2, and the dashed line to the case of
degenerate masses, m1 = m2 = m. A first-order phase transition occurs at = in the case of degenerate quark
masses. The transition is washed out by mass splitting between the quarks. The topological density is measured
in units of m2 , where m2 denotes the mass of the lighter quark.
4.5. Topological susceptibility

The topological susceptibility reads
(Nf =2) (, m1 , m2 , V ) =

1 I2 (12 ( )) 21 22
V I1 (12 ( )) 212 ( )
I2 (12 ( )) I1 (12 ( ))
12 ( ) cos
4
sin2 ( ) .
+
+
1 2
12 ( ) I1 (12 ( )) I2 (12 ( ))
(75)
For equal quark masses, one has

(Nf =2) (, m, V ) =
m 2 cos I2 (( ))
2 | cos(/2)| I1 (( ))

2 m2 V sin2 ( ) I2 (( ))2
1 .
+
4
cos2 (/2) I1 (( ))2
(76)
This reduces at = to
(Nf =2) ( = , m, V ) =
2 m2
V,
4
(77)
272
Fig. 5. The topological susceptibility as a function in the macroscopic limit. The solid, long dashed, and dashed
line represent m1 m2 , m1 /m2 = 2, and m1 = m2 , respectively.
which diverges in the infinite volume limit on account of the first-order phase transition.
For = , however, the limit is finite,
lim (Nf =2) (, m, V ) =

m
cos(/2),
2
= .
(78)
This is completely consistent with the expectation from the flavor singlet WardTakahashi
identity at = 0 which predicts a linear rise in topological susceptibility with mass [51],
lim (Nf ) ( = 0, m, V ) =

m
+ O m2 .
Nf
(79)
Eq. (78) may be considered to be a generalization of Eq. (79) to nonzero values of .

In the limit of large mass splitting, the topological susceptibility reduces to
(Nf =2) (, m1 , m2 , V )
I2 (V m1 )
= m2 cos
+ O(m2 /m1 ),
I1 (V m1 )
m1 m2 ,
(80)
which in the macroscopic limit is

lim (Nf =2) (, m1 , m2 , V ) = m2 cos + O(m2 /m1 ),
See Fig. 5.
m1 m2 .
(81)
273
5. Vacuum properties for Nf = 3

The summation over all topological charges in the partition function for Nf 3 was
performed in Section 3. In this section, we complement the analysis of Section 4 by
examining the dependence of the QCD vacuum for three flavors of quarks. The partition
function for Nf = 3 involves a single integration which we were not able to calculate
analytically. Since the partition function has a grouptheoretic origin, we expect that the
integrals that result after summing over all topological charges provide some functional
representation of the Goldstone manifold. However, since the integrand is a smooth
function of and except possibly at a single point in the interval of integration and the
integral is taken over a compact interval, the numerical evaluation of the partition function
is straightforward. Additionally, the derivatives of the partition function with respect to
and can be commuted past the integration. Explicit expressions are only given for
the chiral condensate and the topological density since the expressions for the chiral and
topological susceptibilities become quite complicated and are not particularly enlightening
even in the limit of degenerate quark masses. We focus on a triplet of quark masses,
(mlight, mlight, mheavy ), between the limits of total degeneracy, mlight = mheavy , and very
large mass splitting, mlight mheavy .
5.1. Partition function
For three degenerate masses, the partition function can be written in the compact form:
Z
(Nf =3)

3 2
e cos
cos
I1 (, )
(, )
2
0

3
I2 (, )
cos
2
2

2
e cos I1 ((, )) I2 ((, ))

2 ,

=
d
cos 3
(, ) cos 3
2
2
2
(, ) =
2
(82)
where (, ) = 2| cos(
2 )|. Unlike the Nf = 2 partition function, we find that
Z (Nf =3) (, ) does not become independent of at = which is consistent with the
discussion in Section 4.1. As discussed in Refs. [28,29], the leading order term in the chiral
expansion does not vanish at = for Nf 3.
The dependence of the energy density as a function of , however, is similar to the two
flavor case. The energy density in the macroscopic limit is shown in Fig. 6 shifted by its
zero point, E0 = mheavy . The global minimum and maximum of the energy density
in the macroscopic limit are at = 0 and , respectively. As long as two quark masses
are degenerate, there seems to be cusp at = . In general, the positions for the global
minimum and maximum are = 0 and , respectively, for any value of . The zero point
energy is dominated by the heavy quark mass. After subtracting this physically irrelevant
zero point contribution, the energy density is relatively insensitive to taking the heavy quark
mass to infinity.
274
Fig. 6. The energy density in the macroscopic limit as a function of . The upper (lower) curve corresponds to
infinite (zero) mass splitting between the two degenerate quark masses and the third quark mass. All finite mass
splittings are contained in the region between the two curves.
5.2. Chiral condensate

For three degenerate quark masses, the chiral condensate is
(Nf =3)
(, ) =
Z (Nf =3) (, )
2
d
0

e cos
a1 I1 (, ) + a2 I2 (, ) ,
(, )
(83a)
where

3
3

cos cos
cos
a1 = cos
,
2
2
2

3
3
3
cos
cos + 2 cos
a2 = cos

.
2
2
2
(83b)
(83c)
In many respects, the behavior of the equal mass chiral condensate for two and three quark
flavors is again similar. The chiral condensate as a function of the scaling variable is plotted
in Fig. 7. The chiral condensate decreases monotonically as increases from zero to like
the Nf = 2 case. However, it is nonzero at = unlike the Nf = 2 chiral condensate. For
small values of , the chiral condensate is linear in and independent of .
Fig. 8 shows the chiral condensate as a function of in the macroscopic limit, .
In this limit, (Nf =3) (, ) has a cusp at = , but otherwise is a smooth function. In the
macroscopic limit with two degenerate quark masses, the value of the chiral condensate is
very insensitive to the mass of the third nondegenerate quark. For any nonzero value of ,
the chiral condensate is always less than the = 0 chiral condensate. We find numerically
275
Fig. 7. The mass dependence of the chiral condensate for various values of for three degenerate quark flavors.
The upper, middle and lower curves correspond to = 0, /2, and , respectively. As is increased from zero
to , the chiral condensate decreases in magnitude for a given .
Fig. 8. The three-flavor chiral condensate in the macroscopic limit as a function of for which two of the quark
masses are degenerate. The upper (lower) curve corresponds to infinite (zero) mass splitting between the two
degenerate quark masses and the third quark mass. All finite mass splittings are in the region between the two
curves.
276
that
( = , mlight, mheavy )
=
1
2,
1
3,
mlight = mheavy,
mlight mheavy
(84)
to a very high precision.

5.3. Topological density: first-order phase transition at =
For degenerate quark masses, the topological density is given by
2
1
1
(, , V ) =
(N
Z f =3) V
2
d

e cos
(85a)
b1 I1 (, ) + b2 I2 (, )
(, )
where the coefficients are given by
3
3
b1 cos
(85b)
sin
+ sin
2
2
2

3
3
b2 cos
cos
sin
2
2
2

1
3
3
tan
.
+ tan
(85c)
2
2
2
2
The topological density as a function of is shown in Fig. 9 in the macroscopic limit. For
three degenerate quark masses, there is a discontinuity at = which is simply Dashens
Fig. 9. The Nf = 3 topological density as a function of in the macroscopic limit for which at least two of
the quark masses are degenerate. The solid, dashed and long-dashed curves correspond to the mass splittings
mheavy mlight , mheavy = 2mlight , and mheavy = mlight , respectively. For completely degenerate masses, there
is a first-order phase transition at = . Even for large mass splittings, however, there is still a very rapid
crossover.
277
Fig. 10. The Nf = 3 topological susceptibility as a function of in the macroscopic limit for which at least two
of the quark masses are degenerate. The solid, dashed and long-dashed curves correspond to the mass splittings
mheavy mlight , mheavy = 2mlight , and mheavy = mlight , respectively. For completely degenerate masses, the
topological susceptibility diverges at = on account the first-order phase transition.
phenomena. This discontinuity is washed out for any nonzero mass splitting, however, for
any mass splitting between mheavy and mlight, the transition is always extremely rapid.
5.4. Topological susceptibility
The topological susceptibility in the macroscopic limit for Nf = 3 is shown in Fig. 10.
For three degenerate masses, the topological susceptibility diverges at = on account of
the first-order phase transition. For three quark flavors with only two degenerate masses,
there is still an extreme drop in the topological susceptibility at = even for a very large
mass splitting. For two degenerate quark flavors, the topological susceptibility is positive
except in the vicinity of = . We find numerically that
m
light
, mlight = mheavy,
(Nf =3)
( = 0, mlight, mheavy, V ) = m3light
lim
(86)
V
, mlight mheavy
2
which is consistent with the WardTakahashi identity (79).
6. Conclusion
In this paper, we investigated the properties of the QCD partition function in the
LeutwylerSmilga finite volume scaling regime. The full partition function including the
contributions from all topological sectors for Nf = 2 has been previously calculated.
278
We extended these results by performing the summation over all topological charges for
arbitrary Nf . For Nf 3, the partition function can be expressed as a (Nf 2)-fold
angular integration over a finite sum of modified Bessel functions. We considered both
the cases of degenerate and nondegenerate quark masses. The partition function remains
2 -periodic in after summing over all topological charges.
We systematically investigated the dependence of the QCD vacuum in the Leutwyler
Smilga regime. In this limit, the partition function only depends on and the scaling
variables, i = mi V . In the limit of degenerate quark masses, the Nf = 2 partition
function is independent of the scaling variable at = . As a consequence, the chiral
condensate vanishes identically at = for all values of . For fixed , the chiral
condensate decreases monotonically as is increased from 0 to .
In the macroscopic limit, i.e., for , the behavior of the two-flavor partition
function is particularly striking. For degenerate quark masses, the first derivative of the
energy density with respect to has a discontinuity at = corresponding to the
spontaneous breaking of the discrete CP symmetry. This phenomena was predicted by
Dashen and was subsequently demonstrated by Di Vecchia and Veneziano and Witten
using large-Nc chiral perturbation theory. The chiral condensate also develops a cusp at
= for degenerate quark masses in the macroscopic limit.
For Nf = 3, we find that all the examined quantities are very insensitive to the mass
splitting between two degenerate flavors and the third heavy flavor in the macroscopic
limit. When compared to those of Nf = 2, the vacuum properties for three flavors are
quite similar. While the partition function for Nf = 2 can be expressed in closed form,
the partition function for Nf = 3 requires the straightforward numerical evaluation of a
single integral. Unlike the Nf = 2 case, the Nf = 3 partition function is not independent
of the scaling variables at = and subsequently the chiral condensate is nonzero at
= even in the macroscopic limit. The chiral condensate does, however, exhibit the
same monotonicity in as the Nf = 2 chiral condensate and in the macroscopic limit
there is a cusp at = . Dashens phenomena, a first-order phase transition at = , is
also realized in this limit.
Many of our results are corroborated by general = 0 field theoretic identities,
for example, the WardTakahashi identity. Our analysis naturally extends these results
to nonzero values of . We have also demonstrated that QCD in the mesoscopic
regime exhibits Dashens phenomena. We examined the physics of QCD at = 0 in a
nonperturbative framework. While the existence of Dashens phenomena has been known
for thirty years and studied using chiral perturbation theory and numerical simulations, we
have demonstrated the spontaneous breaking of the CP symmetry at = in a way that
is both nonperturbative and analytic. We have also provided a further step towards the full
evaluation of the path integral for chiral perturbation theory. We hope that this work helps
elucidate a parameter space of QCD which has been largely unexplored.
Acknowledgements
We especially thank P.H. Damgaard for very useful comments and suggestions. We
acknowledge D. Diakonov for posing a question that led to the present work and thank
279
G. Akemann, S. Chandrasekharan, C. Dawson, T. Fugleberg, A. Jackson, D. Kharzeev,

M. Oswald, O. Philipsen, R. Pisarski and K. Splittorff for useful conversations and
suggestions.
Appendix A. Explicit calculation of the degenerate mass partition function

Starting with the representation
(Nf )
()
dj
2 Nf
=
0 j =1

iNf iN
f , (A.1)
exp cos(j ) ij ij j i1 ,...,iNf eii1 , . . . , e
the summation Z (Nf ) (, ) =

2
0
(Nf )
ei Z
() can be performed by the use of

d1
i(1 Nf )
f (1 , . . . , Nf , )
e
2
=
= f (2 + + Nf , 2 , . . . , Nf , ).
(A.2)
After a change of integration variables

i( ++Nf )
= ei ,
ei2 1 + e 3

i(3 ++Nf )
= ei
ei2 1 + e
with
=
(A.3)
(A.4)
2 + 2 cos(3 + + Nf )
(A.5)
one arrives at
Z (Nf ) (, )
2
2 Nf
d cos i dj cos(j )i(j 3/2)j
i/2
e
e
=e
2
2
0
0 j =3
(, 3 , . . . , Nf , ),
(A.6)
where
(, 3 , . . . , Nf , )

1 e 2i (3 ++Nf )i

i

1 e 2 (3 ++Nf )+i

= 1
ei3
.
..
.
.
.

i
1
e Nf
..
.

)+i(Nf 1)

.
ei(Nf 1)3

..

.

i(N 1)Nf
e f
i(Nf 1)
(3 ++Nf
2
i(Nf 1)
(3 ++Nf
2
)i(Nf 1)
(A.7)
280
Expanding the determinant around the first two rows leads to

Nf 1
(, 3 , . . . , Nf , ) =

(N )
(1)r eir eir r f (3 , . . . , Nf , ),
(A.8)
r=1
(Nf )
where the phases r
are given by
Nf r
(N )
r f (3 , . . . , Nf , ) =
j =1
i(j +r/2)(3 ++Nf ) (Nf )

j,j +r (3 , . . . , Nf )
(A.9)
with
(N )
k,l f (3 , . . . , Nf )

1 ei(k2)3
.
..
= ..
.
i(k2)
1 e
Nf
eik3
..
.
e
ikNf
ei(l2)3
..
.
i(l2)Nf
eil3
..
.
e
ilNf

ei(Nf 1)3

..
.
.

i(Nf 1)Nf
e
(A.10)
The integration over the angle can be easily performed, since

2

d cos ir
2r
e
Ir ().
e eir = Ir+1 () Ir1 () =
2
(A.11)
From that the final expression for the partition function with degenerate masses is
Z (Nf ) (, ) = 2e/2
Nf
2
0 m=3
Nf 1
r(1)r
r=1
dm cos(m )i(m3/2)j
e
2
Ir () (Nf )
r (3 , . . . , Nf , ).
(A.12)
Alternatively, from Eq. (11), the equal mass partition function is
Z (Nf ) (, ) = a1 ,...,aNf
ei
= a1 ,...,aNf
Nf

I+mam ()
(A.13a)
m=1
Nf

dm
2
2
m=1 0

exp cos(m ) + i( + m am )m .
(A.13b)
Using Eq. (44), the summation over gives a delta function which facilitates the integration
a 1 :
Z
(Nf )
(, ) =
2
a1 ,...,aNf
(2)Nf 1
0

d2 dNf exp cos( + 2 + 3 + + Nf )

exp cos(2 ) + cos(3 ) + + cos(Nf )

cos (a1 1)( + 2 + + Nf )

+ (2 a2 )2 + + (Nf aNf )Nf .
281
(A.14)
The integration over 2 can be done analytically [52] by expanding cos( + 2 + +

Nf ) = cos(2 ) cos( + 3 + + Nf ) sin(2 ) sin( + 3 + + Nf ).
The result is
Z (Nf ) (, )
= a1 ,...,aNf

Nf 2

+ 3 + + Nf
dm cos(m )
Ia1 a2 +1 2 cos
e
2
2
m=3

cos (3 a1 a2 )( + 3 + + Nf )/2 (3 a3 )3

(Nf aNf )Nf .
Using the addition rules for modified Bessel functions, one arrives at Eq. (45).
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]
G. t Hooft, Phys. Rev. Lett. 37 (1976) 8.

K.F. Smith et al., Phys. Lett. B 234 (1990) 191.
I.S. Altarev et al., Phys. Lett. B 276 (1992) 242.
E. Witten, Nucl. Phys. B 156 (1979) 269.
G. Veneziano, Nucl. Phys. B 159 (1979) 213.
D. Kharzeev, R.D. Pisarski, M.H. Tytgat, Phys. Rev. Lett. 81 (1998) 512.
K. Buckley, T. Fugleberg, A. Zhitnitsky, Phys. Rev. Lett. 84 (2000) 4814.
D. Ahrensmeier, R. Baier, M. Dirks, Phys. Lett. B 484 (2000) 58.
K. Buckley, T. Fugleberg, A. Zhitnitsky, Phys. Rev. C 63 (2001) 034602.
G. Schierholz, Nucl. Phys. Proc. Suppl. 42 (1995) 270.
S. Chandrasekharan, U. Wiese, Phys. Rev. Lett. 83 (1999) 3116.
M. Alford, S. Chandrasekharan, J. Cox, U.J. Wiese, Nucl. Phys. B 602 (2001) 61.
A. Hart, M. Teper, Nucl. Phys. Proc. Suppl. 83 (2000) 476.
B. Alles, M. DElia, A. Di Giacomo, Phys. Lett. B 483 (2000) 139.
S. Durr, Nucl. Phys. B 611 (2001) 281.
G.S. Bali et al., SESAM Collaboration, Phys. Rev. D 64 (2001) 054502.
R. Bertle, M. Engelhardt, M. Faber, Phys. Rev. D 64 (2001) 074504.
A. Hasenfratz, Phys. Rev. D 64 (2001) 074503.
A. Ali Khan et al., CP-PACS Collaboration, Phys. Rev. D 64 (2001) 114501.
A. Hart, M. Teper, UKQCD Collaboration, hep-lat/0108006.
R. Dashen, Phys. Rev. D 3 (1971) 1879.
P. Di Vecchia, G. Veneziano, Nucl. Phys. B 171 (1980) 253.
E. Witten, Annals Phys. 128 (1980) 363.
M. Creutz, Phys. Rev. D 52 (1995) 2951.
N. Evans, S.D. Hsu, A. Nyffeler, M. Schwetz, Nucl. Phys. B 494 (1997) 200.
I. Halperin, A. Zhitnitsky, Phys. Rev. Lett. 81 (1998) 4071.
T. Fugleberg, I. Halperin, A. Zhitnitsky, Phys. Rev. D 59 (1999) 074023.
A.V. Smilga, Phys. Rev. D 59 (1999) 114021.
M.H. Tytgat, Phys. Rev. D 61 (2000) 114009.
(A.15)
282
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
R.A. Janik, M.A. Nowak, G. Papp, I. Zahed, Acta Phys. Polon. B 32 (2001) 1297.
R.A. Janik, M.A. Nowak, G. Papp, I. Zahed, hep-ph/9905306.
J.C. Plefka, S. Samuel, Phys. Rev. D 55 (1997) 3966.
J.C. Plefka, S. Samuel, Phys. Rev. D 56 (1997) 44.
H. Yoneyama, M. Imachi, S. Kanou, Nucl. Phys. Proc. Suppl. 73 (1999) 644.
R. Burkhalter, M. Imachi, Y. Shinno, H. Yoneyama, Prog. Theor. Phys. 106 (2001) 613.
H. Leutwyler, A. Smilga, Phys. Rev. D 46 (1992) 5607.
For a recent review, see: J.J. Verbaarschot, T. Wettig, Ann. Rev. Nucl. Part. Sci. 50 (2000) 343, references
therein.
J. Gasser, H. Leutwyler, Phys. Lett. B 188 (1987) 477.
A.D. Jackson, M.K. Sener, J.J. Verbaarschot, Phys. Lett. B 387 (1996) 355.
P.H. Damgaard, Phys. Lett. B 424 (1998) 322.
P.H. Damgaard, Nucl. Phys. B 556 (1999) 327.
R. Brower, P. Rossi, C. Tan, Nucl. Phys. B 190 (1981) 699.
A.B. Balantekin, Phys. Rev. D 62 (2000) 085017.
J.C. Osborn, D. Toublan, J.J. Verbaarschot, Nucl. Phys. B 540 (1999) 317.
P.H. Damgaard, J.C. Osborn, D. Toublan, J.J. Verbaarschot, Nucl. Phys. B 547 (1999) 305.
P.H. Damgaard, K. Splittorff, Nucl. Phys. B 572 (2000) 478.
G.N. Watson, Theory of Bessel Functions, Cambridge University Press, London, 1944.
G. Akemann, J.T. Lenaghan, K. Splittorff, hep-th/0110157.
P.H. Damgaard, private communication.
A. Smilga, J.J. Verbaarschot, Phys. Rev. D 51 (1995) 829.
R.J. Crewther, Phys. Lett. B 70 (1977) 349.
I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series and Products, Academic Press, New York, 1980
(Translated from Russian).

Stability of antisymmetric tensor fields of

ChernSimons type in AdS spacetime
Hiroki Hata, Yoshiaki Tanii
Physics Department, Faculty of Science, Saitama University, Saitama 338-8570, Japan
Abstract
Stability of massive antisymmetric tensor fields with the ChernSimons type action in antide Sitter spacetime is studied. It is found that there exists a complete set of solutions whose energy is
conserved and positive definite if the mass is positive. Scalar products of the solutions are shown to
be well-defined and conserved. In contrast to the previously studied scalar field case there is no other
set of stable solutions with a different kind of boundary condition. 2002 Published by Elsevier
Science B.V.
PACS: 04.62.+v; 04.65.+e
1. Introduction
Anti-de Sitter (AdS) spacetime is a maximally symmetric spacetime with a negative
constant curvature. It naturally appears as a solution of Einstein equation with a
negative cosmological constant. It also appears in compactifications of higher-dimensional
supergravities in the KaluzaKlein theory [1]. Recently interest in field theories in AdS
spacetime has been much increased due to their relevance to the AdS/CFT correspondence
in the string/M theory [24] (for a review see Ref. [5]). An important issue of field theories
in AdS spacetime is their stability, which was previously discussed in Refs. [68]. (For
recent studies on the stability see Ref. [9].) Another important issue is a choice of boundary
conditions of fields at spatial infinity. In Ref. [10] boundary conditions are chosen such that
the Cauchy problem for field equations is well-defined by requiring the conservation of the
scalar product of fields.
E-mail addresses: hata@krishna.th.phy.saitama-u.ac.jp (H. Hata), tanii@post.saitama-u.ac.jp (Y. Tanii).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 5 6 - 3
284
H. Hata, Y. Tanii / Nuclear Physics B 624 (2002) 283298
The purpose of this paper is to study the stability of free massive antisymmetric tensor
fields of arbitrary rank n in AdS spacetime. The string/M theory and supergravities contain
antisymmetric tensor fields, which play an important role. Therefore, their stability is an
important issue. There are two types of theories of antisymmetric tensor fields. One type
of theories have an action with the second order kinetic term of the Maxwell type and the
other type of theories have an action with the first order kinetic term of the ChernSimons
type [11]. (Equivalent theories to the latter were studied in Refs. [12,13].) Both types of
theories appear in supergravities [14]. In this paper we only consider the ChernSimons
type theories, which are theories of nth rank antisymmetric tensor fields in d = 2n + 1
dimensions. As a preparation for study of the stability we first obtain the general solution of
the field equation. Then we obtain conditions for the stability by studying the conservation
and the positivity of the energy. We also study scalar products of the solutions.
The conditions for the stability of scalar fields in AdS spacetime were obtained in
Refs. [7,8]. A free massive scalar field theory in d-dimensional AdS spacetime is stable
if the mass m satisfies

2
d 1 2
m
>
,
(1.1)
a
2
where a 1 is the radius of AdS spacetime. More precisely, there exists a complete
set of solutions of the field equation whose energy is conserved and positive definite.
Furthermore, when the mass satisfies

2

d 1 2
m
d 1 2
(1.2)
<
<1
,
2
a
2
there exists another set of stable solutions satisfying a different kind of boundary condition
at spatial infinity. In the latter case the coefficient of the improvement term in the energy
momentum tensor must take a particular value. Scalar fields can be stable even if the mass
squared is negative due to a positive contribution from the kinetic term to the energy.
As in the scalar field case we find that there exists a complete set of solutions for
antisymmetric tensor fields whose energy is conserved and positive definite if the mass
is positive m > 0. There are three kinds of improvement terms in the energymomentum
tensor for antisymmetric tensor fields of rank n 2. The coefficients of these terms can
take arbitrary values although they do not contribute to the energy. The scalar products of
the solutions are shown to be well-defined and conserved.
In contrast to the scalar field case, however, there is no other set of stable solutions.
The conservation of the energy allows another set of solutions satisfying a different kind of
boundary condition at spatial infinity as in the scalar field case but their energy turns out to
be divergent. Therefore, only one kind of boundary condition is possible for antisymmetric
tensor fields with the ChernSimons type action. A ChernSimons type theory of the
second rank antisymmetric tensor fields in five-dimensional AdS spacetime was previously
studied in Ref. [15] in the context of the AdS/CFT correspondence. By using a different
approach it was found there that only one kind of boundary condition is possible, which is
consistent with our result.
In the next section we introduce the first order action of antisymmetric tensor fields in
AdS spacetime and obtain the energymomentum tensor. In Section 3 the general solution
285
of the field equation is obtained in terms of the hypergeometric functions. The conservation
and the positivity of the energy are discussed in Sections 4 and 5, respectively. In Section 6
we show that there exists a well-define and conserved scalar product for the solutions.
In Appendix A we give a construction and useful identities of spherical harmonics for
antisymmetric tensors on the (d 2)-dimensional sphere S d2 .
2. Antisymmetric tensor fields in AdS spacetime

We consider antisymmetric tensor fields in d-dimensional AdS spacetime. AdS
spacetime is a maximally symmetric spacetime and has the metric
g dx dx =
1
a 2 cos2

dt 2 + d 2 + sin2 hab d a d b ,
(2.1)
where , = 0, 1, . . . , d 1, are d-dimensional world indices and the constant a 1

is the radius of AdS spacetime. The time coordinate t has a range < t < ,
which corresponds to considering the universal covering of AdS spacetime. The radial
coordinate has a range 0 < 2 with the spatial infinity at = 2 . a and hab
(a, b = 1, 2, . . . , d 2) are coordinates and the metric of the (d 2)-dimensional unit
sphere S d2 . Non-vanishing components of the Christoffel connection are
0 0 = 0 0 = = tan ,
a b = tan hab ,
a b =
1
a ,
sin cos b
1
b a c = had (b hcd + c hbd d hbc ) = b a c ,
2
where b a c is the Christoffel connection of S d2 . The Riemann tensor is given by

R = a 2 g g .
(2.2)
(2.3)
Our conventions for the curvature tensors are R = + ( ),

R = R , R = g R . For other properties of AdS spacetime see, e.g., Ref. [5].
We consider a free theory of a complex antisymmetric tensor field B1 n of rank n in
d-dimensional AdS spacetime for d = 2n + 1. The ChernSimons type action [11] is

1
1 1 2n+1
d
S = d x (1) 2 (n+1)
B1 n n+1 Bn+2 2n+1
(n!)2

m
g B 1 n B 1 n ,
(2.4)
n!
where 1 2n+1 is the totally antisymmetric tensor and denotes the complex conjugation. The reality of the action requires that the mass m is real. When m = 0, the action
consists of only the kinetic term, which do not depends on the metric but is invariant under
general coordinate transformations. This is an action of a topological field theory. We do
not discuss the stability of the m = 0 case since there is no local degrees of freedom and
286
the energy is zero. The field equation derived from this action is
1
(1) 2 (n+1)
1
n!
1 2n+1
n+1 Bn+2 2n+1 m g B 1 n = 0.
(2.5)
We define the energymomentum tensor of the theory as a variation of the action with
respect to the metric. To do this we need an action for general metric, which reduces to the
original action (2.4) for the AdS metric (2.1). We use the action

1
1 1 2n+1
d
S = d x (1) 2 (n+1)
B1 n n+1 Bn+2 2n+1
(n!)2
RB 1 n B 1 n
g B 1 n B 1 n +
n!
n!a

R B 2 n B 2 n +
R B 3 n B 3 n ,
+
n!a
n!a
(2.6)
where , and are arbitrary constant parameters. The parameter is chosen as

= m d(d 1) + (d 1) + 2 a
(2.7)
so that this action coincides with Eq. (2.4) for the AdS metric. The last three terms in
Eq. (2.6) containing the curvature tensors are generalizations of the well-known R 2 term
in the scalar field theory. These terms give improvement terms in the energymomentum
tensor. Note that we do not need to introduce the term R B 3 n B 3 n since it
is related to the last term in Eq. (2.6) by the Bianchi identity of the Riemann tensor. From
Eq. (2.6) we obtain the energymomentum tensor as
2 S
T =
g g

1

2
nm + (d 1)a + 2 a B(
m 2(d 1)a g B B
=
B)
n!
n!

2
2
D D g D B B
+
n!a

B D 2 B(
B) g D( D ) B B
+
2D D( B)
n!a

4
D D B(
(2.8)
B) ,
n!a
where (B B) = B 1 2 n B 1 2 n , (B B ) = B
B 2 n , etc. There is no
2 n
contribution from the kinetic term.
The energy is defined as follows [6]. The energymomentum tensor (2.8) satisfies
D T = 0 for arbitrary , and when the field equation is used. We can construct
a conserved current

g T = 0
(2.9)
287
for each Killing vector satisfying D + D = 0. The energy is the charge of this
current for a timelike Killing vector = (1, 0, . . . , 0)

E = d d1 x g T t t ,
(2.10)
where the integral is over (d 1)-dimensional space.
3. Solutions of the field equation

We shall obtain the general solution of the field equation (2.5). It is more convenient to
rewrite the field equation in another form. Applying 1 to Eq. (2.5) we obtain a constraint
D1 B 1 n = 0,
(3.1)
where D is the covariant derivative in AdS spacetime. Applying the differential operation
in the first term of Eq. (2.5) to Eq. (2.5) again and using Eq. (2.5) in the second term we
obtain the second order equation

D D + n(n + 1)a 2 m2 B 1 n = 0.
(3.2)
We first solve these equations and then substitute the solutions into the first order equation (2.5) to obtain further conditions.
By the constraint (3.1) the components Bt a2 an and Bta3 an are not independent
but can be expressed by Ba1 an and Ba2 an . The field equation (3.2) for [1 n ] =
[a1 an ], [a2 an ] gives
L1 B a1 an
2(1)n n
[a1 B a2 an ] = 0,
sin3 cos
2
a B a1 a2 an = 0,
L2 B a2 an
tan 1
(3.3)
where a is the covariant derivative on S d2 using the Christoffel connection b a c in

Eq. (2.2). The differential operators L1 and L2 are defined as
4n 1
a a + 3n(n 1) 4n2 (m/a)2
+
,
+
sin cos
cos2
sin2

4n 3
a a + 3n2 7n + 3
L2 = t2 + 2 + 4 tan +
+
sin cos
sin2
L1 = t2 + 2 +
(2n + 1)2 (m/a)2

4.
cos2
(3.4)
It can be shown that when Eqs. (3.1) and (3.3) are satisfied, then the remaining components
of Eq. (3.2) are automatically satisfied.
We decompose the antisymmetric tensor field into transverse and longitudinal modes
( ) for antisymmetric tensor fields on S d2 . The
by using spherical harmonics Ya(l)
1 an
a
1
= 0 and are eigenfunctions of the Laplacian
spherical harmonics are transverse Ya(l)
1 an
288
a a on S d2 with the eigenvalue [l(l + d 3) n]. The quantum number l takes

values l = 0, 1, 2, . . . , for n = 0 and l = 1, 2, 3, . . ., for n 1. In Appendix A we sketch
(l)
how to construct Ya1 an and give some useful identities. Using the spherical harmonics the
components of the antisymmetric tensor field are decomposed as
B a1 an (x) = R1 (t, )Y (l)a1an ( ) + R2 (t, ) [a1 Y (l)a2 an ] ( ),
B a2 an (x) = sin cos R3 (t, )Y (l)a2an ( ) + R4 (t, ) [a2 Y (l)a3an ] ( ),
B t a2 an (x) = R5 (t, )Y (l)a2an ( ) + R6 (t, ) [a2 Y (l)a3an ] ( ),
B ta3 an (x) = R7 (t, )Y (l)a3an ( ) + R8 (t, ) [a3 Y (l)a4an ] ( ),
(3.5)
where Y (l)a1an = ha1 b1 han bn Yb(l)

and a = hab b . The factor sin cos in front of
1 bn
R3 is for later convenience. Substituting Eq. (3.5) into the constraint (3.1) for [2 n ] =
[a2 an ], [a3 an ], R5 , . . . , R8 , are expressed in terms of R1 , . . . , R4 as
1
t R5 = (l + n 1)2 R2 (sin cos + 2n)R3 ,
n

2n 1
R4 ,
t R6 = + 2 tan +
sin cos
1
(l + n)(l + n 2)R4 ,
t R7 =
n1
t R8 = 0.
(3.6)
Other components of the constraint (3.1) are then automatically satisfied. R1 , . . . , R4 , in

turn, are determined by solving Eq. (3.3). To solve Eq. (3.3) it is convenient to change the
variable as
v = sin2 .
(3.7)
Let us first consider R1 . We define the function f1 (v) by

1 (t)v 2 (1 v) 2 f1 (v),
R1 (t, ) = N
1
(3.8)
where
1 (t) = N1 ei1 t + N
1 ei1 t
N
(3.9)
1 are complex constants. The transverse part of the first equation in (3.3)
and N1 and N
gives

4v(1 v)v2 + 2 2 + 4n 2( + + 1)v v + 12 ( + )2
+

( l + n)( + l + 3n 2) ( 2n)2 (m/a)2
f1 (v) = 0.
+
v
1v
We choose the parameters and as

2
m
= l n,
( 2n)2 =
a
(3.10)
(3.11)
289
so that the v 1 and (1 v)1 terms in Eq. (3.10) vanish. There are two possible values
of
|m|
.
= 2n
(3.12)
a
Then, Eq. (3.10) becomes a hypergeometric equation

v(1 v)v2 + c1 (a1 + b1 + 1)v v a1 b1 f1 (v) = 0,
(3.13)
where
1
a1 = ( + l n 1 ),
2
1
b1 = ( + l n + 1 ),
2
c1 = l + n.
(3.14)
The solution which gives Ba1 an regular at = 0 is a hypergeometric function

f1 (v) = 2 F1 (a1 , b1 , c1 ; v)

(c1 )
(a1 + n) (b1 + n) v n
.
=
(a1 ) (b1 )
(c1 + n)
n!
(3.15)
n=0
The equation for R4 can be solved similarly. We put

4 (t)v 2 (ln+1) (1 v) 2 (+1)f4 (v),
R4 (t, ) = N
1
(3.16)
where
4 (t) = N4 ei4 t + N
4 ei4 t
N
(3.17)
and satisfies the second equation in (3.11). The longitudinal part of the second equation
in (3.3) gives a hypergeometric equation on f4 . The solution which gives Ba2 an regular
at = 0 is
f4 (v) = 2 F1 (a4 , b4 , c4 ; v),
(3.18)
where
1
a4 = ( + l n 4 ),
2
1
b4 = ( + l n + 4 ),
2
c4 = l + n.
(3.19)
The equations for R2 and R3 are slightly more complicated since they are coupled
equations. The longitudinal part of the first equation and the transverse part of the second
equation in (3.3) become
2n
R3 = 0,
v
2(l + n 1)2
LR3 +
R2 = 0,
nv
LR2 +
(3.20)
290
where
L = t2 + 4v(1 v)v2 + 4(2n v)v
l(l + 2n 2) (n 1)(3n + 1) 4n2 (m/a)2
+
.
v
1v
2 and R
3 as
These equations can be diagonalized by defining new functions R
2 = (l + n 1)R2 + nR3 ,
R
3 = (l + n 1)R2 + nR3 .
R
(3.21)
(3.22)
Eq. (3.20) then becomes

2
2 = 0,
L (l + n 1) R
v

2
3 = 0.
L + (l + n 1) R
v
(3.23)
Putting
2 (t)v 2 (ln+1) (1 v) 2 f2 (v),
2 (t, ) = N
R
1
3 (t)v 2 (ln1) (1 v) 2 f3 (v),

3 (t, ) = N
R
1
(3.24)
where
2 ei2 t ,
2 (t) = N2 ei2 t + N
N
3 (t) = N3 ei3 t + N
3 ei3 t ,
N
(3.25)
these equations become hypergeometric equations on f2 and f3 . The solutions which give
Ba1 an and Ba2 an regular at = 0 are
f2 (v) = 2 F1 (a2 , b2 , c2 ; v),
f3 (v) = 2 F1 (a3 , b3 , c3 ; v),
(3.26)
where
1
a2 = ( + l n + 1 2 ),
2
1
b2 = ( + l n + 1 + 2 ),
2
c2 = l + n + 1,
1
a3 = ( + l n 1 3 ),
2
1
b3 = ( + l n 1 + 3 ),
2
c3 = l + n 1.
(3.27)
Thus we have obtained the general solution of the second order equation (3.2). We
now consider the first order equation (2.5). Substituting Eq. (3.5) into Eq. (2.5) and using
Eq. (A.11) and the second order equation (3.2) we find that Eq. (2.5) is satisfied if the
functions Rs satisfy
1
n1
m
R4 = (1) 2 n(n1)
sin cos t R1 ,
a
l+n

1
1 l+n1
n
m
R3 = (1) 2 n(n+1)+ 2
t R2 + 2 R5 .
a
n
sin
(3.28)
291
i satisfy
These equations are satisfied if 1 = 4 , 2 = 3 and N
1
l+n m
d
N1 = (1) 2 n(n1)
N4 ,
dt
n1 a
1
1
d
N3 = (1) 2 n(n+1)+ 2
dt
2

a 1
m
2
2

2 N3 2(l + n)(l + n 1)N2 .

(l + n 1)
m 2
a
(3.29)
3 are related to N
1 and N
2 , respectively, and independent
4 and N
By these relations N
degrees of freedom are reduced from four to two.
4. Conservation of the energy

We first consider the conservation of the energy as a condition for the stability. From
Eq. (2.9) the time derivative of the energy (2.10) is given by

d
E = d d2 g g Tt = ,
(4.1)
2
dt
where the integral is over S d2 . Tt can be written as

2

( tan )t B B
Bt ) +
n!Tt = 2nm B(
a

t Ba1 an B a1 an
a sin cos

n1
+
tan
t Bta2 an B t a2 an
a
sin cos

n
tan +
t Ba
B a2 an
+
2 an
a
sin cos

t
n+1
B
+ 2(n 1)
tan +
t Bt
a
2 sin cos

4
t Bta2 an B t a2 an + 4 a B(
Bt )
a sin cos

t
n+1
4
B + a ( )a .
+
tan +
t Bt
a
sin cos
(4.2)
Here, a ( )a represents total derivative terms on S d2 , which vanish in the integral (4.1).
Substituting Eq. (3.5) into Eq. (4.1) it is divided into three independent parts

2

1
d
+ E2 Y (l) 2 + E3 Y (l) 2
E=
,
(4.3)
d E1 Ya(l)
a2 an
a3 an
1 an
= 2
dt
n!
where d = d d2 h is the volume element of S d2 and

(l) 2
a1 b1
Y

han bn Ya(l)
Y (l) .
(4.4)
a1 an = h
1 an b1 bn
292
E 1 , E 2 and E 3 depend only on R1 , (R2 , R3 ) and R4 , respectively.

To evaluate the right-hand side of Eq. (4.1) we need boundary behaviors of the functions
Ri for 2 . They can be obtained from the behavior of the hypergeometric function
2 F1 (a, b, c; v) for v 1. When 2n is not an integer, we find that near the boundary
Rs behave as

1 (t) A1 (cos ) 1 + O cos2
R1 (, t) N

+ B1 (cos )+4n 1 + O cos2 ,

2 (, t) N
2 (t) A2 (cos ) 1 + O cos2
R

+ B2 (cos )+4n 1 + O cos2 ,

3 (, t) N
3 (t) A3 (cos ) 1 + O cos2
R

+ B3 (cos )+4n 1 + O cos2 ,

4 (t) A4 (cos )+1 1 + O cos2
R4 (, t) N

+ B4 (cos )+4n+1 1 + O cos2 ,
(4.5)
where
Ai =
(ci ) (ci ai bi )
,
(ci ai ) (ci bi )
Bi =
(ci ) (ai + bi ci )
(ai ) (bi )
(4.6)
with ai , bi and ci given in Eqs. (3.14), (3.19) and (3.27). We see that the value of
determines boundary behaviors of the solutions. The case in which 2n is an integer
is discussed at the end of this section.
Let us first consider E1 . We obtain

tan 4n1 2
4n
E1 =
( tan ) +
t |R1 |2
a
a
a sin cos
1
1 |2
= 4n t |N
a

A21 2(2 2n + 1) + (cos )24n 2A1 B1 2(2n + 1) +

+ B12 2(2 6n 1) (cos )2+4n

(4.7)
+ C(cos )24n+2 + D(cos )2 + E(cos )2+4n+2 .
= 2
The last three terms represent higher order terms than the first three, respectively. When we
choose = + , the first term automatically vanishes. For the second and the third terms to
vanish we have to require either (i) B1 = 0 or (ii) A1 = 0, 2(2 6n 1) = 0. On the
other hand, when we choose = , the third term vanishes automatically and we have
to require either (iii) A1 = 0 or (iv) B1 = 0, 2(2 2n + 1) + = 0. It can be shown
that the conditions (i) and (ii) are equivalent to (iii) and (iv), respectively. It is enough to
consider two cases (i) and (iv) and set B1 = 0. For B1 = 0 only C term survives among the
293
higher order terms. In the case (i) it vanishes automatically. In the case (iv) we further need
to require 2 4n + 2 > 0, i.e., |m| < a.
The conditions for E2 and E3 to vanish can be obtained in a similar way. It first requires
B2 = B3 = B4 = 0. The remaining terms are

1
3 A3 2
2 A2 N
2(2 2n + 1) + (cos )24n t N
4n
2na

1
+
2n(2 2n + 1) + (2 n) + 4 (cos )24n
4
2
4n
8n 2 a

A + ( + l n 1)N
A 2

t ( l 3n + 1)N
2 2
3 3

1
nm
3 A3
+
2 +
N2 A2 + N
a
4n2 22 a 4n

A + ( + l n 1)N
A + c.c.,

( l 3n + 1)N
E2 =
E3 =
2 2
4 A4 (cos )24n
t N
1
1
(l + n)(l + n 2)( 2n)
n1
12 a 4n

2n(2 2n + 1)( 2n) + (2 n)( 2n)

nm
+ 4( 2n) 2
.
a
(4.8)
= 2
For = + all these terms vanish automatically. For = we have to require the
coefficients to vanish, which fixes the parameters , and .
To summarize, the conservation of the energy leads to one of the two possibilities
(I) = + ,
Bi = 0,
(II) = ,
Bi = 0,
|m| < a.
(4.9)
In the case (II) the coefficients of the improvement terms must be chosen as
m
n
,
=
2( n)(2 2n + 1) a
n m
,
=
n a
nm
.
=
(4.10)
2a
The conditions Bi = 0 require ai = 0 or bi = 0, which lead to the quantization of i
1 = (2k1 + + l n),
2 = (2k2 + + l n + 1),
3 = (2k3 + + l n 1),
4 = (2k4 + + l n),
(4.11)
where ki are non-negative integers. Then, the hypergeometric functions in Eqs. (3.15),
(3.18), (3.26) can be expressed by the Jacobi polynomials as in the scalar field theory [8].
The above analysis does not immediately apply to the case 2n = N for an integer N .
Let us consider the N > 0 case. (The N < 0 case is equivalent to the N > 0 case.) This
occurs only when we choose = + . In such a case the coefficients Ai in Eq. (4.6) are
294
divergent for generic values of i since ci ai bi = N . To make Ai finite we have

to choose i such that ci ai = ki
or ci bi = ki
for non-negative integers ki
. The
conservation of the energy requires Bi = 0 as above, which restrict the values of ki
to
ki
= N, N + 1, . . . . Redefining ki = ki
N = 0, 1, . . . , we recover the values of i in
Eq. (4.11). Therefore, the results obtained for non-integer 2n is also valid for integer
2n.
5. Positivity of the energy

We next consider the second condition of stability, i.e., the positivity of the energy. The
integrand of the energy (2.10) is

tan 2n1 1
t
g T t = 2m h
n Bt Bt gt t B B
a
2

tan 2n1 1

( tan ) 2 B B + Bt B t
+ h
a
a

t
2n 1
B B + 4 Bt
B
+ tan +
sin cos

1
Ba B a + 4 Bta B t a
(5.1)
+ h a ( )a .
sin cos
We see that the contributions from the , and terms in the action (2.6) are total
derivative. Substituting Eq. (3.5) into Eq. (5.1) it is divided into three independent parts

2

1
+ E2 Y (l) 2 + E3 Y (l) 2 .
E=
(5.2)
d E1 Ya(l)
a2 an
a3 an
1 an
n!
E1 , E2 and E3 depend only on R1 , (R2 , R3 ) and R4 , respectively.
E1 is evaluated as

tan 4n+1 1
E1 = m d
(5.3)
|R1 |2 + 7E1 ,
a
sin2
where 7E1 is total derivative terms

tan 4n1
7E1 = d
a

1
2n
2 tan +
|R1 |2
a
sin cos
sin cos

1
1 |2 A21 (cos )24n
= 4n 2(2 2n + 1) + |N
a

+ O (cos )24n+2
.
(5.4)
= 2
The integral of the bulk term in Eq. (5.3) is convergent for = + as seen from the
boundary behavior of R1 in Eq. (4.5). However, it is divergent for = . Therefore,
295
the choice = is not allowed and only = + is possible. In the case of scalar fields
discussed in Refs. [7,8] the mass term in the energy is also divergent but it is canceled by
a divergent contribution from the kinetic term. Both of = + and = are possible in
the scalar field theories. In the present theory there is no kinetic term in the energy since
the kinetic term of the action is a topological term. For = + the bulk term in Eq. (5.3)
is obviously positive definite when m > 0. The boundary term 7E1 vanishes since it has a
positive power of cos .
Similarly, E2 and E3 are given by

tan 4n+1
E2 = m d
a

1
1
1
(l + n 1)2 |R2 |2 + n cos2 |R3 |2 + 2 |R5 |2 + 7E2 ,
2
sin n
sin

4n+1

tan
E3 = m d
a

1
1
4 n(n 1) 2 |R7 |2
sin
sin

n
(l + n)(l + n 2) |R4 |2 + |R6 |2 + 7E3 ,
+
(5.5)
n1
where the boundary terms are

7E2 =

7E3 =

1
2 A2 N
3 A3 2 (cos )24n
2(2 2n + 1) + N
4n
4na

1
2n(2 2n + 1) + (2 n) + 4
4
n2

A + ( + l n 1)N
A 2 (cos )24n

( l 3n + 1)N
2 2
3 3

+ O (cos )24n+2
,
= 2

(l + n)(l + n 2)( 2n)2
2n(2 2n + 1) + (2 n) + 4
4
4n
(n 1)4 a

2 2

24n
24n+2

N4 A4 (cos )
+ O (cos )
.
(5.6)

= 2
For = + the bulk integrals in Eq. (5.5) are convergent and are positive definite when
m > 0. The boundary terms 7E2 and 7E3 vanish in the same way as 7E1 .
It is thus proved that the energy is well-defined and positive definite for the case (I)
in Eq. (4.9) if the additional condition m > 0 is satisfied. The case (II) in Eq. (4.9) is
not allowed since the energy is divergent. Therefore, there exists only one complete set
of solutions corresponding to the case (I) for antisymmetric tensor fields with the Chern
Simons type action.
296
6. Scalar product
Finally we consider a scalar product of the fields. The scalar product of two solutions
B11 n and B21 n is defined as

(B1 , B2 ) = d d1 x g J t
(6.1)
by using the time component of the conserved current

F 1 n F1 1 n B21 n .
J = i B1
(6.2)
1 n 2
The integral in Eq. (6.1) is convergent since g J t = O((cos )24n+1 ) for 2 and
2 4n + 1 > 1 for = + and m > 0. The scalar product (6.1) is also conserved

d
(B1 , B2 ) = d d2 g J = = 0
(6.3)
2
dt
since g J |= 2 = O((cos )24n+2 )|= 2 = 0 for = + and m > 0. Therefore, we

have a well-defined conserved scalar product.
Acknowledgements
This work is partially supported by the Grant-in-Aid from the Ministry of Education,
Culture, Sports, Science and Technology, Japan, Priority Area (#707) Supersymmetry and
Unified Theory of Elementary Particles.
Appendix A. Spherical harmonics on S d2
In this appendix we sketch how to construct spherical harmonics for antisymmetric
tensor fields on S d2 following the approach in Refs. [1618]. See also Ref. [19]. We
embed a (d 2)-dimensional unit sphere S d2 in (d 1)-dimensional Euclidean space
Rd1 with the Cartesian coordinates x i (i = 1, 2, . . . , d 1). The metric of Rd1 is given
by
ds 2 = ij dx i dx j
= dr 2 + r 2 hab ( ) d a d b ,
(A.1)

where r = ij x i x j is a radial coordinate and a (a = 1, 2, . . . , d 2) are angular
coordinates parametrizing the unit sphere S d2 with the metric hab .
Let us consider antisymmetric tensors Ti1 in in Rd1 which satisfy
ni1 Ti1 in = 0,
i1 Ti1 in = 0,
(A.2)
ni ( ) = r 1 x i
is a unit vector normal to the sphere. In the polar coordinates (r, a )

where
these conditions become
Tra2 an = 0,
a1 Ta1 an = 0,
(A.3)
297
where a is the covariant derivative on S d2 . Therefore, restricting to the unit sphere they
represent transverse tensors on the sphere. The relation between Ti1 in and Ta1 an is
nin
ni1
Ti i .
(A.4)
a1
an 1 n
In the Cartesian coordinates spherical harmonics for such transverse antisymmetric
tensors for n 1 are given by
Ta1 an = r n
Yi1 in ( ) = r l C[i1 in j1 ](j2 jl ) x j1 x j2 x jl .

(l)
(A.5)
Here, C[i1 in j1 ](j2 jl ) is a constant coefficient, which is antisymmetric in i1 , . . . , in , j1 and

symmetric in j2 , . . . , jl , and is traceless with respect to any pair of the indices. Spherical
harmonics for n = 0 are given by
Y (l) ( ) = r l C(j1 jl ) x j1 x jl ,
(A.6)
where C(j1 jl ) is symmetric in j1 , . . . , jl and is traceless with respect to any pair of the
indices. Note that l takes values l = 0, 1, 2, . . ., for n = 0 and l = 1, 2, 3, . . . , for n 1.
One can easily check that these tensors indeed satisfy the conditions (A.2). Applying the
Laplacian d1 = ij i j in Rd1 we find
l(l + d 3) (l)
Yi1 in .
r2
On the other hand, in the polar coordinates we have

n
1 a
d n2
n
(l)
d1 Ya1 an = 2 a + r +
r
2 Ya(l)
1 an
r
r
r
r

1
= 2 a a n Ya(l)
.
1 an
r
(l)
d1 Yi1 in =
(A.7)
(A.8)
(l)
In the last line we have used the fact that Ya1 an , which is related to Eqs. (A.5), (A.6)
by the relation (A.4), has r dependence r n . Comparing Eqs. (A.7) and (A.8) we obtain
eigenvalues of the Laplacian on S d2

a a Ya(l)
= l(l + d 3) n Ya(l)
.
(A.9)
1 an
1 an
Useful identities, which can be easily derived from Eq. (A.9) are

a a [a1 Ya(l)
=
l(l
+
d
3)
d
+
n
+
2
[a1 Ya(l)
,
a
]
n
2
2 an ]
1
a1 [a1 Ya(l)
= (l + n 1)(l + d n 2)Ya(l)
.
2 an
2 an ]
n
(A.10)
There is a duality relation between Ya(l)

and Ya(l)
, which we use in Section 3.
1 am
1 adm3
By appropriately choosing the coefficients Cs in Eqs. (A.5) and (A.6) the relation can be
written as
1
(1) 2 (m+1)(dm2) 1 a1 ad2

=
am+1 Ya(l)
.
Y
(A.11)
m+2 ad2
(l + m)(d m 3)! h
One can easily check that both hand sides of this equation are transverse and have the
same eigenvalue of a a . The normalization factor on the right-hand side is determined
by repeated applications of this relation.
(l)a1 am
298
References
[1] M.J. Duff, B.E.W. Nilsson, C.N. Pope, KaluzaKlein supergravity, Phys. Rep. 130 (1986) 1.
[2] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231, hep-th/9711200.
[3] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from noncritical string theory, Phys.
Lett. B 428 (1998) 105, hep-th/9802109.
[4] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[5] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity,
Phys. Rep. 323 (2000) 183, hep-th/9905111.
[6] L.F. Abbott, S. Deser, Stability of gravity with a cosmological constant, Nucl. Phys. B 195 (1982) 76.
[7] P. Breitenlohner, D.Z. Freedman, Positive energy in anti-de Sitter backgrounds and gauged extended
supergravity, Phys. Lett. B 115 (1982) 197;
P. Breitenlohner, D.Z. Freedman, Stability in gauged extended supergravity, Ann. Phys. 144 (1982) 249.
[8] L. Mezincescu, P.K. Townsend, Stability at a local maximum in higher dimensional anti-de Sitter space and
applications to supergravity, Ann. Phys. 160 (1985) 406.
[9] O. DeWolfe, D.Z. Freedman, S.S. Gubser, G.T. Horowitz, I. Mitra, Stability of AdSp Mq compactifications
without supersymmetry, hep-th/0105047.
[10] S.J. Avis, C.J. Isham, D. Storey, Quantum field theory in anti-de Sitter spacetime, Phys. Rev. D 18 (1978)
3565.
[11] P.K. Townsend, K. Pilch, P. van Nieuwenhuizen, Self-duality in odd dimensions, Phys. Lett. B 136 (1984)
38;
P.K. Townsend, K. Pilch, P. van Nieuwenhuizen, Phys. Lett. B 137 (1984) 443, Addendum.
[12] S. Deser, R. Jackiw, S. Templeton, Three-dimensional massive gauge theories, Phys. Rev. Lett. 48 (1982)
975;
S. Deser, R. Jackiw, S. Templeton, Topologically massive gauge theories, Ann. Phys. 140 (1982) 372.
[13] S. Deser, R. Jackiw, Self-duality of topologically massive gauge theories, Phys. Lett. B 139 (1984) 371.
[14] A. Salam, E. Sezgin, Supergravities in Diverse Dimensions, North-Holland, Amsterdam/World Scientific,
Singapore, 1989.
[15] G.E. Arutyunov, S.A. Frolov, Antisymmetric tensor field on AdS5 , Phys. Lett. B 441 (1998) 173, hepth/9807046.
[16] P.S. Gallot, D. Meyer, Oprateur de courbure et Laplacien des formes diffrentielles dune varit
Riemannienne, J. Math. Pures et Appl. 54 (1975) 259.
[17] G.W. Gibbons, M.J. Perry, Quantizing gravitational instantons, Nucl. Phys. B 146 (1978) 90.
[18] M.A. Rubin, C.R. Ordez, Eigenvalues and degeneracies for n-dimensional tensor spherical harmonics,
J. Math. Phys. 25 (1984) 2888;
M.A. Rubin, C.R. Ordez, Symmetric tensor eigenspectrum of the Laplacian on n-spheres, J. Math.
Phys. 26 (1985) 65.
[19] P. van Nieuwenhuizen, An introduction to simple supergravity and the KaluzaKlein program, in: B.S.
DeWitt, R. Stora (Eds.), Relativity, Groups and Topology II, North-Holland, Amsterdam, 1984.

The secret gauging of flavor symmetries in

noncommutative QFT
Ken Intriligator, Jason Kumar
Department of Physics, University of California, San Diego, La Jolla, CA 92093-0354, USA
Abstract
We show that flavor t Hooft anomalies automatically vanish in noncommutative field theories
which are obtained from string theory in the decoupling limit. We claim that this is because the
flavor symmetries are secretly local, because of coupling to closed string bulk modes. An example is
the SU(4) R-symmetry of N = 4 D = 4 NCSYM. The gauge fields, along with all closed string bulk
modes, are not on-shell external states but do appear as off-shell intermediate states in nonplanar
processes; these closed string modes are thereby holographically encoded in the NCFT. 2002
Elsevier Science B.V. All rights reserved.
PACS: 02.40.Gh; 11.25.-w
1. Introduction
Noncommutative field theories (NCFT) exhibit fascinating similarities with string
theory. In the realization of noncommutative field theories via D-branes in a background B
field, one sees that the decoupling limit [1] retains some of the stringy behavior, which is
usually thrown away in the decoupling limit for commutative QFTs. In particular, diagrams
with nonplanar momentum exhibit UV suppression, which can lead to UV/IR mixing [2,3].
This effect was interpreted in [2,3] as nondecoupling of closed string modes (see also [4]),
and can also be viewed as the effect of stretched open strings [59]. A way to think about
this [2] is that the closed string propagator involves ( )2 g , which one might expect to
vanish in the decoupling limit 0; however, holding the open string metric fixed to
E-mail addresses: keni@ucsd.edu (K. Intriligator), j1kumar@ucsd.edu (J. Kumar).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 4 1 - 1
300
K. Intriligator, J. Kumar / Nuclear Physics B 624 (2002) 299306
G = , we actually have in the decoupling limit

( )2 g 2 .
(1)
We claim, based on Eq. (1), that all closed string modes remain coupled to the NCFT.
In the decoupling limit, the closed string modes do not exist as on-shell external states,
but they still remain coupled as intermediate, off-shell contributions to nonplanar loop
diagrams. As a particular case of this, we argue that worldvolume flavor symmetries are
secretly gauged, with the gauge bosons coming from closed strings which remain coupled
via Eq. (1). As evidence, we discuss anomalies.
The UV suppression of nonplanar diagrams in NCFT, which is related to Eq. (1),
ensures that no nonplanar diagrams are anomalous [10,11]. Thus mixed anomalies, which
could only arise via nonplanar diagrams, always automatically vanish in NCFT. The
interpretation of [11] is that these anomalies are cancelled because the NCFT automatically
contains the necessary closed string axion fields which cancel the anomaly by the Green
Schwarz mechanism, precisely as in string theory. This can be seen explicitly in string
theory constructions of chiral NCFTs. As also pointed out in [11], all ABJ type anomalies,
involving both flavor and gauge currents, are also automatically cancelled by a Green
Schwarz type mechanism.
We will here argue that all flavor t Hooft anomalies also automatically vanish in NCFTs
which are obtained via string theory. This is caused by the same nonplanar UV suppression
factors which occur via Eq. (1). Our interpretation of this, as well as the vanishing of the
ABJ anomalies, is that the flavor symmetries are secretly gauged, and thus necessarily
anomaly free. String theory constructions, of course, will always lead to a consistent theory.
The t Hooft anomaly cancellation, and secret gauging, is via closed string modes, which
remain coupled essentially due to Eq. (1).
The flavor symmetries of the NCFT living in the worldvolume of a D-brane are the
normal bundle rotations of the directions transverse to the brane. Before taking the
decoupling limit, these are gauge symmetries of the bulk theory, which includes gravity.
The worldvolume flavor current couples to bulk gauge fields, which can be written in terms
of the bulk metric. For D-branes in a general curved bulk metric background, we will have
terms in the worldvolume action like [14]

d p+1 x str gab a D a D b + fermion terms.
Here str is the symmetrized trace over the worldvolume U (K) gauge indices, D are
the U (K) gauge (and not flavor) covariant derivatives, with = 0 p the worldvolume
spacetime index, and a, b, c run over the bulk transverse directions. Expanding this
around a classical solution for a gives a coupling

d p+1 x gab,c str D a D b c + fermion terms,
(2)
with gab,c the derivative of the classical bulk metric. These terms lead to a coupling
between the worldvolume flavor current J[bc] = tr( [b c] ) + fermion terms (of course,
its the fermion terms which are relevant for anomalies) and a bulk gauge field:

d p+1 xJ[bc] A[bc] with A[bc] = ga[b,c] a ,
(3)
301
where a is a U (K) adjoint and a is its eigenvalue, which we take to be all the same
for K coincident branes. Simply put, the D-brane couples to angular momentum, which is
gauged in the full string theory. The gauge field is the Christoffel symbol, i.e., derivatives
of the graviton, ga[b,c] , pulled back to the worldvolume via a .
The above coupling between the flavor current and the bulk closed string modes goes
away in the commutative 0 decoupling limit. However, in the NCFT noncommutative
decoupling limit, the worldvolume theory retains this coupling to the bulk closed string
modes in intermediate channels of nonplanar loop amplitudes. In particular, for our present
case of interest, the NCFT retains the information that the worldvolume flavor symmetries
correspond to bulk gauge symmetries. Also retained is the cancellation of all gauge
anomalies, including that of the gauged flavor symmetry.
Though our discussion can be generalized, we focus on a particular example, noncommutative N = 4 U (N) gauge theory. Both the commutative and the noncommutative theory respect a SU(4)R flavor symmetry. In the commutative case, the Tr SU(4)3R t Hooft
so the t Hooft
anomaly is nonzero: all of the gauginos are in the 4, with none in the 4,
anomaly is Tr SU(4)3R = |G|, the dimension of the gauge group. This is an obstruction
to gauging SU(4)R in the commutative theory; one would first have to add some additional massless fields to cancel this obstruction. Though t Hooft anomalies are generally
expected to remain unchanged by the dynamics of the theory, we claim that making the
theory noncommutative does effectively change the t Hooft anomaly, actually making it
vanish.
In the full 10d string theory, before taking the decoupling limit, the more precise
statement about the cancellation of the t Hooft anomaly is that the SU(4)R current is
indeed conserved, but there is anomaly inflow of this current between the brane and the
bulk spacetime. The natural holographic analog of this for the NCFT is that current inflow
into the bulk should correspond to current flowing across energy scales.
In Section 2 we compare commutative and noncommutative field theory flavor anomaly
calculations to string theory anomaly calculations, and will demonstrate that the flavor
anomalies vanish in noncommutative field theories obtained from string theory in the
decoupling limit. In Section 3, we argue that this anomaly cancellation is in fact a
consequence of the reintroduction of closed string modes, which cause the symmetry to
be gauged. This is outlined in the matrix model description of NCFT.
2. The cancellation of flavor t Hooft anomalies

Though our discussion applies to any noncommutative gauge theory which arises as
a limit of string theory, we will focus on the particular case of noncommutative N = 4
U (N) gauge theory, which arises via D3-branes in the presence of a background NS
NS B-field. The noncommutative theory preserves the SU(4) flavor symmetry. Before
taking the decoupling limit, the SU(4) symmetry is a gauge symmetry, with the gauge
field corresponding to the graviton, as in Eq. (3). (As seen in 10d IIB sugra on S 5 [12],
the SU(4)R gauge field actually comes from a linear combination of the metric and the
four-form gauge field.)
302
Consider the Tr SU(4)3R t Hooft anomaly first in the commutative theory before taking
the decoupling limit. This anomaly can be computed by a string theory world-sheet annulus
diagram which, as we will discuss, yields the result that the anomaly vanishes. Of course,
this is the expected result, given that SU(4)R is a gauge symmetry before taking the
decoupling limit. Now consider the commutative field theory in the decoupling limit, which
has a nonzero Tr SU(4)3R t Hooft anomaly. This nonzero result is obtained because, in the
commutative decoupling limit one omits the region of the moduli space where the annulus
becomes a long, thin, cylinder. This omitted part is the nonzero bulk inflow contribution
to the anomaly, coming from closed string modes in the bulk, which would cancel the
anomaly if included.
This is similar to the string theory cancellation of reducible anomalies, which arise from
nonplanar diagrams. These diagrams are regularized by string theory, and thus there is no
reducible anomaly, even if the corresponding field theory does have one. Again, the field
theory limit omits the region of the annulus moduli space where it is a long cylinder, which
cancels the anomaly. The spacetime interpretation is that the long cylinder is the closed
string channel, which contains the tree-level diagram involving the GreenSchwarz field
responsible for cancelling the anomaly.
Now NCFT, even in the decoupling limit, retains stringy behavior in nonplanar
diagrams. For example, the stringy automatic GreenSchwarz anomaly cancellation of
reducible anomalies occurs in NCFT [11]. We now argue that in NCFT the Tr SU(4)3R
t Hooft anomaly is also automatically cancelled, even in the decoupling limit, by the bulk
inflow closed string contributions to the annulus diagram. To do this, we consider the full
string theory anomaly diagram, and then take the noncommutative field theory limit. We
will see that, unlike the case of the commutative field theory limit, in the NCFT limit the
diagram is still finite and hence nonanomalous.
The Tr SU(4)3R anomaly diagram is an annulus with three insertions of the SU(4)R
gauge field (essentially the bulk graviton, as in Eq. (3)). Because these gauge fields are
closed string modes, their vertex operators are to be inserted in the bulk of the annulus
diagram, at locations which are to be integrated over. Fortunately, our main conclusion
will not require that we actually compute the diagram in detail. Instead, we only need to
argue that it is regulated in the UV, as this is sufficient to show that the Tr SU(4)3R anomaly
vanishes. To show that the diagram is regulated in the UV, we does not even need to worry
about specific form of the gauge boson vertex operator; all that is needed is the form of the
worldsheet propagator between the closed string vertex operator insertions.
The worldsheet propagator is the Green function on the annulus, with nonzero B-field
and appropriate boundary conditions, which was exhibited, e.g., in [9,13]. The relevant
term is
G (1 , 2 )
1 2 2

1 2t | t
(G)
[Re 1 ]2 + [Re 2 ]2 2

+
=
ln +
1
2
8 2
t
t
2 2t
|t
+ irrelevant terms,
(4)
where t is the modulus of the annulus and 1,2 are the worldsheet bulk insertion points.
If there is nonplanar momentum k flowing between these vertex operator insertions, the
303
annulus diagram will contain a factor of

exp k k G ;
these closed string factors yield the stringy UV regulator for the diagram, which imply
that the diagram is finite, and thus anomaly free, in the full string theory before taking the
0 decoupling limit.
We now consider the NCFT limit, 0 with t fixed. The anomaly will still vanish
because, even in the decoupling limit, the above UV suppression factors continue to
provide a UV regulator for the anomaly diagram. This is because the above propagator,
in the decoupling limit, continues to contain terms of the form
( 2 )
(5)
t
which lead to the UV regulator in the NCFT limit. There is indeed such a term in the
above propagator whenever the vertex operator insertion locations satisfy Re 1 = Re 2 .
A special case of this is all nonplanar open string diagrams, where one vertex operator is
on one boundary of the annulus, Re 1 = 0, and the other is on the other boundary of the
annulus, Re 2 = . But the above UV suppression factor also occurs for closed string
vertex operators, inserted in the bulk of the annulus, as long as Re 1 = Re 2 .
For our SU(4)3R anomaly calculation, we have 3 closed string vertex operators inserted
at locations i , i = 1, . . . , 3, which should be integrated over the bulk of the annulus. As
long as the insertions are at different Re i , the diagram will have the UV damping factors
discussed above, and thus no anomaly. The integration over the i will include regions
where the Re i are not separated, and thus without the UV damping, but such regions
are a set of measure zero. Thus, the integrated diagram remains UV finite, and the SU(4)
R-symmetry anomaly is automatically canceled. It is easy to see that this generalizes to
all global symmetries in string-derived NCFT, since any global symmetry in such a theory
must arise from a gauge symmetry with a closed string gauge boson in the underlying string
theory. The associated one-loop annulus anomaly diagram with closed string insertions will
be finite, exactly as above.
Note that, although the usual U (N) symmetry is noncommutative, the SU(4)R gauge
symmetry will be commutative. This is because SU(4) gauge boson is a closed string, so
the argument in [1] for noncommutivity, based on open string gauge fields, does not apply.
Of course the SU(4)R symmetry could not possibly have been noncommutative in any
case, since its actually Spin(6), with scalars in the 6 and fermions in the spinor 4.
3. Reappearance of the bulk graviton couplings

The cancellation of the flavor anomalies in the previous section was via the closed
string UV suppression factors of the form e t k g k , which do not become trivial in the
NCFT decoupling limit thanks to the scaling of Eq. (1). These suppression factors can be
%x 2
considered as a stretched string effect [59] as they can be written as e t [8].

The anomaly cancellation is clearly very similar to that of the full string theory before
taking the decoupling limit. The natural interpretation, as in string theory, is that the
304
annulus diagram differs from the naive field theory diagram by including the closed
string propagation region of the moduli space, where the cylinder has nonzero height.
Thus the anomaly is naturally regarded as being cancelled by closed string bulk inflow,
involving closed string modes which are re-introduced by UV/IR mixing. This is analogous
to the claim of [11], that the mixed anomalies are canceled by the coupling of the
noncommutative gauge bosons to closed string axion-type fields, which cancel the anomaly
by the GreenSchwarz mechanism. (A difference is that there the closed string field is a
twisted sector axion, living in the brane worldvolume, whereas here the closed string modes
generally live in the bulk.)
So the Tr SU(4)3R anomaly cancellation mechanism is naturally the same as in the full
string theory. Moreover, the reason why this actually had to be the case is also naturally the
same as in the full string theory: the flavor symmetries are secretly gauge symmetries. This
is because the closed string gauge bosons, which couple to the flavor current as in Eq. (3),
actually do not decouple in the NCFT limit. They do not exist as external, on-shell states.
But, nevertheless, their presence is seen in nonplanar NCFT diagrams, which includes all
closed string modes in the intermediate channel via the propagators (or, in other words,
UV/IR mixing) discussed in the previous section. Because NCFT retains the couplings, as
in Eq. (3), to the closed string gauge fields, the SU(4)R symmetry is secretly gauged.
This interpretation also explains why the observation of [11], that all ABJ type
mixed flavor-gauge anomalies automatically vanish (via GreenSchwarz cancellation) in
noncommutative field theories, also actually had to be the case. If the flavor symmetries
were really global symmetries, such anomalies need not vanish. But because the flavor
symmetries are secretly gauged, consistency of the theory requires these anomalies, along
with the t Hooft anomalies, to all be cancelled. String theory, and also the NCFT
decoupling limit, automatically ensure that this is the case.
The secret gauging of the SU(4)R symmetry, along with the supersymmetry algebra,
implies that gravity is also included and one is discussing supergravity. This is consistent
with our picture for NCFT, since all of the closed string modes, and in particular the 10d
supergravity fields, remain coupled as off-shell, intermediate states in nonplanar diagrams.
The fact that 10d bulk closed string modes remain coupled via nonplanar diagrams also fits
with the remark in [15], that the leading nonplanar one-loop effective action will lead to a
10d, nonrelativistic gravitational force between sources of stress-energy.
The coupling of the worldvolume flavor current to the bulk can also be seen in the
matrix model description of NCFT. As an example, consider Dp-branes with p even
via the M-theory matrix model (one can similarly consider p odd in the IIB matrix
model). Follow the discussion in [16], we expand the above action around a solution which
a+p
captures the noncommutivity: XI = X0I + Y I , with X0i = x i and X0 = 0, Y i = ij A j ,
a+p

a
i
j
Y
= 2 , with i = 1, . . . , p, a = 1, . . . , 9 p, and [x , x ] = i ij (with ij of
rank p). For flat gI J , this leads to the action for the NCFT YangMills theory in the
worldvolume of the Dp-brane [16].
To see our couplings to the bulk gravitons, we should expand the matrix model action
for a general curved metric gI J [14]:

1
dt STr gI J (X)Dt XI Dt XJ gI K (X)gJ L (X) XI , XJ XK , XL + , (6)
2
305
where STr is the symmetrized trace, . . . are the fermion terms, and I, J = 1, . . . , 9 run
over the directions transverse to the D0s. Expanding around the solution corresponding to
NCFT, we replace
gI J (X) =

1
gI J,R1 Rn (X0 )Y R1 Y Rn
n!
n=1
for all occurrences of the metric in the above action. Following [16], this then leads to the
NCFT action, along with additional couplings, as in [17], like

1 I J (i1 in )
i1 in gI J (0) + ,
dt
T
n!
n=0
with T I J (i1 in )
moments of the stress tensor. The n = 1 term in the above yields our desired
coupling Eq. (3) between the worldvolume flavor currents and the associated bulk gauge
field. Even though there is no noncommutivity in the time direction, J0[bc] does couple to
an associated bulk gauge field, as seen upon expanding the kinetic term in the action (6).
Finally, we make a general comment about the kinetic terms of the closed string modes.
In the NCFT limit, say for D3-brane with the noncommutivity only in the x 1 and x 2 spatial
directions, [x1 , x2 ] = i , the closed string metric Eq. (1) will have nonzero components
only in the noncommutative directions. Thus, the closed string modes with no momentum
in the noncommutative directions will decouple (these are a set of measure zero). Since
the only closed string modes which couple have momenta along the noncommutative
directions, which are taken to be only space directions, those modes will be space-like.
The exponential factor e t k g k appearing in the nonplanar amplitude will thus have an
exponent which is always negative, and thus always damp the UV limit of the integration
over nonplanar momenta. Though there are no finite energy poles associated with the
closed string degrees of freedom, the coupling in the noncommutative directions between
the closed string gauge boson modes and the flavor currents are sufficient to ensure
finiteness of the one-loop diagram and invariance of the theory under the local flavor
symmetry transformations.
Acknowledgements
We would like to thank J. Gomis, A. Rajaraman, L. Susskind, M.M. Sheikh-Jabbari
and M. Van Raamsdonk for useful discussions and N. Seiberg for several useful
correspondences. J.K. would like to thank the Aspen Center for Physics for its hospitality.
This work is supported by DOE-FG03-97ER40546.
References
[1] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032, hep-th/9908142.
[2] S. Minwalla, M. Van Raamsdonk, N. Seiberg, Noncommutative perturbative dynamics, JHEP 0002 (2000)
020, hep-th/9912072.
306
[3] M. Van Raamsdonk, N. Seiberg, Comments on noncommutative perturbative dynamics, JHEP 0003 (2000)
035, hep-th/0002186.
[4] A. Rajaraman, M. Rozali, Noncommutative gauge theory, divergences and closed strings, JHEP 0004 (2000)
033, hep-th/0003227.
[5] M.M. Sheikh-Jabbari, Open strings in a B-field background as electric dipoles, Phys. Lett. B 455 (1999)
129, hep-th/9901080.
[6] D. Bigatti, L. Susskind, Magnetic fields, branes and noncommutative geometry, Phys. Rev. D 62 (2000)
066004, hep-th/9908056.
[7] Z. Yin, A note on space noncommutativity, Phys. Lett. B 466 (1999) 234, hep-th/9908152.
[8] H. Liu, J. Michelson, Stretched strings in noncommutative field theory, Phys. Rev. D 62 (2000) 066003,
hep-th/0004013.
[9] H. Liu, J. Michelson, *-trek: the one-loop N = 4 noncommutative SYM action, hep-th/0008205.
[10] C.P. Martin, The UV and IR origin of non-Abelian chiral gauge anomalies on noncommutative Minkowski
spacetime, hep-th/0008126.
[11] K. Intriligator, J. Kumar, *-wars episode I: the phantom anomaly, hep-th/0107199.
[12] H.J. Kim, L.J. Romans, P. van Nieuwenhuizen, The mass spectrum of chiral N = 2 D = 10 supergravity on
S**5, Phys. Rev. D 32 (1985) 389.
[13] S. Chaudhuri, E.G. Novak, Effective string tension and renormalizability: string theory in a noncommutative
space, JHEP 0008 (2000) 027, hep-th/0006014.
[14] M.R. Douglas, D-branes in curved space, Adv. Theor. Math. Phys. 1 (1998) 198, hep-th/9703056.
[15] M. Van Raamsdonk, The meaning of infrared singularities in noncommutative gauge theories, hepth/0110093.
[16] N. Seiberg, A note on background independence in noncommutative gauge theories, matrix model and
tachyon condensation, JHEP 0009 (2000) 003, hep-th/0008013.
[17] W.I. Taylor, M. Van Raamsdonk, Supergravity currents and linearized interactions for matrix theory
configurations with fermionic backgrounds, JHEP 9904 (1999) 013, hep-th/9812239.

An associative and noncommutative product for the

low energy effective theory of a D-brane in curved
backgrounds and bi-local fields
Kiyoshi Hayasaka, Ryuichi Nakayama
Division of Physics, Graduate School of Science, Hokkaido University, Sapporo 060-0810, Japan
Abstract
We point out that when a D-brane is placed in an NSNS B field background with nonvanishing
field strength (H = dB) along the D-brane worldvolume, the coordinate of one end of the open
string does not commute with that of the other in the low energy limit. The degrees of the freedom
associated with both ends are not decoupled and accordingly, the effective action must be quite
different from that of the ordinary noncommutative gauge theory for a constant B background.
We construct an associative and noncommutative product which operates on the coordinates of
both ends of the string and propose a new type of noncommutative gauge action for the low energy
effective theory of a Dp-brane. This effective theory is bi-local and lives in twice as large dimensions
(2D = 2(p + 1)) as in the H = 0 case. When viewed as a theory in the D-dimensional space, this
theory is nonlocal and we must force the two ends of the string to coincide. We will then propose a
prescription for reducing this bi-local effective action to that in D dimensions and obtaining a local
effective action. 2002 Elsevier Science B.V. All rights reserved.
PACS: 04.60.Ds; 11.25.-w; 11.15.-q
1. Introduction
Noncommutative field theories [1,2] are obtained in an 0 limit of the open string
theory. This noncommutativity is inherent in the open string theory [3]. By considering
an open string in an NSNS B field background and taking a specific field theory limit
the effects of the B field are encoded into a special multiplication rule of functions, the
E-mail addresses: hayasaka@particle.sci.hokudai.ac.jp (K. Hayasaka),
nakayama@particle.sci.hokudai.ac.jp (R. Nakayama).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 6 1 - 7
308
K. Hayasaka, R. Nakayama / Nuclear Physics B 624 (2002) 307326
product or Moyal product. By studying such field theories we expect that open string
physics may be understood in the framework of field theory. Until recently such investigations are mostly restricted to constant B backgrounds [47]. The algebra of the coordinates
of the end points of the open string was studied, the connection of the commutative and
noncommutative descriptions of the gauge theory was elucidated, and so on.
When the B field is constant, the background metric g is flat. Attempts to extend
the analysis to the curved background, i.e., the nonvanishing B field strength H = dB,
appeared in the context of open string theory in WZW models [8]. In this approach it
was found that the algebra of functions on the D-brane worldvolume is given by the
q-deformation of the Lie algebra. The general framework for general backgrounds was,
however, not yet obtained.
More recently a noncommutative field theory for a weak field strength H and a weakly
curved background was investigated [9].1 Several correlation functions of the string theory
are obtained in the 0 limit and it was concluded that the algebra of functions
on the D-brane worldvolume is nonassociative and noncommutative. Various correlation
functions are expressed in terms of a nonassociative product . In [11] the commutation
relations of the coordinate x of the end point of the open string were studied by using the
approximation of a very short and slowly moving string. It was found that the commutators
of the coordinate x contain a momentum p and it was argued that the space on which the
D-brane lives is not the usual noncommutative space which we are getting used to. In [12]
a new product was defined in order to make the product associative on the functions of
x and p and an attempt to define a gauge transformation in this xp space was presented.
An explicit construction of the gauge theory action, however, was not obtained yet.
The purpose of the present paper is to understand how the low energy effective
theory can be formulated as an associative and noncommutative gauge theory, when the
background space is curved and the field strength H = dB along the direction of the
D-brane worldvolume is nonvanishing. Our analysis will be restricted to the bosonic string.
We assume that the field strength is weak and perform analysis in perturbation series in H
up to O(H 1 ). We will first obtain the commutation relations of the coordinates x , y
of the two ends (at = 0 and ) of the open string in the low energy limit (Section 2).
This will be performed by discarding the oscillators of the string variable X (, ), but
without further approximation. The result is striking. x and y do not commute with
each other in contrast to the H = 0 case. This means that when we take the 0 limit,
the gauge and scalar degrees of freedom at both ends of the string are not decoupled in the
low energy effective theory. We are forced to take into account the degrees of freedom at
both ends in constructing the field theory description and the numbers of gauge and scalar
fields are doubled. Furthermore, the noncommutative product constructed according to
the algebra of x and y is associative, but operates on both coordinates (Section 3). We are
thus lead to consider a bi-local field theory. All fields are functions of x and y. The gauge
theory must be formulated in the 2D-dimensional (x, y) space instead of the ordinary
D-dimensional x space. This bi-local field theory will, however, be acausal and must be
1 Topological sigma models with a H = dB term were also considered in [10] and the related underlying
geometry was called H -Poisson or twisted Poisson one.
309
regarded as an intermediate step toward construction of the local effective field theory,
which will be discussed later.
Construction of this bi-local gauge theory can be performed in a standard way. The
derivatives are generalized in such a way that they satisfy Leibnitz rule with respect to
the product (Section 4). The commutator of the gauge covariant derivatives defined in
terms of these new derivatives gives the gauge field strength (Section 5). The trace or the
integral is defined in such a way that the cyclic property is respected. An obstruction in
this prescription is the curved D-brane worldvolume. The metric of the effective gauge
theory is the open string metric [7], which is curved already at O(H 1 ) [9]. Because the
gauge symmetry of the noncommutative gauge theory is realized by differential operators
on the fields, if the action contains such a nonconstant metric, one anticipates that the
gauge symmetry may be broken. Remarkably, we found that there exists a coordinate
transformation in the (x , y ) space (not in the x space) which effectively makes H = 0.
In other words in a suitably chosen frame (x (x, y), y (x, y)) the coordinates satisfy the
commutation relations of the H = 0 theory. Therefore, it is natural to further assume that
the metric is flat in this frame. Moreover, the derivatives with respect to x and y coincide
with the new derivatives defined above to satisfy Leibnitz rule. By writing down the action
for the noncommutative gauge theory with H = dB = 0 in the flat (x , y ) frame and by
going to the (x, y) frame by means of the inverse coordinate transformation, we obtain the
action for a bi-local, noncommutative gauge theory in curved backgrounds.
As mentioned above, the above bi-local theory is nonlocal and will be acausal. Next,
we will present a prescription to reduce our bi-local theory to a local theory (Section 6).
Actually, in order to obtain a massless string we must consider a limit y x. This is not
automatically achieved in the 0 limit. We will set y = x by hand in the integrand
of the effective action and integrate this over x with a suitable weight function (x). We
will show cyclic property of this reduced effective action which relates various correlation
functions of the fields at x and y, i.e., at both ends of the string. This reduced effective
action is still invariant under a part of the gauge transformation. We propose this as the low
energy effective action of the open string in the H = 0 background. This reduced effective
action, however, contains in the integrand and the manipulation such as differentiation
with respect to x and/or y must be performed before setting y = x. In this sense this
reduced effective action is not the ordinary action in the x space.
Finally in Section 7 we will speculate on the relation of our noncommutative gauge
theory to the Matrix model in curved backgrounds.
2. Noncommutativity of the two end-point coordinates of an open string

We will quantize an open string theory with a Dp-brane in the presence of an NSNS
B field along the Dp-brane, whose worldvolume spans a subspace = 0, 1, 2, . . . , p. The
field strength of the B field, H = dB, is assumed to be small but nonvanishing in the
direction of the Dp-brane worldvolume.
In this case we can perform a perturbative analysis in H . Such an analysis was first
attempted in [9]. We will simplify the problem by taking H constant. By the consistency
of the background due to Weyl invariance the metric tensor g must be a constant up to
310
O(H 1 ) [9]. We will choose a gauge

1
B (X) = b + H X ,
(1)
3
where b is a constant. We assume that H is nonvanishing for , , . . . = 1, 2, . . . , D
( p). When H = 0, b is also in general nonvanishing because we can always shift
X by a constant. We also assume that B is block diagonal, i.e., Bi = Bi = 0 for D,
i > D and D is even and det(b ),=1,2,...,D = 0. In what follows we will restrict our
attention only to this D-dimensional subspace.
The worldsheet action of the open string in the conformal gauge is given by

1
S=
(2)
d d g X X + B X X .

4
The coordinate X must satisfy the following boundary conditions at = 0, .
g X B X = 0.
Inside the worldsheet it must satisfy the equations of motion

g X + X X H X X = 0.
(3)
(4)
For small H the metric g is constant up to O(H 1 ) and equation of motion (4) reads

X + X g H X X = 0 + O H 2 ,
(5)
where the dot and prime stand for / and / , respectively. The boundary condition
(3) is now expressed as
1
X J X g H X X = 0 (at = 0, ),
3
(6)
where J = g b .
We will solve (5) under the condition (6) up to O(H 1 ). We set X = X(0) + X(1) .
The leading order solution is given by
X(0) (, ) = x + p + (Jp) + (1 + J ) f ( + )
+ (1 J ) f ( ),
where
(Jp)
stands for
f ( ) =
(7)
and

1 in
an e
+ an ein .
n
(8)
n=1
x , p are zero modes and an oscillators. We will next plug this in (5) and derive an
equation for the O(H 1 ) correction X(1) . This can easily be solved:

1 2
2 g H p (Jp)
4
1
( )g H p (1 + J ) f ( + )
4
X(1) =
1
+ ( + )g H p (1 J ) f ( )
4
1
+ ( )g H (Jp) (1 + J ) f ( + )
4
1
+ ( + )g H (Jp) (1 J ) f ( )
4
1
+ g H (1 + J ) (1 J ) f ( + )f ( )
2
+ g h ( + ) + g k ( ).
311
(9)
Here h and k are some functions to be determined by (6).
The energies of the oscillation modes are proportional to 1/ and in the 0

limit they can be ignored. In this paper we will restrict attention only to the zero modes
and simply set an = 0 (f = 0). One may speculate that x, p and a may mix up due to
the interaction and the commutators of x, p may contain a. We have explicitly checked
that this is not the case up to O(H 1 ). The full analysis including the oscillators is now in
progress and will be reported elsewhere [13]. Eq. (6) forces h , k to satisfy an equation
(1 + J ) h ( + ) (1 J ) k ( )
1
1
1
= H x p H (Jp) p + J H p (Jp) ,
3
6
2
(10)
at = 0, . The primes on h , k stand for the derivatives with respect to the arguments.
This equation can be solved by subtracting (10) with = 0 from (10) with = and
+ . This determines h and then (10) gives k :

1
2 1 3J
H (Jp) p
H x p ,
24 1 + J
6
1
+
J

2
1
1 + 3J
H (Jp) p +
H x p .
k ( ) =
24 1 J
6
1
h ( ) =
(11)
The next step is to quantize this system. We will use the phase-space integral invariant
of Poincar I to derive Poisson brackets [5]. It is defined by

I=
d X (, ) P (, ),
(12)
where P is the canonical momentum conjugate to X ,

2 P = g X + B X .
(13)
X and P represent differential 1-forms. The quantity (12) is independent of and

defines the Poisson structure. Substitution of (7), (9), (13) into (12) yields I as a 2-form
in the basis x x , x p , p p . The coefficients are a 2D by 2D matrix
and its inverse gives the Poisson brackets. They are then converted to the commutation
312
relations:

i
x , x = 2 i + G G H x
3

i
3 G G H (Jp)
+
9

i
G + G H p ,
i
i
G H x
x , p = 2 G +
3

i
+ 3 G G H p
6

i

G G H (Jp) ,
2

4i
p , p = G G H (Jp) .
3
Here and G are defined by

J 1
1
1
g
,
G =
g
.
=
1 J2
1 J2
(14)
(15)
These are the noncommutative parameter2 and the open string metric for the H = 0 background [7].
Similar commutation relations in the 0 limit are obtained in [11] by using some
approximation of short and slowly moving string. This 0 limit is a specific one taken
with fine tuning [7]. g is adjusted to be O( 2 ), while x O( 0 ), b , H , p
O( 1 ). Hence J O( 1 ), O( 1 ) and G O( 0 ). Defining the total
momentum at = 0 by p 0 d P (0, ) O( 0 ), we get in the 0 limit

2i
H p ,
x , x = 2 i (x) +
3

i
x , p = i + H p ,
(16)
p , p = 0,
3
which agrees with the result of [11,12]. Here (x) is in (15) with J = g 1 b replaced by
g 1 B(x).
In [11] it is argued that the appearance of p on the righthand sides of (14) makes
the construction of the low energy effective action difficult. In [12] some proposal for a
construction of gauge transformations is presented. An explicit form of the action integral
in the xp space is, however, not proposed.
In this paper we will observe the relations (14) from a different view point and construct
a noncommutative gauge theory based on this algebra. Crucial point is that the coordinates
2 This is different from that used in the literature by a factor 2 . When multiplied by this factor, remains
finite in the 0 limit. We used the definition (15) in order to avoid messy expression for (14).
of the end points of the open string (at = 0) are given by

X (0, 0)zero modes = x ,

X (0, )zero modes = x + (Jp) + O H 1 .
313
(17)
Throughout this paper we will use x and y to stand for the coordinates of the two ends of
the open string. The difference of the two end point coordinates is thus proportional to p.

y x = (Jp) + O H 1 .
(18)
Note that this is O( 0 ) and does not automatically go to zero but is fixed in the 0
limit. For large momenta the length of the open string grows [15]. If b is invertible,
which we assume in this paper, so are J and , and we can invert (18):
p =

1 1
J
y x + O H 1 .
(19)
Now the algebra (14) can be reexpressed in terms of x and y :3

i
x , x = i + G G H x
3

i
+ 3 G G H y x
9

i
G + G H J 1 y x ,
(20)
9

i
3 G G H y x
x ,y =
18

i
G + G H J 1 y x ,
+
(21)
18

i
y , y = i G G H y
3

i
+ 3 G G H y x
9

i
G + G H J 1 y x .
(22)
9
Eqs. (20) and (22) shows that the coordinates of the ends of the open string x , y do
not commute. The commutator of ys is obtained from that of xs by interchanging x y
and reversing the overall sign. Furthermore, (21) shows that the coordinate of one end x
does not commute with that of the other y. This is in contrast to the H = 0 case [6,12].
The gauge and scalar degrees of freedom at the = 0 end and those at = are not
decoupled. To construct a low energy effective theory we need to consider both degrees
of freedom. This means that if the two ends of the string are separated, the low energy
effective field theory will become acausal. To avoid this problem we will take a limit

3 In what follows we will set 2 = 1.
314
y x in Section 5. Nevertheless, these two ends must be separated for a while in order to
construct an associative product.
We expect that the full string theory including the oscillators is causal. The reason for
the noncommutativity of x and y may lie in the fact that we discarded oscillators. It will
be interesting to study the commutators of the full string variable X (, ).
3. An associative and noncommutative product

It can be shown that the commutation relations (20)(22) satisfy Jacobi identities. We
will then construct a product which realizes this algebra. The parameter H is treated
as a small perturbation but must be taken into account to the full order. We report only
the results here:
f1 (x, y) f2 (x, y)
f1 (x, y) f2 (x, y)

i
f2
f2

f1
f1
G G H x
6
x x
y y

i
G G 3 H y x
18

f1 f2
1 f1 f2
1 f1 f2
f1 f2
x x 2 x y
2 y x y y

i
+
G + G H G1 1 y x
18

f1 f2
1 f1 f2
1 f1 f2
f1 f2
x x 2 x y
2 y x y y
1
G G H
36

f1
2 f2
2 f1
f2
f1
2 f2
2 f1
f2
x x x x x x y y y
y y y

1
3 + G G + G G + G G H
+
72

2 f2
2 f1
f2
f1
2 f2
2 f1
f2
f1
x x y
x y x y y x y x y
(23)
Here is the ordinary noncommutative product for x and y in the H = 0 case:
f1 (x, y) f2 (x, y)

i

i
exp f1 (x, y)f2 (x , y )
.
2
x x
2
y y
x =x,y =y
The noncommutative parameter is for x and for y.
(24)
The product (23) is associative. The proof of the associativity (up to O(H 1 ))

f1 (x, y) f2 (x, y) f3 (x, y) = f1 (x, y) f2 (x, y) f3 (x, y)
315
(25)
is straightforward.
Because the product contains both x and y, the product of the functions of x only,
f1 (x) f2 (x) will also depend on y. Hence we must regard all fields to be functions of
both x and y, i.e., bi-local fields.
4. Derivatives
The low energy effective theory on the D-brane is expected to be some kind of
noncommutative gauge theory. To construct such a theory we need to define proper
derivatives. Because the product (23) depends on the coordinates explicitly, this will
not satisfy Leibnitz rule:

(f1 f2 )
(26)
f1 f2 f1
f2 = 0.
x
x
x
To circumvent this problem we must modify the derivatives. We define

f
f
f
(x,
y)
f
(x,
y)
+
a
,
+
b
,
,
x
x
x
y

f
f
f
(x,
y)
f
(x,
y)
+
a
,
+
b
,
.
y
y
y
x
(27)
(28)
Here a , b , a and b are linear functions of x, y and are O(H 1 ). { , } is the anticommutator with respect to the product (24). We note that because a, b, a, b are linear
functions, {a, f/x} can be replaced by 2af/x in (27), (28).
These functions must be determined in such a way that the new derivatives (27), (28)
satisfy Leibnitz rule. By (27) we obtain
f1
f2
(f1 f2 )
f2 f1
x
x
x

f2
f1
= a , f1 a , f2
x
x

f2
f1
+ b , f1 b , f2 + .
y
y
(29)
Here [ , ] is the commutator with respect to the product (24). can be computed from the
definitions (23), (26). By requiring (29) to vanish we obtain the equations:

a
1
a
= G G H
x
x
9

1
G + G H G1 1 ,
18
(30)
316

1
a
b

G
H
+
y
x
36

1
G + G H G1 1 ,
+
36

1
b
b

=
G G 3 H
y
y
18

1
G + G H G1 1 .
18
General solution to these equations is given by
(31)
(32)

1 1
x
G G H
12

1
2x + y 1 G G 3 H
+
72

1
2x + y 1 G + G H G1 1
72

x 1 S y 1 U ,
(33)

1
x + 2y 1 G G 3 H
b (x, y) =
72

1
x + 2y 1 G + G H G1 1
+
72

y 1 T + x 1 U .
(34)
a (x, y) =
Here S (= S ), T (= T ) and U are some constants to be discussed later.
Similarly a , b can be determined. By using the x y symmetry we find

a (x, y) = a (y, x),
b (x, y) = b (y, x).
(35)
The field strength of the gauge fields will be defined as the commutator of the gauge
covariant derivatives. For this procedure to be consistent the derivatives (27), (28) must
commute. Otherwise, the commutator becomes a differential operator. For instance, from
(27) we obtain

b b f
a a f
+
2
,
f
=
2
(36)
x x
x
x x
x
x y
and the two coefficients of the derivatives of f must vanish. We obtain 4 more conditions
from the other commutators. It turns out the following three of the six conditions are
independent:
a a
=
,
x
x
b b
=
,
x
x
b
a
=
.
x
y
(37)
These equations impose the following conditions on S, T , U :

S S

1
1
G G G H ,
= G G H +
18
36
(38)
317

U 1 U
1 1 1
1 1 1
G
G
G H
G H
36
36
1 1 1 1 1
1
G
H + H ,
G
36
12

1
U = 1 T
G G 3 H
72
1 1
1

G +G
G
+
H .

72
=
(39)
(40)
If we recall the symmetry T = T , (39) comes out as a result of (40) and hence is
not an independent condition.
The solution to (38) is given by

S = c G G H ( G) + c ( G ) G + ( G ) G H

+ c ( G ) + ( G ) H G1 1

1
G G H G1 1 ,
+
(41)
36
c is a constant. T and U are arbitrary as long as (40) is satisfied. Unfortunately,
further conditions which may determine these values are not yet available to the authors.
The fact that the derivatives (27), (28) commute suggests the existence of a coordinate
transformation x, y x , y by which the derivatives (27), (28) are reexpressed as ordinary
derivatives:
=
=
+ 2a + 2b
(42)
x x x
x
y
(and a similar equation for /y). Here we used the property, {a, h} = 2ah which holds
for a linear function a. In this transformation x, y are treated as commuting variables.
Indeed Eq. (37) guarantees the existence of such a transformation and it is given by
1 1
x x
G G H
18

1
+ x x 1 G + G G1 1 H
36

1
y x 1 G G 3 H
36

1
+ y x 1 G + G G1 1 H
36

1
+ y y 1 G G 3 H
72

1
y y 1 G + G G1 1 H
72

+ x x 1 S + 2x y 1 U y y 1 U ,
x = x +
and a similar equation with the replacement x y,
x
y .
(43)
318
By using (20)(22) it is possible to show that in the primed coordinate system x and y
commute with each other:

x , x = i ,
(44)
y , y = i ,
x , y = 0.
Here the quadratic terms in (43) are assumed to be symmetrized. In other words the
effect of the field strength H can be compensated by a coordinate transformation. This
fact suggests that the metric tensor in the (x , y ) space may be also flat and given
by G . We should note that this is not the ordinary space but a dimensionally-doubled
space combining the coordinates of the two ends of the string and that the transformation
(43) mixes up x and y as a whole. It is tempting to assume that the existence of such a
transformation persists to all orders of H . We do not have a proof at present.
Let us construct a trace operation Tr f (x, y). This must satisfy the cyclic property:
Tr f1 f2 = Tr f2 f1 .
(45)
We assume the form

Tr f d D x d D y G(x, y)f (x, y)
(46)
and determine the function (x, y) by the condition

D
D
d x d y G(f1 f2 ) = d D x d D y G(f2 f1 ).
(47)
For H = 0 we must also have = 1. Note that on the right-hand side of (46) the product
is the ordinary one. G is det G . It is not hard to show that this function is given by

1
1
x + y H x + y G1 1 G1 H . (48)
6
6
It may be natural to expect that this extra factor in the trace has an origin in the
Jacobian from (x, y) to (x , y ). We indeed obtain
(x, y) = 1
det

1
1
(x , y )
=1
x + y H
x + y G1 1 G1 H
(x, y)
12
12

+ 2 x + y 1 S + U
(49)
and we can show from (38)(40) that this coincides with .

5. Noncommutative gauge theory for H = 0
We are now ready to define gauge transformations and write down the gauge theory
action. This action will be an integral over x and y and defines a bi-local theory. We will
construct this bi-local theory as an intermediate step for obtaining a local gauge theory
in the next section.
Let us define the gauge covariant derivatives by
D = /x + iA = /x + iA ,
= /y i A = /y i A .
D
(50)
319
Here A , A are the gauge fields in the (x , y ) frame and transformed to those in the
(x, y) frame like the derivatives (42),
A = A + 2a A 2b A ,
A = A + 2a A 2b A .
(51)
We will use a bar to denote functions associated with the coordinate y. In (50), (51) the
signs in front of A and A are opposite, because the two ends of the open string have
opposite charges. The gauge transformation is defined by
D U D U 1 ,
U D
U 1 ,
D
U U 1 ,
(52)
where U = exp (i) is a gauge function. exp is an exponential defined in terms of .

is a scalar field in the adjoint representation. For an infinitesimal these read

,
A = i , A +
,
= i[, ].
A = i , A
x
y
(53)
Here [ , ] is the commutator with respect to the product (23). An infinitesimal transformation of A , A is consistently obtained:

A = i[, A ] + i , a , A i , b , A ,
(54)
x

A = i[, A ] + + i , a , A i , b , A .
(55)
y
Now by using the standard prescription we can write down the action integral for the
bi-local noncommutative gauge theory:

1
d D x d D y G(x, y)
Sbi-local = 2
4gYM V

G G [D , D ] [D , D ] 2G D , Xm D , Xm

, D
D
, D

+ Xm , Xn + G G D

m n
2

m
m
, X

,X

2G D , X D
+ X

d D x d D y GV 1 L(x, y),
(56)

where V is the volume d D x G and gYM the YangMills coupling. We explicitly write
m are the scalar fields
only the integration measure relevant to our discussions. Xm and X
describing the fluctuations of the D-brane. The scalar fields are doubled, because there are
two such independent fluctuations at both ends of the open string. This action is invariant
under the gauge transformation (52) due to the cyclic property (47) of the trace (46).
Because G is constant, this does not break the gauge symmetry. There is no constraint
on the gauge function (x, y).
By substituting (27), (28), (50), (51) into (56) we will find a nonconstant metric in the
action. For example, we obtain for the commutator of the covariant derivatives:
(x,x)
(y,x)
(x,x)
(x,x)
+ 2a F
+ 2a F
+ 2b F
[D , D ] = i F

(x,y)
+ 2b F + R ,
(57)
320
where F s are the field strength in terms of the ordinary derivatives:
A A + i[A , A ] ,
x
x
(y,x)
(x,y)
= A A + i A , A = F
,
F
y
x
(y,y)
F
= A A i A , A
y
y
(x,x)
=
F
R is a sum of terms like

a
A
i
,
.
A
x
x
(58)
(59)
Here we omit its explicit form. By defining a new curved metric,

(x,x) (x, y) = G + 2G a + 2G a ,
G
(y,y) (x, y) = G + 2G a + 2G a ,
G
(x,y) (x, y) = 4G b ,
G
(y,x) (x, y) = 4G b ,
G
(60)
we obtain the part of the action for the gauge field:

1
y)
Sgauge field = 2
d D x d D y G(x,
4gYM V
(x,x)
(x,x)
(x,x) F (x,x) F

G
G
(y,y)
(y,y) G
(y,y) F (y,y) F
+G
(x,x) G
(x,y) F
+ 2G
(x,y)
F
(y,y) G
(y,x) F
+ 2G
(y,x)
F
(x,x)
(y,y)

(x,x)
(y,y)
+ 2G G F
R + 2G G F
R .
(61)
y) is the determinant of the 2D 2D matrix (G

(x,x)
(x,y)
(y,x)
(y,y)
Here G(x,
, G
/G
, G
) and
2
is a sum of terms similar to those of R . In the above
coincides with {(x, y)} . R
is an ordinary product. In this way the noncommutative gauge theory in the curved space
(,) depends on the coordinates, the action is gauge
is obtained. Although the metric G
can be obtained.
invariant. Similarly, an action for the scalars X, X
In the special case H = 0, x and y commute and the product of functions of x only is
still a function of x only. Thus it is natural to set A = A (x), A = A (y), Xm = Xm (x),
m = X
m (y). The product reduces to . Then the action Sbi-local becomes the sum of two
X
decoupled noncommutative gauge theories with noncommutative parameters and ,
respectively.
In H = 0 noncommutative gauge theory, one obtains unitary and causal S-matrix as
long as the noncommutative parameter has only space components [14]. If we regard (56)
as the action integral in 2D-dimensional space, we may be able to expect that this theory
is also causal, even if there are terms containing the field strength H . If we regard x and y
321
as two distinct points in the same D-dimensional space, however, this theory will be nonlocal and we cannot expect to obtain causal S-matrix. We must reduce Sbi-local to a single
x-integral and make the theory local.
6. The y x limit and the reduced effective action
We need to find a prescription to connect this action to the amplitudes of the string
theory in the low energy limit. Let us recall that the energy of the open string is proportional
to the length of the string divided by and given by
1
1
M G p p G (J 1 ) (y x) (J 1 ) (y x) .
In the low energy limit this goes to a finite value. A massive string will be unstable and
eventually decay into a massless string. To obtain a massless string we must take the y x
limit by hand. If one sends y to x directly in the product , (23), this product becomes nonassociative. In this sense the separation yx works as a kind of regularization to realize the
associative product . At the final step we need to reduce the x, y integral to a single x
integral. The integration measure is not just d D x (x, x) but must be modified.
As a simple example let us first consider the following integral in the (x, y) space:

d D x d D y (x, y) 1 (x, y) 2 (x, y) n (x, y)
(62)
i (x, y)s are scalar fields. We propose that the following reduced integral in the x space
should correspond to this integral in the y x limit:4

d D x (x) 1 (x, y) 2 (x, y) n (x, y) y=x .
(63)
In (63) we set y = x only after all the algebra associated with is finished. Here (x) is a
function which is determined by the requirement of cyclic property. Actually if we set
1
(x) = 1 + x H ,
3
the two point function of the functions of x only,

I2 [1 , 2 ] d D x (x) 1 (x) 2 (x) y=x
is given by
I2 [1 , 2 ] =

1
d D x 1 (x) 2 (x) 1 + x H .
3
(64)
(65)
(66)
This is cyclic invariant:

I2 [1 , 2 ] = I2 [2 , 1 ].
4 At O(H 1 ) this prescription is similar to the one (p = 0) adopted in [12] due to the relation (19).
(67)
322
Fig. 1. Cyclic property on the boundary of a disc.
Contrary to our expectation the function (x) is not proportional to

(x) = G 1 1 x H
det G
3

(x), where G
(x) = (g(1 (g 1 B(x))2 )) is the open string metric in
but to det G
the presence of H .
Furthermore, we can check that this correlation function enjoys the following interesting
property:

d D x 1 (x) 2 (x) y=x = d D x 1 (x) 2 (y) y=x

D

= d x 1 (y) 2 (x) y=x = d D x 2 (y) 1 (y) y=x .
(68)
This property reveals the string theory origin of the product. To see this it is helpful to
visualize the product as an operation adding one rung to a ladder. Let us consider a ladder
and name the two legs as x and y, respectively. The product adds one unit on top of the
ladder. i (x) adds one rung with the x-leg marked, while i (y) one with the y-leg
marked. This operation must be performed step by step and is noncommutative. Finally,
setting y = x and integrating over x join the two legs at the top and bottom, separately.
After this joining the legs are merged into a circle and we expect to have the cyclic property
as in the string amplitude. This is the case here. Eq. (68) is illustrated in Fig. 1.
Similarly, the three-point function

I3 [1 , 2 , 3 ] d D x (x) 1 (x) 2 (x) 3 (x) y=x
(69)
is given by
I3 [1 , 2 , 3 ]

i
= d D x (x) 1 (x) 2 (x) 3 (x) G G H x
6
( 1 2 3 + 1 2 3 + 1 2 3 )

1

+
(70)
G G G H 1 2 3 .
18
In the 0 limit those terms in (66) and (70) which contain the metric G drop and
agree with the results of correlation functions obtained in [9].
323
It is not difficult to show that the three point function (70) satisfies
I3 [1 , 2 , 3 ] = I3 [3 , 1 , 2 ] = I3 [2 , 3 , 1 ].
(71)
We can also show the following identities:

d D x 1 (x) 2 (x) 3 (x) y=x = d D x 3 (y) 2 (y) 1 (y) y=x

= d D x 1 (x) 2 (x) 3 (y) y=x

= d D x 1 (x) 3 (y) 2 (x) y=x

= d D x 3 (y) 1 (x) 2 (x) y=x

= d D x 1 (x) 3 (y) 2 (y) y=x

= d D x 3 (y) 1 (x) 2 (y) y=x

= d D x 3 (y) 2 (y) 1 (x) y=x .
(72)
The locations of the fields can be shifted from one end x of the string to the other y, and
vice versa, as long as the ordering on a circle is unchanged. This is natural from the string
theory point of view.
Higher point functions In [1 , . . . , n ] do not have the cyclic property. In [9] it was
shown that in order to obtain the four point function in terms of the non-associative product
one must take a specific linear combination of the ordered products of the functions with
various positionings of the parenthesis. In our approach the correlation functions will be
obtained by functional derivatives of the effective action and various orderings will be
automatically taken into account due to the Bose symmetry.
The cyclic property (67), (68), (71), (72), and the agreement of (66), (70) with the results
of [9] in the 0 limit justifies the prescription (63).
Let us now turn to the gauge theory action (56). The reduced effective action which
corresponds to this is given by

m
Sreduced effective action A , A , Xm , X

=
d D x G (x) L(x, y) y=x
1
= 2
d D x G (x)
4gYM

G G [D , D ] [D , D ] 2G D , Xm D , Xm

, D
D
, D

+ Xm , Xn + G G D

m n
2
m
m

2G D , X D , X + X , X ,
(73)
y=x
where L(x, y) is the Lagrangian density defined in (56). The restriction y = x here breaks
some part of the gauge symmetry. The reduced effective action (73), however, turned out
324
to be still invariant under the gauge transformation with the gauge function
(x, y) = (x)
(y).
(74)
This follows from the identity

d D x (x) (x)
(y),
f (x, y) y=x = 0,
(75)
which can be proved by expanding f (x, y) as n gn (x) hn (y) and using the identities
(72). Because contains derivatives, the integrand of (75) does not trivially vanish, even if
we set y = x.
Let us note that in (73) the product inside the bracket is . Hence the reduced action is
not an ordinary action in the D-dimensional space, because the product in the integrand
contains both derivatives /x, /y. The gauge transformation is also defined in the (x, y)
space. To write down the gauge invariant action in the ordinary sense we must introduce
a set of coordinates (x, y) and consider 2D-dimensional action integral, as we did in the
previous section.
Finally, in order to derive the correlation functions we must further set A = A (x),
m (x) and X
m = X
m (y) before taking the functional derivatives of
A = A (y), Xm = X
(73), because these describe the gauge particles emitted and/or absorbed from the = 0
and ends of the open string and the fluctuations at the two ends, respectively. The same
functions at both ends because of the Bose symmetry. Then this reduced action becomes a
m .
functional of A , X
We must mention some ambiguity in our prescription. When we perform a partial
integration or use the cyclic property (47) in (56), we will obtain different integrands which
will, however, lead to the same result. The corresponding reduced integral does not have
this property. This is unavoidable, because when we reduce the integral, we insert into the
integrand a function proportional to a delta function. The result will depend on where this
delta function is inserted and we must specify this. Our proposal is to reduce the integral
in the standard form of the action (56).
Because we have not determined S , T , U completely, we cannot decide
whether (73) reproduces the open string amplitudes in the low energy limit. Moreover,
such an analysis will meet difficulties, because one cannot generally define free states
in an asymptotically non-flat space. We, however, expect that once S , T , U
are determined, the gauge invariance is strong enough to restrict the form of the effective
action.
7. Discussions
In this paper we considered an open string theory in an NSNS B field background with
a nonzero constant field strength H = dB. The background space is curved although if H
is small, the metric g is constant up to O(H 1 ). We performed a perturbative analysis of
the commutation relations of the string coordinates without oscillators up to O(H 1 ) and
found that the coordinate x of one end of the string does not commute with the coordinate
y of the other. We then constructed an associative and noncommutative product which
325
realizes the commutation relations of the coordinates. This product is an operation for
functions of both the coordinates x, y and even if two functions of x only are multiplied,
the result is a function of both x and y. In this way we are lead to consider bi-local fields
which depends on the coordinates of both ends of the string.
We then found that derivatives /x , /y do not satisfy Leibnitz rule with the
product and we are forced to modify the derivatives. At present the modified derivatives
are not uniquely determined, but remarkably, the new derivatives /x , /y turned
out to be rewritten as /x , /y , respectively, by some coordinate transformation
x, y x , y . The commutation relations of the primed variables coincide with those
of the unprimed ones with H = 0. Hence if we combine the x and y coordinates and
consider 2D-dimensional space (x, y), the primed coordinate system appears to be flat.
This provides us with a clue to write down the gauge theory action. The open string metric
(x) = (g{1 (g 1 B(x))2 }) is curved and it may seem difficult to write down the
G
action without spoiling the gauge invariance due to the coordinate dependent metric. Our
proposal is to write down the action in the flat (x , y ) frame and then transform it to the
(x, y) frame. In this way we obtained an action integral for the noncommutative gauge
theory in curved backgrounds. This gauge theory lives in a double-dimensional space, i.e.,
the space with coordinates (x , y ), = 1, 2, . . . , D. This action is invariant under gauge
transformations with x, y dependent gauge parameters. We then took a limit y x and
proposed a prescription to reduce the action integral to that in D-dimensional space with a
coordinate x. This reduced action still has a part of the gauge symmetry. This will provide
a local low energy effective action for string amplitudes.
Let us now turn to the application of our results to the Matrix model. The formulation
of the Matrix model for curved backgrounds has been investigated [16] but is not yet well
established. As mentioned above, the background space is curved for nonvanishing H .
(x)
The closed string metric g is flat only up to O(H 1 ) and the open string metric G
is curved already at O(H 1 ) [9]. On the other hand in the action (56) the metric G is flat
and has the ordinary form of the gauge theory action in the flat space. This is based on the
fact that we can deform the algebra of the coordinates x, y to that of the flat background
by some coordinate transformation (x, y) (x , y ). Although we do not have a proof that
this persists to all orders in H , let us assume this here. Then this observation suggests the
following general formulation of the Matrix model for curved backgrounds.
Double the number of the matrices of the ordinary matrix model. Denote these as
. This is natural because the open string has two ends. In this paper we
X and X
considered only the bosonic string. In the case of the superstring the fermion fields on
the D-brane must be also doubled.
The action of the Matrix model is given by

2

1
0 X
m 2 + Xm , Xn 2
dt Tr 2 D0 Xm + 2 D
SMatrix = 2
4gYM
m n
2

+ fermions .
,X
+ X
(76)
The action (76) is background independent. The space-time metric is Minkowskian.
We can introduce backgrounds as follows. By a separate linear coordinate trans-
326
the constant metric G can be put in. The matrix multiformation of X , X

plication is associative and noncommutative and there should be an isomorphism
and the bi-local fields whose multiplication rule obeys
between matrices X , X
i D
( = 1, 2, . . . , D) and
the product
.
By
the
replacement
X iD , X

Tr d D x d D y V 1 G(x, y), we will obtain the action (56). The product and the
will introduce the background dependence.
gauge covariant derivatives D , D
For the above formulation to be valid we must obtain the classical solution for the Matrix
model action (76) and then find a way to systematically determine the star product and
and demonstrate that the gauge theory action (56) can be
the covariant derivatives D, D
obtained. We hope to report on the analysis in the future.
References
[1] A. Conne, M.R. Douglas, A. Schwartz, Noncommutative geometry and matrix theory: compactification on
tori, JHEP 9802 (1998) 003, hep-th/9711162.
[2] M.R. Douglas, C. Hull, D-branes, noncommutative torus, JHEP 9802 (1998) 008, hep-th/9711165.
[3] E. Witten, Noncommutative geometry and string field theory, Nucl. Phys. B 268 (1986) 253.
[4] V. Schomerus, D-branes and deformation quantization, JHEP 9906 (1999) 030, hep-th/9903205.
[5] C.-S. Chu, P.-M. Ho, Noncommutative open string and D-brane, Nucl. Phys. B 550 (1999) 151, hepth/9812219;
C.-S. Chu, P.-M. Ho, Constrained quantization of open string in background B field and noncommutative
D-brane, Nucl. Phys. B 568 (2000) 447, hep-th/9906192.
[6] F. Ardalan, H. Arfai, M.M. Sheikh-Jabbari, Dirac quantization of open strings and noncommutativity in
branes, Nucl. Phys. B 576 (2000) 578, hep-th/9906161.
[7] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032, hep-th/9908142.
[8] A.Y. Alekseev, A. Recknagel, V. Schomerus, Non-commutative world-volume geometries: branes on SU(2)
and fuzzy spheres, JHEP 9909 (1999) 023, hep-th/9908040;
A.Y. Alekseev, A. Recknagel, V. Schomerus, Brane dynamics in background fluxes and non-commutative
geometry, JHEP 0005 (2999) 010, hep-th/0003187.
[9] L. Cornalba, R. Schiappa, Nonassociative star product deformations for D-brane worldvolumes in curved
backgrounds, hep-th/0101219.
[10] C. Klimck, T. Strobl, WZW-Poisson manifolds, math.SG/0104189;
P. evera, A. Weinstein, Poisson geometry with a 3-form background, math.SG/0107133;
J.S. Park, Topological open p-branes, hep-th/0012141.
[11] P.-M. Ho, Y.-T. Yeh, Noncommutative D-brane in non-constant NSNS B field background, Phys. Rev.
Lett. 85 (2000) 5523, hep-th/0005159.
[12] P.-M. Ho, Making non-associative algebra associative, hep-th/0103024.
[13] K. Hayasaka, R. Nakayama, Y. Shimono, in preparation.
[14] J. Gomis, T. Mehen, Space-time noncommutative field theories and unitarity, Nucl. Phys. B 591 (2000) 265,
hep-th/0005129;
O. Aharony, J. Gomis, T. Mehen, On theories with light-like noncommutativity, JHEP 0009 (2000) 023,
hep-th/0006236;
L. Alvarez-Gaum, J.L.F. Barbn, Non-linear vacuum phenomena in non-commutative QED, Int. J. Mod.
Phys. A 16 (2001) 1123, hep-th/0006209.
[15] N. Seiberg, L. Susskind, N. Toumbas, Space/time non-commutativity and causality, JHEP 0006 (2000) 044,
hep-th/0005015;
N. Seiberg, L. Susskind, N. Toumbas, Strings in background electric field, space/time noncommutativity and
a new noncritical string theory, JHEP 0006 (2000) 021, hep-th/0005040.
[16] See, for example, M. Douglas, D-branes in curved space, Adv. Theor. Math. Phys. 1 (1998) 198, hepth/9703056.

Global black branes (extended global defects

surrounded by horizons), brane worlds and
the cosmological constant
Sei-Hoon Moon
Institute of Physics and Applied Physics, Yonsei University, Seoul 120-749, South Korea
Abstract
We study global defects coupled to higher-dimensional gravity with a negative cosmological
constant. This paper is mainly devoted to studying global black brane solutions which are extended
global defects surrounded by horizons. We find series solutions in a few separated regions and
confirm numerically that they can be mutually connected. When the world volume of the brane
is Ricci-flat, the brane is surrounded by a degenerated horizon, while it is surrounded by two
horizons when the world volume has a positive constant curvature. Each solution corresponds to
an extremal and a non-extremal state, respectively. Their causal structures resemble those of the
ReissnerNordstrm black holes in anti-de-Sitter spacetime. However, the non-extremal black brane
is not a static object, but an inflating brane. In addition, we briefly discuss a brane world model in the
context of the global black branes. We comment on a few thermodynamic properties of the global
black branes, and discuss a decrease of the cosmological constant on the brane world through the
thermodynamic instability of the non-extremal global black brane. 2002 Elsevier Science B.V. All
rights reserved.
PACS: 04.50.+h; 04.70.-s; 11.10.Kk; 11.25.Mj; 11.27.+d
1. Introduction
Global defects are topologically stable scalar field configurations with non-trivial
homotopy for some internal symmetry manifolds [1]. Usually, this kind of topological
defect arises in a theory with d scalar fields and a potential which has a vacuum manifold
with the topology of a (d 1)-sphere. The simplest examples are domain walls [2325]
E-mail address: jeollo@phya.yonsei.ac.kr (S.-H. Moon).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 5 7 - 5
328
S.-H. Moon / Nuclear Physics B 624 (2002) 327359
when the vacuum manifold has a discrete symmetry, global strings [2,3] when d = 2,
and global monopoles [4] when d = 3, in four spacetime dimensions. While the global
monopole has a well-defined static metric, the domain wall and the global string do not.
The domain wall spacetime is non-static and the metric in the plane of the domain wall
is that of a (2 + 1)-dimensional de Sitter space. The spacetime of the global string has
a curvature singularity at a finite distance from the string core [2]. Its metric was found
by Cohen and Kaplan [3]. However, Gregory showed that time dependence could remove
the singular nature of the global string spacetime [5]. While these analyses were purely
within the context of Einstein gravity in four dimensions (in the absence of a cosmological
constant), the global string spacetime is always regular in the presence of a negative
cosmological constant [11], no matter how small cosmological constant may be. Because
the negative cosmological constant could do a role to avoid the singularity rounding off the
spacetime before it terminates at the singularity.
Global defects were first considered in four spacetime dimensions, but recently they
have been studied in higher dimensions in the context of the brane world scenarios,
such as the large extra dimension scenario (ADD) [7] and the RandallSundrum (RS)
scenario [8,9]. According to the brane world picture the three space dimensional world
that we appear to be living in is a brane that is embedded in a higher-dimensional world.
The brane world scenarios are really attractive in that the Newtonian gravity is recovered
on the brane despite the gravity is fundamentally higher-dimensional one, even with large
(infinite) extra dimensions. This approach provides a plausible explanation of the hierarchy
between the gravitational and the electroweak mass scales, opening the possibility for
construction of new classes of models for the unified theory. These theories also open new
approaches to the cosmological constant problem, because if our four-dimensional world is
embedded in a higher-dimensional spacetime, the effect of non-zero vacuum energy could
affect only the curvature in the extra dimensions allowing for a flat four-dimensional world
as first proposed in Ref. [3944].
There have been various attempts to realize the brane world scenario in the context of
higher-dimensional gravitating global defects in Refs. [1017]. Cohen and Kaplan [10]
have considered a brane world, which may be viewed as a global string in two extra
dimensions. Its spacetime exhibits a curvature singularity at a finite distance from the
string core, similar to the case in four dimensions. They argued that, imposing a unitary
boundary condition, the location of the singularity may play a role of the boundary of
the extra dimensions, which form a finite but non-compact space of exponentially large
proper size. The phenomenology resembles the ADD scenario, rather than that of the
RS scenario. Gregory [11] has shown that a non-singular global string-like solution with
the required properties for the RS-type scenario exists in the presence of a negative
bulk cosmological constant. Olasagasti and Vilenkin [12] explored a more general case
of a brane carrying a global charge in a higher-dimensional spacetime with a non-zero
cosmological constant. In particular, with a negative cosmological constant there exist
solutions of which the geometry of extra dimensions asymptotes to a cylinder with cross
section being a (d 1)-sphere of a fixed radius (cigar-like geometry). The cigar-like
solutions with an exponentially decaying warp factor is of interest, since they have features
needed for the RS-type scenario.
329
On the other hand, in Ref. [16] the authors investigated the spacetime of higherdimensional global defects in the presence of a negative cosmological constant adopting
Schwarzschild-type metric ansatz. They found extremal black-hole-like p space dimensional defects (global black p-branes), which are Ricci-flat branes surrounded by a degenerated horizon. They showed that the near-horizon geometry of the extremal global black
branes coincides with the cigar-like geometry with an exponentially damping warp factor discovered in Refs. [11,12]. The interior region inside the horizon of the global black
p-brane possesses all of features needed for a brane world with the Newtonian gravity. In
the picture of Ref. [16] the size of the horizon can be interpreted as the compactification
size of d extra dimensions, even though the interior region inside the horizon infinitely
extends and asymptotes to AdSp+2 S d1 . Therefore, the large mass hierarchy is translated into the large size of the horizon, which could be supported by the large magnitude
of charge carried by the black branes. In this picture, the Hawking radiation could be a
possible mechanism for the resolution of various problems associated with brane world
scenarios, such as the observed flatness and the approximate Lorentz invariance of our
world. The bending of the brane and the bulk curvature that violate the SO(3, 1) isometry
on our brane [37,38] would correspond to excitations upon the extremal state, and the excited states that are non-extremal black branes would evolve into an extremal state through
the Hawking radiation process.
The non-zero vacuum energy density (equivalently, the cosmological constant) on the
brane could also be an excitation upon the extremal state, which would be diluted via the
Hawking process if the excited state correspond to a non-extremal black brane. However,
even though the state excited by the vacuum energy density is a non-extremal state, it
will be somehow different from static non-extremal black branes treated usually in the
literatures before in the context of supergravity or string theories [22], because with nonzero vacuum energy density the branes will inflate and the corresponding spacetime will
not be static. Therefore, their thermodynamic properties will also be different from those of
the static ones. In this respect, we may need to find the corresponding black brane solutions
and to examine their thermodynamic properties, to see whether a certain thermodynamic
process can be a dynamical mechanism thinning out the vacuum energy density on the
brane.
This paper will be mainly devoted to finding global black brane solutions which are
extended global defects surrounded by horizons and studying their spacetime structures.
However, we will also focus on a few aspect of a brane world based on the global black
branes and the cosmological constant problem.
This paper is organized as follows. In Section 2 we derive the equations of motion
corresponding to the Schwarzschild-type metric ansatz in a scalar theory with global
internal symmetry coupled to higher-dimensional gravity with a negative cosmological
constant. In Section 3 we explore black-hole-like p space dimensional defect (black
p-brane) solutions, which shall be referred in what follows as global black p-brane
or global black brane. We find series solutions in a few separated regions and confirm
numerically that they can be mutually connected. When the world volume of the branes
is Ricci-flat, the branes are surrounded by a degenerated horizon, while when the world
volume is not Ricci-flat but of a positive constant curvature, they are surrounded by two
horizons. They correspond to extremal black branes and non-extremal ones, respectively.
330
In Section 4 we discuss the spacetime structures of the global black branes. Their
causal structures resemble those of the ReissnerNordstrm black holes in anti-de-Sitter
spacetime, if we consider the transverse slice to the brane. The near-horizon geometries
are an infinitely long anti-de-Sitter throat with the topology of AdSp+2 S d1 for the
extremal black branes, and the spacetime of an inflating p space dimensional domain wall
times (d 1)-dimensional sphere for non-extremal ones. In Section 5 we discuss a few
issues associated to the brane world scenarios. We briefly comment a few thermodynamic
properties of the global black branes, and discuss a decrease of the cosmological constant
on the brane world through the thermodynamic instability of the non-extremal global black
brane. In the last section, we end with some conclusions.
2. The equations of motion

In this section we derive field equations appropriate to isolated gravitating global
defects in higher-dimensional spacetime with a negative cosmological constant. Higherdimensional gravitating global defects are described as topologically stable scalar field
configurations of a field theory with a spontaneously broken continuous global symmetry
coupled to D-dimensional gravity. We consider a scalar theory with global O(d) symmetry
coupled to gravity, of which potential has minimum on the (d 1)-sphere of radius v 2 :
S = Sgravity + Sscalar ,

MD2
Sgravity =
dx D gD (R 2),
16

2
1
Sscalar = dx D gD M a M a b b v 2 ,
2
4
(1)
(2)
where M is the fundamental scale of the higher-dimensional gravity theory. We use a

mostly plus signature. Note that a and v have a mass dimension of (p + d 1)/2
and the negative bulk cosmological constant ( < 0) has a mass dimension of 2. This
theory admits p space dimensional topological solitons, where p + 1 = D d. We
shall refer them as global p-brane in what follows. M, N, . . . denote D-dimensional
spacetime indices and a, b, . . . do d-dimensional internal space indices and run on
1, . . . , d. We shall use the notation {x } with = 0, . . . , p for the coordinates on the
p-brane world volume. {y a } denote the transverse coordinates to the p-brane and are
related to the spherical coordinates of the transverse space by the usual relations: y a =
{r cos 1 , . . . , r sin 1 cos d1 , r sin 1 sin d1 } with r 2 = y a y a . The corresponding
Einstein equations and scalar field equation are
2
,
T N + M N
D2 M
D
2
M

b b
M a
2 a
RM N =
M v = 0,
(3)
(4)
where
TM N M a N a + M N
b b
2
v2 .
2(D 2)
(5)
331
The global defects are defined by a topologically non-trivial mapping of the vacuum
manifold O(d)/O(d 1) to the boundary of transverse dimensions S d1 . We take ansatz
for the scalar field configuration as:
a = vXy a ,
(6)
where the function X could be a function of both t and r in general. The defect solution
should have X(r = 0) = 0 at the center of the defect and approach the radial hedgehog
configuration outside the core.
In this paper we consider a particular case that the energymomentum tensor of the
defects is time independent. As a result, no gravitational and particle radiation exists. The
spacetime section transverse to the defects is static, and the whole spacetime including
longitudinal directions is allowed to be static or stationary as we will see later. The
spacetime of the extended global defects have the topology of R p+2 S d1 . For black
branes, the spacetime transverse to the brane will be a generalization of the 4-dimensional
spherically symmetric solutions, e.g., the Schwarzschild or the ReissnerNordstrm black
hole solutions. To obtain these solutions of the Einstein equations we will assume that
outside any horizon there is a Killing vector which is timelike and that surfaces of constant
t are formed from concentric (d 1)-spheres times p-dimensional space. We can then
quite generally adopt following Schwarzschild-type metric ansatz:
ds 2 = e2A(r)B(r)g (x) dx dx +
dr 2
2
+ r 2 dd1
,
B(r)
(7)
where g (x) is a metric on (p + 1)-dimensional brane world volume which is not

2
necessarily static, dd1
is the line element on the unit (d 1)-sphere, and A(r) and B(r)
are functions of r only. With this metric ansatz, the defect solutions with unit winding
number correspond to the field configurations
a = vf (r)
ya
.
r
(8)
The defect solutions have f (r) = 0 at the center and approach f (r) 1 outside core. The
energymomentum tensor for the field configuration (8) is then given by
(1 f 2 )2
Tt t = Txxi i =
,
2(p + d 1)
(9)
(1 f 2 )2
Tr r = Bf 2 +
,
2(p + d 1)
(10)
f2
(1 f 2 )2
Taa = 2 +
,
r
2(p + d 1)
(11)
where TNM = TNM /v 4 .

Since the metric (7) is a special case of the more general class of metric,
2
ds 2 = F y a d s2 + d sd2 ,
(12)
332
where d sd2 is the metric of the slice in the transverse directions and d s2 is the world volume
metric, the Ricci tensor splits in the following way:

g F
2 F + p(F
)2 ,
R = R
(13)

ab (p + 1) a b F .
Rab = R
F
(14)
Since T g , through Einsteins equations we have that R g and finally, from

g . That is, R,
the curvature associated with the metric g ,
Eq. (13) above, that R
must be constant [12]. Therefore, without losing generality we can assume that the spatial
part of the metric intrinsic to the brane is homogeneous and isotropic, and the directions
parallel to the brane are boost invariant in the strong sense. Einsteins field equations then
reduce to

1
2p + 3 p B 2
1
R
2
(p
+
1)BA
BA
BA B
p + 1 e2A B
2
4 B
2

d 1
1
2

,
BA + B = 8GD Tt t +
r
2
p+d 1
(15)

d 1 B
3
1
2

(p + 1) BA + B A + BA + B
2
2
2 r
2
= 8GD Tr r +
,
p+d 1
(p + 1)
(16)
B 1
2
BA p + 2 B
(d 2) 2 = 8GD Taa +
,
r
2 r
r
p+d 1
(17)
supplemented by the equation for the metric on the p-brane,

=
R

R
g .
p+1
(18)
It can be easily shown that only two of the three Eqs. (15)(17) are independent. The scalar
field equation is written down as

p B
d 1
1 (d 1)
2
f +
+ 1+
+ (1 + p)A f
+
(f
1)
f = 0.
r
2 B
B
r2
(19)
GD v 2 /MD2
Here we have defined a dimensionless Newtonian constant

rescaled the coordinates and redefined the cosmological constant as
vxM xM
and
.
v 2
and we have
(20)
333
3. Solutions
We may be able to find several classes of solutions to the set of field Eqs. (15)
(19). However, we will restrict our discussion to only black-hole-like solutions in this
paper. First, we find extremal black brane solutions which are flat branes surrounded by
a degenerated horizon. Next we shall discuss about non-extremal black brane solutions
which are bent branes surrounded by two horizons.
For the simplest case (of p = 0 and d = 2) which corresponds to a vortex in (2 + 1)dimensional spacetimes, the global vortex solutions were thoroughly analyzed for the
global U (1) scalar theory in Ref. [19] and for the O(3) non-linear model in Ref. [20]
in the presence of a negative cosmological constant. For the global U (1) theory, there
are three types of solutions depending on the relative scales of parameters of the theory:
the cosmological constant, the Planck scale and the symmetry breaking scale. They are
regular vortex, extremal and non-extremal ReissnerNordstrm-type black holes carrying
topological charges. Their spacetimes are perfectly regular and do not involve any physical
curvature singularity. When the magnitude of a negative cosmological constant is larger
than a critical value at a given symmetry breaking scale, the spacetime formed by a isolated
vortex is regular hyperbola with a deficit angle. However, it becomes a charged black hole
when the magnitude of the cosmological constant is less than the critical value. For the
O(3) non-linear model, the usual regular topological lump solution cannot form a black
hole even though the scale of symmetry breaking is increased. There exist non-topological
solitons of half integral winding in the given model, and the corresponding spacetimes
involve charged BTZ black holes.
When p 1 and d 2, the equations of motion, however, become much more nonlinear as can be seen from Eqs. (15)(19). Due to this high nonlinearity of Eqs. (15)
(19), the spacetime around the global p-brane may develop a curvature singularity
in an analogous fashion to the four-dimensional global string metric (with vanishing
cosmological constant) [2,3]. In other side, we may also expect that the bulk negative
cosmological constant could do a role to avoid the singularity rounding off the spacetime
before it terminates at the singularity [11]. As discussed in Ref. [12], the spacetime
structure of global extended defects are depends on the number of extra dimensions d, the
For a negative
bulk cosmological constant , and the p-brane world volume curvature R.
= 0, every spacetime found in Ref. [12] are regular
cosmological constant. When R
= 0, some of the spacetimes has singularity at a
without any genuine singularity. When R
finite proper distance from the core.
= 0 correspond to extremal
We will see that, for black-hole-like solutions, cases R
= 0 to the non-extremal states. The black brane solutions
black p-branes, while cases R
are perfectly regular ones. Moreover, we will show that some of the independent solutions
found in Ref. [12] actually describe different regions of the spacetime of a global black
brane.
3.1. Extreme black branes: flat branes surrounded by a degenerated horizon
Now we will analyze the equations derived in the previous section for the isolated global
= 0. We
defect spacetime and find solutions of which world volume is Ricci-flat, i.e., R
334
will see that such branes correspond to extremal black branes, if it has a horizon. Since
it seems almost impossible to find out exact analytic solutions of equations of motion
Eqs. (15)(19), we will just try to examine the behavior of the global p-brane spacetime at
a few separated regions. Outside the core, the energymomentum tensor is approximated
into
1
Tt t = Txxi i Tr r 0 and Taa 2 ,
(21)
r
with f (r) 1. Therefore, outside the core the behavior of the solutions is determined
solely by the relative magnitude of the scalar field energy density (8GD /r 2 ) to the
cosmological constant (||) through the equation (17).
Near the origin where the field energy density dominates over the cosmological
constant, the series solution up to the leading term is as follows:
f (r) f0 r,
(22)
A(r) A0 + A2 r ,
(23)
B(r) 1 + B2 r ,
(24)
where f0 is a shooting parameter to be determined by the proper behavior of the fields at

the asymptotic region, and A0 can be consistently fixed as A0 = 0 so that the spacetime is
locally Minkowski at the origin. The coefficients A2 and B2 are given as follows:

p||
4GD
p
2
f0
+
,
A2
(25)
d 1
2d(p + d 1)
d(d 1)(p + d 1)

8GD
pd +1
2(p d + 1)||
B2
(26)
f02
+
.
d 1
2d(p + d 1)
d(d 1)(p + d 1)
While the metric function A(r) increases, B(r) decreases near the origin for generic values
of f0 . With this solution the metric (7) near the core takes the form of

4GD 2|| 2
dr 2
2
r g (x) dx dx +
+ r 2 dd1
,
ds 2 1
(27)
d(p + d 1)
1 + B2 r 2
where g (x) is a general Ricci-flat metric on the brane, which satisfies (p + 1) (g)
dimensional vacuum Einstein equations R
= 0. This shows that the metric component
g00 of the spacetime decrease in the transverse directions near the origin when 2GD >
||, regardless value of f0 . This fact implies a possibility that the (00)-th component of the
metric could vanish at a finite coordinate distance and of the formation of black-hole-like
defects, when 2GD > ||.
In the far region from the core where the cosmological constant dominates over the
scalar fields energy density, the metric functions behave up to the leading terms for
sufficiently large r as follows:
f (r) 1,
(p + d)(p + d 1)(8GD d + 2) 1
,
A(r) A +
8(p + 1)||
r2
2||
r 2,
B(r)
(p + d)(p + d 1)
(28)
(29)
(30)
335
where A is a constant to be determined by matching with series solutions in other regions.

The metric (7) then is
ds 2 B r 2 g (x) dx dx +
dr 2
2
+ r 2 dd1
,
B r 2
(31)
where exp(2A ) has been absorbed into the longitudinal coordinates x and B
2||/(p + d)(p + d 1). This asymptotes to D-dimensional anti-de-Sitter spacetime
(AdSD ) with g (x) = , as we will see in next section.
We now turn to the intermediate region, at which the scalar energy density and
the cosmological constant is comparable, between the core and the asymptotic regions
described by metrics (27) and (31), respectively. From the behavior of g00 , we may expect
that a horizon is located in the intermediate region. Since what we are looking for is the
existence of a black hole horizon, it will be convenient to write the metric component g00
in terms of B(r) and f (r) by eliminating A(r) with use of Eq. (17) as
e2A B = Cr
2(d2)
(p+1)
1
(p+1)
2
exp
p+1
r

8GD f 2 d + 2 2|| 4GD (1 f 2 )2 dr
r
,
r
p+d 1
B
(32)
where C is an integral constant and it can be set to be 1 with the boundary condition
A = 0 at r = 0. To examine the existence and the property of the horizon, we begin
assuming that there exists a horizon at a finite coordinate distance rH from the core or
the asymptotic region, so that the timelike Killing vector t becomes null there, that is,
exp[2A(rH )]B(rH ) = 0. Further, we assume that B(r) becomes zero and is analytic at the
horizon. However, we do not put further restriction on A(r) because it could be singular
at r = rH , as can easily be observed from Eq. (17). We also assume that the scalar field
f (r) increases monotonically to unit value and behaves regularly at the horizon. Then the
position of the horizon rH and the value of fH ( f (rH )) are determined in a closed form
from the field equations.
Multiplying e2A B on Eq. (15), eliminating A and A with Eq. (17), and taking the limit
r rH , we obtain
e2A(rH ) [B1 + ][B1 + 2] = 0,
(33)
where B1 B (rH ) and
8GD fH2 d + 2 2|| 4GD (1 fH2 )2

rH .
rH
p+d 1
(34)
On the other hand, eliminating A , A from Eqs. (16) and (17), and taking the limit r rH ,
we get
[(p + 2)B1 + 2][B1 + 2] = 0.
(35)
There are two possibilities for B1 satisfying simultaneously the two relations Eqs. (33) and
(35): B1 = 2 = 0 or B1 = 0 = . However, the former possibility should be excluded
336
because with such condition the metric component g00 (rH ) does not vanish, but approaches
to a non-zero value at r = rH , leading to a contradiction with our starting assumption. This
can easily be checked by inserting following series expansions into Eq. (32)
B(r) B1 (r rH ) + B2 (r rH )2 + B3 (r rH )3 + ,
(36)
f (r) fH + f1 (r rH ) + f2 (r rH ) + .
(37)
The latter possibility only is acceptable and satisfies the condition for the surface at r = rH
to be a horizon. Inserting above series expansions into Eq. (32) and using B1 = 0 = give
up to the leading term
1
e2A B (r rH ) p+1
1+ B4
2||4 GD (1f 2 )2
H
p+d1
(38)
where we have replaced B2 with BH , i.e., B2 BH . Therefore, if BH < 4[2||

4GD (1 fH2 )2 ]/(p + d 1), e2A B touches zero at r = rH . We will see later that this
condition is always satisfied. Since = 0, from Eq. (34) we find the position of the horizon
in a closed form
2
=
rH
(p + d 1)(8GD fH2 d + 2)
2|| 4GD (1 fH2 )2
(39)
The fact B (rH ) = 0 indicates that the horizon is degenerated like that of an extremal black
brane. This tells that the black p-brane solution with Ricci-flat world volume is extremal,
as we have expected.
Inserting series expansions Eqs. (36) and (37) (with B1 = 0) into Eq. (17) and using the
condition = 0, one finds that

+ O (r rH )0 ,
A (r)
(40)
r rH
where

16GD (1 fH2 )fH rH f1
2|| 4GD (1 fH2 )2
1
p
.
p+22
p+1
(p + d 1)BH
(d 1)(p + d 1)BH
(41)
Here, the coefficients BH , fH , and f1 can be determined in terms of parameters given in
the theory using the equations of motion. Thus A(r) is at most logarithmically divergent at
r = rH :

A(r) ln(r rH ) + O (r rH )0 ,
(42)
and the (00)-component of the metric then becomes
g00 = e2A(r)B(r) BH (r rH )2(1).
(43)
With series expressions of B(r), f (r) and A(r) at the horizon, from Eqs. (15) and (17) we
obtain
BH (1 )2 =
2|| 4GD (1 fH2 )2

,
(p + 1)(p + d 1)
(44)
and then simultaneously solving this and Eq. (41) for BH and , we get

1
p + 9 + 4
1=
1
,
2(2 + )
p+1

|| 2GD (1 fH2 )2
BH =
(p + 5 + 2 ) (p + 1)(p + 9 + 4 ) ,
p+d 1
337
(45)
(46)
where
16pGD rH (1 fH2 )fH f1

(d 1)(2|| 4GD (1 fH2 )2 )
(47)
For lower sign, (1 ) is negative. This branch of solutions should be excluded because, in
that case, the norm of t (= e2N B) will diverge at r = rH , leading to a contradiction with
our starting assumption that g00 is zero at r = rH . We will discuss more about this branch
in next section. Whereas, with the upper sign (1 ) is positive saying that the surface of
r = rH is a horizon. BH satisfy our starting assumption that e2A B 0 as r rH , as can
be seen from Eq. (38). The coefficient f1 can be completely determined in terms of rH and
fH using Eqs. (19), (36), (37) and (40):
f1 =
(d 1)fH /rH (8GD fH2 d + 2)

.
2 p 8GD fH
rH
fH2
(p + d 1)rH
(48)
On the other hand, considering limiting behaviors of the scalar field and the metric
functions at the horizon, from the scalar field equation (19) we obtain another relation
for rH and fH
2
rH
=
d 1
.
1 fH2
(49)
Solving simultaneously the two equations (39) and (49), rH and fH can be expressed in
terms of parameters in the theory, and GD , as

(p + d 1)(8GD d + 2)
32GD ||(d 1)(2p + d 1)
2
rH =
, (50)
1 1
4||
(8GD d + 2)2 (p + d 1)2
fH2 =
(p + d 1)
p
+
2p + d 1 8GD (2p + d 1)

32GD ||(d 1)(2p + d 1)
.
(d 2) (8GD d + 2) 1
(8GD d + 2)2 (p + d 1)2
(51)
The complete determination of the position of the horizon rH and the value of the scalar
amplitude fH is an inevitable result since the leading term of Einstein equations (15)
(17) and the scalar field equation (19) lead to two algebraic equations for rH and fH . The
coefficients B3 and f1 are expressed in terms of rH and fH by using the field equations,
so B3 and f1 also are completely determined in terms of the parameters of the theory.
Since the higher order coefficients of the series expansions (36) and (37) can be expressed
338
in terms of lower order coefficients, every coefficients can be determined completely in

closed forms. This implies that the flat brane solution with horizon is unique for given
parameters of the theory.
The size of the horizon (rH ) and the value of scalar field at the horizon (fH ) have two
different expression for parameters given in the theory, that is, we have two different types
of black brane solutions. For the first type of solutions with lower sign, the horizon is
formed inside the brane core and the bulk cosmological constant has to be super-Planckian
for generic values of symmetry breaking scale of the theory. The condition that fH2 should
be positive and real requires that the bulk cosmological constant have a value within a
limited region above the Planck scale:
2GD
(8GD d + 2)2 (p + d 1)2

(d 2)(p + d 1)
< || <
.
2
32GD (d 1)(2p + d 1)
(52)
For this value of the bulk cosmological constant, the core size of a black brane will be
larger than the size of the horizon.
As an example, for p = 3, d = 2 the value of the
scalar field at the horizon is at most 3/7 showing that the core region extends outward
beyond the horizon. Then, solutions of this sort may be viewed as small black branes lying
within larger global branes. However, these solutions may develop a classical instability
in a similar way to the case of the magnetically charged YangMills solitonic black holes
with sufficiently small horizon [21]. This instability may lead to the possibility that such a
black brane could evaporate completely, leaving in its place a non-singular global defect.
The question of whether the black branes with small horizon lead to similar instabilities
discussed in Ref. [21] itself will be an interesting one, but is beyond the scope of this
paper.
On the other hand, for solutions with upper sign the mass scale of the bulk cosmological
constant could be well below the Planck scale and the horizon can be formed far outside
the brane core for generic values of the symmetry breaking scale and the bulk cosmological
constant, as can easily be checked in Eq. (51). Solutions of this sort will be matched
smoothly to a version of thin-wall approximation found in Ref. [16]. In this paper we
will consider only the black brane solutions corresponding to the upper sign, because we
are interested in only solutions with horizon larger than brane core with respect to the brane
world scenario of Ref. [16].
The metric (7) then has the form near r = rH
2(1)

ds 2 BH (r rH )
g (x) dx dx +
dr 2
2
2
+ rH
dd1
,
BH (r rH )2
(53)
where is +1 for the exterior region (r > rH ) and 1 for the interior region (r < rH ).
Note that, so far, we have implicitly assumed that we examine the near-horizon region
from the outside toward the horizon at rH and hence obtained only exterior metric ( = 1).
Had we done from the outside region, we would then obtain the interior metric ( = 1).
Up to now, finding an exact analytic solution to the field equations (15)(19) being
impossible, we have examined behavior of the solution of the global p-brane at a few
separated regions. The spacetime of the global black p-brane will be described via
metric equations (27), (31) and (53) provided that they are connected mutually. For
metric functions, A(r) increases near the origin, diverges at rH and decreases and rapidly
339
Fig. 1. A flat 3-brane solution surrounded by a degenerated horizon with two transverse directions (d = 2) when
GD = 0.15 and || = 0.420872997. With the thick solid line we report the behavior of the metric function B(r)
and with the dashed line the behavior of the scalar field f (r), whereas with the thin solid line the behavior of
A (r).
approaches to a constant value at the asymptotic region, as can be seen from (23), (29)
and (42). B(r) decreases near the origin, vanishes at rH and increases quadratically at
the asymptotic region, as indicated in Eqs. (24), (30) and (36). The scalar field f (r)
starts to grow linearly from zero at the origin, reaches to a fixed value fH of Eq. (51)
and then approaches to its vacuum value at the asymptotic region. From the behaviors
of the metric functions and scalar field, we can easily guess that the fragments of
solution can be connected mutually if they behaves monotonically at each side of the
horizon.
While we have not resolved yet whether the solutions obtained in separated regions
can be connected mutually or not, numerical calculations indicate that the metric functions
and the scalar field behave monotonically at each side of the horizon, and the solutions
obtained in separated regions are connected smoothly showing the existence of the
black p-brane solution. Such an example of the black brane solution is shown in
Fig. 1. The obtained solution is for an extremal black 3-brane with two transverse
spatial direction (d = 2), that is, a global black string-like defect. The scalar field
f (r) (the dashed line) behaves monotonically for every r and connects f (0) = 0 and
f () = 1 smoothly across the horizon at r = rH . The metric function B(r) (the
thick solid line) decreases monotonically inside the horizon and vanishes at the horizon
rH , and then increases monotonically. On the other hand, the derivative of the metric
function A(r) (the thin solid line) shows a singular behavior at the horizon as shown in
Eq. (40).
Before ending this section, we briefly discuss the scales of parameters of the theory:
the symmetry breaking scale v, the Planck scale M and the cosmological constants .
340
The right-hand side of Eq. (39) ought to be positive. This condition yields1 v 2 > ((d
2)/8)MD2 . While the condition does not give any new information for d = 2, for d > 2,
the condition sets the symmetry breaking scale v to be of the order of the fundamental scale
M . On the other hand, in order to trust the theory, v 2/D2 ought to be smaller than the
fundamental scale: v 2/D2 < M . Hence, we assume that (d 2)/8 < v 2 /MD2 < 1.
We also assume that v 2/D2 < M even for d = 2, but this seems valid for naturalness
reasons. The horizon size rH is then determined by the bulk cosmological constant ,
2
viz., rH
||. There is no restriction on the scale of the cosmological constant. It seems
to be natural that the cosmological constant is of the order of the fundamental scale
in the absence of a mechanism to protect a small cosmological constant from radiative
corrections. However, given our ignorance concerning the bulk cosmological constant
problem, there is a priori no reason to expect that is of the same order as the fundamental
scale M2 . Thus, we will treat the cosmological constant simply as an input parameter.
3.2. Non-extremal black branes: bent branes surrounded by two horizons
In the previous subsection we found extremal black brane-like solutions. Such extremal
states are usually considered as critical limits of non-extremal black objects. There may
be various excitations upon the extremal states. An example is the usual static nonextremal black p-branes found in supergravity and superstring theories [22]. The nonextremality breaks the bulk Lorentz invariance [37,38] in the brane world directions, and
the bulk curvature components in the brane world directions depend on the extra dimension
coordinates. Another example could be bent branes surrounded by horizons among them,
= 0). Such objects
of which world volume has non-vanishing constant curvature (i.e., R
may be important, as it makes the brane world scenario on the global defects more likely.
We will try to show the existence of such brane solutions. We find out series solutions at a
few separated regions as we did for flat branes in Subsection 3.1.
At the origin, we find series solutions similar to those of the Ricci-flat brane, Eqs. (22)
(24), up to the leading order. The coefficients A2 and B2 are slightly modified with terms
as
depending on R

4GD
p
A 2
f02
d 1
2d(p + d 1)

p(p + 1) d(d 1)
p||
,
(54)
+
+
R
d(d 1)(p + d 1) 2(p + 1)d(d 1)(p d + 1)

2(p d + 1)||
R
2 8GD f02 p d + 1
+
+
.
B
d 1
2d(p + d 1)
d(d 1)(p + d 1) d(d 1)
(55)
1 Since we are interested in black brane solutions with horizon larger than the size of the brane core, here
we have assumed that fH 1. This assumption will be valid for discussions on the spacetime structure of black
brane solutions with horizon outside the brane core.
341
The metric at the core then takes the form of

2
4GD 2||
R
ds 2 1
r g (x) dx dx
d(p + d 1)
d(p + 1)
+
dr 2
2
+ r 2 dd1
,
2 r 2
1+B
(56)
where g (x) is a metric on the world volume of the brane, which satisfies (p + 1) This shows that the metric
dimensional Einstein equation (18) with non-zero curvature R.
component g00 of the spacetime decrease in the transverse directions near the origin when

4GD > 2|| + [(p + d 1)/(p + 1)]R.
In the far region from the core, the metric functions behave in the same manner as in
flat-brane solutions up to the leading orders regardless the curvature scale of the brane
of Eq. (15)
world volume, as can be easily seen from Eq. (15). The first term containing R
decays out and becomes negligible compared to other terms at the asymptotic region, which
approach to a constant values. The metric is then simply
dr 2
2
+ r 2 dd1
,
(57)
B r 2
where B is the same coefficient with that of the metric (31).
As in the case of flat brane (if they are formed) the horizons will be formed in the
intermediate region between the core and the asymptotic regions. We will probe the
horizons from both the core and the asymptotic regions. We assume that there exist
horizons at finite coordinate distances r = rH from observers living on the core or in the
asymptotic region. We will follow the same procedure as done in the flat brane case to
identify the location of and the properties of the horizon. While Eq. (35) is not modified
= 0, the relation (33) is modified as
with inclusion of non-zero curvature R
ds 2 B r 2 g (x) dx dx +
= e2A(rH ) [B1 + ][B1 + 2].

2R
(58)
The only possibility satisfying simultaneously two equations (35) and (58) is B1 =
2/(p + 2) = 0. Inserting this into Eq. (58) gives
e2A(rH ) B12 =

4R
.
p(p + 1)
(59)
= 0, and the horizons are not degenerated contrary to the

This implies that B1 = 0 unless R
case of flat branes. Since the left-hand side is positive definite, the world volume curvature
has to be positive constant, that is, the corresponding (p + 1)-dimensional brane world has
to be de Sitter spacetime. On the other hand, inserting series expansions Eqs. (36) and (37)
into Eq. (32) and taking the limit r rH we obtain
2(d2)
p+1
e2A(rH ) = rH
|B1 |
p+2
p+1
(60)
where we have set C = 1 in Eq. (32) with the boundary condition A = 0 at r = 0. From
Eqs. (59) and (60) we get the absolute value of B1 at r = rH

(p+1)/p
4R
2(d2)/p
|B1 | = rH
.
(61)
p(p + 1)
342
Then, using the relation B1 = 2/(p + 2) we obtain following two relations between rH
and fH :

2|| 4GD (1 fH2 )2 2
rH 8GD fH2 d + 2
p+d 1

(p+1)/p
+1
4R
p + 2 2(d2)
rH p
,
=
(62)
2
p(p + 1)
where the positive sign of the right-hand side corresponds to B1 > 0 and the negative sign
to B1 < 0. We need another relation between rH and fH to identify the position of the
horizon and the value of the scalar field at the horizon. However, unlike the case of the flat
brane the set of equations of motion doesnt give the additional relation. The leading term
of the scalar field equation (19) that gave the additional relation in the case of the flat brane
does role only to determine f1 in terms of rH and fH .2 So we cannot determine completely
rH and fH unlikely in the case of the flat brane.
Since for each sign the relation Eq. (62) admits a positive real root of rH for a given
fH , we have two distinct values of rH , i.e., rH + (for + sign of r.h.s.) and rH (for
sign of r.h.s.).3 Therefore, we have two distinct solutions near the surfaces at rH
2
ds
e2AH B1 (r rH ) d s2 +
dr 2
2
+ rH 2 dd1
,
B1 (r rH )
(63)
where d s2 g dx dx , and AH and B1 are values at rH + and rH , respectively.

In usual, this type of metrics develops curvature singularities at r = rH except the case
p = 0. However, as we will see in the next subsection, the spacetime described by this
metric is regular at r = rH thanks to the relation (59) and so the surfaces at r = rH are
coordinate singularities. Moreover, the fact that B1+ > 0 but B1 < 0 seems to imply that
surfaces at rH + and rH correspond to the outer horizon and the inner horizon of a nonextremal black brane, respectively. This will be true, if rH + > rH and the two solutions
are mutually connected. The reason is as follows. Since e2A is finite at the horizons and
is positive always, the spacetime structure is determined solely by the behavior of B(r).
Since B(r) is positive near the core and at the asymptotic region, B(r) has to be negative
at an intermediate region if there should be two horizons. Then the tangent of B(r) should
be negative at the inner horizon and positive at the outer horizon, as we have observed in
the case of the ReissnerNordstrm black hole.
If the horizons are far outside the core, then the scalar field has nearly the same values
at the inner and outer horizons (i.e., fH fH + 1), and then we can easily see that
rH + > rH from Eq. (62). In particular, when d = 2, this can be explicitly observed,
because the roots can be written in closed form as follows

(p+1)/p
(p + 1)(p + 2)
4R
rH =
8[|| 2GD (1 fH 2 )2 ] p(p + 1)
2 f is determined in terms of r and f as f = 2f [d 1 (1 f 2 )2 r 2 ]/(p + 2)r 2 |B |.
H
H
H
1
1
H
H
H 1
3 Here we assume that || > 2 G (1 f 2 )2 . Note that this condition was automatically satisfied for flat
D
H
brane solutions as guaranteed by Eq. (44). If we consider the flat brane solution as a critical case of a bent brane
solution, then the assumption will be natural. Remind also that 8 GD fH2 > d 2 from Eq. (39).

+

4R
2

2
p(p + 1)
64 || 2GD (1 fH )2
1/2
(p + 1)8GD fH2
+

.
2 || 2GD (1 fH2 )2
(p + 1)2 (p + 2)2
343
2(p+1)/p
(64)
Clearly, if fH fH + fH , then rH + > rH . The difference between rH + and rH is

given by

4R
rH + rH
4[|| 2GD (1 fH 2 )2 ] p(p + 1)
(p + 1)(p + 2)
(p+1)/p
> 0.
(65)
0, and the position of the degenerated

The two horizons are degenerated in the limit R
horizon coincides with that of the extremal black brane given by Eq. (39). Therefore, our
0, as expected from the
solution seems to behaves correctly in the critical limit that R
usual non-extremal black objects.
The remaining step is now to see whether the solutions obtained in separated regions
can be mutually connected or not. The numerical works indicate that the possibility of
the existence of a smooth configuration to connect the solutions (56), (57), and (63), and
a smooth scalar field configuration to interpolate between f (0) = 0 and f () = 1. An
example of such bent brane solution surrounded by two horizons is illustrated in Fig. 2.
Fig. 2. A non-extremal black 3-brane solution with two transverse directions (d = 2) when GD = 0.15,
= 3 108 . With the thick solid line we report the behavior of the metric function
|| = 0.420872997 and R
B(r) and with the dashed line the behavior of the scalar field f (r), whereas with the thin solid line the behavior
of A (r).
344
4. Spacetime structures
4.1. Extreme global black branes
In this subsection we discuss the spacetime structure induced via mutually connected
metrics (27), (31) and (53). We begin by examining the geometry inside the nearhorizon region. As mentioned in the previous section, outside the core the behavior of
metric functions is controlled by the relative magnitude of the scalar field energy density
8GD /r 2 to the cosmological constant ||. The horizon occurs in the region where
8GD /r 2 and || are comparable. The spacetime of such region is well described by the
metric (31). The shape of the interior region will depend on value of the cosmological
constant. When the cosmological constant is of the same order as the fundamental
scale, both the core radius rc ( v)1/2 and the horizon size rH are of the order
of the fundamental size M1 and so the scalar field energy density 8GD /r 2 and the
cosmological constant || are of comparable magnitude over the entire interior region.
Thus the interior region is well described by the two metrics Eqs. (27) and (31).
However, when the cosmological constant is much less than the fundamental scale, the
geometry of the interior region is more complicated. In this case, the size of the horizon
is much larger than the size of the core. The core region and the near-horizon region
are squeezed into small portions and, in between the two, most portion is occupied by
an intermediate region characterized by 8GD /r 2 ||. Even though we dont have
any analytic solution for the intermediate region between the core and the horizon, we
could expect its geometry to be approximated by that of the global defect spacetime in
the absence of the negative cosmological constant. The spacetime geometry depends on
the transverse space dimensions. When d = 2, the geometry of this region will resemble
that of the CohenKaplan solution [10] with removed curvature singularity. If d 3, the
geometry of this region would become similar to that of the global monopole solution of
Ref. [12].
The exterior region interpolates between the near-horizon region and the asymptotic
region described by metrics (31) and (53), respectively. Since in the far region from the core
the cosmological constant dominates over the scalar field energy density, one may expect
the spacetime asymptotes to a negative constant curvature space, which is D-dimensional
g = . The Riemann curvature tensor corresponding
anti-de-Sitter space (AdSD ) with
to the metric (31) is

1
RMNP Q = B + 2
(66)
(gMP gNQ gMQ gNP ),
r M,N,P ,Q{a }
where the second term in the bracket is non-vanishing only for components Ra ,b ,c ,d .
Thus we easily see that, when d = 2, the spacetime described by the metric (31)
is a
negative constant curvature space, i.e., an anti-de-Sitter space with curvature scale B .
While, for d 3 the space is no longer a constant curvature space, but asymptotes to the
D-dimensional anti-de-Sitter space (AdSD ), as r . This type of spacetimes has been
found in Refs. [11,12]. In fact, the metric (31)
be rewritten to the form found in [11,12]
r can
introducing the proper radial distance dr / B(r ) as
ds 2 e2
g (x) dx dx + d 2 + e2
2
dd1
.
(67)
345
Finally, we examine the near-horizon geometry. We begin by studying the motion of test
particles. Due to the Lorentz invariance in the longitudinal direction with g (x) = and
the spherical symmetry in the transverse direction, it is sufficient to study null and timelike
geodesics in the system

2(1) 2
dt +
ds 2 = BH (r rH )
dr 2
.
BH (r rH )2
Defining the covariant 4-velocity u = g,

dx /d , we find the geodesic equation
2

2(1) dt 2
dr
1

(r
r
u
u
=
B
)
+
,
= g,
H
H

d
BH (r rH )2 d
(68)
(69)
where is an Affine parameter (for timelike geodesics, it is the proper time). Here is 0
for null geodesics and 1 for timelike geodesics. There is a constant of the motion

: = g,
u = g00

2(1) dt
dt
= BH (r rH )
,
d
d
where (/t) denotes the static Killing vector. The equation motion then is
2

2
2
dr
BH (r rH ) : 2 (r rH ) = 0.
d
(70)
(71)
The equation of motion tells that (dr/d ) 0 as r rH . The proper distance from a
point r0 to the horizon on a constant time slice is logarithmically divergent as

r rH
1
.

(r)
(72)
ln
BH r0 rH
However, test particles starting at a point r0 and moving toward the horizon reach r = rH
and t = in the finite Affine parameter
rH
=
r0
dr
< ,
{: 2 [ (r rH )]2 + BH (r rH )2 }1/2
(73)
and, for null geodesics ( = 0), we explicitly get

1 [ (r0 rH )]1
.
(74)
:
1
Therefore, the coordinates {x , r, a } are not geodesically complete in both interior and
exterior regions. Hence, we need to extend each region onto new patches across the
horizon. It is well known that as the extension gets across a Cauchy horizon, it is not
unique. That is, depending on the choice of spacetime identification there are different
possibilities to analytically continue the spacetime across the horizon. For an example,
we would consider extensions in which the interior region is reflected across a Cauchy
horizon so as to have its copy in the next patch. Since the extension has its own horizon,
such extension can be repeated giving rise to an infinite array of the interior region. In the
same manner, we may obtain an infinite array of the exterior region, which is a maximal
rrH
346
Fig. 3. Penrose diagram for the extreme global black p-brane with field configurations shown in Fig. 1. The spatial
coordinates x i and a are suppressed. This diagram is very similar to that of the extreme ReissnerNordstrm
black hole, but now there is a non-singular brane core at r = 0 rather than a curvature singularity. The boundary
at r is timelike rather than null.
analytic extension of the exterior region. Another plausible extension is the covering space
of the black p-brane system, which is an infinite lattice of the exterior regions and the
interior regions as depicted in Fig. 3. This spacetime is obtained by repeating infinitely the
extension procedure of the exterior region to the interior region across the Cauchy horizon
and its inverse. This possibility involves no identifications and thus contains no closed
timelike curves (CTCs). This type of extension seems to be more natural if we consider
the field configuration found in Fig. 1, where the scalar field monotonically interpolates
across horizon between the false vacuum at the core and the true vacuum at asymptotic
region. This spacetime has a causal structure similar to that of the extreme Reissner
Nordstrm (RN) black hole in four-dimensional spacetime, besides that it is perfectly
regular everywhere, while that of the extreme RN black hole has a timelike singularity
at the origin. The spacetime of the global black brane asymptotes to a D-dimensional antide-Sitter space, while that of the extreme RN black hole to a flat Minkowski spacetime.
The near-horizon geometry is very similar to that of the extreme RN black hole, which
geometry has the topology of AdS2 S 2 with constant curvature scales of 1/rH . It is easy
to see that the metric (53), with g (x) = , describes
a constant curvature spacetime
with topology of AdSp+2 S d1 with curvature scales (1 )2 BH and 1/rH for AdSp+2
and S d1 , respectively: components of the Riemann curvature tensor corresponding to the
metric (53) are given by
R = (1 )2 BH (g g g g ),
(75)
Ra b c d =
1
(ga c gb d ga d gb c ),
2
rH
347
(76)
where ,
,
. . . refer to x and r, and a , b , . . . do indices on S d1 .
It will be helpful to rewrite the metric (53) in the form of RobinsonBertotti (RB) metric
as
2
dsRB
2
H
2
2
2
dx
dx
+
d 2 + rH
dd1
,
2
2
H
(77)
where we have introduced the Poincar radial coordinate

1/(1)

1

= BH (r rH )
r = rH +
,
(78)
H
BH H
with H 1/ BH (1 ). If the radial coordinate ranges between 0 and , then

the Poincar coordinates {x , } cover only one half of AdSp+2 , which is named as the
Poincar patch. In this coordinate system, the Cauchy horizon is at = 0 and the boundary
of AdSp+2 at = corresponds to a (p + 1)-dimensional flat Minkowski spacetime,
R 1,p . However, with the definition Eq. (78) has restricted ranges and so each region
covers only a part of the Poincar patch. When (1 ) is positive, ranges from 0 to a
1
/(1 ) (at r 0).
finite value f . For the interior solution ( = 1), roughly f rH
For the exterior region ( = 1), the metric (77) should be replaced with asymptotic solution
(31) or (67) with appropriate coordinate transformation at f . Hence, with positive (1 ),
the metric (77) describes portions including the AdS horizon of AdSp+2 S d1 .
On the other hand, when (1 ) is negative, the boundary at r = rH is not a horizon
and the metric is not that of the near horizon geometry of extremal black brane because
the timelike Killing vector t diverges at r = rH . And, as can be observed from Eq. (74),
the boundary is at a affinely infinite distance. Thus, the boundary at r = rH corresponds
to the timelike infinity of AdSp+2 , which is conformal to flat Minkowski spacetime M1,p .
In this branch, ranges from a finite value to infinity. Hence, the metric (77) describes an
AdS fragment including the (p + 1)-dimensional flat Minkowski boundary but cutting out
the AdS horizon.
Recall that near the horizon of the extremal RN black hole the asymptotic metric is also
written in the form of the RB metric as
2
=
dsRB
2
2 2 rH
2
dt + 2 d 2 + rH
d22 ,
2
rH
(79)
where rH = GM is the size of the horizon of the extremal RN black hole. It

is a kind of KaluzaKlein vacuum in which two directions are compactified and
the effective spacetime is the two-dimensional AdS2 spacetime of constant negative
curvature. Similarly, the AdSp+2 S d1 spacetime described by the near horizon metric
(53) or (77) is also a KaluzaKlein vacuum. Usually, such type of vacuum is realized
as a supersymmetric bosonic configuration, i.e., a supersymmetric vacuum of a higherdimensional supergravity theory. It also appears as the near-horizon geometry of nondilatonic black p-brane solutions of higher-dimensional supergravity or superstring
theories. It will be interesting to compare the metric (53) with that of supersymmetric
non-dilatonic black p-branes.
348
For comparison, we write down the metric of non-dilatonic extremal black p-brane:

2
dr 2
2
ds 2 = (r) p+1 dx dx +
+ r 2 dd1
,
(r)2
(80)
where the extra dimension is fixed to be d = 4p/(p 1) and
d2
rH
.
(r) 1
r
(81)
In the limit r rH , the metric is written as

2
2
2(p + 1) p+1
2
(r rH ) p+1 dx dx
ds
(p 1)rH

2(p + 1) 2 dr 2
2
2
+
+ rH
dd1
,
(p 1)rH
(r rH )2
(82)
which explicitly has the same form as Eq. (53). Even though Eqs. (53) and (82) describe
the same geometry near the horizon, the global spacetime structures of the two system, the
global black brane and the non-dilatonic black brane, are much different. The spacetime of
the global black brane is completely regular everywhere, whereas that of the non-dilatonic
black brane is singular at the center r = 0. The geometry of the spacetime of the global
black brane asymptotes to an AdS space, while that of the non-dilatonic brane does to the
Minkowski spacetime.
Notice that if we used a metric ansatz using a proper radial distance instead the
Schwarzschild coordinates, we would obtain a spacetime described Eq. (53) as an
asymptotic solution because the horizon is sitting at infinite proper distance from the
core. This is the thing done in Refs. [11] and [12], in which the authors discovered cigarlike warped spacetimes which is nothing but the infinitely long AdS throat with topology
of AdSp+2 S d1 . It is easy to see relation between the near-horizon solution and the
cigar-like warped spacetime obtained in Refs. [11,12]. Introducing a new radial coordinate
(> 0) such that

1
exp(k) BH (r rH )
(83)
,
the metric Eq. (53) is rewritten as
2
2
ds 2 exp(2k)g (x) dx dx + d 2 + rH
dd1
,
(84)
where the curvature scale k of the AdSp+2 is defined by k (1 ) BH . For the

interior solution ( = 1), runs from a finite value to (at r = rH ). For the exterior
solution ( = 1), runs between (at r = rH ) and (at r = ), but ought
to be truncated at a finite distance as, sufficiently large r, the near-horizon geometry is
replaced by the asymptotic spacetime Eq. (67). The metric (84) coincides with the cigarlike warped solution of Refs. [11,12] in the thin core limit of fH 1 where the horizon
locates outside the core. Hence, among solutions obtained in [11] and [12], the meaningful
solutions relevant for the RandallSundrum scenario would quite naturally be interpreted
as the near-horizon geometry of extremal global black p-branes. Moreover, we also find
349
that the two solutions, Eqs. (67) and (84), which were apparently treated disjointedly in
Refs. [11,12], can be matched to each other and describe the exterior region together.
In summary, the exterior region outside the horizon interpolates between the nearhorizon region, the infinitely long AdS throat with topology of AdSp+2 S d1 , and
the asymptotic region which asymptotes to AdSd . The interior region can be divided to
three separated regions, core region, near-horizon region and intermediate region between
the two regions. The near-horizon region corresponds to the infinitely long AdS throat
(AdSp+2 S d1 ). As such, the central region surrounded by the AdS throat region
looks like a one-sided RandallSundrum domain-wall embedded in (p + 2)-dimensional
AdS spacetime (AdSp+2 ). When the cosmological constant is of the same order as the
fundamental scale, i.e., M ||1/2 Mpl , most portion of the interior region is occupied
by the core and near-horizon regions and so the interior region is well described by only
the two region. The curvature radii of both AdSp+2 and S d1 are of the order of the Planck
scale. At low-energy below the Planck scale, the extra space is reduced effectively to a
one-dimensional space. Consequently, the global p-brane core looks like a p-dimensional
domain-wall embedded in an AdSp+2 bulk spacetime. In the thin core-approximation
limit, the system is essentially the same as that of the original RandallSundrum scenario.
However, when the cosmological constant is much less than the fundamental scale, the core
and the near-horizon regions are squeezed into small portions and, in between the two most
portion is occupied by an intermediate region. Since the cosmological constant is negligible
in the intermediate region, its geometry can be approximated by that of the global defect
spacetime in the absence of the negative cosmological constant. The geometry of this
region resembles that of the CohenKaplan solution [10] as d = 2 and the global monopole
solution [12] as d 3. Putting the above results together, the spacetime geometry of the
global black p-brane is illustrated in Fig. 4.
Fig. 4. The shape of the slice of the transverse part of the spacetime of the extreme global black brane. The time
and longitudinal directions are suppressed. Circles parallel to the horizon line represent S d1 . The surface is a
transversal slice.
350
4.2. Non-extreme global black branes

In this subsection we discuss the spacetime of the bent brane surrounded by two
horizons, of which a field configuration is shown in Fig. 2. Since the world volume
curvature of the brane core has to be positive constant, the corresponding world volume
metrics are those of (p + 1)-dimensional de Sitter spacetime, e.g.,
d s2 = dt 2 + e2H t dxi2

1 2

2
= 1 H 2 2 dt 2 + 1 H 2 2
d + 2 dp1
,
where 2
(85)
(86)
xi2 , and H
is the expansion rate along the brane. The first metric (85) is defined
in terms of planar coordinates and corresponds to a flat Friedman universe. The second one
is static and is the analog of the Schwarzschild metric for the flat de Sitter spacetime. The
t in these coordinates is not the same as the t in planar coordinates, but we are running out
of letters. In this coordinate system t is a Killing vector and generates the time translation
symmetry. From (86), we see that at = H 1 the norm of t vanishes, so that it becomes
null. Thus, the surface at = H 1 is the usual de Sitter horizon. On the other hand, since
= p(p + 1)H 2 , from Eqs. (59) and (60) it
H is related to the world volume curvature as R
can be read off that
1
H = eAH |B1 |.
(87)
2
The metrics (56), (57), and (63) describe only a few separated regions, but they are
smoothly connected as shown in Fig. 2. Similarly to the extreme black brane case, when
the cosmological constant is of the order of the fundamental scale, both the core size rc
and the sizes of the horizons rH are of the order of M1 and so the spacetime of the
interior region inside rH will be well described by the two metrics (56) and (63). On the
other hand, when the cosmological constant is much less than the fundamental scale, in the
intermediate region the transverse to the brane part of the solution will coincide with the
metric of a global monopole of Ref. [12].
The exterior region outside the outer horizon at rH + interpolates between the nearhorizon region and the asymptotic region described by metrics (57) and (63), respectively.
In the far region, the spacetime described by the metric (57) asymptotes to AdSD , even
though the longitudinal part of the solution is not flat. The Riemann curvature tensor
corresponding to the metric (57) is

H2
1
+
RMNP Q = B +
B r 2 M,N,P ,Q{x }
r 2 M,N,P ,Q{ a }
(gMP gNQ gMQ gNP ),
(88)
where the two terms dressed by square brackets are non-vanishing only if M, N, P , Q
{x } and M, N, P , Q { a }, respectively. Since the two terms in the square bracket decay
quadratically, the spacetime asymptotes to AdSD as r . This type of spacetime has
also been found in Ref. [12], and the metric can be rewritten to the form found in Ref. [12]
as done in the flat brane case
ds 2 e2
g (x) dx dx + d 2 + e2
2
dd1
.
(89)
351
Now we turn to the near-horizon metrics (63). One may check directly by calculating the
geodesics of the metric (63) that the coordinate chart {x , r, a } is geodesically incomplete.
As done in the case of flat brane, we consider the motion of test particles. It will be
sufficient to consider the radial motion of test particles along the line of = 0 in the static
coordinate system to see the causal structure on the transverse slice. That is, it is sufficient
to study null and timelike geodesics in the system
ds 2 =
4H 2
dr 2
,
(r rH ) dt 2 +
B1
B1 (r rH )
(90)
where we have used the relation (87). This metric is reminiscent of the near-horizon
metrics of the non-extremal ReissnerNordstrm black hole. The proper distances from
a point r0 to the horizons in a constant time slice is finite and is given by (r0 )
2 (r0 rH )/B1 . With this metric we find the geodesic equation
2
dt
1
H2
dr
(r rH )
+
.
=
(91)
B1
d
B1 (r rH ) d
With the static world volume metric, there is a constant of motion corresponding to the
Killing vector t
:=
H2
dt
(r rH ) .
B1
d
The equation of motion of the test particles then is
2
B1
dr
: 2 2 B1 (r rH ) = 0.
d
H
(92)
(93)
The equation of motion tells that the test particle reach at the horizon with a finite velocity,
i.e., (dr/d ) :B1 /H as r rH . Therefore, the coordinate chart {x , r, a } is
geodesically incomplete and so the extensions beyond each horizon are required. If the
solutions obtained in separated regions are mutually connected as shown in Fig. 2, the
causal structure induced by the (t, r)-part of the solutions will resemble that of the nonextreme RN black hole. The differences are the non-singular brane core instead a curvature
singularity at r = 0 and the timelike boundary rather than null at r = . The corresponding
Penrose diagram is shown in Fig. 5.
The metric (63) has some very interesting properties. The metrics with the form of (63)
usually have curvature singularities at r = rH . However, when the relation (87) holds,
the curvature invariants are finite and the surfaces at r = rH do indeed appear to be just
coordinate singularities. For example, the corresponding Kretschmann scalar near each
horizon at r = rH is given by
RMNP Q R MNP Q =
2(d 1)(d 2)
.
rH 4
(94)
This is not surprising, since the metric (63), in fact, describes constant curvature spacetimes
with topology of locally (p + 2)-dimensional flat Minkowski spacetime multiplied by
(d 1)-dimensional sphere with radius rH , i.e., M1,p+1 S d1 . This can easily be
352
Fig. 5. Penrose diagram for the non-extreme global black brane with field configurations shown in Fig. 2.
The spatial coordinates x i and a are suppressed. This diagram is very similar to that of the non-extreme
ReissnerNordstrm black hole, but now there is a non-singular brane core at r = 0 rather than a curvature
singularity. The boundary at r = is timelike rather than null.
observed by calculating the Riemann curvature tensor and noticing that the only nonvanishing components of the corresponding Riemann curvature tensor are
Ra b c d =
1
rH 2
(ga c gb d ga d gb c ),
(95)
where a , b , . . . refer to indices on S d1 .

It is interesting to observe that the (p + 2)-dimensional flat part corresponds the
spacetime of a p-dimensional thin domain wall embedded in (p + 2)-dimensional flat
spacetime, which is a generalization of the inflating three-brane in four dimensions [23].
The metric (63) can be rewritten to an analogous form to that of the inflating three-brane
found in Ref. [23] introducing a new radial coordinate e2H z 4H 2[(r rH )/B1 ], the
metric can easily be rewritten as

2
,
ds 2 = e2H z d s2 + dz2 + rH 2 dd1
(96)
where
the horizons are located at z = . Another coordinate transformation

2 B1 (r rH )/B1 + 0 , where 0 is an integral constant, brings Eq. (63) to the form
2
ds 2 = (1 H)2 d s2 + d 2 + rH 2 dd1
,
(97)
where the relation Eq. (87) has been used and 0 has been set as to be 0 = H 1 , so that
the horizons are at = H 1 . The (x , )-parts of metrics explicitly coincide with those
of the vacuum domain wall obtained in Ref. [23]. Thus, the near-horizon spacetime of the
bent brane is nothing but that of (p + 2)-dimensional vacuum domain wall times (d 1)dimensional sphere.
353
The spacetime of domain walls has been extensively studied in many references
[2332]. For the metric Eq. (96), the z = const (or = const for Eq. (97)) hypersurfaces
have the properties of (p + 1)-dimensional de Sitter space. Furthermore, the (t, z)-part (or
(t, )-part) of the metric describes a (1 + 1)-dimensional Rindler space (i.e., flat space
in the reference frame of a uniformly accelerating observer with proper acceleration H ).
Therefore, at each of the (p + 1) spatial directions event horizons exist. The ones in the p
longitudinal directions are de-Sitter-like as discussed at the beginning of this subsection.
The one in the radial direction is Rindler-like, and the various extensions beyond it were
considered in Refs. [23,25,27]. The extensions could be simply obtained by gluing together
two portions of Minkowski spacetime. Such extensions occupy only the interior region of
the Penrose diagram of Fig. 5 without the exterior region interpolating between the horizon
and the timelike infinity. On the other hand, the global spacetime of non-extreme global
black brane corresponding to the field configuration of Fig. 2 bears remarkable similarities
to that of the non-extreme domain walls found in Ref. [30].
When the horizon size rH is of the order of the Planck length scale (this will be true
if ||1/2 M MPl ), at low-energy below the Planck scale the brane core will look
like a inflating domain wall embedded in (p + 2)-dimensional Minkowski spacetime to an
observer residing near the inner horizon. However, when the bulk cosmological constant is
much less than the fundamental scale, the horizon size is much larger than the fundamental
length scale and so, even at low-energy, the core will look like a string, or a monopole,
etc., depending on the number of extra dimensions rather than a domain wall. Interestingly,
when p = 1 and d = 2 the metric (97) itself coincide with the asymptotic solution near the
horizon of the non-static global cosmic string found in Ref. [5], and the interior region
inside the inner horizon resembles the spacetime of the non-static global cosmic string.
On the other hand, the nature of the cosmological event horizon of Refs. [5,6] seems to
be different from the inner horizon at rH , in the sense that across the cosmological event
horizon the coordinates t and of [5,6] remain timelike and spacelike, respectively, while
the roles of t and in the metric (97) is exchanged across the inner horizon.4 The interior
region of the other defects of p > 1 and d 2 are simply the generalizations of the nonstatic global cosmic string spacetime.
5. Brane worlds and a thermal instability for the cosmological constant

In this section we consider the global black branes as an approximation for brane world
scenarios. We will shortly summarize the contents of Ref. [16], in which the extremal
ones only are considered, and discuss the effective gravity on the brane when it is nonextremal. We briefly comment a few thermodynamic properties of the global black branes,
and discuss a decrease of the effective cosmological constant in the brane world through
thermal instability of the non-extremal black brane. The exterior region of the extremal
4 This can be seen explicitly from the metric (63). Note that, if r
H < r < rH + , the coordinate transformation

from r to should be replaced by 2 B1 (r rH )/B1 + 0 and the metric is replaced by ds 2 =
2 .
(1 H)2 d s 2 d 2 + rH 2 dd1
354
global black branes by itself does not seem to provide a suitable framework for the brane
world scenario, because in one direction one will reach the (p + 2)-dimensional AdS
horizon, while in the other direction one will reach the D-dimensional AdS spacetime.
On the other hand, the interior region possesses all the feature necessary for realizing the
RandallSundrum (RS) type brane world scenario. In one direction, it asymptotes to the
infinitely long AdS throat. As such, the central region surrounded by the AdS region looks
like a one-sided RS domain-wall embedded in (AdSp+2 ). As discussed in Ref. [16], the
size of the horizon can be interpreted as the effective size of d extra dimensions in that
the Planck scale in the p-dimensional brane world MPl is determined by the fundamental
p1
p1+d d
scale M and the horizon size rH via the familiar relation MPl M
rH . Moreover,
the gravity on the brane world behaves as expected in a world with d-extra dimensions
compactified with size rH .
When the fundamental scale and the bulk cosmological constant are of the order of the
brane world Planck scale, i.e., M ||1/2 MPl , then the curvature radii of AdSp+2
and S d1 are of the order of the Planck scale. Thus, at low energy below the Planck
scale, the extra space is reduced effectively to a one-dimensional space. Consequently, the
global p-brane core looks like a p-dimensional domain-wall embedded in an AdSp+2 bulk
spacetime. In the thin core-approximation limit, the physics on the brane is essentially the
same as that of the original RS scenario. On the other hand, when the cosmological constant
is hierarchically smaller than the fundamental mass scale, the horizon size is much larger
than the core size or the fundamental length scale, i.e., rH rc M1 , so the physics
is quit analogous to that of the large extra dimension scenario. Then the Planck scale MPl
in the brane world is hierarchically bigger than the fundamental scale M of the higher
p1
p1+d d
rH . And as shown
dimensional gravity, according to the familiar relation MPl M
in Ref. [16], the phenomenology on the brane is imperceptibly different from that of the
usual large extra dimension scenario.
For the bent branes surrounded by two horizons, since the interior region is bounded
by the inner horizon as the interior region of the extreme black brane is bounded by the
Cauchy horizon, the outside of the inner horizon is out of causal contact with the brane
p1
p1+d d
rH . The
world. So the Planck scale on the brane world is given by MPl M
interior region has similar features to the inflating domain wall spacetime or the de Sitter
brane, so the gravity on the brane will resemble that on the de Sitter brane. As discussed
in Refs. [3436], the massless graviton is trapped on the de Sitter brane reproducing the
correct (p + 1)-dimensional gravity on the brane as on the Minkowski brane. A difference
is that in the Minkowski case the massless graviton is a marginal bound state and the
continuum of KaluzaKlein modes starts at zero mass, while in the de Sitter case the
graviton is separated by a finite mass gap of m = (3/2)H from the continuum, that is,
the KaluzaKlein continuum modes starts at m = (3/2)H .
In this picture, the Hawking radiation could be a possible mechanism for the resolution
of various problems associated with the brane world scenarios, such as the observed
flatness and the approximate Lorentz invariance of our world. The bending of the brane
and the bulk curvature that violate the SO(3, 1) isometry on our brane [37,38] would
355
correspond to excitations upon the extremal state,5 and the excited states that correspond
to non-extremal states would evolve into an extremal state through the Hawking radiation
process. The similar argument may be applicable as well for the cosmological constant
problems.
We now turn to the thermal instability of the non-extremal black branes, which could
be a mechanism resolving the cosmological constant problem.6 In the brane world, the
intrinsic curvature of the brane amounts to a net cosmological constant as viewed by an
Since we have found that,
observer pinned on the brane, i.e., phys = [(p 1)/2(p + 1)]R.
in higher-dimensional pure gravity theory with a cosmological constant term explicitly,
there exists a class of solutions with physical (p + 1)-dimensional metrics obeying the

standard Einstein equations with positive arbitrary (including zero) values of phys or R,
the cosmological constant problem may be solved by choosing a solution characterized by
vanishingly small phys , that is, a very near extremal black brane solution.7 However, this
does not seem to tell anything about the cosmological constant problem, because there are
no convincing arguments in favor of this choice. In this respect, we might need a dynamical
mechanism that singles out the solution with an extremely small phys , which would be
a complete solution to the cosmological constant problem. As discussed in Refs. [16,18],
a possible dynamical mechanism may be related to the thermodynamic instability of the
non-extreme black branes, by which the positive vacuum energy density on the brane could
be radiated away causing the brane to evolve into an extremal state.
In this respect, we briefly comment a few thermodynamic properties of the extreme
and non-extreme global black branes, and discuss a decrease of the effective cosmological
constant on the brane. For the extremal black brane, since the surface gravity is zero at the
degenerated horizon, the Hawking temperature is zero and there is no Hawking radiation.
Thus, this system is in fact thermodynamically stable. The entropy of a system is associated
with the number of accessible states. Since the extremal state of the global black brane
seems to be a unique solution to the equations of motion as discussed in the Subsection 3.1,
its entropy would be zero. On the other hand, the entropy associated with a spacetime
containing horizon is given by the area of the event horizon: SBH = AD /4GD , where AD
is the area of the horizon and GD is the D-dimensional Newton constant. The area of the
spacelike surface of constant r reads from the metric Eq. (53) as: AD (r) (r rH )2p(1) ,
which clearly vanishes as r rH . This supports the conclusion of zero entropy property
of the extreme black brane.
5 Here, the corresponding non-extremal black branes are different from those discussed in this paper. Rather,
they will resemble the static black branes, of which world volume metrics depend on the extra dimension
coordinates, e.g., d s 2 = g (y) dx dx = dt 2 + h(r) dxi2 .
6 There have been many interesting ideas in the context of the brane world scenario in attempts to solve
the cosmological constant problem [4044]. The basic idea is self tuning. The brane tension could take on an
extended range of values without jeopardizing the stability of the cosmological constant on the brane. In this way,
the brane could be stable against any radiative corrections to the brane tension. However, the naked singularities,
which lead to inconsistencies in the effective theory on the brane world, is inherent [45,46], or the static self-tuned
solutions are dynamically unstable [4750].
7 Recent cosmological observations [51] place the cosmological constant at a very small value (but not zero)
2
120 M 4 .
in comparison to the Planck scale; in fact, phys MPl
phys /8 10
Pl
356
For the non-extremal case, as mentioned in the previous section there are de Sitter
horizon in p longitudinal directions and the Rindler horizon in radial direction. The
de Sitter horizon is associated with a surface gravity = H and so an observer will
detect an isotropic background in the longitudinal directions of thermal radiation with
temperature T = (2)1 H coming, apparently, from the horizon [33]. The de Sitter
horizon also may be interpreted as the entropy or lack of information that the observer
has about the regions of the universe that he cannot see. If one absorbs the thermal
radiations, one gains energy and entropy at the expanse of this region and so the area
of the horizon will go down according to the first law of the black hole dynamics. As the
area decreases, the temperature of the cosmological radiation goes up (like the black hole
case), so the cosmological event horizon is unstable. On the other hand, at the Rindler
horizon the surface gravity can be identified by the proper acceleration of a uniformly
accelerating observer. Since the metric corresponds to that of the uniformly accelerating
observer with the proper acceleration H , the surface gravity is simply given by = H ,
and the temperature measured
by a static observer near both horizons at rH is just
T = (2)1 H or T = (2)1 2phys/p(p 1). Therefore, the observer will detect two
thermal radiations with the same temperatures of T = (2)1 H coming from the de Sitter
horizon and the Rindler one, respectively. This seems to be well matched with the zeroth
law of black hole mechanics. Here, an interesting point is that the surface gravity at the
horizons is determined by the brane world volume curvature solely regardless the size of
the horizon unlike usual black hole and static black brane cases. An observer residing in
the exterior region at a distance from the outer horizon may observe a steady current of
thermal radiation coming from the outer horizon and going out toward the asymptotically
anti-de-Sitter infinity, even if the observer see an isotropic thermal background in the
longitudinal directions. This implies a possibility that the non-extreme black brane could
evolve into an extreme black brane of which world volume is flat. Since the brane itself
is the source of the radiation in radial direction, this must result in energy loss from
the brane. If the quantum fluctuations of the vacuum energy on the brane is coupled to
bulk fields (including bulk gravitational field), then the vacuum energy density could be
radiated into the bulk outside of the outer horizon through the Hawking radiation process,
resulting in a decrease of the vacuum energy density or, equivalently, the world volume
Thus, the Hawking temperature goes down with time, and the radiation will
curvature R.

last until the temperature reaches to zero. Since the temperature is proportional to R,
= 0. Equivalently, since the temperature is proportional to
the radiation will go on until R

the square root of the effective cosmological constant, phys , the radiation will continue
until phys = 0. Thus, this effect may radiate away all contributions to the cosmological
constant.8
8 Recently, a very interesting proposal on a thermal instability for the cosmological constant was appeared
in Ref. [52]. There are modes of the linearized bulk gravitational field which see the brane as a accelerating
mirror. This give rise to an emission of thermal radiation from the brane into the bulk. The temperature is also

proportional to the square root of the cosmological constant on the brane world, phys .
357
6. Conclusions
In this paper we have mainly examined global black p-brane solutions, which are
black hole-like p-dimensional extended global defect solutions to a scalar theory with
global O(d) internal symmetry coupled to higher-dimensional gravity with a negative
cosmological constant. The spacetimes of them are perfectly regular everywhere. We have
found series solutions in a few separated regions and numerically showed that they can be
smoothly connected. There are two kinds of such solutions, extremal or non-extremal black
branes. The extremal black branes have a Ricci flat world volume and are surrounded by a
degenerated Killing horizon. On the other hand, the world volumes of non-extremal black
branes surrounded by two horizons are not Ricci flat but of non-zero constant curvature.
The extremal black brane solutions seem to be the critical limit of non-extremal ones, in
that the positions of two horizons of non-extremal branes approach to and are degenerated
at the position the horizon of extremal black branes in the limit that the world volume
0. The exterior region of both objects asymptotes to anticurvature goes to zero, i.e., R
de-Sitter infinity.
The near-horizon geometries of both the extremal and the non-extremal branes have
very interesting features. For the extremal black brane the geometry is the infinitely long
d1 . The curvature scales of both
anti-de-Sitter throat with the topology of AdS
p+2 S
d1
AdSp+2 and S
parts are of the order of ||. So the core will look like a one-side flat
domain wall embedded in (p + 2)-dimensional anti-de-Sitter space to an observer residing
in the near-horizon region of the interior region if the size of the sphere is small so that the
observer does not see S d1 -part. The physics on the branes then is essentially the same as
that on the RandallSundrum brane as discussed in Ref. [16]. On the other hand, for the
non-extremal ones it is the (p + 2)-dimensional flat spacetime times (d 1)-dimensional
sphere with radii rH at each horizon. The (p + 2)-dimensional flat part corresponds to
the spacetime of a p-dimensional thin domain wall embedded in (p + 2)-dimensional flat
spacetime. An observer residing in the near-horizon region inside the inner horizon will
see an inflating domain wall if the size of the inner horizon is small enough. Therefore,
the gravitational KaluzaKlein modes have a mass gap in the spectrum and the continuous
spectrum starts above the mass m = (3/2)H , where H is the expansion rate of the brane.
In the brane world, the world volume curvature amounts to the effective cosmological
= 0 (i.e., the extremal black
constant. We obtained a flat brane solution with R
brane) without fine-tuning between parameters in the Lagrangian. However, to solve the
cosmological constant problem completely we need a dynamical mechanism that picks out
We have suggested a possible
the flat brane solution among whole solutions of any R.
thermodynamic mechanism as a candidate for such mechanism. It is provable for a nonextremal black brane to evolve into an extremal black brane owing
to itsthermodynamic
phys resulting
instability. It might emit a thermal radiation with temperature T R
in a decrease of the vacuum energy on the brane, flattening itself and finally reaching
to the extremal limit at temperature T = 0. This effect will dilute all contribution to the
cosmological constant. If it is really possible, then the black brane world model represents
a solution to the cosmological constant problem.
358
Acknowledgements
We would like to thank Yoonbai Kim and Soo-Jong Rey for collaboration on the
previous work and for useful discussions. We also would like to thank Hyeong-Chan Kim
for delightful discussions. It is a great pleasure to thank Yong-Jin Chun for his assistance
to the numerical works. Without his aid, it would have not been finished. This work was
supported by the BK21 project of Ministry of Education.
References
[1] A. Vilenkin, E.P.S. Shellard, Cosmic String and other Topological Defects, Cambridge Univ. Press, 1994.
[2] R. Gregory, Phys. Lett. B 215 (1988) 663;
G.W. Gibbons, M.E. Ortiz, F. Ruiz-Ruiz, Phys. Rev. D 39 (1989) 1546.
[3] A.G. Cohen, D.B. Kaplan, Phys. Lett. B 215 (1988) 67.
[4] M. Barriola, A. Vilenkin, Phys. Rev. Lett. 63 (1989) 341.
[5] R. Gregory, Phys. Rev. D 54 (1996) 4955.
[6] A. Wang, J.A.C. Nogales, Phys. Rev. D 56 (1997) 6217.
[7] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263;
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Rev. D 59 (1999) 086004.
[8] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370.
[9] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690.
[10] A.G. Cohen, D.B. Kaplan, Phys. Lett. B 470 (1999) 52.
[11] R. Gregory, Phys. Rev. Lett. 84 (2000) 2564.
[12] I. Olasagati, A. Vilenkin, Phys. Rev. D 62 (2000) 044014.
[13] T. Gherghetta, M. Shaposhnikov, Phys. Rev. Lett. 85 (2000) 240.
[14] T. Gherghetta, E. Roessl, M. Shaposhnikov, Phys. Lett. B 491 (2000) 353.
[15] I. Oda, Phys. Lett. B 496 (2000) 113.
[16] Y. Kim, S.-H. Moon, S.-J. Rey, Nucl. Phys. B 602 (2001) 467.
[17] K. Benson, I. Cho, Phys. Rev. D 64 (2001) 065026.
[18] W.S. Bae, Y.M. Cho, S.-H. Moon, J. High Energy Phys. 0103 (2001) 039.
[19] N. Kim, Y. Kim, K. Kimm, Phys. Rev. D 56 (1997) 8029;
N. Kim, Y. Kim, K. Kimm, Class. Quantum Grav. 15 (1998) 1513.
[20] Y. Kim, S.-H. Moon, Phys. Rev. D 58 (1998) 105013;
G. Clment, A. Fabbri, Class. Quantum Grav. 17 (2000) 2537.
[21] K. Lee, V.P. Nair, E.J. Weinberg, Phys. Rev. Lett. 68 (1992) 1100;
K. Lee, V.P. Nair, E.J. Weinberg, Phys. Rev. D 45 (1992) 2751.
[22] G.T. Horowitz, A. Strominger, Nucl. Phys. B 360 (1991) 197.
[23] A. Vilenkin, Phys. Lett. B 133 (1983) 177.
[24] L.M. Widrow, Phys. Rev. D 39 (1989) 3571.
[25] G.W. Gibbons, Nucl. Phys. B 394 (1993) 3.
[26] A. Wang, P.S. Letelier, Phys. Rev. D 51 (1995) R6612.
[27] A. Wang, P.S. Letelier, Phys. Rev. D 52 (1995) 1800.
[28] M. Cvetic, S. Griffies, S.-J. Rey, Nucl. Phys. B 381 (1992) 301;
M. Cvetic, S. Griffies, S.-J. Rey, Nucl. Phys. B 389 (1993) 3.
[29] M. Cvetic, R.L. Davis, S. Griffies, H.H. Soleng, Phys. Rev. Lett. 70 (1993) 1191.
[30] M. Cvetic, S. Griffies, H.H. Soleng, Phys. Rev. Lett. 71 (1993) 670.
[31] N. Kaloper, A. Linde, Phys. Rev. D 59 (1999) 101303.
[32] N. Kaloper, Phys. Rev. D 60 (1999) 123506.
[33] G.W. Gibbons, S.W. Hawking, Phys. Rev. D 15 (1977) 2738.
[34] O. DeWolfe, D.Z. Freedman, S.S. Gubser, A. Karch, Phys. Rev. D 62 (2000) 046008.
[35] A. Karch, L. Randall, J. High Energy Phys. 0105 (2001) 008.
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
359
J. Garriga, M. Sasaki, Phys. Rev. D 62 (2000) 043523.

D.J.H. Chung, E.W. Kolb, A. Riotto, Extra dimensions present a new flatness problem, hep-ph/0008126.
J.F. Vzquez-Poritz, Massive gravity on a non-extremal brane, hep-th/0110299.
V.A. Rubakov, M.E. Shaposhnikov, Phys. Lett. B 125 (1983) 139.
N. Arkani-Hamed, S. Dimopoulos, N. Kaloper, R. Sundrum, Phys. Lett. B 480 (2000) 193.
S. Kachru, M. Schulz, E. Silverstein, Phys. Rev. D 62 (2000) 045021.
A. Kenhagias, K. Tamvaski, A self-tuning solution of the cosmological constant problem, hep-th/0011006.
N. Tetradis, Phys. Lett. B 509 (2001) 307.
J.E. Kim, B. Kyae, H.M. Lee, Phys. Rev. Lett. 86 (2001) 4223;
J.E. Kim, B. Kyae, H.M. Lee, Nucl. Phys. B 613 (2000) 306.
S. Forste, Z. Lalak, S. Lavignac, H.P. Nilles, Phys. Lett. B 481 (2000) 360;
S. Forste, Z. Lalak, S. Lavignac, H.P. Nilles, J. High Energy Phys. 0009 (2000) 034.
C. Csaki, J. Erlich, C. Grojean, T. Hollowood, Nucl. Phys. B 584 (2000) 359.
I. Low, A. Zee, Nucl. Phys. B 585 (2000) 395.
P. Binetruy, J.M. Cline, C. Grojean, Phys. Lett. B 489 (2000) 403.
J. Diemand, C. Mathys, D. Wyler, Dynamical instabilities of brane world models, hep-th/0105240.
A.J.M. Medved, Dynamical instability of self-tuning solution with antisymmetric tensor field, hepth/0109180.
N. Bahcall, J.P. Ostriker, S. Perlmutter, P.J. Steinhardt, Science 284 (1999) 1481.
S. Alexander, Yi Ling, Lee Smolin, A thermal instability for positive brane cosmological constant in the
RandallSundrum cosmologies, hep-th/0106097.

Calculating the I = 2 pion scattering length using

tadpole improved clover Wilson action on coarse
anisotropic lattices
Chuan Liu a , Junhua Zhang a , Ying Chen c , J.P. Ma b
a Department of Physics, Peking University, Beijing, 100871, PR China
b Institute of Theoretical Physics, Academia Sinica, Beijing, 100080, PR China
c Institute of High Energy Physics, Academia Sinica, P.O. Box 918, Beijing, 100039, PR China
Abstract
In an exploratory study, using the tadpole improved clover Wilson quark action on small, coarse
and anisotropic lattices, the scattering length in the I = 2 channel is calculated within quenched
approximation. A new method is proposed which enables us to make chiral extrapolation of
our lattice results without calculating the decay constant on the lattice. Finite volume and finite
lattice spacing errors are analyzed and the results are extrapolated towards the infinite volume and
continuum limit. Comparisons of our lattice results with the new experiment and the results from
Chiral Perturbation Theory are made. Good agreements are found. 2002 Elsevier Science B.V. All
rights reserved.
PACS: 12.38.Gc; 11.15.Ha
Keywords: scattering length; Lattice QCD; Improved actions
1. Introduction
It has become clear that anisotropic, coarse lattices and improved lattice actions are ideal
candidates for lattice QCD calculations on small computers [14]. They are particularly
advantageous for heavy objects like the glueballs, one meson states with non-zero spatial
momenta and multi-meson states with or without spatial momenta. The gauge action
E-mail address: liuchuan@mail.phy.pku.edu.cn (C. Liu).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 6 2 - 9
C. Liu et al. / Nuclear Physics B 624 (2002) 360376
employed is the tadpole improved gluonic action on anisotropic lattices:

5 Tr Pij
1 Tr Rij
1 Tr Rj i
S =
9 u4s
36 u6s
36 u6s
i>j

4 Tr P0i
1 Tr Ri0
,
9 u2s
36 u4s
361
(1)
where P0i and Pij represents the usual temporal and spatial plaquette variable, respectively.
Rij and Ri0 designates the 2 1 spatial and temporal Wilson loops, where, in order to
eliminate the spurious states, we have restricted the coupling of fields in the temporal
direction to be within one lattice spacing. The parameter us , which is taken to be the
fourth root of the average spatial plaquette value, is the tadpole improvement parameter
determined self-consistently from the simulation. With this tadpole improvement factor,
the renormalization of the anisotropy (or aspect ratio) = as /at will be small. Therefore,
we will not differentiate the bare aspect ratio and the renormalized one in this work. Using
this action, glueball and hadron spectra have been studied within quenched approximation
[39]. In this paper, we would like to present our results on the pionpion scattering lengths
within quenched approximation. Main results of this paper has already been reported
in [10]. Some details of the calculation are presented in this paper.
Lattice calculations of pion scattering lengths have been performed by various authors
using symmetric lattices without the improvement [1113]. It turned out that, using the
symmetric lattices and Wilson action, large lattices have to be simulated which require
substantial amount of computing resources. It becomes even more challenging if the
chiral, infinite volume and continuum limits are to be studied [13]. In this exploratory
study, we would like to show that, such a calculation is feasible using relatively small
lattices, typically of the size 83 40, with the tadpole improved anisotropic lattice actions.
Attempts are also made to study the chiral, infinite volume and continuum limits
extrapolation in a more systematic fashion. The final extrapolated result on the pionpion
scattering length in the I = 2 channel is compared with the new experimental result from
E865 Collaboration [14] and Chiral Perturbation Theory [1518] with encouraging results.
Of course, more accurate lattice calculations of the pion scattering length will inevitably
require lattices larger than the ones studied in this paper.
The fermion action used in this calculation is the tadpole improved clover Wilson action
on anisotropic lattices [19,20]:
Mxy = xy + Axy ,

3

1
+ t
0i F0i + s (12 F12 + 23 F23 + 31 F31 )
Axy = xy
2max
i=1

(1 )U (x)x+,y + (1 + )U (x )x,y ,
(2)
where various coefficients in the fermion matrix M are given by:

i = /(2us ),
0 = /2,
= 1/(2) 1/(1max ),
2

t = cSW (1 + )/ 4us ,
s = cSW / 2u4s .
(3)
362
Among the parameters which appear in the fermion matrix, cSW is the coefficient of the
clover term and is the so-called bare velocity of light, which has to be tuned nonperturbatively using the single pion dispersion relations [20]. Tuning the clover coefficients
non-perturbatively is a difficult procedure. In this work, only the tadpole improved treelevel value is used instead.
In the fermion matrix (2), the bare quark mass dependence is singled out into the
parameter and the matrix A remains unchanged if the bare quark mass is varied.
Therefore, the shifted structure of the matrix M can be utilized to solve for quark
propagators at various values of valance quark mass m0 (or equivalently ) at the cost
of solving only the lightest valance quark mass value at = max , using the so-called
Multi-mass Minimal Residual (M 3 R for short) algorithm [2123]. This is particularly
advantageous in a quenched calculation since one needs the results at various quark mass
values to perform the chiral extrapolation.
This paper is organized in the following manner. In Section 2, the method to calculate
scattering length is reviewed. In Section 3, some simulation details are described. In
Section 4, our results of the scattering lengths obtained on lattices of various sizes and
lattice spacings are extrapolated towards the chiral limit. A new quantity is proposed which
is easier to measure on the lattice and has a simpler chiral behavior than the scattering
length itself. Finite size effects are studied and two schemes of extrapolating to the infinite
volume limit are studied. Finally, our lattice results are extrapolated towards the continuum
limit. Comparisons with the experimental value and values from chiral perturbation theory
at various orders are also discussed. In Section 5, we conclude with some general remarks.
2. Formulation to extract the scattering length

In order to calculate hadron scattering lengths on the lattice, or the scattering phase
shifts in general, one uses Lschers formula which relates the exact energy level of two
hadron states in a finite box to the scattering phase shift in the continuum. In the case of
two pions which scatter at zero relative three momentum, this formula then relates the exact
(I )
in a finite box of size L and isospin I channel to the corresponding
two pion energy E
(I )
scattering length a0 in the continuum. This formula reads [24]:
(I )
E
2m =
(I ) 2
(I )
(I )

a0
4a0
a0
+
c
1
+
c
+ O L6 ,
1
2
3
L
L
m L
(4)
where c1 = 2.837297, c2 = 6.375183 are numerical constants. In this paper, this formula
(2)
will be utilized to calculate the pionpion scattering length a0 in the I = 2 channel, which
(2)
(2)
then requires the determination of the corresponding energy shift E
E
2m in
that channel.
One issue that one has to keep in mind is that, in a quenched calculation, Lschers
formula (4) has to be modified dramatically as discussed in [25]. Due to the anomalous
contributions in quenched chiral perturbation theory from , the energy shifts are
contaminated by terms that might be of order L0 = 1 and L2 in the I = 0 channel. If these
contributions were substantial, extraction of the corresponding scattering length from the
363
energy shift might become problematic. In the I = 2 channel, the situation is better. The
quenched chiral corrections modify the energy shift by terms that are at least of order L3
with usually small coefficients. Therefore, doing quenched calculation in the I = 2 channel
is safer.
(2)
of two pions with
To measure the pion mass m and to extract the energy shift E
zero relative momentum, we construct the correlation functions from the corresponding
operators in the I = 2 channel. We have used the operators proposed in Ref. [12]. First, the
local operators which create a single pion with appropriate isospin values are:
t)5 u(x, t),
(x, t) = u(x,
t)5 d(x, t),
+ (x, t) = d(x,

1
t)5 d(x, t) ,
0 (x, t) = u(x,
(5)
t)5 u(x, t) d(x,
2
t) are basic local quark fields corresponding to
where u(x, t), d(x, t), u(x,
t) and d(x,
u- and d-quarks, respectively. The operators which correspond to zero momentum single
pions are:
1 a
(x, t),
0a (t) = 3/2
(6)
L
x
where the flavor of pions a = +, , 0 and L3 is the three volume of the lattice. Zero
momentum one pion correlation function can then be formed as:

C (t) = 0a (t)0a (0) .
(7)
From the large t behavior of this correlation function, the pion mass m is obtained.
Similarly, we use the two pion operators in the I = 2 channel defined by:
I =2
(t) = 0+ (t)0+ (t + 1),
O
to construct the two pion correlation functions:

I =2

I =2
I =2
C
(t) = O
(t)O
(0) .
(8)
(9)
I =2 (t), one could infer the exact two pion state energy in
From the large t behavior of C
the finite box. Numerically, it is more advantageous to construct the ratio of the correlation
functions defined above:

I =2
(t) C (t)C (t) .
RI =2 (t) = C
(10)
This ratio thus exhibits the following asymptotic behavior for large t:
t 1
(2)
R(t) eE t ,
(2)
with E
(11)
(2)
= E 2m
is the energy shift in this channel which directly enters Lschers

(2)
formula (4). It is argued [12,25,26] that this formula can only be utilized for small E t
values in a quenched calculation for large enough L. In this case, one could equally use the
linear fitting function:
R(t)
T t 1
(2)
1 E
t,
(12)
364
Fig. 1. Diagrams representing the contributions to the two-pion correlation function in the I = 2 channel.
Diagrams (A) and (D) correspond to the Direct contribution while (B) and (C) belong to the Cross contribution.
(2)
to determine the energy shift E .

Two pion correlation function, or equivalently, the ratio R(t) constructed above can be
transformed into products of quark propagators using Wicks theorem. In order to avoid the
complicated Fierz re-arrangement terms, we used the creation operators at time slices that
are differ by one lattice spacing as suggested in [12]. It can be shown that the I = 2 two
pion correlation function is given by two contributions which are termed Direct and Cross
contributions [11,12]. These contributions are represented by diagrams as shown in Fig. 1,
where the solid and dashed lines represent the u- and d-quark propagators respectively.1
The two pion correlation function in the I = 0 channel is, however, more complicated
which involves vacuum diagrams that require to compute the quark propagators for wall
sources placed at every time-slice, a procedure which is more time-consuming than the
I = 2 channel. In the I = 2 channel, it turns out that the quark propagators have to be
solved only twice with two fixed wall sources placed at t = 0 and t = 1.
3. Simulation details
Simulations are performed on several PCs and two workstations. Configurations are
generated using the pure gauge action (1) for 43 40, 63 40 and 83 40 lattices with the gauge
coupling = 1.7, 2.2, 2.4 and 2.6. The spatial lattice spacing as is roughly between
0.18 fm and 0.39 fm while the physical size of the lattice ranges from 0.7 fm to 3.2 fm.
For each set of parameters, several hundred decorrelated gauge field configurations are
used to measure the fermionic quantities. Statistical errors are all analyzed using the usual
jack-knife method. Basic information of these configurations is summarized in Table 1.
Quark propagators are measured using the Multi-mass Minimal Residue algorithm for
5 different values of bare quark mass. Periodic boundary condition is applied to all three
1 In this work, the u- and d-quark are assumed to be degenerate in mass.
365
Table 1
Simulation parameters for lattices studied in this work. Input aspect ratio parameter is fixed to be 5 for all
lattices being studied. The approximate spatial lattice spacing as in physical units as obtained from [4,9] is also
indicated. Also listed are the maximum value of the hopping parameter max
Lattice
u4s
as (fm)
No. confs
max
1.7
1.7
1.7
43 40
63 40
83 40
0.295
0.295
0.295
0.39
0.39
0.39
700
288
240
0.0570
0.0585
0.0585
0.90
0.90
0.90
2.2
2.2
2.2
43 40
63 40
83 40
0.378
0.378
0.378
0.27
0.27
0.27
700
256
224
0.0590
0.0600
0.0600
0.95
0.95
0.95
2.4
2.4
2.4
43 40
63 40
83 40
0.409
0.409
0.409
0.22
0.22
0.22
700
256
224
0.0620
0.0605
0.0605
0.87
0.92
0.93
2.6
2.6
2.6
43 40
63 40
83 40
0.438
0.438
0.438
0.19
0.19
0.19
600
256
240
0.0630
0.0620
0.0605
0.80
0.88
0.93
spatial directions while in the temporal direction, Dirichlet boundary condition is utilized.
In this calculation, it is advantageous to use the wall sources which greatly enhance the
signal [1113]. Values of the maximum hopping parameter max , which corresponds to the
lowest valance quark mass, are also listed in Table 1. Typically, a few hundred Minimal
Residual iterations are needed to obtain the solution vector for a given source vector. On
small lattices, in particular, those with a low value of , the hopping parameter has to be
kept relatively far away from the critical kappa value in order to avoid the appearance of
exceptional configurations.
The single pseudo-scalar and vector meson, which will be referred to as pion and
rho, correlation functions at zero spatial momentum and three lowest lattice momenta,
namely, (100), (110) and (111) are constructed from the corresponding quark propagators.
Conventional effective mass functions are computed from which the single meson energy
levels with zero and non-zero momentum are extracted from the effective mass plateaus.
The fitting range of the plateau is determined automatically by requiring the minimum of
the 2 per degree of freedom of the fit. Using the anisotropic lattices, we are able to obtain
decent effective mass plateaus from the single pion and rho correlation functions. In Fig. 2,
the effective mass plateaus for the pion and rho are shown for one of our parameter set,
namely, 83 40 lattices with = 1.7. Points with error bars are the effective mass values
from the pion (lower half) and rho (upper half) correlation functions. For simplicity, only
data for one particular value of is shown. The grey shaded horizontal regions represent
the final fitted value of the energy. The starting and ending positions of these shaded regions
correspond to the fitting ranges. The height and thickness of each of these shaded region
designates the fitted value and the corresponding error, respectively. It is seen that single
meson energy values are obtained with good accuracy, especially for the mass. The fitting
quality for other parameter sets is similar. The accuracy of the mass values are so good that
we could neglect the errors of these mass values when we perform the error analysis of the
(2)
are dominating.
scattering length where the errors from the energy shifts E
366
Fig. 2. The single pseudo-scalar (lower) and vector (upper) meson effective mass plot obtained from the meson
correlation functions on 83 40 lattices with = 1.7. Only the results for one value of is shown. The bare velocity
of light is taken to be = 0.90. The open triangles, open squares, open pentagons and open hexagons are data for
the zero and the lowest three non-zero spatial lattice momenta ((000), (100), (110) and (111)), respectively. Grey
shaded horizontal bars represent the fitted values, the errors and the fitting ranges of the corresponding energy
levels, as explained in the text.
Having obtained the single meson energy levels for zero and non-zero spatial
momentum, the dispersion relation of the meson is also at our disposal. The parameter ,
also known as the bare velocity of light, that enters the fermion matrix (2) is determined
non-perturbatively using the single pion dispersion relations as described in Ref. [20]. In
Fig. 3, we show the single pion dispersion relation obtained at the optimal value of = 0.90
for = 1.7, 83 40 lattices. Points with error bars are the single pion energy levels obtained
by fitting the pion effective mass plateaus. All data which correspond to 5 different values
of are plotted. The straight lines are the linear fits to these data. The optimal value of
is such that the slope of these lines are consistent with unity. The optimal value of which
corresponds to each parameter set of the simulation is also tabulated in Table 1.
Two pion correlation functions and the ratio R(t) are constructed from products of
suitable quark propagators according to Wicks theorem. Both the Direct and the Cross
contributions are included. For the ratio R(t), we obtain good signal for all our data sets. In
Fig. 4 we show the ratio R(t) for all five values as a function of the temporal separation t,
together with the corresponding linear fit (12) for = 2.2 on 83 40 lattices. The fitting
range cannot start at a value of t that is too small, particularly in a quenched calculation,
as discussed in [25]. The starting and ending positions of the straight lines indicate the
appropriate fitting range which minimizes the 2 per degree of freedom. Here, especially
on lattices with larger sizes and better signal, we have tried to enlarge the fitting range as
367
Fig. 3. The dispersion relation for the pseudo-scalar meson obtained on 83 40 lattices with = 1.7. The bare
velocity of light is taken to be = 0.90. It is seen that the measured dispersion relations agree with the free
dispersion relation at small lattice momenta with a slope of 1.06(3).
Fig. 4. The ratio R(t) as a function of t obtained from the two pion correlation functions at zero lattice momentum
for 5 different values of the bare quark mass. The lattice size is 83 40 and the gauge coupling = 2.2. The optimal
bare velocity of light is taken to be = 0.95. The straight lines are linear fits according to Eq. (12), from which
(2)
the energy shifts E E 2m are extracted for all 5 values of .
368
Table 2
(2)
Extracted values of the single pseudo-scalar, vector meson mass and the energy shift E in the isospin I = 2
1
3
channel obtained from the simulation on 4 40 lattices, all measured in at . Also listed are the values of the
(2)
quantity F =
a0 m2
(2)
m (see Section 4 for the discussion) obtained by substituting the value of E into Eq. (4)
= 1.7, 43 40 lattices
= 2.2, 43 40 lattices
(2)
E
(104 )
0.057
0.056
0.055
0.054
0.053
0.4129(6)
0.4827(6)
0.5503(5)
0.6163(5)
0.6813(5)
0.512(2)
0.570(2)
0.629(1)
0.689(1)
0.7496(8)
99(3)
89(3)
81(3)
75(3)
69(2)
F
1.29(3)
1.42(3)
1.57(4)
1.72(4)
1.88(5)
(2)
E
(104 )
0.059
0.058
0.057
0.056
0.055
0.275(1)
0.341(1)
0.406(1)
0.4700(9)
0.5330(8)
0.334(2)
0.392(2)
0.451(1)
0.510(1)
0.570(1)
200(3)
175(3)
157(3)
143(2)
131(2)
1.04(1)
1.24(1)
1.44(1)
1.66(2)
1.88(2)
= 2.6, 43 40 lattices
= 2.4, 43 40 lattices
0.062
0.061
0.060
0.059
0.058
0.180(2)
0.245(1)
0.308(1)
0.368(1)
0.428(1)
0.216(3)
0.276(3)
0.336(2)
0.395(2)
0.453(2)
(2)
E
(104 )
E
(104 )
(2)
256(19)
218(17)
186(18)
140(10)
158(21)
0.58(3)
0.78(4)
0.97(6)
1.03(5)
1.45(12)
0.063
0.062
0.061
0.060
0.059
0.215(1)
0.2701(9)
0.3261(8)
0.3821(8)
0.4380(7)
0.246(2)
0.286(2)
0.342(2)
0.398(2)
0.454(1)
143(15)
135(13)
125(12)
116(11)
107(10)
0.45(4)
0.56(4)
0.73(5)
0.89(6)
1.06(7)
Table 3
Same as Fig. 2 but for 63 40 lattices
= 1.7, 63 40 lattices
0.0585
0.0575
0.0565
0.0555
0.0545
m
0.3029(8)
0.3803(7)
0.4472(5)
0.5159(4)
0.5827(4)
m
0.416(2)
0.477(1)
0.537(1)
0.597(1)
0.657(1)
= 2.2, 63 40 lattices
(2)
E
(104 )
43(6)
37(5)
32(4)
30(4)
28(4)
(2)
1.32(14)
1.45(15)
1.62(16)
1.82(18)
2.03(20)
0.060
0.059
0.058
0.057
0.056
0.202(1)
0.2751(7)
0.3440(8)
0.4097(7)
0.4737(6)
(2)
0.0605
0.0595
0.0585
0.0575
0.0565
m
0.205(1)
0.275(1)
0.3402(9)
0.4030(8)
0.4610(6)
m
0.267(3)
0.322(2)
0.379(1)
0.436(1)
0.494(2)
E
(104 )
0.294(2)
0.348(2)
0.404(1)
0.461(1)
0.520(1)
81(6)
66(5)
56(5)
49(5)
43(4)
1.19(7)
1.33(8)
1.51(9)
1.72(12)
1.91(13)
= 2.6, 63 40 lattices
= 2.4, 63 40 lattices
E
(104 )
66(21)
62(8)
48(10)
38(10)
31(10)
(2)
0.82(20)
1.08(10)
1.18(19)
1.25(25)
1.30(32)
0.0615
0.0605
0.0595
0.0585
0.0575
0.199(1)
0.2626(9)
0.3241(6)
0.3840(5)
0.4439(6)
E
(104 )
0.232(2) 130(10) 1.09(6)

0.291(1) 106(8) 1.38(6)
0.351(1) 91(7) 1.70(8)
0.409(2) 79(6) 2.00(10)
0.467(1) 71(6) 2.30(12)
much as possible. But for small lattices, e.g., 43 40 lattices at = 2.6, signals only last a
few time slices.
Finally, the extracted values of the single meson energy and the energy shifts are all
listed in Tables 24 for 43 40, 63 40 and 83 40 lattices, respectively.
369
Table 4
Same as Fig. 2 but for 83 40 lattices
= 1.7, 83 40 lattices
0.0585
0.0575
0.0565
0.0555
0.0545
0.3005(7)
0.3782(6)
0.4495(5)
0.5180(4)
0.5846(4)
0.421(2)
0.481(1)
0.542(1)
0.597(1)
0.657(1)
= 2.2, 83 40 lattices
(2)
E
4
(10 )
(2)
21(3)
14(2)
13(1)
12(1)
11(1)
1.57(21)
1.43(13)
1.62(14)
1.78(17)
1.98(18)
0.060
0.059
0.058
0.057
0.056
0.2026(7)
0.2758(6)
0.3432(5)
0.4084(4)
0.4721(4)
m
0.279(2)
0.338(1)
0.397(1)
0.4558(6)
0.5150(6)
E
(104 )
45(9)
37(8)
32(7)
29(6)
26(6)
F
1.43(23)
1.66(25)
1.97(29)
2.30(34)
2.65(39)
= 2.6, 83 40 lattices
= 2.4, 83 40 lattices
(2)
E
(104 )
0.0605
0.0595
0.0585
0.0575
0.0565
0.1966(7)
0.2662(6)
0.3317(5)
0.3950(5)
0.4569(4)
0.256(1)
0.3130(8)
0.3717(6)
0.4303(6)
0.4887(5)
56(6)
42(6)
34(5)
29(4)
25(5)
(2)
1.43(11)
1.59(15)
1.83(19)
2.09(20)
2.34(31)
0.0605
0.0595
0.0585
0.0575
0.0565
0.1991(5)
0.2652(5)
0.3285(4)
0.3904(3)
0.4513(3)
m
0.2394(9)
0.2988(7)
0.3579(6)
0.4167(6)
0.4753(5)
E
(104 )
64(10)
52(7)
42(9)
36(8)
32(7)
F
1.39(15)
1.73(15)
2.03(27)
2.33(33)
2.65(39)
(2)
After obtaining the energy shifts E
, these values are substituted into Lschers
formula to solve for the scattering length a0(2) for all five values. This is done for
lattices of all sizes being simulated and for all values of . The results for the quantity
(2)
F a0 m2 /m (see the next section for the reason of this choice) are also listed in Table 2
to Table 4. From these results, attempts are made to perform an extrapolation towards the
chiral, infinite volume and zero lattice spacing limit.
4. Extrapolation towards chiral, infinite volume and continuum limits

Since the valance quark mass values being studied are far from the chiral limit, we
first try to make the chiral extrapolation at finite volume and finite lattice spacing. This is
facilitated by results for the scattering length at 5 different values of . In the chiral limit,
the scattering length in the I = 2 channel is given by the current algebra result due to
Weinberg [15]:
(2)
a0 =
(2)
1 m
,
16 f2
(13)
where a0 is the scattering length a0 in the I = 2 channel and f 93 MeV is the pion
decay constant. Chiral Perturbation Theory to one-loop order gives an expression which
includes the next-to-leading order contributions [16]. Even the next-next-to-leading order
corrections have been calculated within Chiral Perturbation Theory [17]. For a summary of
the results in Chiral Perturbation Theory, see Ref. [18] and the references therein. Since the
one-loop and two-loop numerical results on the pionpion scattering length in the I = 2
channel do not differ from the current algebra value substantially, we will mainly compare
with Weinbergs value. Complication arises in the quenched approximation. In principle,
370
the quenched scattering lengths becomes divergent in the chiral limit [25]. However, these
divergent terms only become numerically important when the pion mass is close to zero.
For the parameters used in our simulation, these terms seem to be numerically small in the
I = 2 channel and we could not observe this divergence from our data.
In most of the previous lattice calculations [1113], both the mass and the decay
constant of the pion were calculated on the lattice. Then, the lattice results of m and f
were substituted into the current algebra result (13) to obtain a prediction of the scattering
length. This is to be compared with the scattering length obtained from Lschers formula.
In these studies, some discrepancies between the lattice results and the chiral results was
observed [13]. There were also discrepancies between the staggered fermion results and
the Wilson fermion results. The major disadvantage of the this procedure is the following:
first, it is difficult to make a direct chiral extrapolation towards the chiral limit; second,
the lattice results of the decay constant are usually much less accurate, both statistically
and systematically, than the mass values. In fact, most of the discrepancies are due to
inaccuracy of the decay constants, as realized already in [13]. In this paper, we propose
to use another quantity2 which has a simpler chiral behavior to compare with the current
algebra result (13). This quantity is simply F = a0(2)m2 /m , which in the chiral limit reads:
F
a0(2)m2
m
1 m2
1.3638,
16 f2
(14)
where the final numerical value is obtained by substituting in the experimental values for
m 770 MeV and f 93 MeV. Note that both m and f have finite values when
approaching the chiral limit. Therefore, the above combination will approach a finite value
in the chiral limit. On the lattice, we could extract the scattering length a0(2) from Lschers
formula. More importantly, the mass of the pion and the rho meson can be obtained with
good accuracy on the lattice. So, the factor F can be calculated on the lattice with good
precision without the lattice calculation of f . The error of the factor F obtained on the
lattice will mainly come from the error of the scattering length a0(2) , or equivalently, the
(2)
energy shift E and nowhere else. Since we have calculated the factor F for 5 different
values of valance quark mass, we could make a chiral extrapolation and extract the result
in the chiral limit. Comparisons with Weinbergs result (14) and the experiment will offer
us a cross check among different methods.
In the upper half of Fig. 5, we show the chiral extrapolation of the quantity F =
(2)
a0 m2 /m as a function of the pseudo-scalar mass squared (m2 ) for the simulation on
63 40 lattices at = 2.2. The open squares represent data points for F at various values
of m2 . The straight line is the linear chiral extrapolation and the extrapolated result is
also depicted as a solid square at m = 0. Weinbergs result and the result from chiral
perturbation theory are also shown as a filled triangle and a filled circle at m = 0,
respectively. The vector meson mass values squared are also shown as the open circles
in the upper half of this plot. The pseudo-scalar meson mass squared m2 (open squares)
are plotted in the lower half of the figure as a function of 1/(2), which linearly depends
2 A similar dimensionful, instead of dimensionless, quantity a (2) /m has been proposed in Ref. [13].
371
(2)
Fig. 5. Chiral extrapolation for the quantity F a0 m2 /m for our simulation results at = 2.2 on 63 40
lattices. In the lower half of the plot, the pseudo-scalar mass squared are plotted as open squares as function of
1/(2). The straight line represents the corresponding linear fit (15) for the data. Also shown in the lower half as
open hexagons are the results for the factor F . In the upper half of the plot, the same quantity F is plotted (open
squares) as a function of m2 . Also plotted as open circles are the data points for m2 . The straight line represents
the linear extrapolation towards the chiral limit m2 = 0, where the extrapolated result is also depicted as a solid
square. As a comparison, the corresponding experimental result [14] for this quantity is drawn as a shaded band.
Weinbergs result and the result from chiral perturbation theory are also shown as a filled triangle and a filled
circle at m = 0, respectively.
on the valance quark mass mq via:

1
1
mq = A
.
2 2c
(15)
It is seen that meson mass squared depends on the valance quark mass linearly, as expected
from chiral symmetry. It is also possible to dig out the critical value of the hopping
parameter where the pseudo-scalar mass vanishes. The data points for the factor F are
also shown in the lower half of the plot as open circles. The fitting quality for the pion, rho
and the factor F are reasonable. The quality for the chiral extrapolation of other simulation
points are similar. As is seen, the linear fit gives a reasonable modeling of the data. The
divergent contributions from quenched chiral perturbation theory seem to be numerically
small for the lattices being simulated.
After the chiral extrapolation, we now turn to study the finite volume effects of the
simulation. According to formula (4), the quantity F obtained from finite lattices differ
372
(2)
Fig. 6. Infinite volume extrapolation in scheme I for the quantity F = a0 m2 /m for our simulation results
at = 2.6, 2.4, 2.2 and 1.7. The straight lines represent the corresponding linear extrapolation in 1/L3 . The
extrapolated result is also shown, together with its error.
from its infinite volume value by corrections of the form 1/L3 . However, it was argued
in Refs. [12,25,26] that in a quenched calculation, the form of Lschers formula is
invalidated. Finite volume corrections will have a different dependence on the volume,
e.g., a correction of the form 1/L5 instead of 1/L6 , as predicted by formula (4). This
would mean that the factor F receives finite volume correction of the form 1/L2 . In our
simulation, however, we were unable to judge from our data which extrapolation is more
convincing. We therefore perform our infinite volume extrapolation in both ways, calling
them scheme I (extrapolating according to 1/L3 ) and scheme II (extrapolating according to
1/L2 ), respectively. In fact, extrapolation in these two different schemes yields compatible
results within statistical errors. The fitting quality of scheme II is somewhat, but not
overwhelmingly, better than that of scheme I. In Fig. 6 and Fig. 7, we have shown the
infinite volume extrapolation according to scheme I and scheme II, respectively, for the
simulation points at = 2.6, 2.4, 2.2 and 1.7. The extrapolated results are shown with
open squares at L = , together with the corresponding errors. The straight lines represent
the linear extrapolation in 1/L3 or 1/L2 . It is seen that, on physically small lattices, e.g.,
those with = 2.6 and = 2.4, the finite volume correction is much more significant than
larger lattices. For the lattices with = 1.7, the finite volume dependence of the result is
rather weak, showing that even on 43 40 lattices with this lattice spacing, the finite volume
correction is small. This is also reflected by the slopes of the linear fits. It is seen from the
373
Fig. 7. Same as Fig. 6 but for scheme II.
figures that the slopes of the fitting straight lines increase with , showing a larger finite
volume correction at larger values hence smaller lattices.
Finally, we can make an extrapolation towards the continuum limit by eliminating
the finite lattice spacing errors. Since we have used the tadpole improved clover Wilson
action, all physical quantities differ from their continuum counterparts by terms that are
proportional to as . The physical value of as for each value of can be found from Refs. [4,
9], which is also included in Table 1. This extrapolation is shown in Fig. 8 where the results
from the chiral and infinite volume extrapolation discussed above are indicated as data
points in the plot for all 4 values of that have been simulated. In the lower/upper half
of the plot, results in scheme I/II are shown. The straight lines show the extrapolation
towards the as = 0 limit and the extrapolated results are also shown as solid squares
together with the experimental result from Ref. [14] which is shown as the shaded band.
For comparison with chiral perturbation theory, Weinbergs result (14) and the results
from Chiral perturbation theory are also shown as the filled triangles and filled circles
at as = 0, respectively. It is seen that our lattice calculation gives a compatible result for
(2)
the quantity a0 m2 /m when compared with the experiment. The statistical errors for the
final result is still somewhat large. This is mainly due to lack of results at smaller lattice
spacings. However, the result is promising since the chiral, infinite volume and continxuum
extrapolation for the pion scattering length have not been studied systematically using
small lattices before. Another encouraging sign is that all data points at different lattice
spacing values are about the same which indicates that the O(as ) lattice effects are small,
presumably due to the tadpole improvement of the action. To summarize, we obtain from
374
(2)
Fig. 8. Continuum extrapolation for the quantity a0 m2 /m obtained from our simulation results at = 2.6, 2.4,
2.2 and 1.7. The results for both schemes I and II are shown. The straight lines represent the linear extrapolation
in as /r0 . The extrapolated results are also shown, together with the experimental result from Ref. [14] indicated
by the shaded band. For comparison, Weinbergs result (14) and the results from Chiral perturbation theory are
also shown as the points at as = 0, respectively.
the linear extrapolation the following result for the quantity F = a0(2)m2 /m :
a0(2)m2
m
= 1.05(23) for scheme I,
a0(2)m2
= 1.41(28) for scheme II.

(16)
m
If we substitute in the mass of the mesons from the experiment: m 770 MeV and
(2)
m 139 MeV, we obtain the quantity a0 m :
(2)
a0 m = 0.0342(75) for scheme I,

a0(2)m = 0.0459(91) for scheme II.
(17)
Weinbergs current algebra prediction (13) yields a value of a0(2)m = 0.046. This
quantity has been calculated in Chiral Perturbation Theory to one-loop order with the
(2)
result: a0 m = 0.042 [16] and recently to two-loop order [17,18]. The final result from
Chiral Perturbation Theory gives: a0(2)m = 0.0444(10), where the error comes from
theoretical uncertainties. On the experimental side, a new result from E865 Collaboration
375
(2)
claims a0 m = 0.036(9). It is encouraging to find out that our lattice results in both
schemes are compatible with the experiment. Our result in scheme II also agrees with the
Chiral Perturbation Theory results very well while our result in scheme I is barely within
one standard deviation of the chiral results.
5. Conclusions
In this paper, we have calculated pionpion scattering lengths in isospin I = 2 channel
using quenched lattice QCD. It is shown that such a calculation is feasible using relatively
small, coarse and anisotropic lattices with limited computer resources like several personal
computers and workstations. The calculation is done using the tadpole improved clover
Wilson action on anisotropic lattices. Simulations are performed on lattices with various
sizes, ranging from 0.8 fm to about 3 fm and with different value of lattice spacing. Quark
propagators are measured with 5 different valance quark mass values. These enable us to
explore the finite volume errors and the finite lattice spacing errors in a more systematic
fashion. The infinite volume extrapolation is explored in two different schemes which
yields compatible final results. The lattice result for the scattering length is extrapolated
towards the chiral and continuum limit where a result consistent with the experiment and
the Chiral Perturbation Theory is found. We believe that, using the method described in
this exploratory study, more reliable and accurate results on pion scattering lengths could
be obtained with simulations on larger lattices and computers.
Finally, our method for calculating the pionpion scattering length discussed in this
paper can easily be generalized to calculate the scattering lengths of other hadrons, or
in other channels, e.g., I = 0 channel, where extra care has to be taken due to enhanced
terms coming from quenched chiral loops. The method can also be applied to calculate the
scattering phase shift at non-zero spatial lattice momenta, where presumably lattices larger
than the ones utilized in this exploratory study is needed.
Acknowledgements
This work is supported by the National Natural Science Foundation of China (NSFC)
under grant No. 90103006 and Pandeng fund. C. Liu would like to thank Prof. H.Q. Zheng
for helpful discussions. The authors would like to thank Prof. C. Bernard for bringing our
attention to the issue of enhanced quenched chiral loop contributions.
References
[1]
[2]
[3]
[4]
[5]
[6]
G.P. Lepage, P.B. Mackenzie, Phys. Rev. D 48 (1993) 2250.

M. Alford, W. Dimm, G.P. Lepage, G. Hockney, P.B. Mackenzie, Phys. Lett. B 361 (1995) 87.
C. Morningstar, M. Peardon, Phys. Rev. D 56 (1997) 4043.
C. Morningstar, M. Peardon, Phys. Rev. D 60 (1999) 034509.
C. Liu, Chinese Phys. Lett. 18 (2001) 187.
C. Liu, Commun. Theor. Phys. 35 (2001) 288.
376
[7] C. Liu, in: Proceedings of International Workshop on Nonperturbative Methods and Lattice QCD, World
Scientific, Singapore, 2001, p. 57.
[8] C. Liu, J.P. Ma, in: Proceedings of International Workshop on Nonperturbative Methods and Lattice QCD,
World Scientific, Singapore, 2001, p. 65.
[9] C. Liu, Nucl. Phys. (Proc. Suppl.) B 94 (2001) 255.
[10] C. Liu, J. Zhang, Y. Chen, J.P. Ma, hep-lat/0109010, Phys. Lett. B submitted, 2001.
[11] R. Gupta, A. Patel, S. Sharpe, Phys. Rev. D 48 (1993) 388.
[12] M. Fukugita, Y. Kuramashi, H. Mino, M. Okawa, A. Ukawa, Phys. Rev. D 52 (1995) 3003.
[13] S. Aoki et al., Nucl. Phys. (Proc. Suppl.) B 83 (2000) 241.
[14] P. Trul, hep-ex/0012012.
[15] S. Weinberg, Phys. Rev. Lett. 17 (1966) 616.
[16] J. Gasser, H. Leutwyler, Phys. Lett. B 125 (1983) 325.
[17] J. Bijnens et al., Nucl. Phys. B 508 (1997) 263.
[18] G. Colangelo, J. Gasser, H. Leutwyler, Nucl. Phys. B 603 (2001) 125.
[19] T.R. Klassen, Nucl. Phys. (Proc. Suppl.) B 73 (1999) 918.
[20] J. Zhang, C. Liu, hep-lat/0107005, Mod. Phys. Lett. A, accepted, 2001.
[21] A. Frommer, S. Gsken, T. Lippert, B. Nckel, K. Schilling, Int. J. Mod. Phys. C 6 (1995) 627.
[22] U. Glaessner, S. Guesken, T. Lippert, G. Ritzenhoefer, K. Schilling, A. Frommer, hep-lat/9605008.
[23] B. Jegerlehner, hep-lat/9612014.
[24] M. Lscher, Commun. Math. Phys. 105 (1986) 153.
[25] C.W. Bernard, M.F.L. Golterman, Phys. Rev. D 53 (1996) 476.
[26] S.R. Sharpe, R. Gupta, G.W. Kilcup, Nucl. Phys. B 383 (1992) 309.

Measurement of the lifetime difference of Bd

mesons: possible and worthwhile?
A.S. Dighe a , T. Hurth b , C.S. Kim c , T. Yoshikawa d
a Max-Planck-Institut fr Physik, Fhringer Ring 6, D-80805 Mnchen, Germany
b Theory Division, CERN, CH-1211 Geneva 23, Switzerland
c Department of Physics and IPAP, Yonsei University, Seoul 120-749, South Korea
d Department of Physics and Astronomy, University of North Carolina, Chapel Hill, NC 27599-3255, USA
Abstract
We estimate the decay width difference d /d in the Bd system including 1/mb contributions
and part of the next-to-leading order QCD corrections, and find it to be around 0.3%. We explicitly
show that the time measurements of an untagged Bd decaying to a single final state isotropically
can only be sensitive to quadratic terms in d /d , and hence the use of at least two different final
states is desired. We discuss such pairs of candidate decay channels for the final states and explore the
feasibility of a d /d measurement through them. The measurement of d would be essential
for an accurate measurement of sin(2) at the LHC. The nonzero width difference can also be used
d mixing
to identify new physics effects and to resolve a twofold discrete ambiguity in the Bd B
phase. We also derive an upper bound on the value of d /d in the presence of new physics, and
point out some differences in the phenomenology of width differences in the Bs and Bd systems.
2002 Published by Elsevier Science B.V.
1. Introduction
Within the Standard Model (SM), the difference in the decay widths of Bd mesons is
CKM-suppressed with respect to that in the Bs system. A rough estimate leads to
d
s 2
0.5%,
d
s
(1)
where = 0.225 is the sine of Cabibbo angle, and we have taken s /s 15% [13].
Here d(s) = (L + H )/2 is the average decay width of the light and heavy Bd(s) mesons
(BL and BH , respectively). We denote these decay widths by L , H , respectively, and
E-mail addresses: amol.dighe@cern.ch (A.S. Dighe), tobias.hurth@cern.ch (T. Hurth),
cskim@mail.yonsei.ac.kr (C.S. Kim), tadashi@physics.unc.edu (T. Yoshikawa).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 5 5 - 1
378
A.S. Dighe et al. / Nuclear Physics B 624 (2002) 377404
define d(s) L H . No experimental measurement of d is currently available.

Moreover, no motivation for its measurement (other than just measuring another number
to check against the SM prediction) has been discussed, and hence the study of the lifetime
difference between Bd mesons has hitherto been neglected as compared to that in the
Bs system. The phenomenology of the lifetime difference between Bs mesons has been
explored in detail in [4,5].
With the possibility of experiments with high time resolution and high statistics, it is
worthwhile to have a look at this quantity and make a realistic estimate of the possibility
of its measurement. At LHCb for example, the proper time resolution is expected to be as
good as 0.03 ps. This indeed is a very small fraction of the Bd lifetime (Bd 1.5 ps
[6]), so the time resolution is not a limiting factor in the accuracy of the measurement, the
statistical error plays the dominant role. Taking into account the estimated number of Bd
producedfor example, the number of reconstructed Bd J /KS events at the LHC
is expected to be 5 105 (Table 3 of [7])the measurement of the lifetime difference
does not look too hard at first glance. Naively, one may infer that if the number of relevant
events with the proper time of decay measured with
the precision is N , then the value
of d /d is measured with an accuracy of 1/ N . With a sufficiently large number of
events N , it should be possible to reach the accuracy of 0.5% or better.
The measurement of d /d is in reality harder than what the above naive expectation
may suggest, since most of the quantities that involve the lifetime difference are quadratic
in the small quantity d /d . In fact, as we shall explicitly show in this paper, the time
measurements in the decays of an untagged Bd to a single final state are sensitive only to
(d /d )2 . This implies that in order to discern two different lifetimes, the measurements
need to have an accuracy of (d /d )2 2.5 105 , which is beyond the reach of the
currently planned experiments.
However, the combination of lifetimes measured in two different untagged decay
channels may be sensitive to linear terms in d /d . We explore three pairs of such
untagged measurements in this paper: (i) lifetime measurements through decays to selftagging (e.g., semileptonic) final states and to CP eigenstates, (ii) CP-even and CP-odd
components in the decay mode Bd J /K (Ks 0 ), and (iii) time-dependent untagged
asymmetry between Bd J /KS and Bd J /KL .
The conventional gold-plated decays for measurement, J /KS and J /KL ,
neglect the lifetime difference while determining sin(2). For an accurate determination
of , the systematic errors due to d /d need to be taken into account. Moreover, there is
the possibility that the measurement of the lifetime difference leads to a clear signal for new
physics. Furthermore, if the lifetime difference is neglected, the ambiguity (/2 )
remains unresolved. Observables that are sensitive to the lifetime difference may resolve
this discrete ambiguity under certain conditions.
The observables mentioned above can also give an independent measurement of cos(2)
in principle. In order to be able to do this, however, the theoretical uncertainties on d
need to be minimized. Therefore, we start by presenting in Section 2 a detailed calculation
of d , including 1/mb contributions and part of the next-to-leading order (NLO) QCD
corrections. The NLO precision in the width difference d is also essential for obtaining
379
a proper matching of the Wilson coefficients to the matrix elements of local operators from
the lattice gauge theory.
The rest of the paper is organized as follows. In Section 3 we explicitly demonstrate the
quadratic dependence on d /d of quantities measurable through untagged B decays to a
single final state. We explore the combinations of decay modes that can measure quantities
linear in d /d . We calculate the corrections due to d to the measurement of sin(2)
through Bd J /KS , and also indicate the possibility of the d measurement through
tagged decays to CP eigenstates. In Section 4, we point out important differences in the
upper bounds on s and d in the presence of new physics, and elaborate on the
possibility of detecting new physics and resolving discrete ambiguities in the mixing
phases through them. We summarize our findings in Section 5.
2. Estimation of d
2.1. Basic definitions
We briefly recall the basic definitions: in the WignerWeisskopf approximation the
oscillation and the decay of a general linear combination of the neutral flavour eigenstates
d , a|Bd + b|B
d , is described by the time-dependent Schrdinger equation
Bd and B

d a
i
a
i
(2)
= M
.
b
dt b
2
Here M and are 2 2 Hermitean matrices. CPT invariance leads to the conditions M11 =
M22 and 11 = 22 . Exact CP invariance would imply M21 = M12 and 21 = 12 (a phase
d , CP|B
d = |Bd is made). Independent of the choice
choice, namely CP|Bd = |B
) = 0.
of the unphysical phases, CP invariance (in mixing) would imply Im(M21
21
The mass eigenstates, the light BL and the heavy BH , are given by
d
|BL,H = p|Bd q|B
(3)
with the normalization condition |q|2 + |p|2 = 1. Only the magnitude |q/p| is measurable,
the phase of this quantity is unphysical and can be fixed arbitrarily by convention.
The mass difference and the width difference between the physical states are defined by
m = MH ML ,
= L H ,
(4)
such that m > 0, d > 0 in the SM. The real and imaginary parts of the eigenvalue
equations are the following:

1
(m)2 ( )2 = 4|M21 |2 |21 |2 ,
4

m = 4 Re M21
21 .
(5)
(6)
With the help of the CP-violating parameter
)
2 Im(M21
21
= |p|2 |q|2 = BL |BH .
2
(m) + |21 |2
(7)
380
and = .
Fig. 1. Schematic representation of Feynman diagrams for M12 = M21
12
21
The effect of CP violation due to mixing on the mass difference m and on the lifetime
difference may be explicitly shown:
4|M21 |2 2 |21 |2
(8)
,
1 + 2
4|21|2 16 2|M21 |2
( )2 =
(9)
.
1 + 2
In the limit of exact CP invariance ( = 0) the mass eigenstates coincide with the CP
eigenstates, CP|BH = |BH and CP|BL = +|BL and the mass difference and width
difference are given by m = 2|M21 |, = 2|21|. However, even with a nonzero ,
taking into account that is constrained by the upper bound || |21|/(2|M21 |) and
21 /M21 = O(m2b /m2t ), we can write

4
4
)
mb
2 Re(M21
mb
21
m = 2|M21| 1 + O
,
=
. (10)
1+O
4
|M21 |
mt
m4t
(m)2 =
We shall neglect the terms of O(m4b /m4t ) 106 in our calculations.

2.2. Method of calculation
In the following we consider the two off-diagonal elements M21 and 21 , which
correspond respectively to the dispersive and the absorptive part of the transition amplitude
d . We follow the method of [2,3] which was used there in the Bs B
s system
from Bd to B
(see also [8,9]).
Within the SM the well-known box diagram is the starting point of the calculations.
M21 is related to the real part of this diagram (see Fig. 1). The important QCD corrections
are most easily implemented with the help of the standard operator product expansion.
Because of the dominance of the top quark contribution, M21 can be described by a local
B = 2 Hamiltonian below the mW scale:

2
mb
1
B=2
,
Bd |Heff |Bd 1 + O
M21 =
(11)
2MBd
m2W
2
G2F
Vt b Vt d C Q (mt , mW , )Q() + H.c.,
2
16
Q = (bi di )V A (bj dj )V A .
B=2
Heff
=
CQ
(12)
(13)
contains the short-distance physics. It is known up to NLO

The Wilson coefficient
d |Q( mb )|Bd will be discussed below.
precision [10]. The hadronic matrix element B
381
In the standard model, 21 is related to the imaginary part of the box diagram. Via
the optical theorem it is fixed by the real intermediate states. Therefore, only the box
s
diagrams with internal c and u quarks contribute (see Fig. 1). In contrast to the Bs B
case where the intermediate cc contribution is the dominating one, because of its CKM
V )2 , over the uu,
factor (Vcb
the cu and the uc contribution (see Section 4.1), in the Bd
cs

Bd all four contributions have to be taken into account. In the effective theory where we
integrate out the W boson, 21 is given by:

1
B=1
B=1
21 =
(14)
Bd | Im i d 4 x T Heff
(x)Heff
(0)|Bd ,
2MBd
where

G
F
B=1
Heff
= Vub
Vud
Ci Quu
Ci Qcu
Ci Quc
i + Vcb Vud
i + Vub Vcd
i
2
i=1,2
i=1,2
i=1,2
cc
+ Vcb Vcd
Ci Qi Vt b Vt d
Ci Qi .
(15)
i=1,2
i=3
The operators are (i, j denote color indices)

qq
qq
Q1 = (bi qj )V A (qj di )V A ,
Q2 = (bi qi )V A (qj dj )V A ,
(16)
Q3 = (bi di )V A (qj qj )V A ,
Q4 = (bi dj )V A (qj qi )V A ,
(17)
Q5 = (bi di )V A (qj qj )V +A ,
Q6 = (bi dj )V A (qj qi )V +A .
(18)
The penguin operators Q3 Q6 have small Wilson coefficients and are therefore suppressed
with respect to the four-quark operatorswhich all have the same two Wilson coefficients C1 and C2 . In the leading logarithmic approximation we have:

C = C2 C1 ,
C () =
(MW )
()
(0) /(20 )
C (MW ),
C (MW ) = 1,
(19)
where 0 = (11N 2f )/3 = 23/3 and (0) = 6(1 N)/N . The coefficients to NLO
precision can be found in [11].
Because there is another short-distance scale, the bottom quark mass, the operator
product of two B = 1 operators can be expanded in inverse powers of the bottom quark
mass scale in terms of local B = 2 operators:

1
B=1
B=1
Bd | Im i d 4 x T Heff
(x)Heff
(0)|Bd
21 =
(20)
2MBd
En
B=2 (0)|Bd .
=
(21)
n Bd |On
m
b
n
These matching equations fix the values of the B = 2 Wilson coefficients En . The
corresponding four quark operators On are the following: the operators Q and QS ,
Q = (bi di )V A (bj dj )V A ,
(22)
QS = (bi di )SP (bj dj )SP ,
(23)
382
represent the leading order contributions. Their matrix elements are given in terms of the
bag parameters, B and BS , the mass of the Bd meson MBd , and its decay constant fBd :
N +1
B,
N
MB2 d
2
d |Q|Bd = fB2 MB2 2

B
d
d
d |QS |Bd = fB2 MB

B
d
d
(m
b +m
d )2
(24)
2N 1
BS .
N
(25)
In the naive factorization approximation, B and BS are fixed by B = BS = 1. Reliable

lattice calculations for B and BS are already available [12]. We note that to NLO precision
one has to distinguish between the pole mass mb and the running quantity

2
s
4
ln 2 +
m
b () = mb 1
(26)
mb 3
using the MS scheme.
The 1/mb corrections are given by the operators
md
(27)
(bi di )SP (bj dj )S+P ,
mb

1
R2 = 2 b i D (1 5 )D di bj (1 5 )dj ,
(28)
mb

1
R3 = 2 b i D (1 5 )D di bj (1 5 )dj ,
(29)
mb

1
R4 =
(30)
bi (1 5 )iD di b j (1 5 )dj ,
mb
1
S ,
R0 = QS + Q + Q
(31)
2
S has the interchanged color structure as compared to QS . There are also colorwhere Q
i and Q
corresponding to Ri and Q. We note that these 1/mb
interchanged operators R
R1 =
operators are not independent, the relations between them are in fact the equations of
motion.
s system were estimated in [2]
The matrix elements of these operators within the Bs B
using naive factorization, which means that all the corresponding bag factors were set to 1.
d system the analogous results are:
For the Bd B

M2
d |R0 |Bd = fB2 MB2 N + 1 1 Bd ,
B
(32)
d
d
N
m2b
d |R1 |Bd = fB2 MB2 md 2N + 1 = 0,
B
(33)
d
d
mb N
d |R
1 |Bd = fB2 MB2 md N + 2 = 0,
B
(34)
d
d
mb N
2

MBd
1N
2
2
d |R

2 |Bd ,
= B
Bd |R2 |Bd = fBd MBd
(35)
1
2
N
mb
d |R3 |Bd = fB2 MB2

B
d
d

MB2 d
m2b
MB2 d

1
2N + 1
,
2N
(36)
N +2
,
1
2N
m2b

2
MBd
2
2

1 ,
m2b
2

MBd
1
2
2

.
1
2
N
mb
d |R
3 |Bd = fB2 MB2
B
d
d
383
(37)
(38)
(39)
We neglect terms proportional to md /mb ; the other terms proportional to (MB2 d /m2b ) 1
are of order QCD /mb .
In the matrix elements Ri (Eqs. (32)(39)), we use the pole mass mb . There is a
subtlety involved here: as discussed in [3], there are terms of order s and of leading
power in mb in the matrix element of R0 to NLO precision. In view of the relation (31),
it is not surprising that there are such terms. In the schemewhich was used in [3] and
which is also used herethese terms are subtracted in the matrix element R0 while
taking into account the leading NLO contribution. Then the R0 matrix element is still of a
subleading nature. The specific subtraction scheme for the factorized matrix elements Ri
corresponds to the use of the pole mass in Eqs. (32)(39). Of course, this specific choice
for the matrix elements has to be taken into account if the NLO results are combined with
a lattice calculation of the Ri .
There is an additional remark in order. We estimate 21 by the cut of the partonic
diagrams. The underlying assumption of local quarkhadron duality can be verified in the
s system, in the simultaneous limit of large N and of small velocity [1], therefore one
Bs B
d system the small velocity argument fails
expects no large duality violations. In the Bd B
since the uu,
uc and cu intermediate states contribute significantly, and the larger number
of light intermediate states leads to a larger energy release. We follow Ref. [2] and make the
d system are also not larger than 10%.
assumption that the duality violations in the Bd B
In order to test this assumption one should include all corrections up to that accuracy.
2.3. Analytical results
In this section, we present an analytic expression for 21 including 1/mb , penguin and
part of the NLO corrections. If one takes into account the error inherent in the naive
factorization approach to the matrix elements of the subleading operators R, it seems to be
leading
.
a reasonable approximation to keep at least all terms up to an accuracy of 102 21
We keep also higher order terms in order to check the accuracy of our approximation.
In the effective theory of the B = 2 transitions the matrix elements of the 1/mb
operators (R) are formally suppressed by a factor of the order of 0.1 with respect
to those of the leading operators Q and QS . The natural variable z = m2c /m2b also
formally introduces a suppression factor of approximately 0.1. The NLO contribution has
formally an extra suppression factor (s /4) of order 0.01. Within the effective theory
384
of the B = 1 Hamiltonian, the combination K = C peng C dom and K = C peng C peng are
suppressed by almost a factor 0.01 and 104 , respectively, with respect to the combination
K = C dom C dom , where C peng denotes the Wilson coefficients of the penguin operators
qq
Q3 , . . . , Q6 and C dom that of the dominating operators Q1(2) . The contribution due to
K therefore can be safely neglected. Schematically our analytical result for 21 has the
following form:
21 = KQ
(40)

2 3
+ KR O(1) + O(z) + O z + O z
(41)

2

+ K Q O(1) + O(z) + O z
(42)

2

+ K R O(1) + O z
(43)
s
KQO(1),
+
(44)
4
where Q represents the leading order operators Q and QS . The terms inside the curly
brackets are the ones that we calculate only to estimate the errors. In the presentation of
the results the following combinations of the Wilson coefficients are used:
K1 = 3C12 + 2C1 C2 ,
K1
K3
K2 = C22 ,
= 2(3C1 C3 + C1 C4 + C2 C3 ),
K3 = C12 ,
K2
= 2(3C1 C5 + C1 C6 + C2 C5 + C2 C6 ),
= 2C2 C4 ,
K4 = C1 C2 ,
(45)
(46)
(47)
and the common factor of [G2F m2b /(24MBd )] is implicit in the following Eqs. (48),
(49), (51), (53).
qq
qq
In the leading log approximation we calculate the Q1 and the Q2 contributions
uc,
cu and uu intermediate states,
to 21 . By extracting the absorptive parts of the cc,
we can find the off-diagonal element. For this leading contribution (40), after replacing
V
Vub
ud by the unitarity relation, we get to all orders in z:

2

1
leading
= Vtb Vt d
21
K1 + K2 Q + (K1 K2 )QS
2

+ Vcb Vcd Vt b Vt d 3z(K1 + K2 ) 3z2 K2 z3 (K1 K2 ) Q

+ 6z2 (K1 K2 ) 4z3 (K1 K2 ) QS

2
1
+ Vcb
Vcd
1 4z K1 + K2 z(K1 + 2K2 ) Q
2
+ 1 4z (1 + 2z)(K1 K2 )Qs

1
+ (1 z)2 (2K1 + K2 ) + z(K2 K1 ) + K1 + K2 Q
2

+ (1 z)2 (2 + 4z)(K2 K1 ) (K2 K1 ) QS .
(48)
This result, (48), confirms the findings of [9].
In the Eqs. (49)(52) we give the penguin and 1/mb contributions. The coefficients of
(Vtb Vt d )2 in these results can be checked with the z 0 limit of the results in the literature
385
for the Bs system. In this sense our results are consistent with the findings in the Bs system
given in [2].
qq
qq
The 1/mb corrections to the operators Q1 and Q2 give [see the term (41)]

2

1
1/mb
21 = Vt b Vt d 2 K1 K2 R2 2K1 R1 + 2K2 R4
2

+ Vcb Vcd Vt b Vt d 12z2K1 R1 2R3

+ 6z2 K2 R2 + 4R3 + 2R4

+ KRO z3 .
(49)
The term in curly brackets in (49) can be written as

{ } = Vcb
Vcd Vtb Vt d 4z3 K1 2R1 R2 6R3

4z3 K2 R2 + 6R3 + 2R4

2

Vcd 12z3K1 2R1 R2 6R3
+ Vcb

12z3K2 R2 + 6R3 + 2R4 .
The penguin contributions [terms (42), (43)] are

2

1
peng
21 = Vtb Vt d
K1 + K2 Q + (K1 K2 )QS
2

+ 2R2 2R1 K1 + R2 + 2R4 K2

+ Vcb
Vcd Vtb Vt d 3zK1 + 3zK2 3zK3 Q

+ K QO z2 + K RO(z) ,
where the terms in curly brackets (and the lower order ones) may be written as

{ } = Vcb
Vcd Vtb Vt d

3z2 K2 + 6z2 K3 Q + 6z2 K1 6z2 K2 QS

12z2 R1 24z2R3 K1

+ 6z2 R2 + 24z2 R3 + 12z2R4 K2 + 4z2 R2 K3 .
(50)
(51)
(52)
NLO = ( /(4))KQ [term (44)] is found from [3] by

The NLO QCD correction 21
s
taking the limit z 0 of their results: 1

2

157
(m
)
248
109
s
b
2
NLO
Vt b Vt d
K3
K4
+
21 =
K2 Q
4
6
9
3
18

2
20
8
1
(34K4 + 10K2) ln
Q
+ 10K3 + K4 + K2 ln
3
3
mb
mb
1 We add only the leading contribution of the NLO QCD corrections for the term (V V )2 . The leading terms
tb td
V )(V V ) and (V V )2 cancel out through the GIM mechanism.
of the contributions for the terms (Vcb
cd
td td
cb cd
386
2

128
40
248
8
K3 +
K4
K2 QS
3
9
3
9

2
64
32
1
+ 32K3 K4 + K2 ln
(16K4 + 16K2 ) ln
QS
3
3
mb
mb

2
2
1
1 C8
(53)
+ ln
+
K2 Q 8QS .
27 9
mb
3 C2
The explicit 1 and 2 dependence in (53) cancels against the dependence of the Wilson
B=1 (15) and the dependence of the matrix elements
coefficients of the hamiltonian Heff
of the B = 2 operators at the order in s we take into account. For a proper matching with
lattice evaluations of these matrix elements it is important to note that the results in (53)
are based on the NDR scheme, with the choice of 5 and the evanescent operators as given
in Eqs. (13)(15) of [3].
The net 21 is
leading
21 = 21
1/mb
+ 21
peng
+ 21
NLO
+ 21
,
with the implicit multiplicative factor of
(54)
[G2F m2b /(24MBd )].
2.4. Numerical results

Let us now calculate the numerical value of d . From Eq. (10), d can be
approximately written as
21
21
= m Re
,
M21
M21
[see Eq. (11)] is given by
d 2|M21 | Re
where M21
M21 =
2
2
G2F MW
B
Vt b Vt d S0 (xt )Q.
2
(4) (2MBd )
(55)
(56)
2 , is the QCD correction factor and S is the InamiLim function:

2t /MW
Here xt = m
B
0

3
9
3
1
3
x
S0 (x) = x
(57)
+
log x.
4 4(1 x) 2(1 x)2

2 1x
Using the results obtained in the previous section, we can write down the width difference
(normalized to the average width) in the form

2
m
5 BS MBd t t 3 1 t t

G
=
K Gt t +
GS +
Bd
d
8 Bm
8 B 1/m
2b

2
Vcb Vcd
5 BS MBd ct 3 1 ct
ct
G +
+ Re
GS +
G
Vtb Vt d
8 Bm
8 B 1/m
2b

2
Vcb Vcd 2
5 BS MBd cc
cc
(58)
GS
+ Re
G +
.
Vtb Vt d
8 Bm
2b
The superscripts {tt, ct, cc} correspond to the terms in the expression for d (54)
V )(V V ), (V V )2 }, respectively.
that involve the CKM factors {(Vtb Vt d )2 , (Vcb
cd
tb td
cb cd
387
The subscript S denotes the contribution from the operator QS , and the subscript
1/m denotes the terms that give the 1/mb corrections. The normalizing factor K
2 S (x )) and the value of (m/ ) may be taken from experiments:
(4m2b )/(3MW
B 0 t
d
xd (m/ )d = 0.73 0.03 [6]. The form of Eq. (58) can bring out important features of
the dependence of d on various parameters, as we shall see below. This representation
also has the advantage that within the leading term the CKM dependence cancels out and
the value of xd is available from experiments.
A remark about the penguin contributions is in order. We only include the interference of
the penguin operators C3 , . . . , C6 with the leading operators C1 and C2 . At the NLO, this
approximation can be made consistent (in the sense of scheme independence) by counting
the Wilson coefficients C3 , . . . , C6 as of order s . These Wilson coefficients are modified at
NLO through the mixing of C1 and C2 into C3 , . . . , C6 . For C1 and C2 we use the complete
NLO values. Since the contribution due to C8 starts only at the NLO level, we only have to
use the LO value for that Wilson coefficient. We stress that if one uses the consistent NLO
approximation just described, the corresponding LO approximation includes no penguin
contributions and uses the LO values for C1 and C2 .
The choice of the b-quark mass at LO is ambiguous (it may be taken to be the pole mass
or the running mass at one or two loop level); we take it to be the running mass in the MS
scheme to leading order in mb .
We use the following values of parameters to estimate d :
MBd = 5.28 GeV,
m
b (mb ) = 4.4 GeV,
mb = 4.8 GeV,
mc = 1.4 GeV,
m
t (mb ) = 167 GeV.
(59)
To the NLO precision [we use here the NDR scheme to get B (mb ) = 0.846 and include
the NLO Wilson coefficients [14] and the corrections computed in Eqs. (49), (51)], we get
(in units of 103 )

1
BS

1.38
= 0.73 + 4.93
Bd
B
B

1
BS
1
BS
cos
+ 0.02 + 0.005 0.021
0.003
1.07 + 0.29
Rt
B
B
B
B

cos 2
1
BS
+
(60)
+ 0.01
.
0.02 0.06
B
B
Rt2
Let us perform a conservative estimate of the error on the value of d /d that we
obtain here. The errors arise from the uncertainties in the values of the CKM parameters,
the bag parameters and the mass of the b quark. There are also errors from the scale
dependence, the breaking of the naive factorization approximation, and the neglected
higher-order terms in the z expansion.
In the SM, we have
cos
= 1.03 0.08,
Rt
cos 2
= 0.87 0.15,
Rt2
(61)
388
Table 1
The numerical value of d / for different values of the bag parameters. The column LO (C) shows
the leading (next-to-leading) order result with factorization, i.e., B = BS = 1. The values of the bag
factors in column A are taken from [3] and the ones in column B from the (preliminary) results in an
unquenched (Nf = 2) lattice calculation by the JLQCD Collaboration [12]
B
BS
d /d
LO
1.0
1.0
6.3 103
0.90
0.75
1.9 103
0.83
0.84
2.6 103
1.0
1.0
2.8 103
where we have taken the values of the CKM parameters from the global fit [15]. The
leading term on the first line in (60) is independent of the CKM elements. The quantity
cos /Rt is known to an accuracy of about 10% and appears in (60) with a coefficient 0.3
relative to the leading term. The quantity cos 2/Rt2 , although known to only about 20%,
appears with a very small coefficient ( 102 ) as compared to the leading term in (60).
The net error due to the uncertainty in the CKM elements is thus approximately only 3%,
i.e., about 0.1 103 .
We estimate the effect of the uncertainties in the bag factors by computing (60) with
three sets of values of the bag parameters. The numerical results are as shown in Table 1.
From the table, and using the uncertainties on the values of the bag parameters as given in
[16], we conservatively estimate the corresponding uncertainty in the value of d /d
due to bag factors to be approximately 0.5 103 . The uncertainty in the value of
m
b = 4.4 0.2 also leads to an error of 0.5 103 . The uncertainty due to the scale
3 (where is varied between 2m and m /2
1 dependence is estimated to be +0.5
1
b
b
1.2 10
following the common convention). The error due to the input value of xd is 0.1 103 .
The errors due to the breaking of the naive factorization assumption (which was made
in the calculation of the matrix elements of the 1/mb operators) are hard to quantify.
Assuming an error of 30% in the R matrix elements (as in [16]), we estimate the error
due to this source to be 0.3 103 .
Table 1 also gives the LO value of d /d in the factorization approximation. We
observe that the NLO corrections significantly decrease the value of d /d as computed
at LO, and that there effectively is no real (s /4) suppression of the NLO contribution, as
one naively expects. Therefore, higher-order terms in the z expansion become important.
While we estimate the error due the z expansion in the 1/mb and the penguin contributions
from the terms in curly brackets in (60) to be less than 0.05 103 , the issue of higherorder terms in the NLO contribution (53) is more subtle. We can write 21 in the form (see
Section 4.1 for details)
21 (Bd ) = N

2

Vcb
Vcd f (z, z) 2f (z, 0) + f (0, 0)

2
+ 2 Vcb
Vcd Vtb Vt d f (0, 0) f (z, 0) + Vtb Vt d f (0, 0) .
(62)
389
Here in (53) we have included the complete NLO coefficient of (Vtb Vt d )2 , which includes
all the terms of the order z0 in the hadronic matrix elements f . However, in order to
calculate the corrections due to higher-order terms in z, a complete NLO calculation is
necessary. The contribution of these terms to [(f (0, 0)f (z, z))/f (0, 0)] can be computed
to NLO precision using [3] to be

f (0, 0) f (z, z)
(63)
0.2.
f (0, 0)
NLO
If we estimate the contribution to [(f (0, 0) f (z, 0))/f (0, 0)]NLO to be also of the same
order, this results in the estimation of the net error in d /d due to these terms to be
0.8 103 .
Our net estimation for the width difference is

d
3
(64)
= 2.6+1.2
1.6 10 .
d Bd
We have taken the central value to be the one obtained from the latest preliminary
(unquenched) results from lattice calculations [12]. The dominating theoretical errors are
NLO
that correspond to the nontrivial z dependence
the scale dependence and the terms in 21
of the function f (z, 0) in (62). To take care of the latter, a complete NLO calculation is
definitely desirable.
In the above calculations, we have used the expansion of 21 in the form

2

Vcb Vcd
Vcb Vcd 2
+ Fcc
21 (Bd ) = N Vt b Vt d f (0, 0) 1 + Fct
Vtb Vt d
Vtb Vt d

2
N Vt b Vt d f (0, 0)(1 + fc ),
(65)
where
Fct
2[f (0, 0) f (z, 0)]

f (0, 0)
and Fcc
f (0, 0) 2f (z, 0) + f (z, z)

.
f (0, 0)
Following the suggestion in [13], 2 we have also performed the expansion (and the error
analysis) in the form

2
Vub Vud
Vub Vud 2
21 (Bd ) = N Vt b Vt d f (z, z) 1 + Fut
+ Fuu
Vtb Vt d
Vtb Vt d

2
N Vtb Vt d f (z, z)(1 + fu ),
(66)
where
2[f (z, z) f (z, 0)]
f (0, 0) 2f (z, 0) + f (z, z)
and Fuu
.
f (z, z)
f (z, z)
In this expansion the unknown NLO terms are suppressed by small CKM factors. This
gives the width difference as

d
(67)
103 ,
= 3.0+0.9
1.4
d Bd
Fut
2 We thank Uli Nierste for bringing this to our attention.
390
where, as before, we use the latest preliminary (unquenched) results from lattice
calculations [12] for the bag parameters. The results of both (64) and (67) are consistent.
The errors in (67) are smaller, but it should be noted that in both calculations the errors due
to NLO terms are based on the assumption on the function f (z, 0) which we stated above.
3. Measurement of d /d
It is not possible to find a final state to which the decay of Bd involves only one of
d mixing phase (2) is large, the
the decay widths L and H . Indeed, since the Bd B
CP eigenstates are appreciably different from the lifetime eigenstates. The decay rate to a
CP eigenstate, therefore, involves both the lifetimes. The semileptonic decays are flavortagging, and hence also involve both the lifetimes in equal proportion.
We start by concentrating on the untagged measurements, i.e., the measurements in
which the (mt) oscillations are cancelled out. When the production asymmetry between
d is zero (as is the case at the B factories), this corresponds to not having to
Bd and B
d . Restricting ourselves to untagged
determine whether the decaying meson was Bd or B
measurements is a way of getting rid of tagging inefficiencies and mistagging problems. At
d is necessary.
hadronic machines, a handle on the production asymmetry between Bd and B
In this section we show explicitly that the time measurements of the decay of an
untagged Bd to a single final state can only be sensitive to quadratic terms in d /d .
This would imply that, for determining d /d using only one final state, the accuracy
of the measurement needs to be (d /d )2 105 . This indicates the necessity of
combining measurements from two different final states to be sensitive to a quantity linear
in d /d 0.3 102 . We discuss three pairs of candidate channels for achieving this
task. Finally, we point out the extent of systematic error in the conventional measurement
of due to the neglect of the width difference, and show how the tagged Bd J /KS
mode can also measure d /d by itself.
3.1. Quadratic sensitivity to d /d of untagged measurements
It is common wisdom that the time measurements, in general, are sensitive only
quadratically to d /d . Specific calculations (e.g., see [5]) also get results that can be
clearly seen to obey this rule. Here, we give an explicit derivation of the general statement,
pointing out the exact conditions under which the above statement is valid. Ways of
getting around these conditions lead us to the decay modes that can provide measurements
sensitive linearly to d /d .
The nonoscillating part of the proper time distribution of the decay of Bd can be written
in the most general form as
f (t) =

1
(1 + b)eL t + (1 b)eH t .
2
The nonoscillating part can also be looked upon as the untagged measurement.
(68)
391
For an isotropic decay, the only information available from the experiment is the
time t. This information may be completely encoded in terms of the (infinitely many) time
moments
n
n
t f (t) dt
t
(69)
.
f (t) dt
Expanding in powers of d /d , we get

n

nb d
n!
2
t =
1
+ O (d /d ) .
(d )n
2 d
Defining the effective untagged lifetime as b
information (69) is encoded in

n
t = n!(b )n 1 + O (d /d )2 .
(70)
1
d

b d
2 d ,
all the available

(71)
)2 ,
only the
Thus, when the accuracy of the lifetime measurement is less than (d /d
combination b of d , d and b may be measured through a single final state. This
measurement is insensitive to b (to this order) and hence incapable of even discerning the
presence of two distinct lifetimes (b = 0 and b = 1 would correspond to the presence of
only a single lifetime involved in the decay.) In particular, in order to determine d /d ,
the lifetime measurement through the semileptonic decay needs to be more accurate than
(d /d )2 105 . This task is beyond the capacity of the currently planned experiments.
Combining time measurements from two different final states, however, can enable us
to measure quantities linear in d /d . Indeed, for two final states with different values b
(say b1 and b2 ), we can measure

b2 b1 d
b 1
=1+
+ O (d /d )2 .
(72)
b 2
2
d
In the next subsections, we discuss pairs of decay channels that can measure this
quantity (72) that is linear in d /d .
3.2. Decay widths in semileptonic and CP-specific channels
Let us first develop the formalism that will be applicable for all the decays that we shall
consider below. When the width difference is taken into account, the decay rate of an initial
d , and
Bd to a final state f is given as follows. Let Af f |Bd , A f f |B
f
q A f
,
p Af
(73)
where p and q are as defined in (3). Using the CP-violating parameter d as defined in (7),
we get

d
q
= 1 1 d .
(74)
p
1 + d
The approximation here is valid since we have | d | |d /md | 102 . Henceforth,
we shall only consider terms linear in d .
392
d to a final state f is given by [5]:

The decay rate of an initial tagged Bd or B

1 + |f |2 d t
d t
e
+ Adir
cosh
Bd (t) f = Nf |Af |2
CP cos(mt)
2
2

d t
+ Amix
sin(mt)
,
+ A sinh
CP
2
(75)

2

1
+
|
|

t
f
d
d (t) f = Nf |A f |2
ed t cosh
Adir
B
CP cos(mt)
2
2

d t
mix
ACP sin(mt) ,
+ A sinh
2
(76)
where the CP asymmetries are defined as
Adir
CP =
1 |f |2
,
1 + |f |2
Amix
CP =
2 Im f
,
1 + |f |2
A =
2 Re f
,
1 + |f |2
(77)
and Nf is a time-independent normalization factor.

In the case of semileptonic decays, f {D<+ }, so that A f = 0 and hence f = 0. The
time evolution (75) then becomes

d t
+ cos(mt) ,
Bd (t) f ed t cosh
(78)
2
eL t + eH t + oscillating terms,
(79)
so that for semileptonic decays, we have bSL = 0. Note that b = 0 is true for all self-tagging
modes, so that all the arguments below for semileptonic modes hold true also for all the
self-tagging decay modes.
For the decays to CP eigenstates that proceed only through tree processes (and have zero
or negligible penguin contribution), we have f = (1 d )e2i (the two signs + and
correspond to CP-even and CP-odd final states, respectively). Then (75) gives

Bd (t) f

d t
d t
cos(2) sinh
sin(2) sin(mt) ,
ed t cosh
(80)
2
2

eL t 1 cos(2) + eH t 1 cos(2) + oscillating terms,
(81)
where we have neglected the small corrections due to d . Thus, for CP eigenstates, we have
bCP+ = + cos(2) and bCP = cos(2).
The ratio between the two lifetimes CP and SL is then

SL
cos(2) d
=1
+ O (d /d )2 .
CP
2
d
(82)
The measurement of these two lifetimes should be able to give us a value of |d |, since
| cos(2)| will already be known to a good accuracy by that time.
393
Note that it is also possible to measure the ratio of the lifetimes CP and CP+ :

CP
d
= 1 + cos(2)
+ O (d /d )2 .
CP+
d
(83)
Although the deviation of the ratio from 1.0 in this case is larger by a factor of 2,
using the effective semileptonic lifetime instead of one of the CP eigenstates would
still be the favoured method. This is because the CP specific decay modes of Bd
(e.g., J /KS(L) , D + D ) have smaller branching ratios than the semileptonic modes.
In addition, the semileptonic data sample may be enhanced by including the self()+
tagging decay modes (e.g., Ds D () ) that also have large branching ratios. After 5
years of LHC, we should have about 5 105 events of J /KS , whereas the number
of semileptonic decays at LHCb alone that will be directly useful in the lifetime
measurements is expected to be more than 106 per year, even with conservative estimates
of efficiencies.
3.3. Transversity angle distribution in Bd J /K
The decays Bd V V (where V V is a flavour-blind final state consisting of two
vector mesons) take place both through CP-even and CP-odd channels. Since the angular
information is available here in addition to the time information, these decay modes
are not subject to the constraints of the theorem in Section 3.1, and quantities sensitive
linearly to d /d can be obtained through a single final state. This cancels out many
systematic uncertainties, and hence these modes can be extremely useful as long as the
direct CP violation is negligible, and we can disentangle the CP-even and CP-odd final
states from each other. This separation can indeed be achieved through the transversity
angle distribution [1719].
We illustrate the procedure with the example of Bd J /(<+ < )K (KS 0 ). The
most general amplitude for the decay B J /K is given in terms of the polarizations
>J / , >K of the two vector mesons:

mK L L A T
A
T
> > >J / >K

A Bd J /K = A0
i >J / >K p,
EK J / K
2
2
(84)
where EK is the energy of the K and p the unit vector in the direction of K in the J /
rest frame. The superscripts L and T represent the longitudinal and transverse components,
respectively. Since the direct CP violation in this mode is negligible, the amplitudes A0 and
A are CP-even, whereas A is CP-odd. Let us define the angles as follows. Let the x axis
be the direction of K in the J / rest frame, and the z axis be perpendicular to the decay
plane of K KS 0 , with the positive y direction chosen such that py (KS ) 0. Then we
define (, ) as the decay direction of <+ in the J / rest frame and as the angle made
by KS with the x axis in the K rest frame.
Here is the transversity angle, i.e., the angular distribution in can separate CP-even
and CP-odd components of the final state. The angular distribution is given by [20]
394
2
2
3
d [Bd J /(<+ < )K (KS 0 )] 3
= A+ (t) 1 + cos2 + A (t) sin2 ,
d cos
8
4
(85)
where |A+ (t)|2 |A0 (t)|2 + |A (t)|2 is the CP-even component and |A (t)|2 |A (t)|2
the CP-odd one. These two components can be separated from the angular distribution (85)
through a likelihood fit or through the method of angular moments [20,21]. 3
The time evolutions of the CP-even and CP-odd components are given by

A+ (t)2 = A+ (0)2 cos2 eL t + sin2 eH t + ed t sin(Md t) sin(2) ,

A (t)2 = A (0)2 sin2 eL t + cos2 eH t ed t sin(Md t) sin(2) .
(86)
(87)
These are the same as the time evolutions in (81). The difference in the untagged lifetimes
of the two components,

d
CP
= 1 + cos(2)
+ O (d /d )2 ,
CP+
d
(88)
is linear in the lifetime difference d .

The disentanglement of the CP-even and CP-odd components from the angular
distribution is a statistically efficient process [21]. In fact, in the Bs system, the angular
distribution of Bs J /(<+ < )(K + K ) can be used for determining the lifetime
difference s , and is the preferred mode for measuring this quantity.
The mode J /K suffers from the presence of a 0 in the final state, which may be
missed by the detector, thus introducing a source of systematic error that needs to be
minimized.
3.4. Untagged asymmetry between B J /KS and J /KL
Two of the decay modes of Bd that have been well explored experimentally (because
of their usefulness in measuring ) are B J /KS and J /KL . Here we show that the
time-dependent asymmetry between the decay rates of these modes is a quantity linear in
d /d , and therefore within the domain of experimental feasibility.
Let us define
A(Bd J /KS ) = AS ,
A(Bd J /KL ) = AL ,
d J /KS ) = A S ,
A(B
d J /KL ) = A L ,
A(B
so that using

0
,
|KS = (1 + >)K 0 + (1 >)K

0
,
|KL = (1 + >)K 0 (1 >)K
(89)
3 In [19] we suggested to use the CP-oddCP-even interference in the decay B J /K to measure the
value of d /d . However, it involves tagged measurements in addition to two- or three-angle distributions, and
hence is not as attractive as the untagged measurements described here.
395
we can write (with the phase convention Arg(q/p) = 0)

AS = AL = Aei 1 + ap ei ei (1 + >),

A S = A L = Aei 1 + ap ei ei (1 >),
(90)
V
Vcb
cs
ei ei
where ap
is the ratio of contributions that involve the CKM factors
and
Vtb Vt d , respectively. The latter contribution (penguin) is highly suppressed with respect to
the former one (tree): the value of ap is less than a percent. Here is the strong phase and
V ) 0.015 in the SM. From (73), (74) and (90), we get
Arg(Vtb Vt s /Vcb
cd

S = L 1 d e2i 1 2> 2i sin ap ei e2i (1 2> ), (91)
where > is an effective complex parameter that absorbs all the small theoretical
uncertainties.
d is zero (as is the case at the B
When the production asymmetry between Bd and B
factories), the untagged rate of decay is

Bun J /KS (KL ) N |AS |2 1 2 Re(> ) ed t

d t
d t
+ Ad sinh
.
cosh
(92)
2
2
The only difference between the decay to KS and that to KL is the sign of A :
A (KS ) = A (KL ) = cos(2) 2 Im(> ) sin(2).
(93)
The untagged time-dependent asymmetry between Bun J /KS and KL is

(Bun (t) J /KS ) (Bun (t) J /KL )
(Bun (t) J /KS ) + (Bun (t) J /KL )

d t
1 2 Im(> ) tan(2)
= cos(2) tanh
2

d t
.
cos(2) tanh
2
A(KL , KS )
(94)
(95)
(96)
Thus, the measurement of this asymmetry will enable us to determine |d |, given
sufficient statistics and a measurement of sin 2.
The factor limiting the accuracy of the above asymmetry is the measurement of
(Bun (t) J /KL ). At the B factories, KL may be detected through its hadronic
interactions in the calorimeter, and though its energy is poorly measured, the corresponding
detection of J / can help in reducing the background. However, the number of events
available may be too small for an accurate measurement. In the hadronic machines this
decay has a high background, so the systematic errors in the measurement may be too
large for this method to be of practical use.
3.5. Effect on the measurement of sin(2)
The time-dependent CP asymmetry measured through the gold-plated mode Bd
J /KS is [22,23]
396
d (t) J /KS ] [Bd (t) J /KS ]

[B
d (t) J /KS ] + [Bd (t) J /KS ]
[B
sin(md t) sin(2),
ACP =
(97)
(98)
which is valid when the lifetime difference, the direct CP violation, and the mixing in
the neutral K mesons is neglected. As the accuracy of this measurement increases, the
corrections due to these factors will need to be taken into account. Keeping only linear
terms in the small quantities > and d , we get

d t
ACP = sin(mt) sin(2) 1 sinh
(99)
cos(2)
2

+ 2 Re(> ) 1 + sin2 (2) sin2 (mt) cos(mt)
(100)
+ 2 Im(> ) cos(2) sin(mt).
(101)
The first term in (99) represents the standard approximation used (98) and the correction
due to the lifetime difference d . The rest of the terms [(100) and (101)] include
and KK
mixings, which are of the same
corrections due to the CP violation in BB
order as d /d .
In the future experiments that aim to measure to an accuracy of 0.005 [7], the
correction terms need to be taken into account. The corrections due to > and d will
form a major part of the systematic error, which can be taken care of by a simultaneous
The BaBar Collaboration gives the bound on the coefficient
fit to sin(2), d and >.
of cos(mt) in (100), while neglecting the other correction terms [24]. When the
measurements are accurate enough to measure the cos(mt) term, the rest of the terms
would also have come within the domain of measurability. For a correct treatment at this
level of accuracy, the complete expression for ACP above (99)(101) needs to be used.
3.6. Tagged measurements
Until now, we have discussed only the untagged measurements. Taking into account the
oscillating part of the time evolution of the decay rate, we have the decay rate in general as
g(t) = f (t) + Ced t sin(mt + ),
(102)
where f (t) is the untagged decay rate as defined in (68), C a constant and a phase. The
lifetime of the oscillating part is an additional lifetime measurement, which opens up the
possibility of being able to determine d /d through only one final state (and without
angular distributions as in Section 3.3).
In the case of the semileptonic decays, this strategy fails since the semileptonic width
measured with the untagged sample is
SL =
1 14 (d /d )2
(L + H )L H
=
,
d
(L )2 + (H )2
1 + 14 (d /d )2
(103)
so that

SL /d = 1 + O (d /d )2 .
(104)
397
Thus the semileptonic decays would provide sensitivity only to quadratic terms in d /d ,
even if it were possible to use the tagged measurements efficiently.
However, the untagged lifetime measured through the decay to a CP eigenstate is

1
cos(2) d
1
,
CP
(105)
d
2
d
so that it differs from the lifetime of the oscillating part (d 1/d ) by terms linear in
d /d . Thus, the tagged measurements of a CP-even or CP-odd final state (D + D ,
J /KS , J /KL , etc.) can measure d /d by themselves.
The mistag fraction is the main limiting factor on the accuracy of this measurement, and
the tagging efficiency limits the number of events available. It is indeed possible that the
d measurement through the semileptonic decays will be more accurate than that through
the oscillating part of the CP-specific final state. This then reduces to the method suggested
in Section 3.2. For further experimental details on a tagged measurement of d /d we
refer the reader to Ref. [25].
4. Lifetime differences in Bs and Bd systems

The calculations of the lifetime difference in Bd (as performed here) and in the Bs
system (as in [2,3]) run along similar lines. However, there are some subtle differences
involved, due to the values of the different CKM elements involved, which have significant
consequences. In particular, whereas the upper bound on the value of s (including the
effects of new physics) is the value of s (SM) [26], such an upper bound on d can be
established only under certain conditions and involves a multiplicative factor in addition
to d (SM). Also, whereas the difference in lifetimes of CP-specific final states in the Bs
s mixing phase, the corresponding
system cannot resolve the discrete ambiguity in the Bs B
d mixing
measurement in the Bd system can resolve the discrete ambiguity in the Bd B
phase. Let us elaborate on these two differences in this section.
4.1. Upper bounds on d(s) in the presence of new physics
For convenience, let us define q Arg(21 )q , q Arg(M21 )q , where q {d, s}.
Then we can write
q = 2|21|q cos(q q ).
(106)
Since the contribution to 21 comes only from tree diagrams, we expect the effect of new
physics on this quantity to be very small. We, therefore, take |21 |q and q to be unaffected
by new physics. On the other hand, the mixing phase q appears from loop diagrams and
can therefore be very sensitive to new physics.
Let us first consider the Bs system. Here 21 may be written in the form

2

21 (Bs ) = N Vcb
Vcs f (z, z) + 2 Vcb
Vcs Vub
Vus f (z, 0)

2
+ Vub
Vus f (0, 0) ,
(107)
398
where N is a positive normalization constant and f (x, y) are the hadronic factors that
do not depend on the CKM matrix elements. In the limit z m2c /m2b 0, we get
f (z, z) = f (z, 0) = f (0, 0). Even with z = 0, we expect all the f s to have similar
magnitudes. On the other hand, the CKM elements involved in (107) obey the hierarchy
V )2 4 , (V V )(V V ) 6 , (V V )2 8 . The term involving (V V )2
(Vcb
cs
cb cs
ub us
ub us
cb cs
then dominates in (107), and we can write
2

Vcs f (z, z) 1 + O 2 .
21 (Bs ) = N Vcb
(108)
V )2 . Then,
Since the f s are real positive functions, we have s + Arg(Vcb
cs
2

s = 2|21 |s cos Arg Vcb Vcs s .
In SM, s = Arg(Vtb Vt s )2 , therefore, the argument of the cosine term

V )2 /(V V )2 ] = 2 0.03. Thus in SM, we have
Arg[(Vcb
cs
tb ts
s (SM) = 2|21|s cos(2 ).
(109)
in (109) is given by
(110)
The effect of new physics on s can then be bounded by giving an upper bound on s :
s (SM)
s (SM).
(111)
cos(2 )
Thus, the value of s can only decrease in the presence of new physics [26].
In the case of the Bd system, the situation is slightly different. As in the Bs case, we can
write

2

21 (Bd ) = N Vcb
Vcd f (z, z) + 2 Vcb
Vcd Vub
Vud f (z, 0)

2
+ Vub
Vud f (0, 0) ,
(112)
s
where the normalizing factor N and the hadronic factors f are the same as in the Bs case
in the limit of the U-spin symmetry (see [27]) and therefore have similar magnitudes. The
CKM elements involved in (112) do not obey a hierarchy similar to the Bs case: instead
V )2 (V V )(V V ) (V V )2 6 . Then no single term in (112)
we have (Vcb
cd
cb cd
ub ud
ub ud
can dominate. We can, however, use the unitarity of the CKM matrix 4 to rearrange (112)
in the form

2

21 (Bd ) = N Vcb
Vcd f (z, z) 2f (z, 0) + f (0, 0)

Vcd Vtb Vt d f (0, 0) f (z, 0)
+ 2 Vcb

2

(113)
+ Vtb Vt d f (0, 0) .
Note that in the limit of z 0, all the factors f are identical and hence the coefficients
V )2 and (V V )(V V ) vanish. The last term in (113) is then left over as the
of (Vcb
cd
cb cd
tb td
4 We note that this assumption of the unitarity for a three-generation CKM matrix is quite general, because most
popular new physics models, including supersymmetric models, preserve the three-generation CKM unitarity. The
present CKM values, constrained from various experiments, are completely consistent with the unitarity for the
three-generation CKM matrix. Moreover, one can show that the nonunitary effects within the three-generation
CKM, which can stem from the fourth generation or E(6)-inspired models with one singlet down-type quark, are
4 , once we assume a Wolfenstein-type hierarchical structure for the extended CKM matrix.
dominating one, and we get

2

21 (Bd ) N Vtb Vt d f (0, 0),
399
(114)
thus giving d + Arg(Vtb Vt d )2 . However, the finite value of z 0.1 may give large
corrections to this value.
We have already shown that (114) may be written in the forms (65) or (66). From this,
we get d = + Arg(Vtb Vt d )2 + Arg(1 + fi ), where i {u, c}. Using (106), we then
have
2

d 2|21|d cos Arg Vtb Vt d d + Arg(1 + fi ) .
(115)
In SM, d = Arg(Vtb Vt d )2 , so that

d (SM) 2|21|d cos Arg(1 + fi ) ,
(116)
and an upper bound for d can be written in terms of d (SM) as

d 2|21|d =
d (SM)
.
cos[Arg(1 + fi )]
(117)
We can calculate the bound (117) in terms of the extent of the higher-order NLO
corrections. Estimating |Fct | < 0.4 (corresponding to the error analysis in Section 2.4),
we get | Arg(1 + fu )| < 0.6, so that we have the bound d < 1.2 d (SM). Note that
this bound is valid only in the range of Fct estimated above. A complete NLO calculation
will be able to give a stronger bound.
Thus in the case of the Bd system, we have an upper bound analogous to the one in the Bs
system only under certain conditions. Moreover, the reasons behind the existence of these
two upper bounds differ. Whereas in the Bs case it follows directly from the hierarchy in
the CKM elements, in the Bd case it depends on the values of the hadronic terms. Note that
whereas unitarity was not needed in the Bs case, the assumption that (21 )q is unaffected
by new physics is required in both the cases.
4.2. Mixing phase: new physics and discrete ambiguity
d mixing phase d is efficiently measured through the decay modes J /KS
The Bd B
and J /KL . If we take the new physics effects into account, the time-dependent
asymmetry (97) is ACP = sin(Md t) sin(d ), which reduces to (98) in the SM, where
d = 2. The measurement of sin(d ) still allows for a discrete ambiguity d
d . Whenever a discrete ambiguity in is referred to ( /2 ) in this paper
(or in the literature), strictly speaking we are talking about the discrete ambiguity d
d . In this section, we shall use the notation d instead of 2 in order to illustrate the
comparison with the corresponding quantities in the Bs system.
Getting rid of the above discrete ambiguity is a way of uncovering a possible signal
of new physics. 5 Ways to get rid of this ambiguity have been suggested in literature,
5 In SM, the value of must match with the phase of the b d penguin. However, the direct measurement
d
of the latter phase is not theoretically clean [28], so the preferred way is to compare the measured value of d
with the value of 2 determined through an unitarity fit for all the CKM parameters [15].
400
using the comparison of CP asymmetries in J /KS and [29], time-dependent CP

asymmetries in Bs KS [30] and in Bs K, KK [31], angular distributions and
U-spin symmetry arguments [32], or cascade decays B D K [33]. The measurement
of d through the measurements involving d is unique in the sense that it uses only
untagged measurements.
In Sections 3.2 and 3.3, we have seen that the ratio of two effective lifetimes can enable
us to measure the quantity obs(d) cos(2) d / . In the presence of new physics,
this quantity is in fact (see Eq. (106))
obs(d) = 2
|21|d
cos(d ) cos(d d ).
d
(118)
In SM, we get

|21|d
(119)
cos(2) cos Arg(1 + fi ) .
d
If |fi | < 1.0, we have cos[Arg(1 + f )] > 0 (in fact, from the fit in [15] and our error
estimates, we have cos[Arg(1 + fu )] > 0.8). Then obs(d) (SM) is predicted to be
positive. New physics is not expected to affect d , but it may affect d in such a way as
to make the combination cos(d ) cos(d d ) change sign. A negative sign of obs(d)
is a clear signal of such new physics.
In addition, if d can be determined independently of the mixing in the Bd system, 6
then measuring obs(d) and using (118) gives us two solutions for d in general: 1d
and 2d such that tan(1d + 2d ) = tan(d ). As long as tan(d ) = 0 (as the unitarity
fit suggests), at most one of the solutions will correspond to the value of sin(d )J /KS
obtained through ACP (J /KS ). If none of d1 or d2 corresponds to the value of
sin(d )J /KS , there definitely is new physics in Bd mixing. 7 If one of 1d or 2d matches
with sin(d )J /KS , it gives the actual value of d , and thus resolves the discrete ambiguity
in principle. In practice, this requires the knowledge of 21(d) theoretically to a high
precision and having to measure obs(d) to sufficient accuracy to be able to distinguish
between 1d and 2d . A complete NLO calculation is needed for the former. The latter
may be achieved at the LHC using the effective lifetimes of decays to semileptonic final
states and to J /KS .
Let us contrast this case with that in the Bs system and show these features above are
unique to the Bd system: the corresponding time-dependent asymmetry in the Bs system

is measured through the modes J / or J / ( ) , which give the value of sin(s ), and
therefore leave the discrete ambiguity s s unresolved. The ratio of two effective
lifetimes in the Bs system can enable us to measure the quantity
obs(d) (SM) = 2
obs(s) cos(s )
|21|s
s
= 2
cos(s ) cos(s s ).
(120)
6 For example, if we assume that new physics does not affect the mixing in the neutral K system, the fit
V ) independent of any mixing in the B
to |Vub /Vcb | and >K (a part of the unitarity fit) determines Arg(Vtb
td
V )2 + Arg(1 + f ) where Arg(1 + f ) will be
system. This gives us a measurement of d = + Arg(Vtb
td
i
i
known to a good accuracy with a complete NLO calculation.
7 Or any assumption that goes into the determination of is in question.
d
401
V )2 , we have
Since s + Arg(Vcb
cs
|21|s
cos2 (s ).
(121)
s
This measurement thus still has the same discrete ambiguity s s as in the J /

(or J / ( ) ) case, and the discrete ambiguity in the Bs system is not resolved.
obs(s) 2
5. Summary and conclusions

It has been known for many years that the Bd system is a particularly good place to
test the Standard Model explanation of CP violation through the unitary CKM matrix. The
d mixing is large, and hence the CP violation is expected to
phase 2 involved in the Bd B
be larger in the Bd system in general, as compared to the K or the Bs system. This feature
has already been exploited in various methods for extracting , and , the angles of the
unitarity triangle, by measuring CP-violating rate asymmetries in the decays of neutral Bd
mesons to a variety of final states. In particular, the precise measurement of sin(2) from
the theoretically clean decay modes Bd (t) J /KS (KL ) is a test of the SM, as well as
the opportunity to search for the presence of physics beyond the standard model.
The two mass eigenstates of the neutral Bd systemBH and BL have slightly
different lifetimes: the lifetime difference is less than a percent. At the present accuracy
of measurements, this lifetime difference d can well be ignored. As a result, the
measurement and the phenomenology of d has been neglected so far, as compared
to the lifetime difference in the Bs system for example. However, with the possibility of
experiments with high time resolution and high statistics, such as the electronic asymmetric
B factories of BaBar, BELLE, and hadronic B factories of CDF, LHC and BTeV, this
quantity starts becoming more and more relevant.
Taking the effect of d into account is important in two aspects. On one hand, it affects
the accurate measurements of crucial quantities like the CKM phase and, therefore, must
be measured in order to estimate and correct the error due to it. On the other hand, the
measurement of d can lead to clear signal for new physics.
Thus in addition to being the measurement of a well-defined physical quantity which
can be compared with the theoretical prediction, the value of d is important for getting
a firm grip on our understanding of CP violation. It is, therefore, worthwhile to have a look
at this quantity and make a realistic estimation of the possibility of its measurement, as we
do in this paper.
We estimate d /d including 1/mb contributions and part of the next-to-leading
order QCD corrections. We find that adding the latter corrections decreases the value of
d /d computed at the leading order by almost a factor of two. We get the final result
3
as d /d = (2.6+1.2
1.6 ) 10 , or using another expansion of the NLO QCD corrections
3
d /d = (3.0+0.9
1.4 ) 10 , where for the central value we have used the preliminary
values for the bag factors from the JLQCD Collaboration. In the error estimation, we take
into account the errors from the uncertainties in the values of the CKM parameters, the bag
parameters, the mass of the b quark, and the measured value of xd . The major sources of
402
error are the scale dependence, the breaking of the quarkhadron duality, and the error due
to missing terms in the NLO contribution. We show that a complete NLO calculation is
desirable.
The most obvious way of trying to measure the lifetime difference is through the
semileptonic decays, however it runs into major difficulties. If only the non-oscillating
(untagged) part of the time evolution of the decay is considered, we indeed have a
combination of two exponential decays with different lifetimes. However, as we show
in this paper, there is no observable quantity here that is linear in d /d . The time
measurements allow us to determine the quantity SL (1/d )[1 + O(d /d )2 ]. This
decay mode is thus sensitive only to quantities quadratic in d /d . So this method would
involve measuring a quantity as small as (d /d )2 105 , which is not practical. The
lifetime of the oscillating part is also 1/d , so adding the information from the oscillating
part of the time evolution does not help at all. This problem arises for all self-tagging
decays. Therefore, though self-tagging decays of Bd have significant branching ratios, they
cannot by themselves be expected to give a measurement of d /d .
The time evolutions of Bd decaying into CP eigenstates also involve both the lifetimes,
d mixing phase (2) is large, which implies that the CP eigenstates are
since the Bd B
appreciably different from the lifetime eigenstates. As a result, it is not possible to find
a final state to which the decay of Bd involves only one of the decay widths L and
H . The non-oscillating part of the time evolution of decays to CP eigenstates gives a
quantity CP (1/d )[1 (cos(2)/2)d /d + O(d /d )2 ], but the quantities d
and d cannot be separately determined through this measurement, and sensitivity to
(d /d )2 is necessary. (Indeed, we explicitly prove a general theorem that shows that, for
isotropic decays of Bd to any final state, the untagged measurements can only be sensitive
to (d /d )2 .)
The oscillating part of the time evolution to CP eigenstates has a lifetime 1/d (to an
accuracy of O(d /d )2 ). Therefore, if this lifetime is measured accurately, it can be
combined with the measurement of CP through the untagged part to get a measurement
linear in d /d . However, the need for tagging, and consequent mistagging errors,
reduce the efficiency of this method.
A viable option, perhaps the most efficient among the ones considered here, is to
compare the measurements of the untagged lifetimes SL and CP . Since SL is in fact
the lifetime for all self-tagging decays, and the branching ratios for self-tagging decays
of Bd are much larger than the decays to CP eigenstates, we expect that the most useful
combination will be the measurement of SL through self-tagging decays and that of CP+
through Bd J /KS .
The untagged asymmetry between Bd J /KS and Bd J /KL is a particular
case of using the combination of measurements of CP+ and CP , which we analyze in
detail. The effects of CP violation in the mixing and decay of Bd , as well as the indirect
CP violation in the K system has been taken into account. The usefulness of this method
is limited by the uncertainties in the measurement of (Bd J /KL ).
Since the theorem referred to aboveabout a single untagged decay being sensitive only
to (d /d )2 applies only to isotropic decays, decays of the type B V V can still be
403
used by themselves to determine quantities linear in d /d . A promising example is

Bd J /( <+ < ) K ( Ks 0 ). The CP-odd and CP-even components in the final
state can be disentangled through the transversity angle distribution, and both CP+ and
CP can be determined through the same decay. Since there is only one final state, many
systematic errors are reduced. The only undesirable feature of this decay mode is the
presence of 0 in the final state, which may be missed, especially in the hadronic machines.
The three angle distribution of the same decay mode can also be used to obtain d /d
through the interference between CP-even and CP-odd final states. The three angle method
is, however, not as efficient as the single angle distribution, since one has to use tagged
decays and more number of parameters need to be fitted.
We also point out the interlinked nature of the accurate measurements of and d /d
through the conventional gold-plated decay. In the future experiments that aim to measure
to an accuracy of 0.005 or better, the corrections due to d will form the major part of
the systematic error, which can be taken care of by a simultaneous fit to sin(2), d and
an effective parameter > that comes from a combination of CP violation in mixing in the
Bd and K system.
All the combinations of untagged decay modes discussed here involve measuring the
quantity obs(d) (cos(2)/2)d /d , wherein the value of d /d also depends
on . The complete dependence on is of the form cos(2) cos(d + 2). The sign
of this quantity is positive in SM, but may be changed in the presence of new physics.
There is thus a potential for detecting physics beyond the SM. Moreover, if the value of
d is known independently of B mixing, then since obs(d) is not invariant under
/2 , the discrete ambiguity in is resolved in principle. Note that this feature is unique
to the Bd systemin the Bs system for example, obs(s) does not help in resolving the
s mixing phase.
corresponding discrete ambiguity in the Bs B
It is known that, if (21 )s is unaffected by new physics, then the value of s in the Bs
system is bounded from above by its value as calculated in the SM. In the Bd system, this
statement does not strictly hold true. However, if (21 )d is unaffected by new physics and
the unitarity of the 3 3 CKM matrix holds, then an upper bound on the value of d may
be found. In the absence of a complete NLO calculation, however, this bound is a weak
one.
With the high statistics and accurate time resolution of the upcoming experiments, the
measurement of d seems to be in the domain of measurability. And given the rich
phenomenology that comes with it, it is certainly a worthwhile endeavor.
Acknowledgements
We would like to thank G. Buchalla for many useful discussions and for a reading
of the manuscript. We also thank U. Nierste for critical comments, and C. ShepherdThermistocleous, Y. Wah and H. Yamamoto for useful discussions concerning the
experimental aspects. A.D. would like to thank the CERN Theory Division, where a large
fraction of this work was completed. The work of C.S.K. was supported in part by CHEP-
404
SRC Program, Grant No. 20015-111-02-2 and Grant No. R03-2001-00010 of the KOSEF,
in part by BK21 Program and Grant No. 2001-042-D00022 of the KRF, and in part by
Yonsei Research Fund, Project No. 2001-1-0057. The work of T.Y. was supported in part
by the US Department of Energy under Grant No. DE-FG02-97ER-41036.
References
[1] R. Aleksan, A. Le Yaouanc, L. Oliver, O. Pne, J.C. Raynal, Phys. Lett. B 316 (1993) 567.
[2] M. Beneke, G. Buchalla, I. Dunietz, Phys. Rev. D 54 (1996) 4419, hep-ph/9605259.
[3] M. Beneke, G. Buchalla, C. Greub, A. Lenz, U. Nierste, Phys. Lett. B 459 (1999) 631, hepph/9808385.
[4] I. Dunietz, Phys. Rev. D 52 (1995) 3048, hep-ph/9501287.
[5] I. Dunietz, R. Fleischer, U. Nierste, hep-ph/0012219.
[6] Particle Data Group, D.E. Groom et al., Eur. Phys. J. C 15 (2000) 1.
[7] P. Ball et al., hep-ph/0003238.
[8] R.N. Cahn, M.P. Worah, Phys. Rev. D 60 (1999) 076006, hep-ph/9904480.
[9] J.S. Hagelin, Nucl. Phys. B 193 (1981) 123;
A.J. Buras, W. Slominski, H. Steger, Nucl. Phys. B 245 (1984) 369.
[10] A.J. Buras, M. Jamin, P.H. Weisz, Nucl. Phys. B 347 (1990) 491.
[11] G. Altarelli, G. Curci, G. Martinelli, S. Petrarca, Nucl. Phys. B 187 (1981) 461;
A.J. Buras, P.H. Weisz, Nucl. Phys. B 333 (1990) 66.
[12] S. Hashimoto, N. Yamada, JLQCD Collaboration, hep-ph/0104080.
[13] Report of Workshop on B Physics at the Tevatron: Run II and Beyond (unpublished). Available
at http://www-theory.lbl.gov/~ligeti/Brun2/report/drafts/draft.uu, FERMILAB-Pub-01/197.
[14] G. Buchalla, A.J. Buras, M.E. Lautenbacher, Rev. Mod. Phys. 68 (1996) 1125, hep-ph/9512380.
[15] S. Mele, hep-ph/0103040.
[16] M. Beneke, A. Lenz, J. Phys. G 27 (2001) 1219, hep-ph/0012222.
[17] A.S. Dighe, I. Dunietz, H.J. Lipkin, J.L. Rosner, Phys. Lett. B 369 (1996) 144, hep-ph/9511363.
[18] C.S. Kim, Y.G. Kim, C.-D. Lu, hep-ph/0102168;
C.S. Kim, Y.G. Kim, C.-D. Lu, T. Morozumi, Phys. Rev. D 62 (2000) 034013, hep-ph/0001151.
[19] T. Hurth et al., J. Phys. G 27 (2001) 1277, hep-ph/0102159.
[20] A.S. Dighe, I. Dunietz, R. Fleischer, Eur. Phys. J. C 6 (1999) 647, hep-ph/9804253.
[21] A. Dighe, S. Sen, Phys. Rev. D 59 (1999) 074002, hep-ph/9810381.
[22] B. Aubert et al., BaBar Collaboration, Phys. Rev. Lett. 86 (2001) 2515, hep-ex/0102030.
[23] A. Abashian et al., BELLE Collaboration, Phys. Rev. Lett. 86 (2001) 2509, hep-ex/0102018.
[24] B. Aubert et al., BABAR Collaboration, Phys. Rev. Lett. 87 (2001) 091801.
[25] S. Petrak, Reach of Measurements for Bd0 and Bs0 , BaBar Note 496 (unpublished).
[26] Y. Grossman, Phys. Lett. B 380 (1996) 99, hep-ph/9603244.
[27] R. Fleischer, Phys. Lett. B 459 (1999) 306;
M. Gronau, Phys. Lett. B 492 (2000) 297;
T. Hurth, T. Mannel, Phys. Lett. B 511 (2001) 196, hep-ph/0103331, hep-ph/0109041.
[28] C.S. Kim, D. London, T. Yoshikawa, Phys. Lett. B 458 (1999) 361, hep-ph/9904311;
A. Datta, C.S. Kim, D. London, hep-ph/0105017.
[29] Y. Grossman, Y. Nir, M.P. Worah, Phys. Lett. B 407 (1997) 307, hep-ph/9704287.
[30] Y. Grossman, H.R. Quinn, Phys. Rev. D 56 (1997) 7259, hep-ph/9705356.
[31] C.S. Kim, D. London, T. Yoshikawa, Phys. Rev. D 57 (1998) 4010, hep-ph/9708356.
[32] A.S. Dighe, I. Dunietz, R. Fleischer, Phys. Lett. B 433 (1998) 147, hep-ph/9804254.
[33] B. Kayser, D. London, Phys. Rev. D 61 (2000) 116012, hep-ph/9909560.

Telling three from four neutrinos at the neutrino

factory
A. Donini a , M. Lusignoli b , D. Meloni b
a INFN, Sezione di Roma I and Dipartimento di Fisica, Universit di Roma La Sapienza,
Piazzale Aldo Moro 2, I-00185, Rome, Italy

b Dipartimento di Fisica, Universit di Roma La Sapienza and INFN, Sezione di Roma I,
Piazzale Aldo Moro 2, I-00185, Rome, Italy

Received 31 July 2001; accepted 21 December 2001
Abstract
We upgrade the study of the physical reach of a neutrino factory considering the possibility to
distinguish a three (active) neutrino oscillation scenario from the scenario in which a light sterile
neutrino is also present. The distinction is easily performed in the so-called 2 + 2 scheme, but also
in the more problematic 3 + 1 scheme it can be attained in some regions of the parameter space.
We also discuss the CP violating phase determination, showing that the effects of a large phase in
the three-neutrino theory cannot be reproduced in a four-neutrino, CP conserving, model. 2002
PACS: 14.60.Pq; 14.60.St; 14.80.-j
1. Introduction
Indications in favour of neutrino oscillations [14] have been obtained both in solar
neutrino [510] and atmospheric neutrino [1115] experiments. The atmospheric neutrino
data require m2atm (1.64) 103 eV2 [16] whereas the solar neutrino data prefer
m2 1010 or 107 104 eV2 , depending on the particular oscillation solution chosen
for the solar neutrino deficit. The LSND data [17,18], on the other hand, would indicate
a e oscillation with a third, very distinct, neutrino mass difference: m2LSND
0.36 eV2 . The LSND evidence in favour of neutrino oscillation has not been confirmed
E-mail addresses: andrea.donini@roma1.infn.it (A. Donini), maurizio.lusignoli@roma1.infn.it
(M. Lusignoli), davide.meloni@roma1.infn.it (D. Meloni).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 6 0 - 5
406
A. Donini et al. / Nuclear Physics B 624 (2002) 405422
by other experiments so far [19]; the MiniBooNE experiment [20] will be able to do it in the
near future. If MiniBooNE will confirm the LSND results, we would face three independent
evidence for neutrino oscillations characterized by squared mass differences quite well
separated. To explain the whole ensemble of data four different light neutrino species are
needed; the new light neutrino is denoted as sterile, since it must be an electroweak singlet
to comply with the strong bound [21] on the Z 0 invisible decay width. We stress that all
the present experimental results (LSND included) cannot be explained with three massive
light neutrinos only, as it has been shown with detailed calculations in [22].
In order to improve our present knowledge about neutrino masses and mixings, it has
been proposed to build a neutrino factory [23,24], with two long baseline experiments
running at the same time at two different distances. One of its main goal would be the
discovery of leptonic CP violation and, possibly, its study [2528]. Previous analyses
[29,30] on the foreseeable outcome of experiments at a neutrino factory have shown that
the determination of the two still unknown parameters in the three-neutrino mixing matrix,
13 and , will be possible, while with conventional neutrino beams it would not be so [31].
The scenario in which the LSND result has been confirmed (and therefore a short baseline
experiment to study the four-neutrino parameter space is required) has been considered in
[32,33].
If the result of the MiniBooNE experiment will not be conclusive, it seems anyhow
quite relevant to understand if three- and four-family models are distinguishable, and to
what extent. Moreover, it is a sensible question to ask if the effect of CP violation in a
three-family world could be mimicked in a four-family, CP conserving, neutrino scenario.
There are two very different classes of spectra with four massive neutrinos: three almost
degenerate neutrinos and an isolated fourth one, or two pairs of almost degenerate neutrinos
divided by the large LSND mass gap. The two classes of mass spectra are called for obvious
reasons the 3 + 1 and 2 + 2 scheme, respectively. The experimental results were strongly
in favour of the 2 + 2 scheme [34] until the latest LSND results [18]. The new analysis
of the experimental data results in a shift of the allowed region towards smaller values
of the mixing angle, sin2 (2LSND), reconciling the 3 + 1 scheme with exclusion bounds
coming from other reactor and accelerator experiments [3539]. As a consequence, the
3 + 1 model is (marginally) compatible with the data [4043]. Although the 2 + 2 scheme
gave a better fit to those data (as was shown in [44]), the recent SNO results [10] will
certainly restrict the allowed parameter region and give a considerably worse fit for this
model. This is understandable from the analysis presented in Refs. [45,46], prior to the
SNO data publication, since the SMA solution for solar neutrinos, which was more easily
compatible with the limit on sterile components in atmospheric oscillations, is at present
strongly disfavoured (see also [47,48]).
It will certainly be impossible to tell a four-family 3 + 1 from a three-family scheme
in the case of very small activesterile mixing angles: the 3 + 1 model reduces to a threefamily scheme for vanishing mixing with the isolated state. A discrimination will therefore
be possible only if those angles are large enough. On the other hand, the four-family 2 + 2
scheme has the LSND result built-in (in the sense that it cannot stand if LSND is not
confirmed by MiniBooNE) and it does not go smoothly into a three-family model.
We address the problem of distinction between four and three families in this paper.
In order to do that, we generated experimental data [28] according to a four-family
407
theoretical scheme without CP violation and tried to fit these data with a three-family
model having a free CP violating phase . As a cross-check, we also proceeded in the
opposite direction, fitting with four-family formulae the data generated in a three-family
theory. The data have been generated at three distances, L = 732, 3500, 7332 km, using
the detector and machine parameters as defined in [28].
The main conclusions of this analysis are:
The data generated in four-family without CP violation can be described by the threefamily formulae in some particular zones of the four-family parameter space, not
restricted to the obvious case of very small angles. These zones are reduced in size
for increasing gap-crossing angles.
Combining data taken at different distances the zones in which a description with the
other theory is successful are generally reduced.
An increase of the energy resolution of the detector may have a positive effect
(reducing the zones of ambiguity), but only in the case of rather large gap-crossing
angles.
Whenever a particular zone in the four-family parameter space gives acceptable 2
when fitted in the three-family scheme, the fitted value of the three-family CP violating
phase, , is generally not large. In particular, this is true for L = 3500 km, whereas for
L = 7332 km the determination of is somewhat looser, due to the overwhelming
matter effects. For the shortest baseline, L = 732 km, we have the largest spread in the
values of , although the most probable value is still close to zero.
The cross-check, fitting three-family generated data in a four-family model, is
consistent with the results previously obtained. Fitting data generated with a CP phase
close to 90 in a CP conserving 3 + 1 theory is practically impossible if one combines
data at two different distances.
In the 2 + 2 scheme (as opposed to 3 + 1) the ambiguity with a three-neutrino theory
is essentially absent.
In Section 2 we present the relevant formulae for the oscillation probabilities; in
Section 3 we discuss the possibility to have a fit in the three-neutrino theory of data
generated with the 3 + 1 model; in Section 4 we give a discussion of the previous fits,
particularly regarding the value of the CP violating phase . Section 5 is devoted to
conclusions.
2. Three and four neutrino oscillations

Our ordering of the mass eigenstates corresponds to increasing absolute value of the
mass, m2i < m2j for i < j . We restrict ourselves, for simplicity, to the hierarchical 3 + 1
scheme (m221 = m2 , m232 = m2atm and m243 = m2LSND ) and to the so-called
class II-B 2 + 2 scheme (m221 = m2 , m232 = m2LSND and m243 = m2atm ). All
the other spectra may be obtained changing the sign to one or more of the squared mass
differences.
408
Fig. 1. Transition probability in vacuum and in matter (for constant = 2.8 g cm3 ) in the e ( e )
channel for E = 38 GeV. The following parameters have been assumed: 12 = 22.5 , 23 = 45 , 13 = 13 , and
vanishing CP violating phase.
The PontecorvoMakiNakagawaSakata (PMNS) mixing matrix [14] for three

(active) neutrinos is usually parameterized in terms of three angles ij and one CP violating
phase (for Majorana neutrinos, two additional CP violating phases appear, but with no
effect on the oscillations):
(3)
= U23 (23)U13 (13, )U12 (12 ).
UPMNS
(1)
Four neutrino oscillations imply a 4 4 mixing matrix, with six rotation angles and
three phases (for Majorana neutrinos, three additional phases are allowed). It is convenient
to use two slightly different parameterizations. In the 3 + 1 case we write the mixing matrix
as
(3+1)
= U14 (14)U24 (24)U34 (34 )U23 (23 , 3 )U13 (13, 2 )U12 (12 , 1 ),
UPMNS
(2)
which reduces to the three-family case in the limit i4 0.

A more convenient parameterization for the 2 + 2 case, as shown in [32], is
(2+2)
UPMNS = U14 (14)U13 (13)U24 (24 )U23 (23 , 3 )U34 (34, 2 )U12 (12 , 1 ).
(3)
The general formula for the oscillation probability is well-known:

2 m2ij L
P ( ) = 4
Re U,i U,j U,j U,i sin
4E
i>j

i>j

m2ij L

Im U,i U,j U,j U,i sin

2E
(4)
(the lower sign applies to the antineutrino case).

In all of our numerical calculations we used the exact formulae, including matter effects,
that are only important for pathlengths greater than about 2000 km (see Fig. 1 for an
example in the three-family model).
409
First, we recall the three-family oscillation probabilities in vacuum at distances short

enough, so that the oscillations of longest length, due to m2 , do not have time to develop [49]. In the three-neutrino case, neglect of solar m2 entails no CP violation, and
the oscillation probabilities are:

2
2
2
2 matm L
,
P3 (e ) = sin (23 ) sin (213) sin
(5)
4E

m2atm L
P3 (e ) = cos2 (23) sin2 (213 ) sin2
(6)
,
4E

m2atm L
P3 ( ) = cos4 (13) sin2 (223 ) sin2
(7)
,
4E

2
2
2 matm L
,
P3 (e e ) = 1 sin (213) sin
(8)
4E

m2atm L
P3 ( ) = 1 cos4 (13 ) sin2 (223) + sin2 (23 ) sin2 (213) sin2
.
4E
(9)
The explicit expressions for the exact formulae in the four-family case are quite heavy.
It may be of interest to present approximate formulae, that allow better intuition at least in
their range of validity. We consider therefore the shortest distance from the neutrino source,
L = 732 km (the CERNLNGS and FermilabMINOS distance), where matter effects are
negligible and we present approximate formulae for the 3 + 1 four-family mixing in vacuum (see also [33]). At this distance the effects of the large m2LSND oscillations are averaged and the effects of the small m2 are negligible. We also assume, in agreement with
present bounds [40], 14 = 24 = and we expand in power series in to deduce simple expressions to be compared with Eqs. (5)(9). The result for vanishing CP violating phases is:
P3+1 (e )
2 2 2
s13 s23 sin2
= 4c13
m232 L
4E
2
8c13
c23 s13 s23 (s13
P3+1 (e )
2 2 2 2
= 4c34
c23 s13 c13 sin2

+ c13 s23 )s34 sin
m232L
4E
(10)

m232 L
+ O 2 ,
4E
(11)
2 2
2
+ 4c13 c23 s13 (1 2c13
c23 )c34
s34 sin2
P3+1 ( )

m232L
+ O 2 ,
4E

m232L
4E

m232 L
2 2
2 2
+ O 2 ,
+ 4c13
c34 s34 c23 s23 1 2c13
c23 sin2
4E
4 2 2 2
c23 c34 s23 sin2
= 4c13
(12)
410
P3+1 (e e )

m232 L
4E

2

2

2
2 m32 L
+ O 2 ,
+ 8c13 c23 s13 c13 s13 s34 sin
4E
P3+1 ( )

m232 L
2 2
2 2
= 1 4c13
s23 1 c13
s23 sin2
4E

2

2
2

2
2
2 m32 L
+ O 2 .
+ 8c13 c23 s23 c23 + 1 + 2s13 s23 s34 sin
4E
2 2
= 1 4c13
s13 sin2
(13)
(14)
The three-family limit is plainly reached for and 34 0. We notice that the 34 dependence is generally suppressed with powers of , except in the channel, that is
therefore the most sensitive to this angle. From the above formulae it appears that for small
enough it will be extremely difficult to distinguish a four-family 3 + 1 scheme from three
families looking at e transitions, regardless of the value of 34 .
When the data generated in the 3 + 1 model are fitted in the three-neutrino theory,
the result (if successful) fixes the three-neutrino parameters to some values, that may be
called effective. It will be useful for our discussion to present explicit formulae for some
of them, keeping the small approximation. They can be easily derived from the equations
given above, and are:
eff

= sin(223 ) sin(223 ) 4s34 cos(223) + O 2 ,
sin2 223
(15)
eff

2
2
2
sin 213 = sin(213 ) sin(213 ) 4s34c23 c13 + O .
(16)
Contrary to the 3 + 1 model, the other four-neutrino option, 2 + 2, does not have the
three-neutrino theory as a limit for small gap-crossing mixing angles. In this scheme, some
of the oscillation probabilities have an expression sufficiently simple to be written down
explicitly. In the following formulae, the only approximation made is to neglect the solar
neutrino mass difference m2 and to average the rapid oscillations due to m2LSND :
P2+2 (e )
2 2 2 2
= 2c23
s23 c13 c24

2 2
m243 L
2 2
2
+ sin(434)c24 s24 s23 sin2
c13
c23 sin2 (234) c24
s23 s24
,
4E
P2+2 (e e )
2 2

2 2
2
c24 c24
s23 + s24
= 1 2c23

2
2 2
m243L
2
sin(234) c24
+ 2 cos(234)c24 s24 s23 sin2
s23 s24
,
4E
P2+2 ( )

2

m243 L
2 2
2 2
4 4 2 2
.
c23 s13
+ c13
s23 4c13
c23 c34 s34 sin2
= 1 2c13
(17)
4E
411
The oscillation probabilities involving tau neutrinos may be easily obtained: their expressions are however rather cumbersome, and we omit them here for simplicity.
It may be noted that in the limit of vanishing gap-crossing angles Eq. (17) reduces
to two independent two-neutrinos oscillations, and e s (that is ineffective
since m2 = 0). Therefore, the possibility of confusion with a three-neutrino scheme is
practically inexistent, as we will discuss later.
3. Three or four families?

In a few years from now the LSND results will be confirmed by MiniBooNE or not.
If the former is the case, a short baseline experiment that could explore the LSND gapcrossing parameter space would be mandatory. In case of a non-conclusive result, the
three-family mixing model will be considered the most plausible extension of the Standard
Model. If this is the case, long baseline experiments will be preferred with respect to
the (four-family inspired) short baseline ones. This is the scenario that we would like to
explore, trying to answer the following question: will a neutrino factory and corresponding
detectors (designed to explore the three-family mixing model) be able to tell three neutrinos
from four neutrinos?
We consider the following reference set-up: neutrino beams resulting from the decay
of 2 1020+ s and s per year in a straight section of an E = 50 GeV muon
accumulator. An experiment with a realistic 40 kton detector of magnetized iron and five
years of data taking for each polarity is envisaged. Detailed estimates of the corresponding
expected backgrounds and efficiencies have been included in the analysis [50]. Three
reference baselines are discussed: 732, 3500 and 7332 km. We follow the analysis in energy
bins as made in [28,51,52]. The e channel, the so-called golden channel, will be the
main subject of our investigations.
be the total number of wrong-sign muons detected when the factory is run in
Let Ni,p
polarity p = + , , grouped in energy bins specified by the index i, and three possible
distances, = 1, 2, 3 (corresponding to L = 732 km, L = 3500 km and L = 7332 km,
respectively). In order to simulate a typical experimental situation we generate a set of
data ni,p as follows: for a given value of the oscillation parameters, the expected number
, is computed; taking into account backgrounds and detection efficiencies
of events, Ni,p
and , we then perform a Gaussian (or Poissonian, depending on the number
per bin, bi,p
i,p
of events) smearing to mimic the statistical uncertainty:
ni,p
+ b ) b
Smear(Ni,p
i,p
i,p
i,p
i,p
(18)
Finally, the data are fitted to the theoretical expectation as a function of the neutrino
parameters under study, using a 2 minimization:
2
2
ni,p Ni,p
p
ni,p
(19)
412
where ni,p is the statistical error for ni,p (errors on background and efficiencies
are neglected). We verified that the fitting of theoretical numbers to the smeared
(experimental) ones is able to reproduce the values of the input parameters.
We then proceed to fit with a three-neutrino model the data obtained smearing the
(3 + 1)-neutrino theoretical input. We fixed all but two of the input parameters (in particular
the CP violating phases have been put to zero) and calculated the 2 fitting with the 3
theory for a grid of values of the two varying parameters. In detail, we fixed the values
of m2LSND = 1 eV2 , m2atm = 2.8 103 eV2 , m2 = 1 104 eV2 , 12 = 22.5 and
23 = 45 , we also assumed 14 = 24 = 2 , 5 , 10 , and allowed a variation of 13 from
one to ten degrees in steps of 0.5 and of 34 up to 50 [53] in steps of 1 . The matter
effects have been evaluated assuming an average constant1 density = 2.8 (3.8) g cm3
for L = 732 or 3500 (7332) km and a neutron fraction Yn = 1/2 along the neutrino path.
In order to obtain results in a form that helps intuition, after checking that the conclusions
of a general fit are similar, we kept two of the four parameters in the three-neutrino model
to fixed values,2 12 = 22.5 and 23 = 45 . We allowed the remaining parameters to vary
in the intervals 1 13 10 and 180 < 180 .
In order to better understand the outcome of our fits, we consider the e
conversion probability as a function of the pathlength for a fixed neutrino energy
E = 38 GeV, corresponding to the effective mean energy (including the effect of the
cross section) in a neutrino factory with 50 GeV muons, in both scenarios. We stick
to the simplified approach of Eqs. (10)(14) and assume equal values for the angles
14 = 24 = , after checking that possible sign changes in these quantities do not modify
the general behaviour of our results at the distances considered. In Fig. 2a (Fig. 2b) this
common value is taken to be 2 (5 ). The results for two different values of 34 , 2 and
40 , for 13 = 8 and vanishing CP phases have been plotted in these figures (the wiggly
eff
lines). The smooth lines give the predictions of the three-neutrino model for = 0 and 13
as given by Eq. (16).
It is convenient to write in somewhat greater detail Eq. (10),
P3+1 (e )
2 2 4
= 4s14
s24 c34 sin2
m243 L
4E

2 2 2

m232 L
2
+ 4c13
s13 s23 8c13
c23 s13 s23 (s13 s24 + c13 s23 s14 )s34 sin2
4E

2 2
+ O s14 , s24 , s14 s24 ,
(20)
1 In principle, at distances above 40005000 km a more accurate earth density profile (such as the PREM
[54]) should be used instead of the constant density approximation adopted here. This point has been thoroughly
discussed in [55]. The qualitative features of our results, however, do not depend on this approximation.
2 Note that these parameter values are equal to the input (3 + 1 model) values. This is understandable, since at
the considered distances the effects of m2 = 0 are subleading, and the actual value of 12 is therefore almost
irrelevant; on the other hand, from Eq. (15) it appears that the maximal mixing condition (23 = 45 ) holds in
both theories up to corrections that are of second order in the small quantities 14 , 24 and 13 .
413
Fig. 2. Oscillation probability for e as a function of the pathlength for E = 38 GeV and (a) = 2 or
(b) = 5 . See text for further details.
Fig. 3. Oscillation probability for e as a function of the pathlength for E = 38 GeV and (a) = 2 or
(b) = 5 . See text for further details.
where we explicitly show all the involved angles. We also include the term oscillating
with the shortest length, due to the LSND mass difference, although it is of higher order
in the expansion for small s14 and s24 . For small L this term dominates the transition
probability and it makes the four-neutrino prediction larger than the three-neutrino result,
where this term does not exist. This effect is numerically relevant for large enough gapcrossing angles only, as it may be seen comparing Fig. 2a and Fig. 2b. At intermediate
distances (L 3000 km) the oscillation probabilities in the two models are quite similar:
at this distance we expect therefore that it will be difficult to tell three from four neutrinos.
At larger distances ( L > 5000 km) the distinction will in general be possible for 34 large
enough.
In the case of antineutrinos, a similar pattern is observed. The term oscillating with the
atmospheric mass difference in matter shows a dip at L 8000 km, evident in Fig. 3a. At
this distance the term oscillating with LSND frequency dominates: the average probability
for (3 + 1)-antineutrinos in matter is non-zero for = 5 and 34 = 40 , see Fig. 3b.
We also note that the oscillation probability in the antineutrino case is very small and
this fact, together with the further lowering factor due to the ratio of the fluxes and cross
414
Fig. 4. Plots at 68% in the four-family plane for different baselines and their combinations for 14 = 24 = 2 .
From left to right and top to bottom: = 1, = 2, = 3, = 1 + 2, = 1 + 3, = 2 + 3, = 1 + 2 + 3.
sections, makes the total number of oscillated events essentially dominated by the neutrino
contribution.
We discuss now the fits that we made to data generated in a (3 + 1)-neutrino model
using a three-neutrino theory. The results depend heavily on the values of the small gapcrossing angles 14 and 24 , that we have assumed to be equal for simplicity. If their value
is small (2 ), since the 3 + 1 model has a smooth limit to the three-neutrino theory, the fit
is possible for almost every value of the other parameters. This is shown in Fig. 4, where
in the dark regions of the dalmatian dog hair plot [56] the three-neutrino model is able to
fit at 68% c.l. the data generated with those parameter values. The data include five energy
bins, with muons of both signs, but the first bin in each polarity has been discarded, since
the expected efficiency is very low [50].
415
Fig. 5. Plots at 68% in the four-family plane for different baselines for 14 = 24 = 5 . From left to right and top
to bottom: = 1, = 2, = 3, = 2 + 3.
The reason of the blotted behaviour stands in statistical fluctuations in the smearing
of the input distributions. In the first row the data collected at three different distances
(L = 732, 3500 and 7332 km) are fitted separately: the number of degrees of freedom in
these cases is 6 (the 68% c.l. bound is at 2 = 1.17 6). Notice that the largest distance
gives a somewhat better possibility to tell three from four. In fact, for small values of 34 ,
the e transition probabilities in matter for both schemes are essentially the same
whereas for larger 34 they become different at large distances (see Fig. 2 for E = 38 GeV).
The second row in Fig. 4 shows the combined fit to two different distances ( = 1 and 2,
1 and 3, 2 and 3, respectively). Here the number of degrees of freedom is 14 (the 68% c.l.
bound is at 2 = 1.14 14). Notice that the longest baseline causes a slight improvement
in the discrimination when combined with the other distances. The last plot refers to using
data at = 1, 2 and 3 simultaneously (here we have 22 d.o.f. and the 68% c.l. bound at
2 = 1.12 22).
Increasing the number of energy bins from five to ten (i.e., assuming a detector with
better resolution) for = 2 , we have not observed a sensible reduction of the blotted
regions; we interpret this result as due to the extremely poor statistics per energy bin in this
case. We do not present the corresponding figure.
On the contrary, increasing the value of 14 and 24 , the extension of blotted regions
decreases. This is shown for 14 = 24 = 5 in Fig. 5 for five energy bins. In the first row
we again present a separate fit for each baseline. Notice that for the shortest baseline we
can always tell 3 from 3 + 1. Therefore, in the second row we present only the combination
of = 2 and 3.
416
Fig. 6. Plots at 68% in the four-family plane for different baselines for ten bins and 14 = 24 = 5 . From left to
right and top to bottom: = 1, = 2, = 3, = 2 + 3.
The interpretation of the results shown in Fig. 5 can be easily done following our
previous considerations and Fig. 2b. At the shortest distance, the term in the oscillation
probability varying with the atmospheric mass difference becomes negligible, and the
LSND term dominates: the 3 + 1 model gives a probability larger than the three-neutrino
theory and confusion is not possible. The oscillation probabilities at the intermediate
distance are very similar in the two models, and the confusion is therefore maximal in
this case.
This situation might be improved if a larger number of energy bins could be attained,
in that the different energy dependence in matter would help in the distinction of the two
models. In Fig. 6 we present the dalmatian dog hair plot when the data consist of
ten energy bins. The regions of confusion in this case are considerably reduced. As a
consequence, a fit with the wrong theory to the largest baseline data is only possible for
low ( 10 ) values of 34 .
A further increase of the gap-crossing angles to 14 = 24 = 10 makes the distinction
(even at 95% c.l.) of the two models possible for all values of the other, variable parameters.
A quantitative feeling of the possibility of confusion (or, with a more optimistic attitude,
of discrimination) can be given by the numbers reported in Table 1. They correspond to the
fraction of points, among the 969 data sets fitted, in which the three-neutrino model
can give a good fit (at 68% c.l.). The same points have been smoothly joined to obtain the
dalmatian dog hair plots in Figs. 46. The empty cells in Table 1 correspond to situations
of no possible confusion.
To complete our discussion regarding the 3 + 1 scheme against the three-family theory
we consider now the transition. We notice that the three-family probability is not
recovered for 14 = 24 = 0. In this limit the four-family transition probability equals
417
Table 1
Fraction of data points generated with a four-family theory that can be fitted with a three-neutrino model
=1
=2
=3
=1+2
=1+3
=2+3
=1+2+3
5 bins
= 2
Fig. 4
44%
54%
33%
42%
22%
27%
20%
5 bins
= 5
Fig. 5
54%
32%
26%
10 bins
= 5
Fig. 6
36%
11%
8%
Fig. 7. Plots at 68% c.l. in the four-family plane for different baselines in the channel, for
14 = 24 = 2 . From left to right: = 1, = 2, = 3.
2 , see Eq. (12):
the three-family one times a correction factor c34
2
P3 ( ).
P3+1 ( ) c34
(21)
Due to unitarity, we also have

2
P3+1 ( s ) s34
P3 ( ).
(22)
The region where confusion between the 3 + 1 model and the three-family theory is
possible is that of low 34 . In this case, we considered an idealized detector with constant
efficiency (35%) and a very low level background (105). The mass of the detector is
4 kton and again five energy bins have been assumed. In Fig. 7 we show the 68% c.l.
plots for this channel, where the expected behaviour is clearly observable, for any of the
considered baselines.
Up to this point, we have only discussed in this section the 3 + 1 scheme. We noticed in
the previous section that it is much more difficult to reproduce data generated in the 2 + 2
four neutrino scheme with the three-family theory. We tried to do this in much the same
manner as in the 3 + 1 case. The result can be described giving the numbers corresponding
to the entries of Table 1, that are all smaller than 5%. This maximum value is obtained for
the intermediate distance, L = 3500 km, and corresponds to a narrow region for quite small
418
values of all the gap-crossing angles. Therefore we do not present the relative dalmatian
plots, which in this case are affected by albinism.
4. CP violation vs. more neutrinos

In the previous section, we have extensively illustrated when confusion between the
three-neutrino and the (3 + 1)-neutrino models is possible. However, we have not presented
in detail the results of a fit of the four-family data in the parameter space of the threefamily model: in fact, the dalmatian plots in the previous section represent the regions of
confusion in four-family parameter space.
In three families, the free parameters that we considered are one angle, 13 , and the CP
violating phase, . These parameters are expected to be still unknown (or poorly known)
when the neutrino factory will be operative. In particular, we are interested here in the
following questions:
Is it possible to find a non-vanishing CP violating phase when fitting with the threefamily model the 3 + 1 (CP conserving) data?
Is it possible to fit with a 3 + 1 (CP conserving) theory the data if their fit in the
three-neutrino model gives a large CP violating phase ?
In order to answer the first question, we present in Fig. 8 the 68%, 90% and 99%
confidence level contours3 for typical three-neutrino fits (one for each distance) to 3 + 1
neutrino data generated with 14 = 24 = 2 and five energy bins. The parameters are
best determined for the intermediate distance, 3500 km, that is close to the optimal distance
for studies of CP violation in the three-neutrino scenario [28,5759]. In fact, the value of
the fitted phase stays close to zero, except for a possible mirror region around 160 .
The existence of such mirror regions has been discussed in [52,59]: they may be removed
increasing the energy resolution [60] and/or combining data at two different distances. At
Fig. 8. Confidence level contours for typical points where the three-neutrino theory can well reproduce
(3 + 1)-neutrino data at the three distances studied: from left to right, L = 732, 3500 and 7332 km.
3 The c.l. contours correspond to 2 = 2.30, 4.61, 9.21, respectively, since there are two free parameters.
The best fit points are represented by stars.
419
Fig. 9. Examples of regions in the (13 , ) plane where the (3 + 1)-neutrino theory with vanishing CP violating
phases can reproduce at 90% c.l. three-neutrino data. From left to right and top to bottom: = 1, = 2, = 3,
= 1 + 2, = 1 + 3, = 2 + 3.
the largest distance studied, 7332 km, the uncertainty in the value of is higher, since the
matter effects dominate over the possible effect of CP violation in the oscillations and
is less easily determined. Finally, at the shortest distance and at 99% c.l. any value of is
allowed. Plots at this distance with a similar behaviour have been also presented in [52]:
this distance is clearly not suited for a precise determination of the CP violating phase,
at least for the set-up of neutrino factory and detector considered here. To illustrate the
representativeness of the values chosen for the plots in Fig. 8 we note that, out of about
6000 successful fits, 31% (50%, 37%) for = 1 (2, 3) give for the CP violating phase a
value 15 < < 15 .
In order to answer the second question, and as a cross-check of our previous results, we
adopted the converse procedure, namely we fitted data generated in a three-neutrino, CP
violating, scenario with the formulae of the 3 + 1 model with CP violating phases fixed
to zero. We chose for this purpose to vary the three-neutrino parameters in a region where
the confusion is often not possible in the previous fit, 1 13 10 and 60 120 .
This region has only a 3% (< 0.02%, 1.5%) of successful fits falling in it for = 1 (2,
3). Among the parameters in the 3 + 1 theory, 13 and 34 have been left free to vary,
while 14 = 24 = have been kept fixed to 2, 5 and 10 . The corresponding plots for five
energy bins and for = 2 are presented in Fig. 9, where the dark regions correspond to an
acceptable fit (at 90% c.l.) in the 3 + 1 theory.
The results are precisely what was expected on the basis of the previous fits. The fit
with the wrong model, with the chosen values of parameters, is possible in many a case for
420
L = 732 km, it is very difficult to obtain for the intermediate distance, L = 3500 km, and
at the largest L an intermediate situation holds. Fitting simultaneously data at two different
distances the possibility of confusion is strongly reduced and in fact it vanishes if the data
at intermediate distance are used in any combination with the others. We also observed that
no confusion would be possible assuming the larger values = 5 or 10 .
Our result seems to imply that if the data would point to a maximal CP violating phase
in the three-neutrino theory, it would be very difficult to describe them in a theory without
CP violation, even if with more neutrinos.
5. Conclusions
The present experimental data for solar and atmospheric neutrinos give quite strong
indications in favour of neutrino oscillations and of non-zero neutrino masses. In addition,
the results of the LSND experiment would imply the existence of a puzzling fourth, sterile,
neutrino state. The comprehensive analysis of all the other data (including the recent SNO
results) seems however to disfavour both possible classes of four-neutrino spectra, the socalled 2 + 2 and 3 + 1 schemes.
An experimental set-up with the ambitious goal of precision measurement of the
whole three-neutrino mixing parameter space (including the extremely important quest
for leptonic CP violation) is under study. This experimental programme consists of the
development of a neutrino factory (high-energy muons decaying in the straight section
of a storage ring and producing a very pure and intense neutrino beam) and of suitably
optimized detectors located very far away from the neutrino source. The effort to prepare
such very long baseline neutrino experiments will require a time period covering this and
the beginning of the following decade, in the absence of a conclusive confirmation of the
LSND results (that would call instead for a short baseline experiment to investigate the
four-neutrino mixing parameter space).
It is of interest to explore the capability of such an experimental set-up to discriminate
between a three-neutrino model and a possible scenario where a fourth neutrino is included.
Some relevant questions came to our mind. Can the experimental data be described equally
well with a three-neutrino theory or a four-neutrino model? Even worse, is it possible for
a CP conserving three active and one sterile neutrino model to be confused with a CP
violating three (active) neutrino theory?
In this paper we tried to answer these questions in the framework of an LSND inspired
four neutrino model, assuming a large squared mass difference m2 1 eV2 . However,
the approach is in principle independent of the choice of the third mass gap (in addition to
the solar and atmospheric ones) and can be easily repeated for different values of m2 .
The results of our analysis can be summarized as follows. Data that could be described
in four-family without CP violation can also be fitted with the three-family formulae in
some particular zones of the four-family parameter space, not restricted to the obvious case
of very small angles. These zones are reduced in size for increasing gap-crossing angles
and energy resolution of the detector. The use of oscillations through direct or
indirect detection of tau lepton decays would allow a much easier discrimination of the
two theoretical hypotheses.
421
In the zones of the four-family parameter space that can be also described with the
three-family theory, the fitted value of the CP violating phase, , is generally not large. In
particular, this is true for L = 3500 km, whereas for L = 7332 km the determination of
is somewhat looser. For the shortest baseline, L = 732 km, we have the largest spread in
the values of , although the most probable value is still close to zero.
Data that can be fitted with a CP phase close to 90 in the three-neutrino theory cannot
be described in a CP conserving 3 + 1 theory, provided that data at two different distances
are used.
Finally, in the 2 + 2 scheme (as opposed to 3 + 1) the ambiguity with a three-neutrino
theory is essentially absent.
Acknowledgements
We acknowledge useful conversations with M.B. Gavela, P. Hernandez, C. Pea-Garay
and J. Sato. We are particularly indebted with P. Lipari for discussions on many different
aspects of this paper.
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]
B. Pontecorvo, Sov. Phys. JETP 6 (1957) 429, Zh. Eksp. Teor. Fiz. 33 (1957) 549 (in Russian).
Z. Maki, M. Nakagawa, S. Sakata, Prog. Theor. Phys. 28 (1962) 870.
B. Pontecorvo, Sov. Phys. JETP 26 (1968) 984.
V.N. Gribov, B. Pontecorvo, Phys. Lett. B 28 (1969) 493.
B.T. Cleveland et al., Astrophys. J. 496 (1998) 505.
Y. Fukuda et al., Kamiokande Collaboration, Phys. Rev. Lett. 77 (1996) 1683.
W. Hampel et al., GALLEX Collaboration, Phys. Lett. B 447 (1999) 127.
J.N. Abdurashitov et al., SAGE Collaboration, Phys. Rev. C 60 (1999) 055801.
Y. Suzuki, Super-Kamiokande Collaboration, Nucl. Phys. Proc. Suppl. 77 (1999) 35.
Q.R. Ahmad et al., SNO Collaboration, Phys. Rev. Lett. 87 (2001) 071301, nucl-ex/0106015.
Y. Fukuda et al., Kamiokande Collaboration, Phys. Lett. B 335 (1994) 237.
R. Becker-Szendy et al., Nucl. Phys. Proc. Suppl. 38 (1995) 331.
Y. Fukuda et al., Super-Kamiokande Collaboration, Phys. Rev. Lett. 82 (1999) 2644.
W.W. Allison et al., Soudan-2 Collaboration, Phys. Lett. B 449 (1999) 137.
M. Ambrosio et al., MACRO Collaboration, Phys. Lett. B 434 (1998) 451.
T. Toshito, Super-Kamiokande Collaboration, hep-ex/0105023.
C. Athanassopoulos et al., LSND Collaboration, Phys. Rev. Lett. 81 (1998) 1774.
A. Aguilar, LSND Collaboration, Phys. Rev. D 64 (2001) 112007, hep-ex/0104049.
J. Kleinfeller, KARMEN Collaboration, Nucl. Phys. Proc. Suppl. 87 (2000) 281.
E. Church et al., BooNe Collaboration, nucl-ex/9706011.
C. Caso, A. Gurtu, Eur. Phys. J. C 15 (2000) 256.
G.L. Fogli, E. Lisi, A. Marrone, G. Scioscia, hep-ph/9906450.
S. Geer, Phys. Rev. D 57 (1998) 6989;
S. Geer, Phys. Rev. D 59 (1998) 039903 (Erratum), hep-ph/9712290.
A. De Rujula, M.B. Gavela, P. Hernandez, Nucl. Phys. B 547 (1999) 21.
V. Barger, S. Geer, K. Whisnant, Phys. Rev. D 61 (2000) 053004, hep-ph/9906487.
A. Bueno, M. Campanelli, A. Rubbia, Nucl. Phys. B 573 (2000) 27.
K. Dick, M. Freund, M. Lindner, A. Romanino, Nucl. Phys. B 562 (1999) 29, hep-ph/9903308.
A. Cervera et al., Nucl. Phys. B 579 (2000) 17.
422
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
C. Albright et al., hep-ex/0008064.

A. Blondel et al., Nucl. Instrum. Methods A 451 (2000) 102.
V. Barger et al., hep-ph/0103052.
A. Donini, M.B. Gavela, P. Hernandez, S. Rigolin, Nucl. Phys. 574 (2000) 23;
A. Donini, M.B. Gavela, P. Hernandez, S. Rigolin, Nucl. Instrum. Methods A 451 (2000) 58, hepph/9910516.
A. Donini, D. Meloni, Eur. Phys. J. C 22 (2001) 179, hep-ph/0105089.
S.M. Bilenkii, C. Giunti, W. Grimus, T. Schwetz, Phys. Rev. D 60 (1999) 073007.
F. Dydak et al., Phys. Lett. B 134 (1984) 281.
I.E. Stockdale et al., Z. Phys. C 27 (1985) 53.
Y. Declais et al., Nucl. Phys. B 434 (1995) 503.
E. Eskut et al., CHORUS Collaboration, Phys. Lett. B 497 (2001) 8.
P. Astier et al., NOMAD Collaboration, Nucl. Phys. B 611 (2001) 3, hep-ex/0106102.
V. Barger, B. Kayser, J. Learned, T. Weiler, K. Whisnant, Phys. Lett. B 489 (2000) 345, hep-ph/0008019.
O. Yasuda, hep-ph/0102166.
C. Giunti, M. Laveder, JHEP 0102 (2001) 001, hep-ph/0010009.
O.L. Peres, A.Y. Smirnov, Nucl. Phys. B 599 (2001) 3, hep-ph/0011054.
W. Grimus, T. Schwetz, Eur. Phys. J. C 20 (2001) 1, hep-ph/0102252.
M.C. Gonzalez-Garcia, C. Pena-Garay, Phys. Rev. D 63 (2001) 073013, hep-ph/0011245.
M.C. Gonzalez-Garcia, M. Maltoni, C. Pena-Garay, Phys. Rev. D 64 (2001) 093001, hep-ph/0105269.
V. Barger, D. Marfatia, K. Whisnant, Phys. Rev. Lett. 88 (2002) 011302, hep-ph/0106207.
J.N. Bahcall, M.C. Gonzalez-Garcia, C. Pena-Garay, JHEP 0108 (2001) 014, hep-ph/0106258.
A. De Rujula, M. Lusignoli, L. Maiani, S.T. Petcov, R. Petronzio, Nucl. Phys. B 168 (1980) 54.
A. Cervera, F. Dydak, J. Gomez-Cadenas, Nucl. Instrum. Methods A 451 (2000) 123.
A. Donini, O. Mena, S. Rigolin, Nucl. Instrum. Methods A 472 (2000) 403, hep-ph/0007281.
J. Burguet-Castell, M.B. Gavela, J.J. Gomez-Cadenas, P. Hernandez, O. Mena, Nucl. Phys. B 608 (2001)
301, hep-ph/0103258.
G.L. Fogli, E. Lisi, A. Marrone, Phys. Rev. D 63 (2001) 053008, hep-ph/0009299.
A.M. Dziewonski, D.L. Anderson, Phys. Earth Planet. Inter. 25 (1981) 297.
T. Ota, J. Sato, Phys. Rev. D 63 (2001) 093004, hep-ph/0011234.
W. Disney et al., One Hundred and One Dalmatians, 1961.
V. Barger, S. Geer, R. Raja, K. Whisnant, Phys. Rev. D 62 (2000) 073002, hep-ph/0003184.
A. Bueno, M. Campanelli, A. Rubbia, Nucl. Phys. B 589 (2000) 577, hep-ph/0005007.
M. Freund, P. Huber, M. Lindner, Nucl. Phys. B 615 (2001) 331, hep-ph/0105071.
A. Rubbia, hep-ph/0106088.
Nuclear Physics B 624 [PM] (2002) 431451

LundRegge vortex strings in terms

of Weierstrass elliptic functions
H.J. Shin
Department of Physics and Research Institute of Basic Science, Kyung Hee University,
Seoul 130-701, South Korea
Abstract
We study quasi-periodic solutions of the LundRegge model in terms of the elliptic functions of
Weierstrass. They describe the Kida-class motions of relativistic strings in an external antisymmetric
tensor field. Our solution includes various vortex string shapes such as the closed vortex ring, the
helicoidal filament, the generalized Zees solution and the Hasimoto type 1-soliton filament. 2002
PACS: 47.32.Cc; 02.30.Gp; 67.57.Fg
1. Introduction
The dynamics of a string coupled to an antisymmetric tensor field was first studied
by Kalb and Ramond in the context of string theory [1]. It also appears in the study
of the relativistic motion of vortex strings in a superfluid by Lund and Regge [24]. In
particular, Lund and Regge has proven the integrability of the model by recognizing the
vortex equation as the integrability equation of Gauss and Codazzi. In this scheme the
vortex equation is identified with the complex sine-Gordon equation. This approach was
further developed by Sym et al., leading to the theory of soliton surfaces of LundRegge
PohlmeyerGetmanov system [5]. These studies [2,5] also found vortex configurations
corresponding to the N -soliton solutions of the complex sine-Gordon equation. More
recently, using a simple ansatz, Zee found a generalized smoke ring solution of the vortex
equation [6]. He also point out the importance of the problem of analyzing the general
motion of a string in a background field.
By making use of exact solutions of the complex sine-Gordon equation, various string
motions are in fact constructed in Refs. [7,8] using the duality between the LundRegge
E-mail address: hjshin@khu.ac.kr (H.J. Shin).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 4 5 - 9
432
H.J. Shin / Nuclear Physics B 624 [PM] (2002) 431451
model and the complex sine-Gordon equation. Interestingly they have some analogies with
the vortex filament configurations in fluid dynamics such as the Kelvin wave, the smoke
ring, and the Hasimoto soliton [912]. In addition to these, they also found a new breathertype vortex solution. But it was found that their smoke ring solutions are only a special
case of those of Zee. Motivated by this fact, we here study more general string motion,
which includes the known solutions of [7]. In addition, it contains the smoke ring solution
of Zee. There indeed exists interesting solutions of string motion whose existence is based
on deep structure of integrable theories.
The main ingredient of the present work is the finite-band integration method, which
gives periodic solutions of integrable models [13,14]. In fact, Date [15] have studied
periodic solutions of the complex sine-Gordon equation. But the use of duality between
the LundRegge model and the complex sine-Gordon equation to calculate the periodic
solutions of the LundRegge model is quite difficult, even the case of N -soliton solutions
seems not to be realistic [16]. Moreover the result of Date is incomplete because of the
absence of a constraint condition on the auxiliary variables which appear in the solution
[17,18]. In this paper we deal with the LundRegge model directly rather than the complex
sine-Gordon equation and obtain (quasi-)periodic vortex string configurations of the Lund
Regge model. For this, we first derive the Lax pair of the LundRegge model, which is
explained in Appendix A. Then the general setup for the periodic problem of the Lund
Regge model is given in terms of the squared eigenfunctions in Section 2. Explicit
construction of the 1-phase periodic solution is followed using a modified version of
the finite integration method in Section 3. The resulting formulae involve Weierstrass
elliptic functions. In Section 4, various string configurations from the periodic solutions
are explained in terms of functional relations of the Weierstrass functions. Especially
the reduction of the solution to typical string motions is achieved. Section 5 gives the
discussion.
2. The LundRegge model

Lund and Regge have shown that the relativistic motion of strings in a uniform static
field H is described by the following equation; in a Lorentz frame in which X0 = , the
system is governed by the gauge fixed equation of motion,

2 2 Xi + Hij k Xj Xk = 0, i = 1, 2, 3,
(1)
where Hij k = ij k with the string tension and , are local coordinates on the
string world-sheet. Here, Xi (, ); i = 1, 2, 3 are the vortex coordinates which satisfy the
quadratic constraints:

2
2

Xi + Xi = 1,
(2)
Xi Xi = 0.
In the no-coupling limit ( = 0), this equation describes the transverse modes of the
4-dimensional NambuGoto string in the orthonormal gauge.
The LundRegge vortex equation can be equally described in a matrix form which is
more suitable for the dualization procedure with the complex sine-Gordon equation. Let us
433
define a matrix F by
F i
3

Xi i ,
4
(3)
i=1
where i are Pauli matrices. Then it is easy to verify that the matrix F satisfies the
following equation of motion;
],
F
= [F, F
(4)

, z , z 18 ( + ), z 8 ( ). The constraint equation (2) can
where z
also be rewritten with F as
Tr(F )2 = 2 2 ,
)2 = 2.
Tr(F
In other words, Xi are given by

1
4i
Xi = Fi z = ( + ), z = ( ) .
8
8
(5)
(6)
Thus, the vortex coordinate Xi can be obtained explicitly by solving the equation of F ,
which we now begin research.
First we notice that the equation of motion (4) can be rewritten as a zero curvature
condition

1
1 F
+ (1 )F, +
(7)
= 0,
where is an arbitrary spectral parameter. In turn, this zero curvature condition can be
understood as the integrability condition of the linear system
( + U ) = 0,
( + V ) = 0,
(8)
where

3 +
,
U = (1 )F i(1 )
3

1
1
3 +
1 F
1 i
V=
,
3
(9)
= , = , = , = . The equation of motion (4) is rewritten as

where +
3
+
3
3
3 = i3 = + + ,
i
= i = 23 2 3 ,
i
(10)
while the constraint (5) becomes 32 + + = 2 , 32 + + = 1. (Do not confuse the

coupling constant with the field variables 3 , .) We obtain the linear equation (8) using
the so-called R-transformation [19], which is explained in Appendix A. As we will see in
the next section, this form is well-suited for the finite-band integration method.
434
3. FiniteBand integration method

In mathematical literature the problem of finding the periodic solutions of integrable
equations has been considered for a long time. It was solved first for the case of the
Kortewegde Vries equation and was extended to the case of sine-Gordon equation
and nonlinear Schrdinger equation later on [13]. Here we use the simplest method for
generating the periodic solutions using the method of the squared eigenfunctions [14,18].
We start with the linear equation (8). Let the systems (8) have two basic solutions,
(1 , 2 ) and (1 , 2 ), which is used to build a vector such that
f = (i/2)(1 2 + 2 1 ),
g = 1 1 ,
h = 2 2 .
(11)
Using this definition and the linear equation (8), it can be explicitly checked that P ()
f 2 gh is independent of z and z and is only the function of . The N -phase periodic
solution is obtained by specially taking the form of
P () = f 2 gh = 2
2N+2

( i ),
(12)
i=1
where i are zeros of the polynomial which characterize the periodic solution of the
Bloch wave problem. The zeros i have to consist of complex conjugate pairs j =
Rj + iIj , j +N+1 = Rj iIj , j = 1, N + 1 such that the obtained solutions i , i =
1, 3 are real. It can be seen that this form is consistent with the following form of f, g
and h,
f = 3 N+1 +
N
i fi ,
g = i+ ( 1)
i=0
h = i ( 1)
N

( i ),
i=1
N

i ,
(13)
i=1
where i are functions of z, z . From the definition of f, g and h in Eq. (11), we can obtain
the following equations:
f = ( 1) g ( 1)+ h,
g = 2( 1)+ f + 2i( 1)3 g,

= 1 1 g 1 1 + h,
f

1
1
= 2 1
+ f + 2i 1
3 g.
g
(14)
Equations for h are similarly given.

If we evaluate Eq. (12) at m , which is one of the zeros of g, then we arrive at the
identity

f (z, z ; m ) = m P (m ),
(15)
435
where m indicates the sheet of the Riemann surface where the complex m lies on. Using
Eqs. (13) and (14), we can evaluate g at one of the zeros of g to find

m = 2im P (m )
(16)
(m i ), m = 1, . . . , N.
i=m
N+1
term for f in Eq. (14) we find

3 = i+
i i = i 2 32
i i .
If we look at the
(17)
On the other hand, O(1 ) term of Eq. (14) gives

+
+ f0 = 3 + (i ).
i = +
i ,
Using these results, O(0 ) term of P () gives

+
f02 + +
|i |2 =
|i |2 = 2 (i ).
+
(18)
(19)
Thus we obtain (We assume > 0.)

+

(20)
i =
i =
i .
+
and f
in Eq. (14) and using the results in Eq. (20), we
Applying the same approach to g
can show

m = 2i 1 + m P (m )
(m i )
m +
i=m

1
i
m P (m )
= 2i
(m i ),
m
i
i=m

3 = i( + + ) = i 1
i + .
i
i
(21)
(22)
The remaining problem to solve Eqs. (16)(22) is beautifully accomplished by using the
solution of what is called Jacobis inversion problem. We will not go into details here but
sketch the broad outlines and occurring difficulties of the procedure. The following steps
are needed for solving above problem. First, take the parameters i , i = 1, . . . , 2N + 2,
as known. Second, choose initial conditions i (0, 0) and 3 (0, 0) such that they satisfy
Eq. (12). It is at this point that a technical difficulty arises. An effective method to avoid
this problem was introduced by Kamchatnov [18] to obtain the 1-phase solution. We adopt
this method in the following section to obtain an explicit periodic solution of the string
configuration.
4. 1-phase periodic solution

As explained in the previous section, in the sine-Gordon and NLS equations cases the
additional problem of extraction of the real solutions arises. The periodic solutions
436
obtained by this method have rather complicated form, which prevents their application
to real physical situation. Kamchatnov devised a simple modification of the finite-band
integration method which solves the effectivization problem in the simplest and the
important one-phase case. In this section we follow his method to find the periodic
solutions of the LundRegge model. The one-phase periodic solution is obtained by taking
the simplest nontrivial N = 1 case in Eqs. (12) and (13) which gives
P () = 2
4

( i ),
i=1
g = i+ ( 1)( ),
f = 3 2 + f1 + f0 ,

h = i ( 1) .
(23)
And Eq. (20) becomes

V
3
+
1
=
=
= .
f0
+
i
(24)
The other equations from Eq. (14) for N = 1 case can be expressed as

1
3 = f1 = f0 = i+ ,
= 2i 3 2 + f1 + f0 ,
2
1 = V f1 ,
0 = V f0 ,
= V ,
3 = V 3 ,
f
f
+ = V (2i+ 3 + 2i+ f0 ).
+ = 2i+ f1 2i3 + ( + 1),
(25)
It can be explicitly checked that Eqs. (24) and (25) are consistent with each other.
+ , 3 = (Vf0 ) =
3 . Note that
Especially we can check that + = (V + ) =
3 , f1 , f0 and are functions of W z + V z only.
4.1. , f1 , f0 in terms of 3 , si
To prevent the effectivization problem, we start with Eqs. (12) or (23) to solve , f1
and f0 . First, we introduce constants of motion si , i = 1, 4 which are defined
P () = f 2 gh = 2 4 s1 3 + s2 2 s3 + s4 .
(26)
Inserting f, g and h in Eq. (23) into Eq. (26), we can obtain

s1 2
i = 2f1 3 + 2 2 32 + X,

s2 2
i j = 2f0 3 + f12 + 2 32 + 2X + Y,
i<j
s3
i j k = 2f1 f0 + X + 2Y,
i<j <k
s4 2
i = f02 + Y,
(27)
where X + ( + ) = ( 2 32 )( + ) and Y + . Now by solving for

f0 , f1 , and in Eq. (27), we obtain (This is one particular solution, but other solutions
offer essentially the same result. To obtain this solution we use M ATHEMATICA for
437
symbolic manipulation. In addition, M ATHEMATICA was used to check various formulae

appeared in this paper. For example, it was used to check that Eqs. (33) and (50) satisfies
the equation of motion (4).)
4 2 3s1 + 2s2 s3
2 2 s1 + s3 2s4
,
f0 = 3 +
,
(28)
2 A
2 A

1
1
4 2 3s1 + 2s2 s3
2
2
3 + s1 2 23 3/2 R ,
, =
A
2( 2 32 )
A
(29)
f1 = 23
where A = 2 s1 + s2 s3 + s4 and
R = A2 (x x1 )(x x2 )(x x3 ),
x1 = 2 (2 + 1 + 2 + 3 + 4 2 3 1 4 ),
x2 = 2 (2 + 1 + 2 + 3 + 4 1 3 2 4 ),
x3 = 2 (2 + 1 + 2 + 3 + 4 1 2 3 4 ),
x = 2 A 3 .
(30)
Note that this solution is consistent with Eq. (25).

4.2. Derivation of F3
Using these results, Eq. (25) for 3 becomes
3
d3
i
= iA 2 R =
(x x1 )(x x2 )(x x3 ) .
dW
A
(31)
This equation can be integrated in terms of Weierstrass P(u, g2 , g3 ) function. As far

as Weierstrass elliptic functions are involved, we employ terminology and notation of
Ref. [20] without further explanations. Explicitly

x1 + x2 + x3
1
P(W + w3 , g2 , g3 ) ,
3 =
(32)
3
2 A
8 3
where w3 is an integration constant and g2 = 43 s22 4s1 s3 + 16 2s4 , g3 = 27
s2 43 s1 s2 s3 +
2
2
4 2 s32 32
3 s2 s4 + 4s1 s4 . The integration constant w3 is determined by the initial
condition, which we shall choose as follows; P(w3 ) = e3 at W = 0, where e3 is the
smallest root of the equation 4z3 g2 z g3 = 0. (Other two roots are denoted by e1
and e2 with e1 > e2 > e3 . w3 as well as w1 are called the half period of the P function.
They satisfy P(w1 ) = e1 , P(w2 ) = e2 , e1 + e2 + e3 = w1 + w2 + w3 = 0.) This condition
3 , we can
guarantees |3 | . Now using 3 = iF3 and 3 = Vf0 = 1s f0 = i F
4
obtain

2 2 s1 + s3 2s4
z
F3 = i 3 dW + i
2 As4

x1 + x2 + x3
i
1
= (W + w3 , g2 , g3 ) +
z + z
3
s4
2 A
438

2 2 s1 + s3 2s4
+
z 3
s4

i
s2
= (W + w3 , g2 , g3 ) + 2 2 + s1
z
3
2 A

z
s2
+ + s3 2s4 3 ,
3
s4
(33)
where (u, g2 , g3 ) is the Weierstrass zeta function and the integration constant 3 is taken
as (w3 , g2 , g3 ). (We will use the notation i = (wi , g2 , g3 ), i = 1, 2, 3 in the following.)
4.3. Derivation of +
Now we try to obtain F+ of the LundRegge equation. Using Eqs. (28) and (29), the last
two equations of (25) for + becomes

4 2 3s1 + 2s2 s3
+ = 2i+ 3 +
3 ,
2 A

2
+ s3 2s4
+ = 1 2i+ 3 + 2 s1
3 .
(34)
s4
2 A
They give the solution

z
+ = exp 2 A i
+ (W ),
s4
(35)
where + satisfies the following differential equation,

d +
4 2 3s1 + 2s2 s3
(36)
3 .
= 2i + 3 +
dW
2 A
The following identity, which can be obtained using Eqs. (29) and (31), is required to
integrate + :

4 2 3s1 + 2s2 s3
2i 3 +
3
2 A

2
i
4 2 3s1 + 2s2 s3 2
1 d( 2 32 )
s
+
4
+
1
3
2 ( 2 32 ) dW
2 32
A
= iM1
1
1
1 d( 2 32 )
,
+ iM2
+
+ 3
3 2 ( 2 32 ) dW
(37)
where
4 2 3s1 + 2s2 s3
4 2 s1
(38)
.
2
2 A
Now using equations (32), (36), (37) and the following identities of the Weierstrass elliptic
function:

1
( W )
dW
ln
=
+ 2()W ,
(39)
P(W ) P() P ()
( + W )
M1 , M2 =
439
and
P(W ) P() =
(W + ) (W )
,
2 (W ) 2 ()
(40)
we can obtain

i (W + w3 + 1 ) (W + w3 + 2 )
+ =
)W
(
)W
+
c
, (41)
exp
(
1
2
(1 ) (2 ) 2 (W + w3 )
2 A
where two constants 1 , 2 are defined by the following two relations,
P(1 ), P(2 ) = (x1 + x2 + x3 )/3 2 A ,

d3
R

P (1 ), P (2 ) = 2 A
= 2i
= 4 A iM1,2.

dW
A
3 =
(42)
(43)
3 =
We fix the integration constant c by requiring 32 + |+ |2 = 2 . Especially at W = 0, we

find

1
1
x1 + x2 + x3
P(1 ) + P(2 )
e3 =
e3 ,
3 (0) =
3
2
2 A
2 A
i (1 + w3 ) (2 + w3 )
exp(c)
+ (0) =
2 A (1 ) (2 ) 2 (w3 )

1
=
(44)
(P(1 ) e3 )(P(2 ) + e3 ) exp (w3 )(1 + 2 ) + c ,
2 A
where we have used the identity
P(u) e3 =

2 (u + w3 )
exp 2(w3 )u .
2 (u) 2 (w3 )
(45)
Using these relations together with Eq. (42), we can obtain c = 3 (1 + 2 ).

4.4. Derivation of F+
We now derive F+ . Using F+ = i+ in Eq. (9), we obtain

z
F+ = i + dz + M(z) = i exp 2 A i
+ dW + M(z),
s4
(46)
where M(z) is a function to be determined. To evaluate the integration explicitly and to

determine M(z), we substitute Eq. (46) into the following equation,
+
+ = + =
i F
.
s4
Then a little algebra with Eqs. (29), (31), (32), (35), (41)(43) gives M(z) = 0 and

2 A i + dW

1
4 2 3s1 + 2s2 s3
R
2
3 + s1 4 +
=
+
2+
A3
A
(47)
440

i P (W + w3 ) P (1 ) P (W + w3 ) P (2 )
=
+
4 P(W + w3 ) P(1 )
P(W + w3 ) P(2 )

1
= (W + w3 + 1 ) (W + w3 + 2 ) (1 ) + (2 )
4 A
(W + w3 + 1 ) (W + w3 + 2 )
(1 ) (2 ) 2 (W + w3 )

exp (1 )W (2 )W 3 (1 + 2 ) ,
(48)
where we use the following relation in the last part of derivation,

P (W ) P ()
= 2 (W + ) 2(W ) 2().
P(W ) P()
(49)
Then collecting all these results, we finally obtain

1
(W + w3 + 1 ) (W + w3 + 2 ) (1 ) + (2 )
8A
(W + w3 + 1 ) (W + w3 + 2 )
(1 ) (2 ) 2 (W + w3 )

z
exp (1 )W (2 )W 3 (1 + 2 ) 2 A i
.
s4
F+ =
(50)
5. Special cases
In this section we study some special cases of the obtained solution by taking degenerate
limit of i values. It contains the straight line, the Kelvin wave and the Hasimoto soliton
solution which are obtained previously, as well as some new configurations such as the
generalized Zees solution and the closed ring.
5.1. Straight line
The simplest solution that corresponds to the straight line is obtained by taking 1 =
3 = , 2 = 4 = or more simply 1 = 2 = 3 = 4 = (real). In this case, g2 =
g3 = e3 = 0, x1 = x2 = x3 = 2 2 (1 )2 and A = 2 (1 )4 . The Weierstrass functions
in this limit are given by P(u, 0, 0) = 1/u2 , (u, 0, 0) = 1/u, (u, 0, 0) = u. And w3 =
1 = . Using these relations, we can easily obtain F3 = iz i z , F+ = F1 iF2 = 0.
The rotational and translational symmetry of the vortex equation can give more general
configuration of the straight line [7].
5.2. Kelvin wave or Smoke ring
The next example that is known as the Kelvin wave in the nonrelativistic vortex motion
of fluid mechanics corresponds to taking 1 = 3 = , 2 = + i , 4 = i (0 <
8 6 6
< 1, 0 < ). In this case g2 = 43 4 4 , g3 = 27
, g23 27g32 = 0, and the
441
Weierstrass functions are given in a simple form;

P(u) = 2 2 /3 + 2 2 csc2 ( u),
(u) = 2 2 u/3 + cot( u),

(u) = exp 2 2 u2 /6 sin( u)/( ).
And e3 = 2 2 /3, w3 = i,

sin 1 , sin 2 =
(1 ) (1 )2 + 2 (1 )

i
cot 1 , cot 2 =
(1 )2 + 2 (1 ) .
(51)
,
(52)
(53)
Now, a straightforward calculation gives

2 + 2

z+i
z ,
(1 )2 + 2
(1 )2 + 2 2 + 2
F+ =
2{(1 )2 + 2 }

(1 )2 + 2
2
2
exp 2i (1 ) + z + 2i
z .
2 + 2
F3 = i
(54)
Using the fact that F+ = F1 iF2 C exp(iD), we can obtain F1 = (F+ F+ )/2 =
iC sin D, F2 = i(F+ + F+ )/2 = iC cos D. (Remember the relation between Fi and Xi in
Eq. (6). Especially Fi , i = 1, 2, 3 are pure imaginary.) Which describes a smoke ring. It is
a circular ring having constant radius and traveling along the z-axis. 1
5.3. Hasimoto 1-soliton
To obtain the 1-soliton solution from our periodic solution, we take 1 = 2 = + i ,
512 6 6
4 4
3 = 4 = i in Eqs. (33) and (50). In this case g2 = 64
3 , g3 = 27
2
2
and w3 = i/(4 ), e3 = P(w3 ) = 8 /3, 3 = (w3 ) = i /3, (w3 ) =
i exp( 2 /24)/(2 ). The Weierstrass functions are given by
P(W + w3 ) = 8 2 2 /3 + 4 2 2 tanh2 (2 W ),
(W + w3 ) = 4 2 2 W/3 + 2 tanh(2 W ) + 3 ,

(W + w3 ) = exp 2 2 2 W 2 /3 + 3 W cosh(2 W ) (w3 ).
(55)

And 1 = , sinh 2 2 = i / (1 )2 + 2 , coth 2 2 = i(1 )/ . It then
gives

2 z
tanh 2 z + 2
,
F3 = i z + z
(56)
( 1)2 + 2
+2
1 By using (1 ) A, B, /4,

2 + 2 2 , we can reproduce the results given in [7].
442
and

F+ =
sech(2 W ) exp 2i(1 )z 2i 1 2
z .
(1 )2 + 2
+2
(57)
This solution can be compared with the known form in [7] by taking 1 = 2 = 3 = 4
iD exp(i )/ .
5.4. Generalized form of Zees solution
This case corresponds to a circular ring configuration of varying radius with time, which
is described by X1 = a( ) cos(b ), X2 = a( ) sin(b ), X3 = c( ). It was considered by
Zee for the case of b = 1, but without explicit form of the solution [6]. This configuration
is interesting as it can be easily observed in an experiment. We here analyze detailed
characteristics of this motion using the obtained solution.
First, note that the radius of F+ is a function of W = {( + ) + ( )/ s4 }/8.

Thus it becomes a function of only by taking 2 = s4 , that is 1 = 3 = rei , 2 = 4 =
ei /r. Similarly X3 (or F3 ) becomes -dependent only, by taking 2 2 s1 + s3 2s4 = 0.
It then requires r = 1 or = . In the following, we study each case separately.
5.4.1. Case A: r = 1
In this case, we take 1 = 3 = ei , 2 = 4 = ei . Then

g2 = 16 4 1 + 2 cos cos cos2 cos2 3 sin 2 sin2 3,

g3 = 16 6 1 + 3 cos cos 3 cos2 cos2 + 9 sin2 sin2

+ cos3 cos3 9 cos cos sin2 sin2 27.
And

e1 = 2 2 2 + cos( + ) + cos( ) 3,

e2 = 2 2 1 + cos( + ) 2 cos( ) 3,

e3 = 2 2 1 2 cos( + ) + cos( ) 3.
(58)
(59)
Now the vortex ring is described by

i
1
F3 =
+ w3 2 4 + 3 cos + 3 cos cos( )
4
6
4 H

cos( + ) 3 ,

1
F+ =
+ w3 + 1 + w3 + 2 (1 ) + (2 )
32 3H
4
4
( 4 + w3 + 1 ) ( 4 + w3 + 2 )
(1 ) (2 ) 2 ( 4 + w3 )

exp (1 ) + (2 ) + H i ( ) 3 (1 + 2 ) ,
(60)
4
4
2
where H = (1 cos )(1 cos ).
443
The characteristics of the ring are obtained using the pseudoperiodicity of Weierstrass
elliptic functions:
(u + 2wi ) = (u) + 2i ,

(u + 2wi ) = exp 2(u + wi )i (u).
As
(61)

(e2 e3 )/(e1 e3 )
e1 e3 ,

e1 e3 ,
w3 = iK (e1 e2 )/(e1 e3 )
w1 = K
w1 is real while w3 is pure imaginary for the choice in Eq. (59). Thus the physical
characteristics are described by the real period >W = 2w1 , i.e., > = 8w1 . For example,
F+ obtains a factor

exp 2(1 + 2 )1 2 (1 )w1 2(2 )w1
(62)
after one period >W = 2w1 . But this factor can be shown to be pure imaginary with
the help of the addition theorem of function and the relation 1 = w1 2 which
is valid in this case. So the radius is periodically oscillating between the maximum and
minimum values. The maximum value of the radius is reached at W = 0 in Eq. (50) which
is calculated as follows. With the help of the relation
(w3 + 1 ) (w3 + 2 ) (1 ) + (2 )
1 P (w3 ) P (1 ) 1 P (w3 ) P (2 )
=
2 P(w3 ) P(1 )
2 P(w3 ) P(2 )

1 P (2 )
1 P (1 )
+
=
2 e3 P(1 ) 2 e3 P(2 )
= 4i,
and the identity in Eq. (45), we obtain

F+ (W = 0) = 1
{P(1 ) e3 }{e3 P(2 )}
2
8 H

1
=
sin + sin sin( + ).
4H
Similarly we obtain the minimum of the radius which is attained at W = w1 :

P(
)
e
P(1 ) e2
1
2
2

F+ (W = w1 ) =
P (2 )
P (1 )

16A
P(1 ) e2
P(2 ) e2

1

= cot cot .
4
2
2
(63)
(64)
(65)
For a time period, X3 = 4iF3 / is increased by

4
1
21 + 2 w1 4 + 3 cos + 3 cos cos( ) cos( + ) .

3
2 H
(66)
444
Fig. 1. A breathing circular ring (case A) advances along the z-axis with 0 < < 28.5. A circle is
completed by 0 < < 7.2 for a fixed . The parameters are = 2.9, = 2.2, = 1.
In Fig. 1, we show an example of the breathing circular ring with = 2.9, = 2.2,
= 1. For this parameters, the time period is given by > = 14.16, while the radius |X+ |
oscillates between the maximum = 0.63 and minimum = 0.39 values. X3 is increased by
12.62 during this time.
5.4.2. Case B: =
In this case, we take 1 = 3 = rei , 2 = 4 = ei /r. Then
g2 =
g3 =
And
4 4
2r 2 + 1 + r 8 4r 4 cos 2 2r 6 cos 2 + 4r 4 cos2 2
3r 4

2r 6 2r 2 cos 2 + 6r 4 ,

8 6 2
r r 4 1 + r 2 cos 2 2r 2 r 4 1 + 4r 2 cos 2
27r 6

4r 2 + r 4 1 + 2r 2 cos 2 .

1
2
e1 = 2 1 r 2 + cos 2
3,
r

1
r2
3,
e2 = 2 2 2 +
+ 2 + cos 2
2
2r

1
r2
2
3.
e3 = 2 1 +
+ 2 2 cos 2
2
2r
(67)
(68)
The characteristics of the ring are similarly obtained as in case 1. Especially 1 =

w1 + w3 2 = w2 2 in this case. The maximum value of the radius is
445
Fig. 2. A breathing circular ring of (case B) for 0 < < 13 and 4 < < 4. The parameters are
r = 2, = 1.9, = 1.

F+ (W = 0) =
r
| sin |.
r 2 2r cos + 1
(69)
The minimum of the radius at W = w1 is 0, which can be seen with the following relation:
(w1 + w3 + 1 ) (w1 + w3 + 2 ) (1 ) + (2 )
= (2w2 2 ) (w2 + 2 ) (w2 2 ) + (2 ) = 0.
(70)
During a time period, X3 is increased by
4r
2r cos + 1)

1 2 4
3
2
2
1 2 w1 r 6r cos + 8r + 2r cos 2 6r cos + 1 .
3r
2 (r 2
(71)
In Fig. 2, we show an example of the breathing circular ring with r = 2, = 1.9, = 1.

For this parameters, the time period is given by > = 6.51, while the radius |X+ | oscillates
between the maximum = 1.20 and minimum = 0 values. X3 is increased by 4.18 during
this time.
5.5. Helicoidal curve
When W is a function of only, the z-coordinate of the vortex depends linearly on and
the radius |F+ | depends only on , while the phase of F+ depends linearly on . Thus it
becomes a helix of varying radius with time. The z-coordinate of the vortex point at = 0
depends on such that the height of the vortex along the z-coordinate fluctuates. In this
case, we should take 1 = 3 = r exp(i ), 2 = 4 = exp(i)/r such that s4 = 2 . As an
example, we plot the fluctuating helix in Fig. 3 by taking r = 2, = 0.9, = 1.7, = 1.
446
Fig. 3. (a) A typical helix drawn with 0 < < 30 at = 0. (b) Its radius and height of the z-axis
fluctuate with 0 < < 15 ( = 0.1). The parameters of the helix are r = 2, = 0.9, = 1.7, = 1.
This figure shows the fluctuating radius with time as well as the fluctuating height of the
helix.
5.6. Closed ring
Remember that the pseudoperiodicity of the Weierstrass functions in Eq. (61) results
in the additional factor of Eq. (62) for F+ after a period >W = 2w1 . Thus a necessary

condition for a closed ring, i.e., F+ ( = m> ) = F+ ( = 0) with > = 16w1/ 1

, is
s
4

n
1 (1 + 2 ) w1 (1 ) + (2 ) A i > = i ,
8 s4
m
(72)
where m, n are arbitrary integers. Another condition comes from the closeness of the
z-coordinate, i.e., F3 ( = m> ) = F3 ( = 0), which is
21 +
x1 + x2 + x3
2 2 s1 + s3 2s4
2w1 +
> = 0.
3
s4
8
(73)
The quasiperiodic property

of
the closed ring along the time is following. During a time

period > = 16w1 / 1 + s , the ring returns to its original shape but rotates around the
4
z-axis by an angle

Im 21 (1 + 2 ) 2w1 (1 ) + (2 ) + A >
(74)
4 s4
and moves a distance

2
x1 + x2 + x3
2 2 s1 + s3 2s4
2w1
>

21 +
3
s4
8
A
(75)
along the z-axis. The two conditions, (72) and (73), can be solved numerically using
a software package like M ATHEMATICA. For example, 2 = 4 = 0.9697 + 0.4927i was
found to be the solution of the conditions for a choice 1 = 3 = 0.5 + 0.5i, n = 3, m = 2.
447
Fig. 4. (a) A typical closed ring plotted for 0 < < 223 with parameters 1 = 0.5 + 0.5i,
2 = 0.9697 + 0.4927i at = 0, (b) at = 5, (c) at = 10, (d) and at = 14.554 (time period).
Fig. 5. (a) A typical open string at = 0 with 200 < < 200, and (b) at = 100 with
200 < < 200. The parameters of the string are 1 = 0.8 + 0.7i, 2 = 0.6 + 0.1i.
448
Fig. 4 shows the motion of a closed ring at = 0, 5, 10 and 14.554 (= > , time period).
During a time period, the ring rotates 1.576 radian and advances 2.735 along the z-axis.
It returns completely to its original figure when = 2> . The period is > = 111.51 in
this configuration.
5.7. General vortex string
Most generally, the solution in Eqs. (50) and (33) describes a vortex of open string type.
Fig. 5 shows one just example which is obtained using 1 = 3 = 0.8 + 0.7i, 2 = 4 =
0.6 + 0.1i, = 2. This is one of the most general configurations described by the 1-phase
periodic solution.
6. Concluding remarks
In this paper, we calculate the 1-phase periodic solution of the relativistic vortex string
described by the LundRegge model. The solution was explicitly given in terms of
Weierstrass elliptic functions. We study various string configurations resulting from the
degenerate limit of the solution. It contains already known configurations like the straight
line, the Kelvin wave, the smoke ring and the Hasimoto 1-soliton solution. In addition, it
gives new configurations such as the breathing circular ring, the closed string and curling
open strings.
Our formalism can be adapted to describe the nonrelativistic vortex dynamics of fluid
mechanics. It is described by the Da RiosBetchov equation, X = X X , which is the
dynamical equation for a vortex X(, ) [21]. Thus we can find vortex configurations of
fluid mechanics having quasiperiodic property. This will generalize the Kidas work [22]
without relying on a specific assumption on the form of the solution [23].
Using the result of Section 3, we can obtain higher phase periodic solution which should
be described in terms of Riemanns theta functions. The difficulty in this program is the
effectivization problem such that the obtained solution should satisfy Eq. (12). This
together with the difficulty in treating the Riemann theta function would be the obstacle to
apply it to real physical problems.
The stability analysis of the solution is another important physical problem to be done.
A similar study for the case of nonrelativistic vortex motions was conducted using the
perturbation method in [24]. Our formalism could be useful in that there exists a method
for the analysis of long-time behavior of instabilities using the (quasi-)periodic solution of
the integrable system [14].
Acknowledgements
This work was supported by the Brain Korea 21 Project.
449
Appendix A. Derivation of the linear equation

The derivation of the linear equation (8) needs the duality relation between the
complex sine-Gordon equation and the LundRegge vortex equation. Firstly we briefly
review the essential property of duality using the presentation of [7]. We start with
the complex sine-Gordon equation formulated upon the group theory [25,26]. It was
called SU(2)/U (1)-gauged WessZuminoNovikovWitten model and its field variable
is denoted by g SU(2). Then the equation of motion for the complex sine-Gordon theory
was shown to be expressed as a zero curvature condition

1 1
1
+ g g + T , + g T g = 0,
(A.1)
where T = T = i3 and 3 is the Pauli matrix. The zero curvature condition can be
understood as the integrability condition of the linear system
( + Usg ) = 0,
g 1 g
( + Vsg ) = 0,
(A.2)
1 1
g T g.
where Usg
+ T and Vsg
Now we introduce the duality relation by writing down the vortex variables F in Eq. (3)
in terms of the complex sine-Gordon variables g and showing that the complex sineGordon equation (A.1) implies the LundRegge equation (4) [7,8]. This can be achieved
easily in terms of the associated linear equations in Eq. (A.2) as follows; let (z, z , ) be
a solution of the linear equation of the complex sine-Gordon model. Let us define a matrix
F as
(t ln ).
F 1 = 1
(A.3)
t
To check it gives the LundRegge equation, we first notice that
F = 1 1
+ 1 ()
t
t
Usg
(A.4)
= 1 T .
t
= 1 1 g 1 Tg. This shows that the constraint equation
In a similar way, we find that F
= 1
is automatically satisfied,
Tr(F )2 = 2 2 2 ,
Since
)2 =
Tr(F
2
.
2

Usg
Usg
= 1
, Vsg ,
F
= 1
t
t
it is now easy to get the LundRegge equation

F = 1 T , 1 g 1 Tg = [F, F
].
(A.5)
(A.6)
(A.7)
We now use the modified form of the R-transformation [19] to obtain the linear equation
where ()
is a solution of
(8) of the LundRegge equation. First we define 1
450
Define
the linear equation (A.2) with the spectral parameter ().

1
M 1 = 1 + 1

=
= 1 g 1 g + T 1 g 1 g + T
1 F,

1 = 1 1 g 1 Tg 1 1 g 1 Tg = 1 F.
(A.8)
Then the required linear equations are

( M) = 0,
( N) = 0.
(A.9)
Note that the compatibility condition of the linear equations, i.e., [ M, N] = 0, gives
the LundRegge equation in a form of the zero-curvature condition. In the main text, we
take = 1 without any loss of generality. And we change for notational simplicity.
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]
M. Kalb, P. Ramond, Phys. Rev. D 9 (1974) 2273.

F. Lund, T. Regge, Phys. Rev. D 14 (1976) 1524.
F. Lund, Phys. Rev. Lett. 38 (1977) 1175.
F. Lund, Ann. Phys. (New York) 115 (1978) 251.
A. Sym, O. Ragnisco, D. Levi, M. Bruschi, Lett. Nuovo Cimento 44 (1985) 529.
A. Zee, Nucl. Phys. B 421 (1994) 111.
K. Lee, Q. Park, H.J. Shin, Nucl. Phys. B 563 (1999) 461.
Q. Park, H.J. Shin, Phys. Lett. B 454 (1999) 259.
F.R. Hama, Phys. Fluids 5 (1962) 1156;
F.R. Hama, Phys. Fluids 6 (1963) 526;
R.J. Arms, F.R. Hama, Phys. Fluids 8 (1965) 553.
H. Hasimoto, J. Fluid Mech. 51 (1972) 477.
E.J. Hopfinger, F.K. Browand, Y. Gagner, J. Fluid Mech. 125 (1982) 505.
T. Maxworthy, M. Mory, E.J. Hopfinger, Proc. AGARD Conf. CPP-342, 1983, p. 29.
See, for example, B.A. Dubrovin, I.M. Krichever, S.P. Novikov, Progress in Science and
Engineering, Contemp. Problems in Math., Vol. 4, VINITI, Moscow, 1985.
E.R. Tracy, H.H. Chen, Phys. Rev. A 37 (1988) 815.
E. Date, Osaka J. Math. 19 (1982) 125.
Y. Fukumoto, T. Miyazaki, J. Phys. Soc. Jpn. 55 (1986) 4152.
E.R. Tracy, H.H. Chen, Y.C. Lee, Phys. Rev. Lett. 53 (1984) 218.
A.M. Kamchatnov, Phys. Rep. 286 (1997) 199.
S.J. Orfanidis, Phys. Lett. A 75 (1980) 304, and references therein.
K. It (Ed.), Encyclopedic Dictionary of Mathematics by the Mathematical Society of Japan,
MIT Press, Cambridge, MA, 1993.
L.S. Da Rios, Rend. Circ. Mat. Palermo 22 (1906) 117;
R. Betchov, J. Fluid. Mech. 22 (1965) 471;
R.L. Ricca, Nature 352 (1991).
S. Kida, J. Fluid Mech. 112 (1981) 397.
H.J. Shin, Vortex Filament Motion in Terms of Weierstrass Elliptic Functions, to be published.
S.E. Widnall, J. Fluid Mech. 54 (1972) 641, and references therein.
[25] Q. Park, Phys. Lett. B 328 (1994) 329;

Q. Park, H.J. Shin, Phys. Lett. B 347 (1995) 73;
Q. Park, H.J. Shin, Phys. Lett. B 359 (1995) 125.
[26] I. Bakas, Q. Park, H.J. Shin, Phys. Lett. B 372 (1996) 45.
451

Conformal correlation functions, Frobenius algebras

and triangulations
Jrgen Fuchs a , Ingo Runkel b , Christoph Schweigert b
a Institutionen fr fysik, Universitetsgatan 1, S-65188 Karlstad, Sweden
b LPTHE, Universit Paris VI, 4 place Jussieu, F-75252 Paris Cedex 05, France
Abstract
We formulate two-dimensional rational conformal field theory as a natural generalization of twodimensional lattice topological field theory. To this end we lift various structures from complex vector
spaces to modular tensor categories. The central ingredient is a special Frobenius algebra object A in
the modular category that encodes the MooreSeiberg data of the underlying chiral CFT. Just like for
lattice TFTs, this algebra is itself not an observable quantity. Rather, Morita equivalent algebras give
rise to equivalent theories. Morita equivalence also allows for a simple understanding of T-duality.
We present a construction of correlators, based on a triangulation of the world sheet, that
generalizes the one in lattice TFTs. These correlators are modular invariant and satisfy factorization
rules. The construction works for arbitrary orientable world sheets, in particular, for surfaces
with boundary. Boundary conditions correspond to representations of the algebra A. The partition
functions on the torus and on the annulus provide modular invariants and NIM-reps of the fusion
rules, respectively. 2002 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Hf
1. Introduction
Attempts to understand the spectrum of bulk fields in two-dimensional conformal field
theories gave rise to the formulation of a mathematical problem: classify modular invariant
torus partition functions. Considerable effort has been spent on this problem over the
past 15 years. In theories of closed strings, modular invariance of the partition function
guarantees the absence of anomalies in the low-energy effective field theory. Modular
E-mail address: schweige@lpthe.jussieu.fr (C. Schweigert).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 3 8 - 1
J. Fuchs et al. / Nuclear Physics B 624 [PM] (2002) 452468
453
invariance is a much stronger property than anomaly freedom, i.e., string theory imposes
strictly stronger conditions on the field content than field theory. It comes, therefore, as a
disappointment that there exist modular invariants that obey all the usual constraints
positivity, integrality, and uniqueness of the vacuumbut are nevertheless unphysical
(see, e.g., [1,2]). Thus, albeit a mathematically well-posed problem, classifying modular
invariants is not exactly what is desired from a physical point of view.
The study of the open string field content of conformal field theories resulted in the
formulation of a similar problem [3,4]: classify NIM-reps, that is, representations of the
fusion rules by matrices with non-negative integral entries. Again, this classification yields
(plenty of) spurious solutions that cannot appear in a consistent conformal field theory [5].
Motivated by these observations we pose the following questions: first, what is the
correct structure that allows to classify full rational conformal field theories with given
MooreSeiberg data? And second, how does this structure determine the correlation
functions of the full CFT, in particular, how does one obtain a modular invariant partition
function (as the 0-point correlator on the torus) and NIM-reps (as the 0-point correlators on
the annulus)? In this note we present a natural structure that allows to construct correlation
functions and that can be expected to arise in every RCFT. The structure in question is the
one of a symmetric special Frobenius algebra in a modular tensor category. The resulting
correlation functions are modular invariant and satisfy factorization rules.
Neither the presence of this structure nor its relevance to the questions above is a priori
obvious. Moreover, as the choice of words indicates, to formulate our prescription we
need to employ some tools which are not absolutely standard. In particular, we work in the
context of modular tensor categories. Their relevance to our goal emerges from the fact that
they provide a powerful graphical calculus and a convenient basis-free formalization of the
MooreSeiberg data of a chiral conformal field theory, like braiding and fusing matrices
and the modular S-matrix. A well-known example of a modular tensor category, which can
serve as a guide to the general theory, is the category of fininite-dimensional vector spaces
over the complex numbers. This category is also relevant to the analysis of two-dimensional
lattice topological theory [6], so that one may suspect a relation between those theories and
our prescription. Indeed, one of our central observations is that much of the structure of
(rational) conformal field theories can be understood in terms of constructions familiar
from lattice TFT, provided that one adapts them so as to be valid in general modular tensor
categories as well.
The rest of the paper is organized as follows. In Section 2 we provide the necessary
background on modular tensor categories and motivate the appearance of Frobenius
algebras. In Section 3 we develop the representation theory of these algebras. In Section 4
it is shown how to obtain consistent torus and annulus partition functions from a symmetric
special Frobenius algebra. In Section 5 we give a prescription for general correlators that
generalizes the one for two-dimensional lattice TFTs in the category of vector spaces. It
turns out that the Frobenius algebra itself is not observable; Morita equivalent algebras
give rise to equivalent theories. When combined with orbifold techniques, this fact allows
for a general understanding of T-duality in rational CFT. This is discussed in Section 6.
Section 7 contains our conclusions.
A much more detailed account of our results, including proofs, will appear elsewhere.
454
2. Frobenius algebras
As already pointed out, our considerations are formulated in the language of modular
tensor categories [7], a formalization of MooreSeiberg [8] data. A modular tensor
category C may be thought of as the category of representations of some chiral algebra A,
which in turn correspond to the primary fields of a chiral conformal field theory.
Accordingly the data of C can be summarized as follows. The (simple) objects of C are
the (irreducible) representations of A, and the morphisms of C are A-intertwiners. For
a rational conformal field theory, the category is semisimple, so that every object is a
finite direct sum of simple objects. There is a tensor product which corresponds to
the (fusion) tensor product of A-representations, and the vacuum representation (identity
primary field) provides a unit element 1 for this tensor product. The existence of conjugate
A-representations gives rise to a duality on C, in particular each object V of C has a
dual object V . The (exponentiated) conformal weight provides a twist on C: to every
object there is associated a twist endomorphism V ; for a simple object V corresponding
to a primary field of conformal weight V , V is a multiple of the identity morphism,
exp(2iV )id V . Finally, the presence of braid group statistics in two dimensions is
accounted for by a braiding, i.e., for every pair V , W of objects there is an isomorphism
cV ,W Hom(V W, W V ) that exchanges the two objects. The modular S-matrix of
the CFT is obtained from the trace of the endomorphisms cV ,W cW,V with simple V , W .
These data are subject to a number of axioms that can be understood as formalizations
of various properties of primary fields in rational CFT (see, e.g., Appendix A of [9]).
Essentially, the axioms guarantee that these morphisms can be visualized via ribbons and
that the graphs obtained by their composition share the properties of the corresponding
ribbon graphs, and they also include the requirement that the S-matrix is invertible. (That
one really must use ribbons rather than lines is closely related to the fact that Wilson lines
in ChernSimons gauge theories require a framing.)
A particularly simple example of a modular tensor category is the category of fininitedimensional complex vector spaces. It has a single isomorphism class of simple objects
the class of the one-dimensional vector space Cand has trivial twist and braiding. In
conformal field theory, this category arises for meromorphic models, i.e., models with a
single primary field, such as the E8 WZW theory at level 1. It is also the category in which
one discusses two-dimensional lattice topological theory, and as we will discuss below, this
is not a coincidence.
Central to our construction are certain objects in modular tensor categories that can
be endowed with much further structure. To get an idea on the role of these objects, let
us collect a few more results from CFT. First, recall that modular invariants are either
of extension type or of automorphism type, or a combination thereof [10]. A simple
example of extension type arises for the sl(2) WZW theory at level 4:
2

2

Z( ) = 0 ( ) + 4 ( ) + 22 ( ) .
(1)
This invariant corresponds to the conformal embedding of sl(2) level 4 into sl(3) at level 1,
and can therefore be interpreted as follows: the vacuum sector of the extended theory sl(3)1
gives rise to a reducible sector A = (0) (4) of the sl(2)4 theory. In category theoretic
terms, A is a reducible object of the modular tensor category C C(sl(2)4 ). Now A is not
455
merely an object, but comes with plenty of additional structure. First, the operator product
provides an associative multiplication on A. Second, since in a chiral algebra the vacuum is
unique, the sl(2)4 -descendants of the vacuum provide a distinguished subobject (0) of A.
Finally, crossing symmetry of the 4-point conformal blocks of the vacuum on the sphere,
which relates the s-channel and the t-channel, results in additional properties of the product
on A.
The formalization of these properties of A as an object of C reads as follows [11]. There
is a multiplication morphism m Hom(A A, A) that is associative and for which there
exists a unit Hom(1, A). There exists a coassociative coproduct Hom(A, A A)
as well, along with a co-unit Hom(A, 1). It is convenient to depict these morphisms as
follows:
m=
(2)
Such pictures are to be read from bottom to top, and composition of morphisms amounts
to concatenation of lines; a single line labelled by an object V stands for the identity
morphism id V End(V ) Hom(V , V ). Lines labelled by the tensor unit 1 can be omitted
since End(1) = C so that id1 = 1. In (2) we have suppressed the label A which decorates
each of the lines of the picture. (Also, throughout the paper the fattening of the lines to
ribbons is implicitly understood.)
The (co-)associativity and (co-)unit properties, i.e.,
m (m id A ) = m (idA m),
m ( idA ) = id A = m (idA ),
( idA ) = (id A ) ,
( idA ) = id A = (id A ) ,
are then drawn as
(3)

= =

= =
(4)
Further, the crossing symmetry is taken into account by demanding A to be a Frobenius

algebra, i.e., to satisfy
(5)
456
We also require A to be a special Frobenius algebra, i.e., impose that (after suitably fixing
the normalization of the co-unit)
= dim(A)id1
and m = idA ,
(6)
where dim(V ) stands for the quantum dimension of an object V , which is defined as the
trace of the identity morphism idV . To account for the uniqueness of the vacuum, one
would impose that dim Hom(1, A) = 1, a property that has been termed haploidity of A in
[11]. But with an eye on the relation with two-dimensional lattice TFT we prefer to relax
this requirement and only require that A is symmetric in the sense that the morphisms
1 :=
2 :=
(7)
from A to its dual A are isomorphisms and coincide. It can be shown that haploidity
implies symmetry, hence this is indeed a weaker requirement. But as it turns out, it still
leads to a meaningful theory.
The presence of a braiding allows us to introduce a notion of commutativity: an algebra
A is called commutative iff m cA,A = m. But it is well known that certain off-diagonal
modular invariants like the Dodd series of sl(2) correspond to a hidden chiral superalgebra,
which is not commutative. Correspondingly we do not impose commutativity of A.
Let us now present more examples. The category of vector spaces contains just a single
haploid algebra, the ground field C itself. C is obviously a special Frobenius algebra. Thus
in a sense haploid Frobenius algebras generalize the ground field. Now as already pointed
out, instead of haploidity we impose the weaker condition of symmetry, i.e., equality of
the two morphisms (7). This proves to be most reasonable, as symmetric special Frobenius
algebras in the category of complex vector spaces are known to describe two-dimensional
lattice TFTs [6,12,13]. Since a topological field theory is, in particular, a (rather degenerate)
conformal field theory, it is gratifying that our formalism covers this case. More generally,
an algebra that is symmetric but not haploid should be thought of as containing several
vacua that constitute a topological subsector of the theory. In fact, this topological sector
gives itself rise to the structure of a symmetric special Frobenius algebra Atop over the
complex numbers. A is symmetric in the sense introduced above if and only if Atop is a
symmetric Frobenius algebra in the usual [14] sense.
For general modular tensor categories the situation is much more involved. But one
haploid special Frobenius algebra is always present, namely, the tensor unit 1. The
associated torus parition function is the charge conjugation invariant, and the boundary
conditions are precisely those which preserve the full chiral symmetry (Cardy case). The
construction of correlation functions presented below then reduces to the one already given
in [9].
457
Another large class of examples is supplied by simple currents, i.e., simple objects J of
C of quantum dimension dim(J ) = 1. Simple currents (more precisely, their isomorphism
classes) form a subgroup of the fusion ring of the CFT that organizes the set of primary
fields into orbits. Haploid special Frobenius algebras all of whose simple subobjects are
simple currents can be classified. As objects, they are direct sums of the form

J,
A=
(8)
J H
with H a subgroup of the group of simple currents. But not all objects of the form (8)
can be endowed with an associative product; such a product exists if and only if the
corresponding 6j -symbols yield the trivial class in the cohomology H 3 (H, C ). It follows
from proposition 7.5.4 in [15] that this requirement restricts H to be a subgroup of the socalled effective center, i.e., to consist of simple currents for which the product of the
conformal weight J and the order NJ (the smallest natural number such that J N
= 1),
is an integer.
Allowed objects of the form (8) can admit several different products m Hom(A
A, A); the possible products are classified by the second cohomology H 2 (H, C ). It
is a fortunate fact about Abelian groups that classes in H 2 (H, C ) are in one-to-one
correspondence with alternating bihomomorphisms on H, i.e., with maps from H H
to C that are homomorphisms (i.e., compatible with the product on H) in each argument
and that obey (J, J ) = 1 for all J H. Employing this result, the isomorphism class
of the algebra structure m can be encoded in the KreuzerSchellekens bihomomorphism
(KSB) [16], which graphically is represented as
(J, K)
(9)
(Here the twist morphism K appears; were we drawing ribbons instead of lines, this
would amount to a full 2 rotation of the K-ribbon.) The possible KSBs are in one-toone correspondence with the associative products m on H; different KSBs for a simple
current group H are related by alternating bihomomorphisms. The choice of a KSB has
been called a choice of discrete torsion in [16]; there are indeed models where the KSB
precisely describes discrete torsion. In the case of a free boson compactified on a lattice, the
choice of a multiplication on the algebra object that characterizes the lattice corresponds
to the choice of a background value of the B-field.
We finally quote examples of algebra objects for the exceptional modular invariants
of the sl(2) WZW theory. For the E6 -type modular invariant at level 10, one has A =
(0) (6), for E7 at level 16, one finds A = (0) (8) (16), and for the E8 -type invariant
at level 28, A = (0) (10) (18) (28) [17].
458
3. Representations
Just like for ordinary algebras, the next step to be taken in the analysis of the algebra
A is the study of its representation theory. Precisely as in the case of vector spaces,
an A-representation M consists of two data: an object M of C, corresponding to the
vector space that underlies the module on which A acts, and a representation morphism
M)
that specifies the action of A on the module. And further, these
M Hom(A M,
data are subject to exactly the same constraint as in the vector space case; pictorially:
(10)
(In more detailed terminology, what we get this way is a left A-module. Right modules,
as well as comodules, can be introduced analogously. One can develop the whole theory
equivalently also in terms of those.)
The representations of A are in one-to-one correspondence with conformally invariant
boundary conditions. For any algebra A in a modular tensor category, all A-modules are
fully reducible. Irreducible modules correspond to elementary boundary conditions, while
direct sums of irreducible modules describe boundary conditions with non-trivial Chan
Paton multiplicities. Commutative algebras A with trivial twist, i.e., A = id A , give rise
to modular invariants of extension type. Irreducible modules then fall into two different
classes, local and solitonic modules. Local modules can be characterized [17] by the fact
that their twist is a multiple of the identity, i.e., that their irreducible subobjects all have
the same conformal weight modulo integers. They correspond to boundary conditions
that preserve all symmetries of the extension; solitonic representations describe symmetry
breaking boundary conditions.
The standard representation theoretic tools, like induced modules and reciprocity
theorems, generalize to the category theoretic setting (see, e.g., [11,17,18]) and allow to
work out the representation theory in concrete examples. The case of the E6 modular
invariant of the sl(2) WZW theory has been presented in [11,17]; here we briefly comment
on the representations of simple current algebras (8). One starts from the observation that
every irreducible module is a submodule of some induced module A V . A acts on the
induced module A V from the left by multiplication, i.e., AV = m idV . We only need
to consider irreducible V , and abbreviate the irreducible Vi , with i labelling the primary
fields, by i. As an object in C, the induced module A i is just the H-orbit [i] of i under
fusion, with the order of the stabilizer Si = {J H|J i
= i} of i as the multiplicity:

J i.
A i = |Si |
(11)
J H/Si
To study the decomposition of (11) into irreducible modules, one needs to classify
projectors in the endomorphism space EndA (A i). This vector space possesses the
structure of a twisted group algebra over the Abelian group Si . The twist can be computed
459
explicitly in terms of the KSB (9) and of certain gauge invariant 6j -symbols. This way
one recovers the list of boundary conditions proposed in [19], which was used in [20] to
compute the correct B-type boundary states in Gepner models.
4. Partition functions
We now demonstrate how to extract partition functions, i.e., torus and annulus
amplitudes, from a given symmetric special Frobenius algebra. To this end we make use
of the fact that to every modular tensor category one can associate a three-dimensional
topological field theory [7,8]. Such a 3d TFT can be thought of as a machine that
associates vector spacesthe spaces of conformal blocksto two-manifolds, and linear
maps between these vector spaces to three-manifolds. When a path integral formulation is
available, such as the ChernSimons theory in the case of WZW models [21], the vector
spaces can be thought of as the spaces of possible initial conditions, while the linear maps
describe transition amplitudes between given initial and final conditions.
with a finite number of embedded
More precisely, to each oriented two-manifold X
arcs, which are labelled by primary fields (and certain additional data that will not concern
of conformal
us here, see, e.g., [9] for details), the TFT associates a vector space H(X)
blocks. This assignment behaves multiplicatively under disjoint union, i.e., H(X1 X2 ) =
H(X1 ) H(X2 ), and assigns C to the empty set, H() = C. To an oriented three-manifold
M containing a Wilson graph, with a decomposition M = + M M of its boundary
into two disjoint components, the TFT associates a linear map
Z(M, M, + M) :
H( M) H(+ M).
(12)
These linear maps are multiplicative, compatible with the gluing of surfaces, and obey a
few further functoriality and naturality axioms (see, e.g., chapter 4 of [22]).
As explained in [23], correlation functions of a full conformal field theory on an
of conformal
arbitrary world sheet X are nothing but special elements in the space H(X)

blocks on the complex cover X of X. This is an oriented two-manifold doubly covering X,
= X. Following the strategy
endowed with an anticonformal involution such that X/
of [9,24], we determine the correlators using the connecting three-manifold MX , which
has the complex cover as its boundary,

MX = X.
(13)
[1, 1] by the involution that

The connecting manifold is obtained as the quotient of X
and as t t on the interval [1, 1]. We can and will identify the image
acts as on X
{0} in MX with X. The world sheet X is a retract of the connecting manifold MX ,
of X
and hence we may think of MX as a fattening of X. Upon choosing an appropriate Wilson
graph in MX , the three-dimensional TFT then provides us with a map
:
C(X) := Z(MX , , X)

C H(X),
(14)
and it is this map that yields the correlators on X.

Let us now have a look at a specific situation, and explain how to obtain a modular
invariant torus partition function from a symmetric special Frobenius algebra. The
460
connecting manifold for a torus T is just a cylinder over T , i.e., MT = T [1, 1]. In
this three-manifold, we consider the following Wilson graph, labelled by A:
(15)
Here (as well as in Eqs. (16), (18) and (19) below) the top and bottom are to be identified,
and the lower trivalent vertex denotes the coproduct and the upper vertex the product of A.
To determine the coefficients Zij of the torus partition function, we glue to both sides of
(15) two solid tori, containing a Wilson line labelled by i and j , respectively. This yields
the following Wilson graph in the three-manifold S 2 S 1 :
(16)
It is useful to interpret this result as Zij = tr Pij , where Pij is the linear map given by
the same graph in S 2 [0, 1]. One can show that Pij is a projector, which implies that
the numbers Zij are non-negative integers, as befits a partition function. Furthermore, Z00
equals the dimension of the center of the C-algebra Hom(1, A); in particular, when A is
haploid, then the vacuum sector appears precisely once, Z00 = 1. Finally, one can also
prove modular invariance. We illustrate invariance under the U -transformation +1

(the pictures correspond to Eq. (15) with the interval [1, 1] suppressed):
U
=
(17)
461
Here in the first equality the Frobenius property (5) of A is used, while in the second
step the periodicity of the graph is taken into account; equality with the original picture
then follows by applying the Frobenius property once again. In the case of simple current
algebras (8), our result (16) reproduces the formula of [16] that expresses the most
general simple current modular invariant in terms of the KSB and monodromy charges.
In agreement with [17], for commutative algebras A with A = id A the partition function
can be shown to be of pure extension type in which only local modules appear.
For the annulus, a similar reasoning applies. The connecting manifold looks as follows:
(18)
It is a solid torus with two Wilson lines along its non-contractible cycle. These Wilson
lines have opposite orientation; each of them is is labelled by a (left) A-module that
characterizes the boundary condition. In addition there are Wilson lines labelled by A,
which are attached to the modules by trivalent vertices that are given by the respective
representation morphisms. Again, the annulus coefficients AiM N , i.e., the coefficients in
an expansion of (18) in the characters i , are obtained by gluing, in this case with a single
torus containing a Wilson line labelled by i. This leads to
(19)
As in the torus case, one then shows that the annulus coefficients are non-negative integers.
Moreover, we can show that the matrices Ai with entries (Ai )M N := AiM N indeed furnish
a NIM-rep of the fusion rules of C. It also follows that if the algebra A is haploid and
462
M = N is an irreducible A-module, i.e., an elementary boundary condition, then the

vacuum 1 appears just once in the annulus, i.e., A1 M
M = 1. From the annuli, one can read off
the boundary states, and show that their coefficients provide the classifying algebra [26].
Performing a modular transformation, one obtains the annuli in the closed string channel;
one can check that only fields i appear that are compatible with the torus partition function
(15), and that they appear with the correct multiplicity Zi,i .
5. General amplitudes
The results presented above can be summarized by the statement that every symmetric
special Frobenius algebra gives rise to a modular invariant torus partition function and to
a NIM-rep of the fusion rules that are compatible with each other. To show that we even
obtain a complete consistent conformal field theory, we should construct all correlators on
world sheets X of arbitrary genus, including an arbitrary number of boundary components,
and show that they are invariant under the relevant mapping class group and possess the
correct factorization properties. (In the present note, we restrict ourselves to orientable
world sheets. The unorientable case requires additional structure, so as to account for the
possibility of having several different Klein bottle amplitudes.)
The construction of correlators consists in a beautiful combination of the construction
of [9,24] with structures familiar from lattice TFTs in two dimensions. As is already
apparent from our prescription for the torus and annulus amplitudes, what is needed is
a Wilson graph in the connecting three-manifold MX , which besides the Wilson lines that
correspond to field insertions and to the presence of a boundary contains in addition a
suitable network of A-lines. The general prescription is the following:
(1) To start, we fix once and for all an orientation of the world sheet X. The complex
of X can then be decomposed as X
= X+ X X , with the involution acting
double X
+
as (X ) = X and (X) = X.
p + on X+ and p on
(2) Each bulk insertion point p, on X has two preimages on X,
,
,
X , which are joined by an interval in the connecting manifold MX . We put a Wilson line
along each such interval. The image of X in MX intersects the interval in a unique point,
which we identify with p, X.
(3) Each component of the boundary of X gives rise to a line of fixed points of MX
under the Z2 action. We place a circular Wilson line along each such line of fixed points.
We join q, by a
(4) Boundary insertion points q, on X have a unique preimage q, on X.
short Wilson line to the image of q, in MX . This results in a trivalent vertex on the circular
Wilson line for the relevant boundary component.
(5) We pick a triangulation of the image of X in MX . Without loss of generality, we
do this in the following manner. First, only trivalent vertices may appear. (But we allow
for arbitrary polygonal faces rather than just triangles, which is completely equivalent [25]
to the case with only triangular faces; for brevity we still use the term triangulation.)
Moreover, each segment of the boundary must contain the end point of an edge of the
triangulation, and each of the bulk points p, must lie in the interior of a separate face of
the triangulation, while each boundary point q, must lie in the interior of a separate edge.
Finally, for each bulk insertion we join the perpendicular bulk Wilson line in p, by an
463
additional line to an interior point of an arbitrary edge of the face to which p, belongs.
(More precisely, in terms of ribbons, we first bend the horizontal ribbon straight up, at a
right angle to X, so that it points towards p, and is parallel to the vertical bulk ribbon, and
then join it to the bulk ribbon.)
This completes the geometric part of our prescription. It is illustrated, in the case of a
disk with three bulk and three boundary insertions, in the following figure:
(20)
The remaining task is now to decorate the Wilson lines with labels specifying the
corresponding object of the category and choose couplings (morphisms) for the trivalent
vertices. Our prescription is inspired by the situation in two-dimensional lattice TFTs and
contains that case as a special example.
(6) The short Wilson lines that connect the bulk insertion points p, to the triangulation
are labelled by the algebra A, with the A-line pointing towards p, .
(7) The three lines that join at any vertex of the triangulation are interpreted as outgoing
and are labelled with the algebra object A. The morphisms at the vertices are constructed
from the coproduct in A: at every trivalent vertex at which these A-lines join we put the
morphism 11 , with 1 given by (7). (The condition that A is symmetric ensures that
the resulting correlators do not depend on the choice of which of the three lines carries
the 11 .) In the lattice case, this prescription corresponds to the assignment of structure
constants with only lower indices.
(8) In the case of lattice TFT one also needs a metric. In the general case this
corresponds to reversing the direction of the A-line. Accordingly we place the morphism
(7) in Hom(A, A ) in the middle of each edge of the triangulation.
(9) To characterize the bulk field inserted at p, , we need two irreducible objects j, ; they
To account for the coupling to the
label the Wilson lines that start from the points p , on X.
short A-lines, we need as a third datum for a bulk field a morphism in Hom(A j,+ , j, ).
It turns out that some of the latter morphisms completely decouple, i.e., that every
correlator containing a bulk field labelled by such a coupling vanishes. Physical bulk fields
therefore correspond to a subspace of the couplings Hom(A j,+ , j, ); its dimension
matches the value Zj + j of the torus partition function.
, ,
(10) The segments of the circular Wilson line correspond to boundary conditions; they
are to be labelled by a module of A. Wilson lines of the triangulation that end on such a
464
M).
For
boundary result in a trivalent vertex, which requires a morphism in Hom(A M,
this morphism we choose the representation morphism M .
(11) The last ingredient in our construction are the boundary fields. They have a single
chiral label k, , which we use to label the short Wilson line from q, to q, . The trivalent
vertex that is formed by this Wilson line and the two adjacent boundary conditions M, N
requires the choice of a coupling in Hom(M k, , N).
Again, only a subspace of these couplings is relevant. In this case the subspace can easily
be characterized: M k, is a left A-module, and only the subspace HomA (M k, , N), of
dimension Ak, M N , of A-morphisms gives rise to physical boundary fields that change the
boundary condition from M to N .
In the example of the disk, the picture then looks as follows (after rearranging, in one
of several possible ways, the vertices that result from the presence of 1 and its inverse, so
as to eliminate all occurrences of unit and co-unit):
(21)
Topological invariance of a two-dimensional lattice model means that all correlation

functions are independent of the choice of triangulation. The same type of arguments as in
the topological case can be used to deduce independence from the triangulation also in the
CFT case. Concretely, triangulation independence can be reduced (in a dual formulation)
to invariance under two local moves, the so-called fusion and bubble moves [6,25]. These
moves look like
and
(22)
Our construction amounts to regard the lines in this picture as morphisms and decorate all
of them with the label A, and to interpret the trivalent vertices as products and coproducts,
respectively. The fusion move is, in CFT terms, just crossing symmetry between the
s- and t-channel, and hence invariance under this move is guaranteed by the associativity
and the Frobenius property (5) of A. Similarly, the second the equality (22) is nothing but
the second part of the formula (6), i.e., is guaranteed by the fact that A is special. Together
465
with the naturality properties of 3d topological field theory, the independence from the
triangulation implies that the correlators are invariant under the relative modular group of
[27]. Furthermore, the correct factorization rule for boundary fields follows directly from
dominance properties of the category C. Bulk factorisation requires in addition a surgery
move on the connecting three-manifold.
6. Morita equivalence and T-duality

We have shown how to obtain a full conformal field theory from a symmetric special
Frobenius algebra A in the modular tensor category of the underlying chiral CFT. The
relation between such algebras and full conformal field theories is, however, not one-toone. Rather, different algebras can describe one and the same conformal field theory. In
other words, the algebra A itself should not be thought of as an observable quantity.
This is already familiar from the degenerate realization of our construction that is
provided by topological lattice theories. In that case, symmetric Frobenius algebras over
the complex numbers with the same number of simple ideals (and thus isomorphic
centers) give identical theories [6]. In particular, the algebra can therefore be chosen to
be commutative.
The general case is treated as follows. We call two symmetric special Frobenius algebras
A and B Morita equivalent iff there exist two bimodules A MB and B MA (the first a left
module of A and a right module of B, the second a left module of B and a right module of
A) such that
(A MB ) B (B MA ) = A
and (B MA ) A (A MB ) = B.
(23)
Using these relations, one can replace a triangulation of the world sheet X that is
labelled by A-lines with the dual triangulation labelled by B-lines. Moreover, by standard
arguments [28] it follows that Morita equivalent algebras have the same representation
theory, so that A-modules and B-modules, and hence boundary conditions, are in one-toone correspondence. Since the correlation functions do not depend on the triangulation, this
implies that our prescription yields, upon a relabelling of fields and boundary conditions,
the same correlators when performed with either of the two algebras A and B.
These observations also allow for a general derivation of T-duality in rational CFT.
Suppose we are given a symmetry of the chiral data. Technically, this is a group
G together with a functor from C[G]a category whose objects are pairs of objects of
C and an action of G by automorphisms of the objectsto C; it includes, in particular,
a permutation i i of the primary fields such that 1 = 1, Ti = Ti and Sij = Sij . It
is straightforward to check that along with Zij also Zij is a modular invariant partition
function. (A familiar realization is the standard T-duality in the CFT of a free boson. The
relevant symmetry is the one that reverses the sign of the U(1) charges.) The idea is to
associate to the symmetric special Frobenius algebra A that gives the partition function
Zij another algebra A that yields the partition function Zij . The construction is based on
Morita equivalence in the orbifold theory C that is obtained by modding out the symmetry
on C. One can show that with respect to this orbifold theory, the algebras A and A give,
466
upon appropriate identification of the fields and of the boundary conditions, isomorphic
correlation functions on all surfaces.
7. Conclusions
We have shown how to obtain full rational conformal field theories based on the
chiral data that are encoded in a modular tensor category C from (Morita equivalence
classes of) symmetric special Frobenius algebras in C. We gave a general prescription
for the construction of correlation functions on orientable surfaces, including surfaces with
boundary. We will demonstrate elsewhere that the so obtained correlators give rise to a
fully consistent CFT.
Our results open up several lines of research. First, the generalization of these results
to unorientable surfaces, a necessary step to a deeper understanding of type I string
compactifications, will require more structure that has to account, in particular, for the
different possible choices of orientifold projections.
Another important goal is classification. In the framework we propose, the problem
of classifying rational conformal field theories can be divided in three independent tasks,
each of which deals with a clearly posed (albeit very difficult) problem. The first is to
classify modular tensor categories. Second, given a modular tensor category C, two distinct
problems must be attacked. On one hand we must determine the Morita equivalence classes
of symmetric special Frobenius algebras in C. Note that this issue can be studied entirely at
a topological level, i.e., no analytic properties of chiral blocks are involved. On the other
hand, one also needs to get a handle on the control data that allow to completely reconstruct
a chiral CFT from its modular tensor category. (The correspondence between chiral CFT
and modular tensor categories is not one-to-one. For example, all WZW theories based on
so(2n+1) at level 1 with values of n that differ by a multiple of 24 have equivalent tensor
categories.) This part of the problem is a purely chiral issue. It is closely related to the
classification of chiral algebras, which at present remains the least accessible part of the
classification programme.
In order that this programme is complete, we also still must establish that every full
rational CFT can be obtained from some symmetric special Frobenius algebra A. In all
examples known to usin particular, for all pure extensions as well as for all simple current
modular invariantsthis is indeed the case, but at present we cannot give a general proof.
What is desirable is a general prescription for reconstructing the algebra object A from
some collection of correlation functions of the full CFT. It would also be interesting to
see whether this reconstruction is related to other algebraic structures, such as the double
triangle algebras [29] which in the work of [30,31] are regarded as a structure underlying
every rational CFT. This would also allow for a comparison between the expressions for
correlators obtained there and those which follow from our prescription.
We are confident that the framework we propose is flexible enough to allow for an
extension to non-rational conformal field theories. In fact, rationality implies that the
category C is semi-simple. Generalizations of TFT for non-semi-simple tensor categories
have been studied in the literature, compare the recent book [32]. It can be expected that
the concepts which underlie these generalizations will play a role in the analysis of non-
467
rational theories. Finally, we point out that our results suggest an intimate relation between
the problem of deforming algebras and the problem of deforming conformal field theories.
Note added in proof

In [33], which appeared after publication of the present paper, a reconstruction theorem
for module categories is proven. When combined with the general expectation that
boundary conditions of a rational conformal field theory lead to a module category over
the category C that describes the MooreSeiberg data of the chiral CFT, the results of [33]
provide an additional argument for the existence of an algebra object A, and thereby for
the generality of the approach proposed in the present paper.
Acknowledgements
We are grateful to Giovanni Felder and Albert Schwarz for helpful discussions and
comments, and to the Erwin Schrdinger Institute and to Karlstads universitetet for
hospitality. IR is supported by the EU grant HPMF-CT-2000-00747.
References
[1] A.N. Schellekens, S. Yankielowicz, Field identification fixed points in the coset construction, Nucl. Phys.
B 334 (1990) 67, Appendix B.
[2] J. Fuchs, A.N. Schellekens, C. Schweigert, Galois modular invariants of WZW models, Nucl. Phys. B 437
(1995) 667.
[3] G. Pradisi, A. Sagnotti, Ya.S. Stanev, Completeness conditions for boundary operators in 2D conformal field
theory, Phys. Lett. B 381 (1996) 97;
A. Sagnotti, Ya.S. Stanev, Open descendants in conformal field theory, Fortschritte d. Phys. 44 (1996) 585.
[4] R.E. Behrend, P.A. Pearce, V.B. Petkova, J.-B. Zuber, On the classification of bulk and boundary conformal
field theories, Phys. Lett. B 444 (1998) 163;
V.B. Petkova, J.-B. Zuber, BCFT: from the boundary to the bulk, Proc. Hep-tmr2000 (2000) 038.
[5] T. Gannon, Boundary conformal field theory and fusion ring representations, hep-th/0106105.
[6] M. Fukuma, S. Hosono, H. Kawai, Lattice topological field theory in two dimensions, Commun. Math.
Phys. 161 (1994) 157, hep-th/9212154.
[7] V.G. Turaev, Quantum Invariants of Knots and 3-Manifolds, de Gruyter, New York, 1994.
[8] G. Moore, N. Seiberg, Lectures on RCFT, in: H.C. Lee (Ed.), Physics, Geometry, and Topology, Plenum,
New York, 1990, p. 263.
[9] G. Felder, J. Frhlich, J. Fuchs, C. Schweigert, Correlation functions and boundary conditions in RCFT and
three-dimensional topology, hep-th/9912239, Compos. Math., in press.
[10] G. Moore, N. Seiberg, Naturality in conformal field theory, Nucl. Phys. B 313 (1989) 16.
[11] J. Fuchs, C. Schweigert, Category theory for conformal boundary conditions, math.CT/0106050, to appear
in Fields Institute Commun.
[12] B. Durhuus, T. Jonsson, Classification and construction of unitary topological field theories in two
dimensions, J. Math. Phys. 35 (1994) 5306.
[13] L. Abrams, Two-dimemsional topological quantum field theories and Frobenius algebras, J. Knot Theory
and its Ramif. 5 (1996) 569.
[14] C.W. Curtis, I. Reiner, Representation Theory of Finite Groups and Associative Algebras, Wiley
Interscience, New York, 1962.
468
[15] J. Frhlich, T. Kerler, Quantum Groups, Quantum Categories and Quantum Field Theory, Springer Lecture
Notes in Mathematics, Vol. 1542, Springer-Verlag, Berlin, 1993.
[16] M. Kreuzer, A.N. Schellekens, Simple currents versus orbifolds with discrete torsiona complete
classification, Nucl. Phys. B 411 (1994) 97.
[17] A.A. Kirillov, V. Ostrik, On q-analog of McKay correspondence and ADE classification of
sl(2) conformal
field theories, math.QA/0101219.
[18] A.A. Kirillov, Modular categories and orbifold models, math.QA/0104242.
[19] J. Fuchs, L.R. Huiszoon, A.N. Schellekens, C. Schweigert, J. Walcher, Boundaries, crosscaps and simple
currents, Phys. Lett. B 495 (2000) 427, hep-th/0007174.
[20] J. Fuchs, P. Kaste, W. Lerche, C. Ltken, C. Schweigert, J. Walcher, Boundary fixed points, enhanced gauge
symmetry and singular bundles on K3, Nucl. Phys. B 598 (2001) 57.
[21] E. Witten, Quantum field theory and the Jones polynomial, Commun. Math. Phys. 121 (1989) 351.
[22] B. Bakalov, A.A. Kirillov, Lectures on Tensor Categories and Modular Functors, American Mathematical Society, Providence, 2000, http://www.math.sunysb.edu/~kirillov/tensor/tensor.
html.
[23] J. Fuchs, C. Schweigert, Branes: from free fields to general backgrounds, Nucl. Phys. B 530 (1998) 99.
[24] G. Felder, J. Frhlich, J. Fuchs, C. Schweigert, Conformal boundary conditions and three-dimensional
topological field theory, Phys. Rev. Lett. 84 (2000) 1659, hep-th/9909140.
[25] S.-W. Chung, M. Fukuma, A. Shapere, Structure of topological lattice field theories in three dimensions, Int.
J. Mod. Phys. A 9 (1994) 1305.
[26] J. Fuchs, C. Schweigert, A classifying algebra for boundary conditions, Phys. Lett. B 414 (1997) 251.
[27] M. Bianchi, A. Sagnotti, Open strings and the relative modular group, Phys. Lett. B 231 (1989) 389.
[28] J.C. Vrilly, An introduction to noncommutative geometry, physics/9709045.
[29] A. Ocneanu, (notes by S. Goto), Paths on coxeter diagrams: from platonic solids and singularities to minimal
models and subfactors, in: R. Bhat, G.A. Elliott, P.A. Fillmore (Eds.), Lectures on Operator Theory, Fields
Institute Monographs 13, American Mathematical Society, Providence, 1999, p. 243.
[30] V.B. Petkova, J.-B. Zuber, The many faces of ocneanu cells, Nucl. Phys. B 603 (2001) 449, hep-th/0101151.
[31] V.B. Petkova, J.-B. Zuber, Quantum field theories, graphs and quantum algebras, hep-th/0108236.
[32] T. Kerler, V.V. Lyubashenko, Non-semisimple topological quantum field theories for 3-manifolds with
corners, Springer Lecture Notes in Mathematics, Vol. 1765, Springer-Verlag, New York, 2001.
[33] V. Ostrik, Module categories, weak Hopf algebras and modular invariants, math.QA/0111139.

Families index theory for Overlap lattice

Dirac operator. I
David H. Adams a,b
a Physics department, National Taiwan University, Taipei, 106, Taiwan, ROC
b Mathematics department and Centre for the Subatomic Structure of Matter, University of Adelaide,
Adelaide S.A. 5005, Australia

Abstract
The index bundle of the Overlap lattice Dirac operator over the orbit space of lattice gauge fields is
introduced and studied. Obstructions to the vanishing of gauge anomalies in the Overlap formulation
of lattice chiral gauge theory have a natural description in this context. Our main result is a formula
for the topological charge (integrated Chern character) of the index bundle over even-dimensional
spheres in the orbit space. It reduces under suitable conditions to the topological charge of the usual
(continuum) index bundle in the classical continuum limit (this is announced and sketched here; the
details will be given in a forthcoming paper). Thus we see that topology of the index bundle of the
Dirac operator over the gauge field orbit space can be captured in a finite-dimensional lattice setting.
2002 Elsevier Science B.V. All rights reserved.
PACS: 02.40.-k; 11.15.Ha
1. Introduction
The index bundle of the Dirac operator over the orbit space of gauge fields was studied
many years ago by Atiyah and Singer [1].1 Obstructions to the vanishing of chiral gauge
anomalies have a natural description in this context [1,3]. More recently, the overlap
formalism [4], developed by Narayanan and Neuberger as a way to formulate chiral gauge
theories on the lattice, has provided the ingredients for a lattice version of Dirac operator
index theory when the base manifold (spacetime) is an even-dimensional torus. A lattice
version of the index arose there as the fermionic topological charge of the lattice gauge
E-mail address: dadams@maths.adelaide.edu.au, adams@twcguest.phys.ntu.edu.tw (D.H. Adams).
1 Detailed treatments of Dirac operator index theory can be found in [2].
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 2 3 - X
470
D.H. Adams / Nuclear Physics B 624 [PM] (2002) 469484
field, and can be expressed as the index of the Overlap lattice Dirac operator, introduced
in [5]. The fact that the overlap formulation successfully reproduces the global gauge
anomaly and obstructions to the vanishing of local gauge anomalies [610] indicates that
it should also lead to a lattice version of families index theory for the Dirac operator. The
purpose of the present paper is to show that this is indeed the case.2
We develop a families index theory for the Overlap lattice Dirac operator coupled
to SU(N) lattice gauge fields on the 4-torus. (The specialisation to 4 dimensions is for
simplicity and physical relevance; everything generalises straightforwardly to arbitrary
even-dimensional torus [12].) After recalling the lattice setup in Section 2, we show in
Section 3 that an index bundle over the orbit space of lattice gauge fields can be defined
in a natural way. In Section 4 we derive our main result: a formula (lattice families index
theorem) for the topological charge (integrated Chern character) of this index bundle over
a generic even-dimensional sphere in the orbit space of lattice gauge fields (Theorem 1,
Eq. (4.1) below). It contains an integer-valued part which has no continuum analogue; in
Section 5 we outline the argument in the forthcoming paper [12] showing that the rest of
the expression reduces to the topological charge of the usual (continuum) index bundle in
the classical continuum limit (Theorem 2).
When the group G of gauge transformations is restricted to G0 = { G | (x0 ) = 1}
(where x0 is some basepoint in the 4-torus) the orbit space U/G0 is a smooth manifold
(since G0 acts freely on the space U of lattice gauge fields). In this case the abovementioned part of the families index formula which has no continuum analogue vanishes
in the classical continuum limit, as we will see in Section 5. The above-mentioned classical
continuum limit result then implies that the finite-dimensional lattice setting is capturing
topology of the continuum index bundle over A/G0 when the lattice is sufficiently fine.
In Section 6 we discuss how obstructions to the vanishing of gauge anomalies in the
overlap formulation of lattice chiral gauge theories have natural descriptions in this context.
We conclude in Section 7 by discussing some implications and possible generalisations of
our results, and some related results to be shown in subsequent papers. The background
material in Sections 2 and 3 is intended to make this paper accessible to nonspecialists in
lattice gauge theory.
This is the first in a series of papers. In the follow-up paper [12], we treat the
general case of arbitrary even-dimensional torus and prove the classical continuum
limit result announced and outlined in Section 5, as well as a result mentioned in the
concluding Remark 2. In [13], we apply the results of the present paper to demonstrate
an interplay between topological features of the space of SU(N) lattice gauge fields
on T 4 and the existence question for G0 gauge fixings on the lattice which do not have
the Gribov problem. We find that certain obstructions to the existence of such gauge
fixings in the continuum are absent on the lattice, and that instead the topological sectors
(specified by the fermionic topological charge) contain noncontractible spheres of various
even dimensions when N 3, and noncontractible circles in the N = 2 case. Further
applications of the lattice families index theory developed here to the global gauge anomaly
on the lattice and related issues will be given in [14].
2 A first step in this direction was made earlier in [11].
471
2. Lattice setup
We take the spacetime manifold to be the Euclidean 4-torus T 4 and the gauge group
to be SU(N). A hypercubic lattice on T 4 determines the space C of lattice spinor fields,
the space U of lattice gauge fields, and the group G of lattice gauge transformations. C is
the finite-dimensional complex vector space consisting of the spinor-valued functions on
the lattice sites, i.e., the functions on the lattice sites with values in C4 CN . (C4 and
CN are the spin and SU(N) representation spaces; for simplicity we are specialising to the
fundamental representation of SU(N).) It has the inner product

1 (x) 2 (x)
1 , 2 = a 4
(2.1)
x
(a = lattice spacing). U consists of the SU(N)-valued functions on the links of the lattice.
It has the finite-dimensional manifold structure
U
= SU(N) SU(N) SU(N)
(2.2)
(one copy for each link). The group G consists of the maps : {lattice sites} SU(N). It
acts on C and U by
(x) = (x)(x),
(2.3)
( U ) (x) = (x)U (x)(x + ae )
(2.4)
(e = unit vector in the positive -direction). Continuum spinor fields, gauge fields and
gauge transformations have lattice transcripts, defined in a natural way. In the case of spinor
field (x) or gauge transformation (x) we restrict x to the lattice sites to get elements
in C or G, respectively. In the case of gauge field A (x) the lattice transcript U (x) is
the parallel transport from x + ae to x. Then (2.3)(2.4) are the lattice transcripts of
the continuum gauge transformations. For a given continuum gauge field A, the classical
continuum limit of a quantity depending on the lattice gauge field U (e.g., the index of
a lattice Dirac operator) is the limit of infinitely many subdivisions of the hyper-cubic
lattice (i.e., lattice spacing a 0) with U being the lattice transcript of A. The classical
continuum limit of quantities depending on families of lattice gauge fields is defined
analogously.
The covariant derivative A = + A can be approximated on the lattice by the
covariant finite difference operators a1 , a1 , where
+ (x) = U (x)(x + ae ) (x),
(2.5)
(x) = (x) U (x ae )1 (x ae ),
and =
1
+
2 (
+ ).
Note that
( )
= ,
so
(2.6)
= .
3. Lattice version of Dirac operator index theory

The Dirac operator /
A = A has the naive lattice approximation
1
/ = .
a
(3.1)
472
However, the index theory for this operator is trivial: index

/ U = 0 U U . This wellknown fact is a simple consequence of the chiral symmetry
5
/ +
/ 5 = 0
(3.2)
and the finite-dimensionality of C. A nontrivial lattice version of the Dirac operator

index can instead be constructed by noting that index /
A is minus the spectral flow of
A
A
H m = 5 ( /
m) as m increases from m < 0 to m > 0. (We are following the physics
convention where the s are hermitian; then /
A is antihermitian and HmA is hermitian.)
A
A lattice analogue of Hm is

U
HmU = 5 Dw
(3.3)
m ,
where
r
U
Dw
=
/ U + a U ,
2
(3.4)
is the WilsonDirac operator [15]. The Wilson term a 2r , where = a12 ( ) =

1
+
( ) is the lattice Laplace operator and r > 0 the Wilson parameter, breaks
a2
the chiral symmetry and thereby allows for a nontrivial index theory. It is known that when
m < 0 the hermitian operator HmU has symmetric spectrum and no zero-modes [4]. Hence
the spectral flow as m increases from any m1 < 0 to some m2 > 0 is equal to half the
spectral asymmetry of HmU at m = m2 , i.e., 12 Tr( U ), where
U =
HU
.
|H U |
(3.5)
This suggests defining the lattice version of the index by

1
index /
(3.6)
A Tr U ,
2
for some suitable value of the parameter m in (3.3) (which we have suppressed in the
notation in (3.5) and (3.6)). In fact the right-hand side of (3.6), with
rm0
, 0 < m0 < 2,
(3.7)
a
is precisely the definition of the topological charge of U which arose in the overlap
formulation of chiral gauge theories on the lattice [4]. In [16] this was shown to reduce
to index /
A in the classical continuum limit for any choice of m satisfying (3.7) above. It
is also these choices of m which give the correct classical continuum limit for the families
index theory that we develop in the following.
For this definition of the lattice version of the index we must exclude from U the
lattice gauge fields U for which H U has zero-modes, so that U is well-defined. This
determines a decomposition of U into topological sectors labelled by the lattice index.
It has been shown in [17] (see also [18]) that H U has no zero-modes when each U (p)
is sufficiently close to the identity, where U (p) is the product of the U (x)s around a
plaquette p in the lattice. (We refer to [18] for the currently best bound on 1 U (p)
which ensures this.) Consequently, the absence of zero-modes for H U is guaranteed close
m=
473
to the classical continuum limit, since when U is the lattice transcript of a continuum field
A, and p is the plaquette specified by a lattice site x and directions and , we have
1 U (p) = a 2 F (x) + O(a 3). (See [16] for a more detailed discussion of this point.)
From now on, we take m as in (3.7) and assume that a choice of m0 (0, 2) has been
made, and U denotes the space of lattice gauge fields with the U s for which H U has
zero-modes excluded.
A can be expressed as the index of the Overlap lattice
The lattice version (3.6) of index /
Dirac operator [5], given by
1
D = (1 + 5 ),
(3.8)
a
with as in (3.5). The nullspace of this operator is invariant under 5 and, therefore, has a
chiral decomposition

ker D U = ker D U + ker D U .
(3.9)
The index can then be defined as

index D U = dim ker D U + dim ker D U ,
(3.10)
and coincides with the above lattice version of index /

A:
1
index D U = Tr U .
(3.11)
2
While this was obtained directly in [5], it can also be seen using the fact that (3.8) satisfies
[19]
5 D + D5 = aD5 D
D = 5 D5
(GW relation),
(3.12)
(5 -hermiticity).
(3.13)
The first relation was originally studied by Ginsparg and Wilson [20], and later
rediscovered by Hasenfratz and collaborators [21], who noted that the nullspace of a
solution D is invariant under 5 (since if D = 0 then D5 = (aD5 D 5 D) = 0).
Furthermore, they showed that solutions of (3.12)(3.13) satisfy the following index
formula (see also [22]):

a
index D U = Tr 5 D U .
(3.14)
2
Substituting (3.8) in (3.14) leads to (3.11).
Having reviewed the previously developed lattice index theory for the Dirac operator
we now proceed to develop a lattice version of the families index theory. Rather than the
index (3.10), the central object here is the index bundle, formally given, in analogy with
the continuum case, by
index D = (ker D)+ (ker D) ,
(3.15)
where
(ker D) =

ker D U U U .
(3.16)
474
If (ker D)+ and (ker D) were vector bundles then (3.15) would be a well-defined element
in K(U), the K-theory of U . However, as in the continuum setting, the dimensions of
(ker D U )+ and (ker D U ) can jump as U varies, even though their difference, index D U ,
remains constant. In the continuum some trickery is required to deal with this aspect so
as to make the index bundle into a well-defined element in the K-theory (see, e.g., [2]).
But in the lattice setting this aspect can be dealt with in a simple way, by exploiting the
finite-dimensionality of C, as follows. Besides the usual chiral decomposition,
C = C+ C
(5 = 1 on C ),
(3.17)
there is another decomposition determined by U (which played a central role in the

overlap formalism [4], see also [23,24]):
U

C = C+U CU
(3.18)
= 1 on CU .
By (3.8),

U = 5 1 aD U ,
(3.19)
which looks formally like a gauge field-dependent lattice deformation of 5 , hence the
convention in (3.18). It follows from (3.19) that

ker D U CU C .
(3.20)
U,
In light of this we can define V U to be the orthogonal complement of (ker D U )+ in C+
U
U
and W the orthogonal complement of (ker D ) in C , i.e.,

C+U = ker D U + V U ,
(3.21)

U
U
C = ker D W .
(3.22)
From (3.8) (or (3.12) and (3.19)) we get

5 D U = D U U ,
(3.23)
which implies that D U maps CU to C .
Lemma 1. The map D U : C+U C restricts to an isomorphism D U : V U W U .

Proof. It suffices to show that the orthogonal complement of the image of D U on C is
precisely ker D U . But this is a simple consequence of the 5 -hermiticity property (3.13)
and the invariance of ker D U under 5 .
Now set

V = V U U U ,

W = W U U U .
(3.24)
These are not vector bundles in general, since dim V U and dim W U can jump as U varies,
but we will formally treat them as vector bundles in the following. By Lemma 1, V and W
475
are isomorphic. Using this and (3.21)(3.22), formal K-theoretic manipulations give
index D = (ker D)+ (ker D)
(ker D)+ V (ker D) V
=
= (ker D)+ V (ker D) W

= C+ C ,
(3.25)
where

C+ = C+U U U
(3.26)
and C in (3.25) is to be interpreted as the trivial vector bundle over U with fibre C . Unlike
the initial expression, the final expression (3.25) for index D is a well-defined element in
K(U) due to the following:
Lemma 2. C+ is a smooth vector bundle over U .
Proof. Since dim V U = dim W U we have
index D U = dim C+U dim C ,
(3.27)
dim C+U
hence
is locally constant under variations of U . Furthermore, C+ = P(C) where C
is to be interpreted as the trivial vector bundle over U with fibre C and P : C C is the
smooth vector bundle map defined on the fibres by

1
1 U .
(3.28)
2
Then by a standard mathematical result (see, e.g., Proposition 1.3.2.(ii) of [25]), C+ is a
smooth vector bundle over U .
PU =
We, therefore, take (3.25) as the definition of the index bundle of D. Due to gauge
covariance it descends to an element in the K-theory of the orbit space, i.e.,
index D K(U/G).
(3.29)
Due to the triviality of C in (3.25), the topology of the index bundle is determined
solely by C+ . In particular, for the nonzero degree parts of the Chern character we have
chn (C ) = 0 and3

chn (index D) = chn C+ chn (C ) = chn C+ (n 1).
(3.30)
Thus the topological charge of the index bundle over general 2n-dimensional spheres in
the orbit space, which is obtained by integrating the Chern character, is the same as for
the vector bundle C+ for n 1. We remark that, since C+ is a unitary vector bundle (with
unitary structure determined by the inner product (2.1)), its topological charge over odddimensional spheres always vanishes. So it is only the even-dimensional case that is of
interest.
3 The zero degree part of the Chern character is just ch (index D) = dim C
U dim C = index D U .
0
+
476
4. Topological charge of the index bundle over 2n-spheres in the orbit space
Generically, a 2n-sphere in U/G can be presented as a 2n-dimensional family U (,t )
of lattice gauge fields, parameterised by the 2n-ball B 2n with boundary S 2n1 ( S 2n1
and t is the radial coordinate in B 2n ), with all the boundary points U (,1) being gauge
equivalent, i.e., U (,1) = U (U U (0,1) ) for some family { }S 2n1 of lattice gauge
(,t)
transformations. Set P (,t ) = P U
(given by (3.28)). The main result of this paper is the
following formula, which can be regarded as a lattice families index theorem:
Theorem 1. The topological charge of C+ and, therefore, of the index bundleover the
above 2n-sphere in U/G (n 1) is

2n
1
1
Tr P (,t ) dP (,t )
Q2n =
n
(2i) n!
B 2n
(1)n (n 1)!
2 (2n 1)!
+2

deg (x) ,
2n1

Tr U 1 d
S 2n1
(4.1)
where Tr is the trace for linear operators on C, d = d + dt is the exterior derivative on

B 2n and deg((x)) is the degree of the map S 2n1 SU(N), (x).

Remark. The last term 2 x deg((x)) vanishes in the n = 1 case (since SU(N)
contains no noncontractible circles), but can be nonvanishing for 2 n N . This
even integer-valued term has no continuum analogue. The rest of (4.1) reduces to the
topological charge of the continuum Dirac index bundle in the classical continuum limit,
cf. Section 5. We will also see there that when G is replaced by the constrained group
ofgauge transformations G0 , for which the orbit space is a smooth manifold, the term
2 x deg((x)) gives no contribution to the classical continuum limit.
Proof of Theorem 1. The restriction of the vector bundle C+ to the 2n-sphere in U/G has
(,t)
the following characterisation. Set C+(,t ) := C+U
and define the vector bundle
(,t )
E := C+
(4.2)
(,t )B 2n
over B 2n . The action of the gauge transformation on C restricts to an isomorphism
=
: C+U C+(,1) .
(4.3)
This determines an equivalence relation identifying each C+(,1) with C+U . The bundle
C+ over the 2n-sphere in U/G is then given by E/ . Topologically, this is equivalent to
2n denote another copy of the 2n-ball,
the bundle over S 2n constructed as follows. Let B
denote the trivial bundle over B
2n with fibre C+U . Then E
with coordinates (, s), and let E
2n1
2n via (4.3)
can be glued together along the common boundary S
of B 2n and B
and E
477
2n with the same topology as the restriction of

to get a bundle E over S 2n = B 2n S 2n1 B
C+ to the 2n-sphere in U/G.
To get a formula for the topological charge we introduce a covariant derivative
(connection) in E. With d = d + dt ,
(,t ) = P (,t ) d P (,t )
(4.4)
is a connection in the bundle E over B 2n , and, with d = d + ds ,

(,s) = P U d + f (s) 1 d P U
(4.5)
over B
2n , where f is some cutoff function with f (s) = 1
is a connection in the bundle E
and f (s) = 0 in neighborhoods of s = 1 and s = 0, respectively. A simple calculation
shows that
(,1) 1 .
(,1) =
(4.6)
fit together to give a connection in E. Hence, the topological charge

Consequently, and
of this bundle is

1
n
n
Q2n =
(4.7)
Tr(F )
Tr(F
) ,
(2i)n n!
B 2n
2n
B
respectively.
where F and F
are the curvatures of and ,
Using the fact that for general connection we have = F (i.e., wedge
multiplication by F ), simple calculations in the present case give
F(,t ) = P (,t ) dP (,t ) dP (,t ) ,
(4.8)
(F ) = P (dP )
(4.9)
and
2n

2
(,s)
= f (s)ds 1 d + f (s)(f (s) 1) 1 d P U ,
F

2n1 U

n1

n

ds 1 d
P ,
(F
) = nf (s) f (s)(f (s) 1)
(4.10)
(4.11)
where we have used the fact that 1 d maps C+U to itself, i.e., [1 d , P U ] = 0.
After substituting these in (4.7), the radial parameter in the second integral can be
integrated out to obtain

1
2n1
, (4.12)
Q2n =
Tr P (dP )2n + (n)
Tr P U 1 d
(2i)n n!
B 2n
S 2n1
where
1
(n) = n
0

n1
n!(n 1)!
.
f (s) f (s)(f (s) 1)
ds = (1)n
(2n 1)!
(4.13)
To get the second equality in (4.13), note that the integral depends only on the values of
f (s) at s = 0 and s = 1; to evaluate it we can therefore replace f (s) by s.
478
The first term in (4.12) is the first term in (4.1). Substituting P U = 12 (1 U ) the second
term in (4.12) becomes

1
U 1
2n1
2n1
(n)
.
(4.14)
Tr d
Tr d
2(2i)n n!
S 2n1
S 2n1
The second term here is the second term in (4.1). The first term here can be rewritten as

2n1
(n)
(4.15)
,
4 tr (x)1 d (x)
n
2(2i) n!
x
S 2n1
where tr is the trace for the fundamental representation of SU(N) and the factor 4 comes
from the trivial trace over spinor indices. To show that this coincides with the last term
in (4.1), thereby completing the proof of Theorem 1, it remains to show that

2n1
(n)
(4.16)
= deg((x)).
tr (x)1 d (x)
n
(2i) n!
S 2n1
To get this, note that

deg((x)) =
1
(2i)n n!
n ,
tr F
(4.17)
2n
B
is the curvature of the gauge field A(,

s) = f (s) 1 (x) d (x) on B
2n , since
where F
2n via the gauge transformation

this is gauge-equivalent to zero at the boundary S 2n1 of B
2n , (4.17) reduces to the left 1 (x). After integrating out the radial parameter in B
4
hand side of (4.16). This completes the proof of Theorem 1.
5. Classical continuum limit

In this section we consider the situation where the 2n-sphere {U (,t )} in U/G is
the lattice transcript of a generic 2n-sphere {A(,t )} in the orbit space A/G of smooth
continuum SU(N) gauge fields on the 4-torus. I.e., A(,1) = A (A A(0,1) ), where
{ : T 4 SU(N), S 2n1 } corresponds to a smooth map : S 2n1 T 4 SU(N)
via (x) = (, x). In this case it is well-known that the topological charge of the
continuum Dirac index bundle over the 2n-sphere in A/G equals the degree deg() of the
map . (This follows, e.g., from the result in [1].) We announce and outline the argument
for the following result, which will be shown in [12]:
Theorem 2. (i) The first two terms in the lattice families index formula (4.1) reduce to the
topological charge of the continuum Dirac index bundle over the 2n-sphere in A/G in the
4 Note that F
n is given by (4.11) without the P U and with (x).
479
classical continuum limit, i.e.,

1
1
Tr P (dP )2n
lim
n
a0 (2i)
n!
B 2n
(1)n (n 1)!
2 (2n 1)!
U 1
2n1
Tr d
S 2n1
= deg().
(5.1)
(ii) When the gauge transformations are constrained to belong to

G0 := { G
| (x0 ) = 1}, where x0 is an arbitrary basepoint in T 4 , the remaining term 2 x deg((x))
in the lattice families index formula (4.1) vanishes.
Corollary. The topological charge of the index bundle of the Overlap lattice Dirac operator
over the lattice transcript of a generic 2n-sphere in the continuum orbit space A/G0
coincides with the continuum topological charge deg() when the lattice is sufficiently
fine.
Theorem 2 (i) and (ii) together imply that the topological charge of the lattice
index bundle reduces to that of the continuum index bundle over the 2n-sphere in
A/G0 in the a 0 limit (classical continuum limit). Since the lattice and continuum
topological charges are both integers, the Corollary follows. We remark that the gauge
field orbit space specified by the constrained gauge transformations is a smooth manifold,
since G0 (unlike G) acts freely on the space of gauge fields. (For this reason the gauge
transformations were constrained to belong to G0 in the study of the continuum Dirac
index bundle in [1].)
In the following we outline the
proof of Theorem 2 (the details will be given in [12]). We
begin by considering the term 2 x deg((x)) in the lattice families index formula (4.1) in
the present case where the 2n-sphere in the lattice orbit space is the lattice transcript of the
2n-sphere in the continuum orbit space. I.e., is the lattice transcript of a S 2n1 -family
of continuum gauge transformations, also denoted , and U is the lattice transcript of A.
Since (x) depends smoothly on x, and T 4 is connected, the degree deg((x)) of the map
S 2n1
SU(N), (x) is independent of x. Denoting this by deg(), we therefore
have 2 x deg((x)) = 2N deg() where N = the number of lattice sites. Thus this term
diverges in the continuum limit if deg() #= 0, but vanishes when deg() = 0. The latter
occurs when the gauge transformations belong to G0 , since in this case (x0 ) = 1 for all
S 2n1 , hence deg() = deg((x0 )) = 0. This shows part (ii) of Theorem 2.
The proof of part (i) of the Theorem 2 is based on the following technical result, which
we announce here and prove in [12]:

Tr (dP )2n P
lim
a0
B 2n
1
(2i)2

B 2n T 4
n1

)
(,t )
tr dt A(,t
(x) dx
F (,t ) (x)
(x) dx d A
(5.2)
480
and

lim
a0

2n1 U
Tr 1 d

S 2n1
1
=
2(2i)2

2n1

tr 1 (x) d (x)
F (x)2 ,
(5.3)
S 2n1 T 4
where F (,t ) and F are the curvatures of A(,t ) and A (= A(0,1) ), respectively. In the n = 1
case this has essentially already been derived in a related context in [10,24]. The derivation
for general n is essentially a generalisation of the one given there. Now, substituting these
into the left-hand side of (5.1) gives

n1

1
1
)
(,t )
tr dt A(,t
(x) dx
F (,t ) (x)
(x) dx d A
n+2
n!
(2i)
B 2n T 4
(1)n (n 1)!
4 (2n 1)!
1
2n1

tr (x) d (x)
F (x)2 . (5.4)
S 2n1 T 4
On the other hand, the degree of can be expressed as

1
n+2
deg() =
tr
F
(2i)n+2 (n + 2)!
B 2n T 4

n+2
tr F
(5.5)
2n T 4
B
are the curvatures of the gauge fields

where F , F
)
A(, t, x) = A(,t
(x) dx ,
s, x) = A (x) dx + f (s) 1 (x) d (x)

A(,
(5.6)
2n T 4 , respectively, since these are related at the common boundary

on B 2n T 4 and B
S 2n1 T 4 by the gauge transformation : S 2n1 T 4 SU(N). From (5.6) we
calculate
n1

)
(,t )
F n+2 = (n + 1)(n + 2) dt A(,t
(x) dx
F (,t )(x), (5.7)
(x) dx d A

n+2 = (n + 1)(n + 2) n f (s) f (s)(f (s) 1) n1
F
2

2n1
2
F (x) ds 1 (x) d (x)
.
(5.8)
2n in the
After substituting these in (5.5) and integrating out the radial parameter in B
second integral, (5.4) is obtained, thereby establishing (5.1).
6. Relation to gauge anomalies in lattice chiral gauge theory

In the continuum, the chiral fermion determinant is not really a function of the gauge
field, but rather a section in the U(1) determinant line bundle associated with the index
481
bundle of the Dirac operator over the space of gauge fields (see, e.g., [1,3] for the details
of this and the following remarks). To get the determinant as a function of the gauge field,
a choice of trivialisation of the determinant line bundle must be made. Different choices
of trivialisations correspond to different (gauge field-dependent) complex phase choices
for the determinant. By gauge covariance, the determinant line bundle descends to a line
bundle over the orbit space of gauge fields, and the trivialisations which lead to a gaugeinvariant chiral fermion determinant are precisely those which descend to trivialisations of
the line bundle over the orbit space. Therefore, the obstructions to getting a gauge-invariant
chiral fermion determinant are precisely the obstructions to trivialising the determinant line
bundle over the orbit space. The primary obstructions are the obstructions to trivialising
the line bundle over the 2-spheres in the orbit space (cf. the discussion in the final section
of [3]). Since the Chern character of the determinant line bundle coincides with the first
) of the index bundle (i.e., the degree 2 part of the total
Chern character ch1 (index /
character ch(index /
)), these obstructions are the topological charges of the index bundle
over the 2-spheres in the orbit space. For the generic 2-spheres considered in the preceding
section, which correspond to maps : S 1 T 4 SU(N), the topological charge coincides
with the winding number obstruction studied in [3] (i.e., the winding number of the phase
A
of det(/
+ ) as goes around S 1 ).
The situation for the lattice version of the chiral fermion determinant arising in the
overlap formulation [4] is completely analogous. It is again a section in a U(1) determinant
line bundle [4,7] (see also [10] for further discussion). The line bundle is 1max C+
(1max C ) .5 But this is precisely the determinant line bundle associated with the index
bundle (3.25) of the Overlap lattice Dirac operator that we have introduced in Section 3.
Thus the primary obstructions to the existence of a gauge-invariant phase choice for the
overlap chiral fermion determinant are the topological charges of the lattice index bundle
over the 2-spheres in the orbit space of lattice gauge fields. For generic 2-spheres, these
coincide with the winding number obstructions for the overlap determinant studied in [10].
Indeed, the formula for the topological charge Q2 , given by the 2n = 2 case our families
index formula (4.1), coincides with the winding number formula Eq. (3.11) of [10].
When the gauge group is SU(N) with N 3 there are always nonvanishing obstructions
of the primary type discussed above. This is because of the existence of topologically
nontrivial maps : S 1 T 4 SU(N) for all N 3.6 Hence the determinant line bundle
over the orbit space cannot be trivialised in these cases and no gauge-invariant phase choice
for the chiral fermion determinant (with fermion in the fundamental representation) exists.
The same is true in the lattice setting, at least when the lattice is sufficiently fine, due to the
classical continuum limit result of [10] (or equivalently, the 2n = 2 case of Theorem 2 of
the present paper).
On the other hand, for gauge group SU(2) there are no topologically nontrivial maps
: S 1 T 4 SU(2). Hence, the primary obstructions all vanish in the continuum setting.
5 In fact, the lattice version of the chiral fermion determinant in the overlap formulation equals the determinant
of D : C+ C , where D is the overlap Dirac operator [10,26].
6 The topological structure of SU(N ) is S 3 S 5 S 2N1 modulo a finite set of equivalence relations.
Thus when N 3 these is always an S 5 factor and topologically nontrivial maps : S 1 T 4 SU(N ) can be
constructed from maps S 1 T 4 S 5 with nonvanishing degree.
482
There might still be nonvanishing primary obstructions in the lattice setting though; all we
can say for certain at present is that they vanish in the classical continuum limit. However,
there is another obstruction, namely Wittens global gauge anomaly [27].7 The presence
of this obstruction in the SU(2) theory in the lattice setting has been demonstrated both
numerically [6,8] and analytically (in the classical continuum limit) in [9]. This obstruction
also has a natural description in the context of families index theory for the Dirac operator:
there is a canonical trivialisation of the U(1) determinant line bundle over 1-dimensional
balls (i.e., line segments) in the space of gauge fields; it specifies a real line bundle
which descends to a real line bundle over circles in the orbit space with structure group
O(1)
= Z2 . (The circles come from line segments with gauge equivalent end points.) The
global gauge anomaly is the obstruction to trivialising this bundle. In the continuum setting
this description is implicit in [27]. This viewpoint on the global anomaly, and related issues,
will be discussed in detail in the lattice setting in [14].
Remark 1. We have shown that the index bundle of the Overlap lattice Dirac operator over
the orbit space of SU(N) lattice gauge fields can be constructed in a natural way, and have
derived a formula (lattice families index theorem) for its topological charge over generic
2n-spheres in the orbit space. An unanticipated feature is the presence (when 2 n N )
of an integer-valued term which has no continuum analogue and which generally diverges
in the classical continuum limit. However, when the gauge transformations are constrained
to belong to G0 = { G | (x0 ) = 1} (as was also done in [1] to make the orbit space
a smooth manifold), this term vanishes for the 2n-spheres in U/G0 which are lattice
transcripts of 2n-spheres in the continuum orbit space A/G0 . The rest of the formula
reduces in the classical continuum limit to the topological charge of the continuum index
bundle over 2n-sphere in the continuum orbit space. (This was announced and outlined
here; the key technical part of the argument will be given in [12].) Thus we have seen
how topology of the continuum Dirac index bundle over A/G0 can be captured in a finitedimensional lattice setting.
An implication of this is that 2n-spheres in U/G0 which arise as lattice transcripts
of noncontractible 2n-spheres in A/G0 (i.e., those with deg() #= 0) are again noncontractible. (For if this were not the case, the lattice index bundle over the 2n-sphere would
be trivialisable and hence have vanishing topological charge, in contradiction with the classical continuum limit result, Theorem 2.) All this provides further evidence that the orbit
space of lattice gauge fields is a good finite-dimensional model for the orbit space of continuum SU(N) gauge fields on an even-dimensional torus,8 the topology of which is of
considerable mathematical interest and potential physical relevance. We emphasize that,
7 The nonvanishing of Wittens global anomaly in the continuum in the SU(2) case is related to the fact that
the space of maps T 4 SU(2) has two connected components, i.e., there are maps (gauge transformations)
which cannot be continuously deformed to the trivial map (see [27]).
8 Earlier evidence for this came from the demonstrations that the lattice theory reproduces the global gauge
anomaly and obstructions to the vanishing of local gauge anomalies [610].
483
a priori, it is not at all clear that this should be the case. The situation is complicated by
the fact that the measure-zero subspace of lattice gauge fields U for which H U has zeromodes needs to be excised from U , and it is difficult to say anything at all about what
the resulting topological sectors of U look like. Although the lattice orbit space seems to
be reproducing the topology of the continuum one, the situation before modding out by
the gauge transformations is quite different: in the continuum, the topological sectors are
just affine -dimensional vector spaces with no nontrivial topology; however, in [13] we
will show that the topological sectors of U do have nontrivial topology no matter how fine
the lattice is: Using the results of the present paper we will show that they contain noncontractible spheres of various dimensions. These are not simply due to the presence of
noncontractible spheres in SU(N) but directly reflect the topological structure of U resulting from excising the U s for which H U has zero-modes. (E.g., in the SU(2) case we find
that there are noncontractible circles.) This is closely connected with the existence question
for G0 gauge-fixings in the lattice theory, which we also discuss in [13].
Remark 2. The topological charge of the continuum Dirac index bundle over a 2n-sphere
in the orbit space can be expressed as the index of a Dirac operator in 2n + 4 dimensions,
or 2n + 2m dimensions when the dimension of the spacetime is 2m (see, e.g., [3]). There
is an analogous result in the present lattice setting. It has already been given in the n = 1
case in [11]; the generalisation to arbitrary n is straightforward and will be given in [12].
Remark 3. The construction of the lattice index bundle, and the derivation of the families
index formula (4.1), go through for general Overlap-type lattice Dirac operators of the
form D U = a1 (1 +5 U ) where U : C C is any operator depending smoothly and gaugecovariantly on U and with the properties ( U )2 = 1 and ( U ) = U . Such D are precisely
the solutions to the GW relation (3.12) with the 5 hermiticity property (3.13). Besides
the Overlap Dirac operator (3.8), another solution is the lattice Dirac operator resulting
from the perfect action approach of Ref. [28], although this is given via recursion relations
and no closed form expression is known. Other solutions have been presented in [29].
However, at present we can only say with certainty that the classical continuum limit result,
Theorem 2, holds for the case where D is the Overlap Dirac operator.
Remark 4. From a mathematical viewpoint, an obvious question is how to generalise the
constructions and results of this paper to spacetime manifolds other than the tori. The
/ U in (3.1)
problem is to find suitable generalisations of the naive lattice Dirac operator
U
and the lattice Laplace operator in (3.4); these can then be fed into the formulae (3.3)
(3.8) to get the discrete Dirac operator D U and the constructions and derivation of the
families index formula go through as before, cf. Remark 3 above. (A suitable range for the
parameter m would also need to be determined.) Given a polyhedral cell decomposition of
a general spacetime manifold, a discrete Laplace operator is easy to construct (e.g., along
/ seems less easy though. For this one needs
the lines of [30]). Constructing the discrete
to find a way to incorporate the spin structure into the discrete setting. In the tori case the
spin structure is particularly simple and easily incorporated into the discrete setting with
hyper-cubic cell decomposition, cf. (3.1). How to do this in the general case is less obvious.
This is a problem for future work.
484
Acknowledgements
I thank Ting-Wai Chiu and his students for their kind hospitality at NTU. During this
work the author was supported by a postdoctoral fellowship from the Australian Research
Council and a visiting position at NTU funded by the Taiwan NSC (grant numbers NSC892112-M-002-079 and NSC90-2811-M-002-001).
References
[1] M.F. Atiyah, I.M. Singer, Proc. Nat. Acad. Sci. USA 81 (1984) 2597.
[2] N. Berline, E. Getzler, M. Vergne, Heat Kernels and Dirac Operators, Springer-Verlag, Berlin, 1992;
H.B. Lawson, M.-L. Michelsohn, Spin Geometry, Princeton Univ. Press, Princeton, 1989.
[3] L. Alvarez-Gaum, P. Ginsparg, Nucl. Phys. B 243 (1984) 449.
[4] R. Narayanan, H. Neuberger, Phys. Lett. B 302 (1993) 62;
R. Narayanan, H. Neuberger, Phys. Rev. Lett 71 (1993) 3251;
R. Narayanan, H. Neuberger, Nucl. Phys. B 412 (1994) 574;
R. Narayanan, H. Neuberger, Nucl. Phys. B 443 (1995) 305.
[5] H. Neuberger, Phys. Lett. B 417 (1998) 141.
[7] H. Neuberger, Phys. Rev. D 59 (1999) 085006.
[8] O. Br, I. Campos, Nucl. Phys. (Proc. Suppl.) 83 (2000) 594.
[9] O. Br, I. Campos, Nucl. Phys. B 581 (2000) 499.
[10] D.H. Adams, Nucl. Phys. B 589 (2000) 633.
[11] D.H. Adams, Phys. Rev. Lett 86 (2001) 200.
[12] D.H. Adams, Families index theory for Overlap lattice Dirac operator II, in preparation.
[13] D.H. Adams, Families index theory, gauge fixing, and topological features of the space of lattice gauge
fields, in preparation.
[14] D.H. Adams, Some applications of families index theory to chiral gauge theories on the lattice, in
preparation.
[15] K.G. Wilson, Phys. Rev. D 10 (1974) 2445.
[16] D.H. Adams, J. Math. Phys. 42 (2001) 5522.
[17] P. Hernndez, K. Jansen, M. Lscher, Nucl. Phys. B 552 (1999) 363.
[18] H. Neuberger, Phys. Rev. D 61 (2000) 085015.
[20] P. Ginsparg, K.G. Wilson, Phys. Rev. D 25 (1982) 2649.
[21] P. Hasenfratz, V. Laliena, F. Niedermayer, Phys. Lett. B 427 (1998) 125.
[22] M. Lscher, Phys. Lett. B 428 (1998) 342.
[23] M. Lscher, Nucl. Phys. B 549 (1999) 295.
[24] M. Lscher, Nucl. Phys. B 568 (2000) 162.
[25] M.F. Atiyah, K-Theory, Benjamin, New York, 1967.
[26] Y. Kikukawa, A. Yamada, hep-lat/9810024.
[27] E. Witten, Phys. Lett. B 117 (1982) 324.
[28] P. Hasenfratz, in: M. Asorey, A. Dobado (Eds.), Proceedings of the Advanced School of Nonperturbative
Quantum Field Physics, Peiscola, 1997, World Scientific, Singapore, 1998.
[29] T.-W. Chiu, hep-lat/0106012;
T.-W. Chiu, Phys. Lett. B 488 (2000) 31.
[30] J. Dodziuk, Am. J. Math. 98 (1976) 79.

Fermionic determinant with a linear domain wall

in 2 + 1 dimensions
L. Da Rold a , C.D. Fosco a , A.P.C. Malbouisson b
a Centro Atmico Bariloche, Instituto Balseiro, Comisin Nacional de Energa Atmica,
8400 Bariloche, Argentina

b Centro Brasileiro de Pesquisas Fsicas, CBPF/MCT, Rua Dr. Xavier Sigaud, 150,
22290-180 Rio de Janeiro, RJ, Brazil

Abstract
We consider a Dirac field in 2 + 1 Euclidean dimensions, in the presence of a linear domain wall
defect in its mass, and a constant electromagnetic field. We evaluate the exact fermionic determinant
for the situation where the defect is assumed to be rectilinear, static, and the gauge field is minimally
coupled to the fermions. We discuss the dependence of the result on the (unique) independent
geometrical parameter of this system, namely, the relative orientation of the wall and the direction
of the external field. We apply the result for the determinant to the evaluation of the vacuum energy.
2002 Published by Elsevier Science B.V.
PACS: 11.10.Wx; 11.15.Pg
1. Introduction
In general, the fermionic action S is (or may be transformed into) a quadratic expression
in the Grassmann fields, which is in turn determined by a Dirac operator D. This Dirac
operator, except for some trivial cases, carries a dependence on some parameters and fields.
At some stage in the resolution of a dynamical problem, it may be useful to regard those
fields as external, either because they dont have a proper dynamics, or because they have
not yet been quantized.
E-mail addresses: daroldl@ib.cnea.gov.ar (L. Da Rold), fosco@cab.cnea.gov.ar (C.D. Fosco),

adolfo@cbpf.br (A.P.C. Malbouisson).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 5 1 - 4
486
L. Da Rold et al. / Nuclear Physics B 624 [PM] (2002) 485499
The fermionic determinant is formally given by the product of the eigenvalues of

the Dirac operator. The explicit calculation of those eigenvalues is, however, impossible
to achieve for an arbitrarily general situation, and one has to resort to some form
of approximate expansion. Nevertheless, there are many highly symmetrical external
field configurations where this explicit diagonalization can be successfully carried out.
In particular, the constant F case in several spacetime dimensions, allows for the
determination of the exact eigenvalues, since the diagonalization of D reduces in this case
to finding the spectrum of a one dimensional harmonic oscillator [1]. Many interesting
results have been found in this area, and they have application, for example, to the
determination of effective actions [2].
In most of these cases, the method of resolution amounts to expressing first the Dirac
operator D as a quadratic form in the operators1 x , p (as in the FockSchwinger
method [1]). The diagonalization procedure is then equivalent to defining a transformation
from x and p to a new set of canonical operators, and thus may in general be expressed
as a Bogoliubov transformation in some suitable creation and annihilation operators. The
original problem becomes then an algebraic one, with a complexity which depends of
course on the kind of situation considered.
In this article we will evaluate the fermionic determinant corresponding to a Dirac field
in 2 + 1 dimensions, coupled to an external uniform electromagnetic field F , and with
a mass term which is a linear functional of the coordinates. This kind of mass term is
a simple example of a situation where a domain wall like defect (of constant slope) is
present, in this case in the parity-breaking mass term. We note that, if the electromagnetic
field were parallel to the defect, the situation would fall into the well-known CallanHarvey
phenomenon [3], by which a chiral fermionic zero mode is induced on the defect. We
will, however, consider situations where the field has a different orientation. It should be
remarked that a linear increasing background field as we consider here, can arise only in
theories having flat directions. This is the case in globally supersymmetric theories which
admit topologically stable solutions. In this case linear solutions of the type of Eq. (4)
below arise naturally. We refer the interested reader to Ref. [4] to a recent account on the
subject in the context of dynamical supersymmetry breaking.
We will show that the modulus of the determinant can be found exactly for the above
configurations, what represents a non-trivial generalization of the well-known situation
corresponding to a uniform electromagnetic field and a constant mass term.
The Euclidean action S for this system is

D,
S = d 3x
(1)
where D = / + iA
/ + M, and we have absorbed the electric charge into the definition of the
gauge field A. The Hermitian -matrices are in the irreducible 2 2 representation of the
Dirac algebra:
{ , } = 2 .
1 We work with Euclidean coordinates: x = (x , x , x ), where x denotes the Euclidean time.
0 1 2
0
(2)
487
The domain wall, defined as the region where the mass changes sign, will in our case
have only two defining properties: its location and its slope. The latter determines the
localization (or not) of the defect, and is quantitatively measured by the normal derivative
of the mass along the curve of the defect. The situation we consider in this paper may be
considered as an approximate treatment of localized defects; a general discussion on the
localization of modes for this system may be found in [5].
The Dirac operator D appearing in (1) is not the more suitable one to deal with the
eigenvalue problem, since it is not Hermitian. It is convenient to consider, instead, a
related operator H which is always Hermitian, has a complete set of eigenstates, and its
eigenvalues are the squares of the Dirac operators eigenvalues. This operator H is, of
course,
H = D D,
(3)
which is indeed the operator defining the positive Hermitian part of D in its polar
decomposition. We can see from the structure of H, that for a mass linear in the
coordinates, and with a constant electromagnetic potential, it will be a quadratic expression
in x and p . The mass will be regarded as a linear functional of the coordinates.
This kind of configuration for the mass and the electromagnetic field can be
characterized by two vectors in Euclidean spacetime. For the mass, vanishing along a line
that passes through the origin (what is always possible by a proper choice of coordinates),
can be defined in terms of a vector ,
M = x ,
(4)
so that points in the direction normal to the plane that defines the defect hyperplane.
Regarding the electromagnetic potential A , a linear function of the coordinates suffices
to generate a constant electromagnetic field F , since the gauge potential can be written
in a symmetric gauge as follows:
1
A = F x .
(5)
2
We have found it convenient to use, rather than the antisymmetric tensor, F , its dual
,
F
= 1 F ,
F
(6)
2
where is the totally antisymmetric tensor in 3 dimensions. This puts both the defect
and the gauge field on an equal footing, simplifying the study of the dependence of the
fermionic determinant on the geometric invariants that may be built out of those external
fields.
x , one can present the Dirac
Writing A in terms of the dual of F , A = 12 F
operator parametrized by two constant vectors, thus summarizing all its dependence on the
external parameters:
i
+ x,
D= xF
2
(7)
488
where we have used the notation

(a b) a b ,
a b a b .
(8)
covers all the freedom to describe the

The relative orientation between and F
different geometrical configurations allowed within this model for the Dirac operator.
Different particular cases will describe rather different situations, both from the physical
,
and the mathematical points of view. The simplest situation corresponds to parallel to F
as in the Callan and Harvey mechanism, since the gauge field is then entirely contained in
the defect worldsheet. Then the determinant of the three-dimensional system immediately
factorizes into the product of an infinite number of (decoupled) determinants in 2 + 1
dimensions [5]. Our presentation begins, in Section 2, with the more interesting case
= 0. Namely, the electromagnetic field is normal to the defect hypersurface. The
F
interest in this case stems from the fact that it takes into account the competition between
the localizing effect due to the defect and the effect of the electromagnetic field, which will
tend to drag any charge distribution in a direction normal to the defect. An application to
the calculation of the vacuum energy, for this case, is presented in Section 3. Finally, we
conclude by deriving the eigenvalues of the equivalent diagonal operator for the general
case in Section 4.
2. Fermionic determinant for F = 0
and define two orthogonal directions under the scalar Euclidean
In this situation, F
product defined in (8). Thus, in terms of these two vector fields, we may construct two
families of curves in the Euclidean 3-dimensional space that may be used to define
coordinates, varying along their integral lines. A third family of curves, orthogonal to the
two previous ones, can then be uniquely defined by selecting a right-handed orientation.
. It is obvious from Eq. (7) that in
We define t to be the unit vector in the direction of F
the particular case we are dealing with, the operator D does not depend on the coordinate xt
(which parametrizes the integral curves of t ). We see that it is then sufficient to diagonalize
the part of H which depends on the other two coordinates. Indeed, using Eq. (7), we can
express H in terms of the external parameters,

1 2 2
)2 + (x )2 i(x ) F
+ (F
),
x F (x F
4
which, in the coordinates defined by the unit vectors
H = 2 +

F
,

F
b t,
(9)
(10)
becomes

2
2 2
F
F
+ 2 x2 +
x
4
4 b
(xb x b ) + (F
).
iF
H = 2 t2 b2 +
(11)
489
In this expression, H is invariant under translations in xt , thus the diagonalization of H

amounts to solving the reduced problem defined by a different Hamiltonian H acting on
functions depending on the two coordinates x and xb ,
H t2 + 2H,
(12)
where

2
1
1
2
1
F
).
H = p2 + pb2 + n x2 + b xb2 + (xb p x pb ) + (F
2
2
2
2
2
2
In (13) we have introduced the constants
(13)
2
2
F
F
+ 2 ,
.
b2 =
(14)
4
4
To study the diagonalization of H , it is first convenient to define suitable creation and
annihilation operators for each coordinate,
2 =
1 1/2
1/2
aj = j xj + ij pj ,
2
1/2
1
1/2
aj = j xj ij pj ,
2
j = , b
(15)
(no sum over j ), since the expression for H becomes more symmetric, and its
diagonalization can be done by a generalization of the usual procedure for a quadratic
Hamiltonian.
In terms of these operators, H contains both bilinear and constant terms:
b

F
H = a a + b ab ab +
a ab a ab
4i
b

1
b
F
) ,
a ab a ab + + b + (F
+
+
(16)
4i
b
2
which has the important property of containing terms which are quadratic in the creation
and annihilation operators, and not just the standard one with creation and annihilation
operators mixed by a Hermitian matrix. This makes the diagonalization more cumbersome.
Indeed, to diagonalize H , we first introduce some 2 2 matrices, defined by

i
i
1
) ,
H = aj Aj l al + aj Bj l al aj Bj l al + + b + (F
2
2
2
where A is Hermitian while B is real and symmetric

iu
0 t
A=
,
B=
,
t 0
iu b
with the constants u and t defined by
b
b
F
F
u=
+
,
t=
.
4i
b
4i
b
(17)
(18)
(19)
490
The diagonalization is non trivial due to the presence of the B matrix, which only vanishes
for = 0. To diagonalize the bilinear term in Eq. (17), we shall need to define new
creation and annihilation operators, which will be linear combinations of the ones defined
in Eq. (15). We first note that the aj , aj operators have been defined in such a way that
they verify the commutation relations:

aj , al = [aj , al ] = 0,
aj , al = j l ,

[H, al ] = aj Aj l + iajBj l ,
H, al = aj Aj l + iaj Bj l ,
(20)
where A is the complex conjugate of A. We require the new operators bj , bj to be also
canonical and to diagonalize the bilinear form of (17). Following [6], we require them to
satisfy the commutation relations,

bj , bl = j l ,
bj , bl = [bj , bl ] = 0,

[H, bl ] = bl El
H, bl = blEl ,
(21)
(no sum over l). We can write the new operators in terms of the old ones by using a general
Bogoliubov transformation of the following kind:
bl = aj Uj l aj Vj l ,
bl = aj Vjl + aj Ujl ,
(22)
where, to fulfill the commutation relations above, the coefficients U and V must verify the
matrix equations
U U V V = I,
(23)
U V V U = 0,
(24)
AU iBV = U E,
(25)
and
iBU + A V = V E,
(26)
U ,
as usual,
E being the diagonal matrix whose elements are the eigenvalues of H and
stands for the adjoint of U . Similar redefinitions, although for a different case, where used
in [6] and [7].
The system of equations involving the coefficient matrices defines a generalized
diagonalization problem, which involves matrices rather than vectors. The next step is
to obtain E explicitly in terms of the parameters of the theory. We first note that not all
the equations that define this eigensystem are independent. To determine E we may, for
example, use Eq. (25) to express the coefficients Vj l in terms of Uj l , the parameters of the
theory and the eigenvalues El . This yields

r11 U11 + s11 U21 r12 U12 + s12 U22
,
V=
(27)
r21 U21 + s21 U11 r22 U22 + s22 U12
where the coefficients rj l and sj l are defined by
i(E1 b ) i(E2 b )
u
u
t
t
t
t
r=
,
s = i(E ) i(E ) .
1
2
ut ut
t
(28)
491
Introducing this result into Eq. (26) yields two equations, now involving U11 and U21 , plus
two others involving U12 and U22 ,
U21 = U11
2i u
E12 + t 2 E1 b u2 + E1 b
= iU11
E12 + t 2 E1 u2 + E1 b b
,
2b u
(29)
and
U12 = U22
2ib u
E22 + t 2 E2 u2 + E2 b b
= iU22
E22 + t 2 E2 b u2 + E2 b
.
2 u
(30)
Demanding Eqs. (29) and (30) to be consistent, implies a set of constraints for the
eigenvalues El . After some algebra, those constraints may be written in terms of just one
equation, that determines the possible values of El in terms of the external parameters.
There are 4 solutions to this equation: besides the double eigenvalue E1 = 0, we find:

E2 = E3 = 2 + 3b2 .
(31)
This set of solutions contains the true eigenvalues; however, to discard the spurious ones,
we should check whether they are consistent with (23) and (24), which guarantee the
canonical commutation relations for the new operators. In terms of the variables defined
in Eq. (27), (24) can be recast in the form,
pq(s22 s11 ) + q(r22 r21 ) + p(r12 r11 ) + (s12 s21 ) = 0,
(32)
where p and q are defined by:

p = i
Ej2 + t 2 Ej b u2 + Ej b
2 u
(33)
and
q =i
2 u
Ek2
+ t 2 Ek b u2 + Ek b
(34)
where the labels j and k stand for the two different eigenvalues that can be chosen from
the four possibilities in Eq. (31).
Eq. (23) splits into three algebraic equations, one corresponding to the non-diagonal
elements,
p + q (r11 + qs11 ) (r12 p + s12 ) (r21 q + S21 ) (r22 + ps22 ) = 0,
plus two others for the diagonal terms,

|U11 |2 1 + |q|2 |r11 + qs11 |2 |r21 q + s21 |2 = 1,

|U22 |2 1 + |p|2 |r12 p + s12 |2 |r22 + ps22 |2 = 1.
(35)
(36)
(37)
492
Eqs. (32) and (35) do not contain the coefficients of the matrices U and V , and thus they
are equations for the eigenvalues El . In terms of the original parameters, we obtain for the
eigenvalues of H ,

2 + 2 .
E2 = 2 + 3b2 = F
E1 = 0,
(38)
Note that (36) and (37) only fix the moduli of U11 and U22 ; the phase of these coefficients
is not fixed by the basis choice, and on the other hand no physical magnitude will depend
on it.
We note that one of the eigenvalues vanishes, what implies |U11 | . This means
that, in the direction corresponding to the operators b1 and b1 , there is no harmonic mode,
but rather a free motion. Thus we define the corresponding conjugate x1 and p1 variables,
such that H in (11) becomes
).
H = pt2 + p12 + 2E2 b2 b2 + + b + (F
(39)
We have thus obtained the diagonal form for the operator H. From this expression, we can
obtain the modulus of the determinant of the Dirac operator D.
3. Vacuum energy in a constant electromagnetic field

In this section we evaluate the vacuum energy for the fermionic system studied in the
= 0); we shall see that this physical magnitude is entirely
previous section (i.e., when F
determined by (9).
To that end we first consider the vacuum energy density for a fermionic system in the
presence of an external electromagnetic field. The vacuum to vacuum transition probability
amplitude is given by the S matrix expectation value between vacuum states, which
depends on the external potential A,
S0 (A) = 0 in |S|0 in.
(40)
This object is usually normalized with respect to the transition amplitude in the absence of
external fields, S0 (A):
2

S (A) 2 |S0 (A)| .
0
|S0 (0)|2
(41)
In terms of the normalized transition amplitude S0 (A), one then defines a local function
w(x) [8]

S (A) 2 exp d n x w(x) ,
(42)
0
where n denotes the spacetime dimension. In the Euclidean formulation we can interpret
w(x) as half the vacuum energy density. To see this we consider the Euclidean evolution
operator in the interaction representation U (, 0 ), which obeys the equations [9],

lim Tr U
(43)
,
= 0, in |U (, )|0, in = S0 (A)
2
2
and

1
lim ln Tr U
,
= E0 ,
2
2
493
(44)
where E0 is the vacuum energy. From these equations and the definition (42) we see that
the integral of w(x) over the spatial coordinates is half the vacuum energy. In order to
obtain w(x), we first have to evaluate S0 (A). In the functional integral representation, we
may write

D e d 3 x D[A] ,
S0 (A) = |N |2 D
(45)
where the notation D[A] indicates the Dirac operator dependence on the external field A.
Since the integral is over Grassmannian variables, we write
S0 (A) = |N |2 det(D[A]),
(46)
where det stands for the determinant over both spinor and functional spaces. Inserting (46)
into the definition (41), we see that

|S0 (A)|2 = exp Tr ln H[A] Tr ln H[0] .
(47)
Now, we use the Frullanis identity:
a
ln = lim
b 0

ds sb
e
esa
s
(48)
to rewrite Eq. (47) in a particularly convenient integral representation:

log |S0 (A)|2 = Tr lim
0

ds s H[A]
e
es H[0] ,
s
(49)
whence we can obtain the vacuum energy density as:

w(x) = tr lim
0

ds s H[A]
e
es H[0] ,
s
(50)
where now the trace only affects the spinor space indices.
We evaluate now the vacuum energy, the integral of w(x) over all the Euclidean space,
for the previously described case. The operators H[A] and H[0] are, respectively, given
by,
)
H[A] = pt2 + p12 + 2b2b2 E2 + n + b + (F
(51)
H[0] = p2 + M 2 .
(52)
and
494
We can write H[0] in terms of some creation and annihilation operators c and c , defined
as in Eq. (15), with frequency = , as follows:

H[0] = pt2 + pb2 + 2c c + 1 .
(53)
Evaluating the trace, we see that:

w(x) = 2 lim
0

ds
2
2
x|es[pt +p1 +2b2 b2 + +b ] esE2 esE2 |x
s

2
2
es[pt +pb +2c c+] es es |x .
(54)
The system we are considering is not translation invariant, and moreover, we may write
the explicit coordinate dependence of w(x) by using the completeness of the eigenstates
of the operators b2b2 and c c for the respective Hilbert spaces. Denoting by x2 |m2 and
x |m the respective eigenstates of the operators b2 b2 and c c, which are as usual labeled
by a non-negative integer m = 0, 1, . . . , , then we may write w(x) as

w(x) = 2 lim
0

2

ds
2
2
x2 |m2 x xb |es[pt +p1 +2m+ +b ] |x xb esE2 esE2
s
m=0
2

2
2
x |m xt xb |es[pt +pb +2m+] |xt xb es es .
(55)
It should be obvious then that each term in w(x) in in fact a function of only one coordinate,
x2 . Indeed, one may even check that each term in the series above may be regarded as a
(1 + 1)-dimensional determinant times the square of the amplitude for the corresponding
harmonic oscillator mode.
The vacuum energy density in terms of the eigenstates of harmonic modes b2 b2 and c c.
To obtain the total vacuum energy, W , we have to integrate over the phase space. After this
step the sum over m can be performed and we obtain,
L2
L2
lim w
lim
W=
2 0
2 0

ds s( +b E2 )
e
coth(sE2 ) coth(s) .
s2
(56)
The L2 factor appears because we have considered the space volume to be a square
box. This equation expresses the vacuum energy of a fermionic system with a domain
wall that can be approximated linearly near the defect, and in the presence of a constant
= 0). We can see that in the case
electromagnetic field, for a particular configuration (F
where the electromagnetic field is absent the vacuum energy vanishes, which is consistent
with our non interacting fermions normalization condition.
The regularized expression for W diverges as goes to zero. The divergences in
Eq. (56) should be isolated in some terms as a first step and afterwards suppressed by some
renormalization procedure. To determine the divergent parts in , we use the expansion,
coth(u) =
1 22k B2k 2k1

+
u
,
u
(2k)!
k=1
(57)
495
valid for u2 < 2 , B2k , being the Bernoulli numbers. We obtain,

A
A
L2 A2
W =
(2, A) 2,
+ B2 E2 (0, A) 0,
2 E2
E2
E2

2k

2 B2k E22k1 A22k
A
||
(2k 2, A) 2k 2,
+
(2k)!
E2
2 2
k=2
B2 ||
2k

2 B2k 2k2 ||
1
log
+
B
||
log

2
||
(2k)!
2|| 2
k=2

+ F1 (w , wb , E2 ) + F2 (||) ,
(58)
where

F1 (w , wb , E2 ) =
ds s 2 esA coth(sE2 ),
/E2

F2 (||) =
ds s 2 coth(s||),
(59)
/||
A = w + wb E2 and (, x) is the incomplete Gamma-function. In the expression (58)

singularities are still present in the terms (2, A) and (0, A). To isolate them we use
the formula,
(, x) = ex x

Ln
,
n+1
(60)
n=0
We get at the limit 0, (2, A)

where Ln are the Laguerre
polynomials.
1
1
and
(0,
A)
.
This
last
series diverges logarithmically. After some
2
2
0
n+1
2A
rearrangements, we obtain,

L2
1
1
1
1
+ B2 E2
+ B2 || log
W =
2
2E2 2|| 2
n+1
n=0

B2 ||

A
A
A2
B2 E2 0,
log
2,
E2
E2
E2
||

2k
2

B2k 2E2
A
A
+
(2k 2, A) 2k 2,
E2
(2k)! A
E2
k=2

||
1 (2)2k B2k
2 +
(61)
+ F1 (w , wb , E2 ) + F2 (||) .
2
(2k)!
k=2
The two first terms in (61), contain quadratic and logarithmic divergences which may
be suppressed by a renormalization procedure. It may be, at first sight, shocking to see that
496
the divergent part of W is in fact not a finite degree polynomial in the external field F
and its derivatives, as the usual divergence theorems for a standard Quantum Field Theory
imply [8]. The reason for this seemingly contradictory result is that the hypothesis for those
theorems do not hold in the present case. Firstly, the mass term of the free fermionic field
is not a constant, thus the inhomogeneity of space already changes the main assumptions,
like power counting behaviour. Note that in our case there is no unique large momentum
appears (after
behaviour for the propagator. And secondly, the would be external field F
diagonalization) in such a way that it plays a similar role to the inhomogeneous mass.
Then the external field is more likely to appear in a similar way to a mass in a standard
divergent expression, which is indeed a non-polynomial dependence. In spite of this, a
renormalization prescription may indeed be used for this quantity. For example, we may
realize that W , if regarded as a function of E2 , has divergences where E2 appear as a pole,
a constant or a linear term in E2 . Thus we may cancel all the divergences of this system by
including three counterterms, namely, by adding to W the 1-loop counterterm action W
defined by:
W = 1 (E2 )1 + 0 + 1 E2 ,
(62)
where the j s are divergent constants. These three constants require the use of some
renormalization conditions to fix them; in our case one could of course use the Laurent
expansion of the actually measured W in order to fix those coefficients. It is amusing to
, since
note that this procedure requires the knowledge of the full dependence of W on F
.
the divergent part is not just a single polynomial in F
4. The general case

and F
, respectively, the projections of F
onto the direction of and onto
Let us call F
a direction orthogonal to (with respect to the scalar product defined in Section 2). Then
the square modulus of the Dirac operator for general configuration of the external vectors,
can be written in the form,

1 2 2 2
) 2 + (x.)2
x F + F x.(F + F
4
+ F
) + .(F
).
ix .(F
H = 2 +
(63)
where now t is a unit vector in

In the same way as before we define versors ,
t and b,
, i.e., in the direction
the direction of F
F
.
F
(64)
We can now define creation and destruction operators a and a , analogously as it has
been done in Eq. (15), with = , t, b, and defined by
2 =
2
F
+ 2 ,
4
t2 =
2
F

4
b2 =
2
2 + F
F
(65)
497
In terms of these, H can be written as,

t (b2 t2 ) 1/2
H=
at a + at a + at a + at a
2a a + 1 4
n

1/2

b
ab a ab a
i b2 t2

2

b
1/2
ab a ab a
i b t2
+
b

b
t
b
t
at ab at ab it
at ab at ab .
+
it
t
b
t
b
(66)
We see that H is again a bilinear form in the creation and destruction operators a and
a :
i
i
H = aj Aj l al aj Bj l al + aj Bjl al + constant,
(67)
2
2
where Aj l and Bj l are components of 3 3 matrices, A Hermitian and B symmetric:

iv
0 it u
2 t
B = it
A = t 2t ir ,
(68)
0
s ,
iv ir 2b
u
s
0
with the constants defined by

b
t
b
t
t
+
s=
t =4
b2 t2 ,
t ,
t ,
r=
t
b
t
b

b
b
u=
(69)
b2 t2 ,
v=
+
b2 t2 .
b
We also see that in this case it is possible to use a transformation, defined in terms of
3 3 matrices U and V , such that the Hamiltonian is diagonalized. The properties of this
transformation can be summarized by the equations:
U U V V = I,
UtV V tU = 0
(70)
and
AU + iB V = U E,
iBU + A V = V E.
(71)
As in the simpler 2 2 case, one may obtain linear equations that determine the
transformation matrices U and V . In order for those equations to have a non-trivial
solution, it is necessary to demand the condition:

det A (B )1 A B + (B )1 A A (B )1 E (B )1 E 2 = 0,
(72)
where E are the energies, i.e., the numbers appearing in the diagonal of the matrix E.
Eq. (72) may be explicitly solved, what yields 6 solutions. Of these 6 solutions, we should
eliminate 3 spurious ones, since they are not consistent with the algebraic relations that are
498
verified by U and V . This leads to the following three solutions. One of them vanishes:
E1 = 0, and the other two have a rather cumbersome expression, which may be written in
terms of the previously defined parameters as follows:

E2 = 2 2 + t2 + 3b2 1 b2 t2 64t3 + 96t2 t

1/2 1/2
(73)
,
+ 4 142 t2 + 13t4 + 6 2 t2 b2 + 9b4

E3 = 2 2 + t2 + 3b2 + 1 b2 t2 64t3 + 96t2 t

1/2 1/2
+ 4 142 t2 + 13t4 + 6 2 t2 b2 + 9b4
(74)
.
From these solutions the Hamiltonian operator can be written in diagonal form and the
and . Calculations in this
determinant of the Dirac operator can be obtained for general F
case are more involved, but following along the same lines as we have done in Section 3, it
should be possible to obtain the vacuum energy in the general case. Progress on this subject
will be reported elsewhere.
In this article we have evaluated the fermionic determinant corresponding to a Dirac
field in 2 + 1 dimensions, coupled to an external uniform electromagnetic field F , in
presence of a domain wall like defect (of constant slope), i.e., a mass term which is a
linear functional of the coordinates. This kind of mass term is a simple example of a
situation where a domain wall like defect is present in the parity-breaking mass term.
We note that, if the electromagnetic field is parallel to the defect, the situation fall into
the well-known CallanHarvey phenomenon [3], by which a chiral fermionic zero mode
is induced on the defect. We have considered situations where the field has a different
orientation. In particular, we have considered the case in which the electromagnetic field
is normal to the defect hypersurface. The interest in this case stems from the fact that
it takes into account the competition between the localizing effect due to the defect and
the effect of the electromagnetic field, which tend to drag any charge distribution in a
direction normal to the defect. An application to the calculation of the vacuum energy, for
this case, is presented in Section 3. Finally, we conclude by deriving the eigenvalues of
the equivalent diagonal operator for the general case in Section 4. We have shown that the
modulus of the determinant can be found exactly for the above mentioned configurations,
what represents a non-trivial generalization of the well-known situation corresponding to
a uniform electromagnetic field and a constant mass term.
Acknowledgements
This work was part of a collaboration supported by CLAF (Centro Latinoamericano de
Fsica). We also acknowledge financial support by Fundacin Antorchas, Argentina.
499
References
[1] J. Schwinger, Phys. Rev. 82 (1951) 664.
[2] See, for example, W. Dittrich, M. Reuter, Effective Lagrangians in Quantum Electrodynamics, Springer,
Berlin, 1985.
[3] C.G. Callan, J.A. Harvey, Nucl. Phys. B 250 (1985) 427.
[4] G. Dvali, M. Shifman, Nucl. Phys. 504 (1997) 127.
[5] C.D. Fosco, A. Lpez, Nucl. Phys. B 538 (1999) 685.
[6] J. Avery, Creation and Annihilation Operators, McGrawHill, 1976.
[7] P. Jrgensen, J. Linderberg, Int. J. Quantum Chem. 4 (1970) 587;
S.V. Tyablikov, Methods in the Quantum Theory of Magnetism, Plenum, 1970.
[8] C. Itzykson, J.-B. Zuber, Quantum Field Theory, McGrawHill, 1986, Section 4.3.
[9] J. Zinn-Justin, Quantum Field Theory and Critical Phenomena, 2nd edition, Oxford, 1993, Chapter 2,
Section 30.3.

On a Z3 -graded generalization of the Witten index

Ali Mostafazadeh
Department of Mathematics, Ko University, Rumelifeneri Yolu, 80910 Sariyer, Istanbul, Turkey
Abstract
We construct a realization of the algebra of the Z3 -graded topological symmetry of type (1, 1, 1) in
terms of a pair of operators D1 : H1 H2 , D2 : H2 H3 satisfying [D1 D1 , D2 D2 ] = 0. We show
that the sequence of the restriction of these operators to the zero-energy subspace forms a complex
and establish the equality of the corresponding topological invariants with the analytic indices of
these operators. 2002 Published by Elsevier Science B.V.
1. Introduction
Since the introduction of the supersymmetric quantum mechanics, there has been an
ongoing effort to develop and study its generalizations. Recently, we have offered in [1]
a complete characterization of a class of generalizations of supersymmetry that share its
intriguing topological properties. These symmetries that are called topological symmetries
involve certain integer-valued topological invariants that are the analogs of the Witten index
of supersymmetry. In Ref. [1], we have presented a crude description of the topological
invariants associated with a class of topological symmetries in terms of certain operators,
D : H H+1 . These operators are however subject to a number of rather complicated
compatibility conditions. The purpose of this article is to give a nontrivial solution of these
compatibility conditions for the case of Z3 -graded topological symmetry of type (1, 1, 1).
In view of the fact that supersymmetry coincides with the Z2 -graded topological symmetry
of type (1, 1), Z3 -graded topological symmetry of type (1, 1, 1) is a simplest generalization
of supersymmetry.
The main motivation for the study of these symmetries is to find out whether they
lead to more general topological invariants than the Witten index. The latter is known
to be identical with the analytic index of a Fredholm operator [2]. This has been the
E-mail address: amostafazadeh@ku.edu.tr (A. Mostafazadeh).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 3 0 - 7
A. Mostafazadeh / Nuclear Physics B 624 [PM] (2002) 500508
501
key observation in developing the supersymmetric proofs of the AtiyahSinger index

theorem [3]. Historically, the fact that the analytic index of a Fredholm operator is a
topological invariant was known much earlier than the actual proof of the index theorem.
See, for example, [4]. Therefore, a relevant question is whether the Z3 -graded topological
symmetry of type (1, 1, 1) leads to a topological invariant that is more general than the
analytic index of a Fredholm operator.
The organization of this article is as follows. In Section 2 we recall the definition of
the topological symmetries and review their basic properties. In Section 3 we consider the
Z3 -graded topological symmetry of type (1, 1, 1) in its three-component representation and
derive the compatibility conditions for the operators appearing in this representation. In
Section 4 we discuss a particular nontrivial solution of these compatibility conditions and
describe the corresponding topological invariants. In Section 5 we present our concluding
remarks.
2. Topological symmetries
We begin our analysis by a review of the basic properties of topological symmetries.
A more detailed discussion can be found in [1].
A quantum system is said to possess a Zn -graded topological symmetry (TS) of type
(m1 , m2 , . . . , mn ) iff the following conditions are satisfied:
1. The quantum system is Zn -graded. This means that the Hilbert space H of the quantum
system is the direct sum of n of its (nontrivial) subspaces H , and its Hamiltonian has
a complete set of eigenvectors with definite color or grading. (A state is said to have a
definite grading c iff it belongs to H );
2. The energy spectrum is nonnegative;
3. For
n every eigenvalue E there corresponds a positive integer E such that E is
=1 m E -fold degenerate, and the corresponding eigenspaces are spanned by m1 E
vectors of grade c1 , m2 E vectors of grade c2 , . . . , and mn E vectors of grade cn .
For a system with a TS we can introduce a set of integer-valued topological invariants,
namely
(0)
(0)
ij := mi nj mj ni ,
(1)
where i, j {1, 2, . . . , n} and n(0)

denotes the number of zero-energy states of grade c .
It turns out that using the above definition of TS, one can obtain the underlying operator
algebras supporting these symmetries. In particular, Z2 -graded TS of type (1, 1) coincides
with supersymmetry and 11 yields the Witten index. Similarly, one obtains the algebras of
parasupersymmetry of order 2 and fractional supersymmetry of arbitrary order as special
cases of TSs.
In this article we shall confine our attention to the case of Z3 -graded TS of type (1, 1, 1).
Following [1], we grade the state vectors by associating them with a third root of unity, i.e.,
502
we use a grading operator satisfying

= 1 ,
3 = 1,
[H, ] = 0,
[, Q]q = 0,
(2)
where Q is the generator of TS, q := e2i/3 , and [ , ]q stands for the q-commutator,
[O1 , O2 ]q := O1 O2 qO2O1 . The operator algebra for Z3 -graded TS of type (1, 1, 1) has
the form
Q3 = K,
Q31
Q32

K +K ,
+ MQ1 = 2

+ MQ2 = 23/2i 3 K + K ,
3/2
(3)
(4)
(5)
where K and M are operators commuting with all other operators, M is self-adjoint, and
Q1 :=
Q + Q
,
Q2 :=
Q Q
.
2i
(6)
For K = H , Eq. (3) coincides with the defining relation for the fractional supersymmetry
of order 3 [5].
It is not difficult to show [1] that Eq. (3) implies the particular degeneracy structure
of the Z3 -graded TS of type (1, 1, 1) provided that ker(K) ker(H ). In particular, a
quantum system with a Hamiltonian H has a Z3 -graded TS of type (1, 1, 1) if the following
conditions are satisfied:
1. There are operators and Q satisfying Eqs. (2);
2. The spectrum of H is nonnegative;
3. Eq. (3) holds for K = H .
The presence of this particular topological symmetry in turn implies the existence of a
self-adjoint operator M that commutes with and Q and fulfills Eqs. (4) and (5).
Hereafter, we shall set K = H .
3. Three-component realization of Z3 -graded TS of type (1, 1, 1)

Similarly to the case of supersymmetric quantum mechanics, we can obtain a realization
of the Z3 -graded TS by identifying the state vectors with column vectors whose rows
have definite grade. We first let H1 , H2 , and H
3 be three Hilbert spaces and define the
total Hilbert space H as their inner sum, H := 3=1 H . Every state vector | H can
be written as the sum of its components | belonging to H . In the three-component
representation of H, this is expressed as

| =

|1
|2 .
|3
In this representation, is defined by

q 0 0
2
:= 0 q 0 ,
0 0 1
503
(7)
and the operators Q, H, M take the form

0
0 D3
Q = D1 0
0 ,
0 D2 0

H1 0
0
H = 0 H2 0 ,
0
0 H3

M1 0
0
M = 0 M2 0 ,
0
0 M3
(8)
(9)
(10)
where D1 : H1 H2 , D2 : H2 H3 , D3 : H3 H1 , H : H H , and M : H
H , with {1, 2, 3}, are linear operators, and H and M are self-adjoint. Eqs. (8), (9),
and (10) are direct consequences of Eq. (2).
The condition that H and M commute with Q is equivalent to the statement that
(H1 , H2 , H3 ) and (M1 , M2 , M3 ) are triplets of isospectral operators, as substituting (9),
(8), and (10) in [H, Q] = 0 and [M, Q] = 0 yields
D1 H1 = H2 D1 ,
D2 H2 = H3 D2 ,
D3 H3 = H1 D3 ,
(11)
D1 M1 = M2 D1 ,
(12)
D2 M2 = M3 D2 ,
(13)
D3 M3 = M1 D3 .
(14)
Note that, in view of Eqs. (9) and (11), in order to ensure that the spectrum of H is
nonnegative it is sufficient to show that one of H has a nonnegative spectrum.
Next, we substitute (8) and (9) in Eq. (3) with K = H , i.e., set Q3 = H . This yields
H1 = D3 D2 D1 ,
H2 = D1 D3 D2 ,
H3 = D2 D1 D3 .
(15)
Clearly, Eqs. (15) agree with Eqs. (11). However, the condition that H is self-adjoint
implies H = H . This relation restricts the choice of the operators D and leads to the
first set of compatibility conditions for D , namely,
D1 D2 D3 = D3 D2 D1 ,
(16)
D2 D3 D1
D3 D1 D2
= D1 D3 D2 ,
(17)
= D2 D1 D3 .
(18)
Next, we use Eqs. (8) and (6) to express Q1 and Q2 in the three-component
representation. Substituting the result for Q1 in Eq. (4) and making use of (10), we find

M1 D3 = D3 D2 D2 + D3 D3 + D1 D1 D3 ,
(19)
504

M3 D2 = D2 D1 D1 + D2 D2 + D3 D3 D2 ,

M2 D1 = D2 D2 D3 + D1 D1 + D2 D2 D1 .
(20)
(21)
Combining these equations with Eqs. (12)(14), we find the second set of compatibility
conditions for D :
D2 D2 D1 D3 = D1 D3 D2 D2 ,
(22)
D3 D3 D2 D1
D1 D1 D3 D2
(23)
= D2 D1 D3 D3 ,
= D3 D2 D1 D1 .
(24)
Finally, we substitute the matrix representation of Q2 and Eq. (10) in Eq. (5).
The resulting matrix equation is trivially satisfied provided that we enforce (22)(24).
Therefore, Eq. (5) does not lead to further conditions on D .
We wish to emphasize that so far we obtained a set of necessary conditions, that is (16)
(18) and (22)(24), for having a Z3 -graded TS of type (1, 1, 1). These conditions are not
sufficient as we must also make sure that the energy spectrum is nonnegative. Furthermore,
as we mentioned above, Eq. (3) with the choice of K = H is sufficient for having the
necessary degeneracy structure for the Z3 -graded TS of type (1, 1, 1). The existence of M
and consequently M follows from the definition of the TS. Hence, the conditions (22),
(23) should be automatically satisfied provided that we can fulfill (16)(18). This in turns
means that if we can obtain a solution of (16)(18) and show that the energy spectrum is
nonnegative, then
(0)
(0)
ij = nj ni
(25)
are topological invariants.

In view of Eqs. (9) and (15), we also have n(0)
= dim(ker H ) and
12 = 21 = dim(ker D1 D3 D2 ) dim(ker D3 D2 D1 ),
(26)
23 = 32 = dim(ker D2 D1 D3 ) dim(ker D1 D3 D2 ),
(27)
13 = 31 = 12 + 23 ,
(28)
where dim and ker abbreviate dimension and kernel, respectively.
4. Solution of the compatibility conditions

In Ref. [1], we gave a simple solution of the compatibility conditions, namely
D2 = D1 ,
D3 = 1,
(29)
which applies to the case that H3 = H1 and D1 is a Fredholm operator. One can
easily check that all the compatibility conditions are satisfied for this choice of D . The
corresponding independent invariants are given by

12 = 23 = dim(ker D1 ) dim ker D1 = Analytic Index (D1 ),
13 = 0,
505
where we have made use of Eqs. (26)(29) and the fact that for any Fredholm operator D,

ker D D = ker(D).
(30)
In the following we construct a more interesting solution of the compatibility conditions
for D which applies for arbitrary choices of H . This solution corresponds to the choice
D3 = (D2 D1 ) = D1 D2 .
(31)
One can check that for this choice of D3 conditions (16) and (18) are automatically
fulfilled, whereas (17) imposes the following condition on D1 and D2 .

D1 D1 , D2 D2 = 0.
(32)
It turns out that this is the only condition on the operators D1 and D2 , as Eqs. (22), (23)
are trivially satisfied provided that (32) holds. This confirms our earlier remark that (22),
(23) should follow if we can satisfy (16)(18).
Furthermore, using (12) and (13) and setting
M2 = D1 D1 + D2 D2 + D1 D1 D2 D2 ,
(33)
we can also satisfy (19)(21).

Eq. (31) also implies that the Hamiltonian H has a nonnegative spectrum. In order to
see this we substitute Eq. (31) in Eq. (15). This yields
H1 = D3 D3 ,
H2 = D1 D1 D2 D2 ,
H3 = D3 D3 .
(34)
Clearly these isospectral operators have a nonnegative spectrum, and the energy spectrum
is nonnegative.
Next we compute the topological invariants associated with this solution. In view of
Eqs. (30), (32) and (26)(28), and the identity

ker D1 D1 D2 D2 = ker D1 D2 D2 ,
we have

12 = dim ker D1 D2 D2 dim ker D2 D1 ,

23 = dim ker(D2 D1 ) dim ker D1 D2 D2 ,
(35)
13 = Analytic Index (D2 D1 ).
(37)
(36)
In summary, we have so far shown that for any two operators D1 and D2 satisfying
condition (32), the quantities (35)(37) are topological invariants.
Next, we study the properties of the restrictions of the operators Di to the zero-energy
subspaces
Hi(0) := ker(Hi )
(38)
of Hi . First we note that according to Eqs. (15)(18) and (38),

(0)
ker(D1 ) ker(H1 ) =: H1 ,

ker D1 ker(H2 ) =: H2(0),
(0)
ker(D2 ) ker(H2 ) =: H2 ,

ker D2 ker(H3 ) =: H3(0) .
(39)
(40)
506

(0)
Furthermore, in view of (29), (34), and (38), we can easily show that for every |i Hi ,
H1 |1 = D3 D3 |1 = 0
H2 |2 = D1 D1 D2 D2 |2 = 0
H3 |3 = D3 D3 |3 = 0
D2 D1 |1 = D3 |1 = 0,
D1 D2 D2 |2 = 0,
(D2 D1 ) |3 = D3 |3 = 0.
(41)
(42)
(43)
According to these identities and relations (39), D1 |1 ker(D2 ), D2 |2 ker(D3 ).

Therefore, if we denote the restriction of Di to Hi(0) by Di(0) , we have

im D1(0) ker D2(0) ,
(44)
im D2(0) ker D3(0) ,
im D3(0) = {0},
where im abbreviates image. In other words, the sequence
(0)
(0) D1
(0)
(0) D2
(0)
(0) D3
{0} H1 H2 H3 {0}
is indeed a complex, and we can define the
(0)
(45)
(0)
corresponding cohomology groups ker(Di+1
)/
im(Di ) and Betti numbers

(0)

(0)

bi := dim ker Di+1
/ im Di(0) = dim ker Di+1
dim im Di(0) .
(46)
Next, we note that Hi(0) are assumed to be finite-dimensional. This implies that for
i {1, 2}

dim ker Di(0) + dim im Di(0) = dim Hi(0) ,
(47)

(0)
(0)
(0)
dim coker Di
(48)
+ dim im Di
= dim Hi+1 ,
where coker stands for cokernel. In view of Eqs. (25), (47) and (48), we then obtain
(0)

(0)
(0)
(0)
dim ker Di
i,i+1 = ni+1 ni = dim coker Di
(0)
(0)
= dim ker Di
(49)
dim ker Di
.
(0)
(0)
Note however that because ker(Di ) Hi and ker(Di ) Hi+1 , we have

(0)
(0)
= ker Di .
ker Di = ker(Di ),
ker Di
(50)
Combining Eqs. (49) and (50) we finally obtain, for i {1, 2},
i,i+1 = Analytic Index (Di ).
(51)
This is our main result. It indicates that for the solution (29) the Z3 -graded generalization
of the Witten index is nothing but the analytic index of a Fredholm operator. A corollary
of this result is the following.
Corollary. Let Di : Hi Hi+1 , for i {1, 2}, be Fredholm operators satisfying
[D1 D1 , D2 D2 ] = 0. Then

dim[ker D2 D1 ] dim ker D1 D2 D2 = Analytic Index (D1 ),
(52)

dim ker D1 D2 D2 dim ker(D2 D1 ) = Analytic Index (D2 ),

(53)
Analytic Index (D1 ) + Analytic Index (D2 ) = Analytic Index (D2 D1 ).
(54)
507
Proof. Eqs. (52) and (53) are direct consequences of Eqs. (35), (36), (28), and (51).
Eq. (54) follows from Eqs. (28), (52) and (53).
5. Conclusion
In this article we have explored the topological invariants of a certain class of
Z3 -graded topological symmetries of type (1, 1, 1). The operator algebra of Z3 -graded
topological symmetries of type (1, 1, 1) together with the choice K = H and the condition
of the nonnegativity of the energy spectrum imply the presence of this topological
symmetry and ensures the topological invariance of ij . This makes the construction of
concrete realizations of this symmetry more tractable. Using the algebraic structure and
manipulating the resulting compatibility conditions, we found a nontrivial solution of these
conditions. We explored the meaning of the associated topological invariants and showed
that they are related to the analytic indices of Fredholm operators. Therefore, for the
realization of the Z3 -graded topological symmetries of type (1, 1, 1) that we considered,
the generalized Witten index does not yield a more general topological invariant. This
situation is very similar to the topological invariants of p = 2 parasupersymmetry
(alternatively Z2 -graded topological symmetry of type (1, 2)) studied in Ref. [6]. There
also the known realization of the p = 2 parasupersymmetry does not lead to a more general
invariant.
Our investigation of the nature of the topological invariants associated with Z3 -graded
topological symmetries of type (1, 1, 1)which is a graded fractional supersymmetry of
order 3has also revealed the following interesting properties:
1. The graded Hamiltonians H are isospectral. This is very similar to the case of
supersymmetry and may be useful in obtaining exactly solvable potentials.
2. The restriction of the square of the symmetry generator Q to the zero-energy
eigensubspace H0 vanishes identically. This in turn implies ker Q2 = H0 .
Acknowledgements
I wish to thank K. Aghababaei Samani for reading the first draft of this paper and making
invaluable comments. This project was supported by the Young Researcher Award Program
of the Turkish Academy of Sciences (GEBIP).
References
[1] K. Aghababaei Samani, A. Mostafazadeh, Nucl. Phys. B 595 (2001) 467.
[2] E. Witten, Nucl. Phys. B 202 (1982) 253.
[3] L. Alvarez-Gaum, Commun. Math. Phys. 90 (1983) 161;
L. Alvarez-Gaum, J. Phys. A: Math. Gen. 16 (1983) 4177;
P. Windey, Acta Phys. Pol. B 15 (1984) 453;
A. Mostafazadeh, J. Math. Phys. 35 (1994) 1095.
508
[4] R.S. Palais, in: Seminar on the AtiyahSinger Index Theorem, Ann. of Math. Study, Vol. 57, Princeton Univ.
Press, Princeton, 1965.
[5] C. Ahn, D. Bernard, A. Leclair, Nucl. Phys. B 346 (1990) 409;
L. Baulieu, E.G. Floratos, Phys. Lett. B 258 (1991) 171;
R. Kerner, J. Math. Phys. 33 (1992) 403;
S. Durand, Phys. Lett. B 312 (1993) 115;
S. Durand, Mod. Phys. Lett. A 8 (1993) 1795;
S. Durand, Mod. Phys. Lett. A 8 (1993) 2323;
A.T. Filippov, A.P. Isaev, R.D. Kurdikov, Mod. Phys. Lett. A 7 (1993) 2129;
N. Mohammedi, Mod. Phys. Lett. A 10 (1995) 1287;
N. Fleury, M. Rausch de Traubenberg, Mod. Phys. Lett. A 11 (1996) 2899;
J.A. de Azcarraga, A. Macfarlane, J. Math. Phys. 37 (1996) 1115;
R.S. Dunne, A. Macfarlane, J.A. de Azcarraga, J.C. Prez Bueno, Int. J. Mod. Phys. Lett. A 12 (1997) 3275.
[6] A. Mostafazadeh, Int. J. Mod. Phys. Lett. A 12 (1997) 2725.

CUMULATIVE AUTHOR INDEX B621B624
Adams, D.H.
Aliev, T.M.
Alpha Collaboration
Altarelli, G.
Alvarez-Gaum, L.
Ambrosanio, S.
Asorey, M.
Astier, P.
Autiero, D.
B624 (2002) 469

B621 (2002) 479
B623 (2002) 271
B621 (2002) 359
B623 (2002) 165
B624 (2002) 3
B622 (2002) 593
B621 (2002) 3
B621 (2002) 3
Bajnok, Z.
Bajnok, Z.
Bak, D.
Baldisseri, A.
Baldo-Ceolin, M.
Ball, R.D.
Banner, M.
Barbieri, R.
Barbn, J.L.F.
Bassett, B.A.
Bassompierre, G.
Bazhanov, V.V.
Benslama, K.
Berends, F.A.
Berkovits, N.
Bertolini, M.
Besson, N.
Bird, I.
Blumenfeld, B.
Bobisut, F.
Bosch, S.W.
Bouchez, J.
Boyd, S.
Brandenberger, R.
Buchalla, G.
Buchbinder, I.L.
Bueno, A.
Bunyatov, S.
B622 (2002) 548

B622 (2002) 565
B622 (2002) 95
B621 (2002) 3
B621 (2002) 3
B621 (2002) 359
B621 (2002) 3
B624 (2002) 63
B623 (2002) 165
B622 (2002) 393
B621 (2002) 3
B622 (2002) 475
B621 (2002) 3
B623 (2002) 220
B624 (2002) 45
B621 (2002) 157
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 459
B621 (2002) 3
B621 (2002) 3
B623 (2002) 421
B621 (2002) 459
B621 (2002) 179
B621 (2002) 3
B621 (2002) 3
Caldas, H.C.G.
Camilleri, L.
B623 (2002) 503

B621 (2002) 3
0550-3213/2002 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 0 4 1 - X
Cardini, A.
Carlini Vallilo, B.
Caselle, M.
Cattaneo, P.W.
Cavasinni, V.
Cervera-Villanueva, A.
Challis, R.
Chamoun, N.
Chapovsky, A.P.
Chen, G.-H.
Chen, Y.
Cho, Y.M.
Chu, C.-S.
Chukanov, A.
Cieza Montalvo, J.E.
Collazuol, G.
Conforto, G.
Conta, C.
Contalbrigo, M.
Contino, R.
Corcella, G.
Costa-Santos, R.
Cousins, R.
B621 (2002) 3
B624 (2002) 45
B623 (2002) 474
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B624 (2002) 81
B621 (2002) 257
B622 (2002) 189
B624 (2002) 360
B621 (2002) 388
B621 (2002) 101
B621 (2002) 3
B623 (2002) 325
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B622 (2002) 227
B623 (2002) 247
B623 (2002) 439
B621 (2002) 3
Daniels, D.
Da Rold, L.
De Campos, F.
Dedes, A.
Degaudenzi, H.
De Gouva, A.
Del Prete, T.
Dent, T.
De Santo, A.
Daz, M.A.
Dighe, A.S.
Dignan, T.
Di Lella, L.
Di Vecchia, P.
B621 (2002) 3
B624 (2002) 485
B623 (2002) 47
B624 (2002) 3
B621 (2002) 3
B623 (2002) 395
B621 (2002) 3
B623 (2002) 73
B621 (2002) 3
B623 (2002) 47
B624 (2002) 377
B621 (2002) 3
B621 (2002) 3
B621 (2002) 157
510
Do Couto e Silva, E.
Donini, A.
Dorey, N.
Dudas, E.
Dumarchez, J.
Durhuus, B.
B621 (2002) 3
B624 (2002) 405
B624 (2002) 95
B622 (2002) 46
B621 (2002) 3
B623 (2002) 201
Easson, D.A.
boli, O.J.P.
Ellis, J.
Ellis, M.
B623 (2002) 421

B623 (2002) 47
B621 (2002) 208
B621 (2002) 3
Falceto, F.
Faraggi, A.E.
Fazio, T.
Fehr, L.
Feldman, G.J.
Ferrari, R.
Ferrre, D.
Flaminio, V.
Forger, M.
Forte, S.
Fortunato, S.
Fosco, C.D.
Fraternali, M.
Frau, M.
Fuchs, J.
Fujikawa, K.
Fyodorov, Y.V.
B622 (2002) 593

B624 (2002) 163
B621 (2002) 3
B621 (2002) 622
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 523
B621 (2002) 359
B623 (2002) 493
B624 (2002) 485
B621 (2002) 3
B621 (2002) 157
B624 (2002) 452
B622 (2002) 115
B621 (2002) 643
Gaillard, J.-M.
Gangler, E.
Gardi, E.
Gattnar, J.
Geiser, A.
Geppert, D.
Gherghetta, T.
Ghilencea, D.M.
Gibbons, G.W.
Gibin, D.
Giudice, G.F.
Giveon, A.
Gninenko, S.
Gckeler, M.
Godley, A.
Gomez-Cadenas, J.-J.
Gomis, J.
Gosset, J.
Gling, C.
Gouanre, M.
Grant, A.
Graziani, G.
Guglielmi, A.
B621 (2002) 3
B621 (2002) 3
B622 (2002) 365
B621 (2002) 131
B621 (2002) 3
B621 (2002) 3
B623 (2002) 97
B622 (2002) 215
B623 (2002) 3
B621 (2002) 3
B623 (2002) 395
B621 (2002) 303
B621 (2002) 3
B623 (2002) 287
B621 (2002) 3
B621 (2002) 3
B624 (2002) 181
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
Hagner, C.
Hall, L.J.
Hasenbusch, M.
Hata, H.
Hayasaka, K.
Heinemeyer, S.
Hernando, J.
Hibberd, A.N.
Higashijima, K.
Hisano, J.
Hofmann, R.
Hollowood, T.J.
Horsley, R.
Huang, C.-S.
Hubbard, D.
Huiszoon, L.R.
Hurst, P.
Hurth, T.
Hyett, N.
B621 (2002) 3
B624 (2002) 63
B623 (2002) 474
B624 (2002) 283
B624 (2002) 307
B624 (2002) 3
B621 (2002) 3
B622 (2002) 475
B623 (2002) 133
B621 (2002) 208
B623 (2002) 301
B624 (2002) 95
B623 (2002) 287
B624 (2002) 81
B621 (2002) 3
B624 (2002) 219
B621 (2002) 3
B624 (2002) 377
B621 (2002) 3
Iacopini, E.
Intriligator, K.
Ishibashi, M.
B621 (2002) 3
B624 (2002) 299
B622 (2002) 115
Jamin, M.
Joseph, C.
Juget, F.
B622 (2002) 279

B621 (2002) 3
B621 (2002) 3
Kazakov, V.
Kent, N.
Khoroshkin, S.M.
Khoze, V.A.
Khoze, V.V.
Kim, C.S.
Kim, S.-W.
Kimberly, D.
Kimura, T.
Kirsanov, M.
Klaus, B.
Klimov, O.
Kniehl, B.A.
Kofinas, G.
Kokkonen, J.
Kostov, I.K.
Kostov, I.K.
Kovzelev, A.
Krasnoperov, A.
Kumar, J.
Kumar, S.P.
Kurth, M.
Kustov, D.
Kutasov, D.
Kutasov, D.
Kuznetsov, V.
B622 (2002) 141

B621 (2002) 3
B622 (2002) 475
B621 (2002) 257
B621 (2002) 101
B624 (2002) 377
B622 (2002) 95
B623 (2002) 421
B623 (2002) 133
B621 (2002) 3
B623 (2002) 287
B621 (2002) 3
B621 (2002) 337
B622 (2002) 347
B621 (2002) 3
B622 (2002) 141
B624 (2002) 146
B621 (2002) 3
B621 (2002) 3
B624 (2002) 299
B624 (2002) 95
B623 (2002) 271
B621 (2002) 3
B621 (2002) 303
B622 (2002) 141
B621 (2002) 3
Lacaprara, S.
Lachaud, C.
Lakic, B.
Langfeld, K.
Lanza, A.
La Rotonda, L.
Lashkevich, M.
Laveder, M.
Lee, C.-W.H.
Lenaghan, J.T.
Lerche, W.
Lerda, A.
Letessier-Selvon, A.
Levy, J.-M.
Lindner, M.
Linssen, L.
Liu, C.
Liu, C.
Ljubicic, A.
Lola, S.
Long, J.
L, H.
Lunin, O.
Lupi, A.
Lusignoli, M.
Ltken, C.A.
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 131
B621 (2002) 3
B621 (2002) 3
B621 (2002) 587
B621 (2002) 3
B623 (2002) 201
B624 (2002) 253
B622 (2002) 269
B621 (2002) 157
B621 (2002) 3
B621 (2002) 3
B622 (2002) 429
B621 (2002) 3
B624 (2002) 81
B624 (2002) 360
B621 (2002) 3
B621 (2002) 208
B621 (2002) 3
B623 (2002) 3
B623 (2002) 342
B621 (2002) 3
B624 (2002) 405
B622 (2002) 269
Ma, E.
Ma, J.P.
Ma, J.P.
Macfarlane, A.J.
Magro, M.B.
Malbouisson, A.P.C.
Maes, J.L.
Marchionni, A.
Marotta, R.
Martelli, F.
Martn, C.P.
Martin, I.
Mathur, S.D.
McCoy, B.M.
Mchain, X.
Meloni, D.
Mendiburu, J.-P.
Mercadante, P.G.
Merlatti, P.
Meyer, J.-P.
Mezzetto, M.
Mielke, E.W.
Mishra, S.R.
Mitov, A.D.
Moch, S.
Moon, S.-H.
Moorhead, G.F.
B623 (2002) 126

B622 (2002) 416
B624 (2002) 360
B621 (2002) 712
B623 (2002) 47
B624 (2002) 485
B621 (2002) 37
B621 (2002) 3
B621 (2002) 157
B621 (2002) 3
B623 (2002) 150
B622 (2002) 240
B623 (2002) 342
B623 (2002) 439
B621 (2002) 3
B624 (2002) 405
B621 (2002) 3
B623 (2002) 47
B624 (2002) 200
B621 (2002) 3
B621 (2002) 3
B622 (2002) 457
B621 (2002) 3
B623 (2002) 247
B621 (2002) 413
B624 (2002) 327
B621 (2002) 3
511
Mosconi, P.
Mostafazadeh, A.
Mourad, J.
Mussardo, G.
B621 (2002) 571

B624 (2002) 500
B622 (2002) 46
B621 (2002) 571
Nakayama, R.
Naumov, D.
Necco, S.
Ndlec, P.
Nefedov, Yu.
Nepomechie, R.I.
Neupane, I.P.
Nguyen-Mau, C.
Nitta, M.
NOMAD Collaboration
Nomura, Y.
B624 (2002) 307

B621 (2002) 3
B622 (2002) 328
B621 (2002) 3
B621 (2002) 3
B622 (2002) 615
B621 (2002) 388
B621 (2002) 3
B623 (2002) 133
B621 (2002) 3
B624 (2002) 63
Ohlsson, T.
Oller, J.A.
Orestano, D.
zpineci, A.
B622 (2002) 429

B622 (2002) 279
B621 (2002) 3
B621 (2002) 479
Palla, L.
Palla, L.
Pando Zayas, L.A.
Park, J.-S.
Pastore, F.
Peak, L.S.
Peloso, M.
Pennacchio, E.
Pessard, H.
Petrov, A.Yu.
Petti, R.
Pich, A.
Pilo, L.
Placci, A.
Pleiter, D.
Polesello, G.
Pollmann, D.
Polyarush, A.
Ponsot, B.
Pope, C.N.
Popov, B.
Poulsen, C.
Provero, P.
Pusztai, B.G.
B622 (2002) 548

B622 (2002) 565
B622 (2002) 257
B621 (2002) 689
B621 (2002) 3
B621 (2002) 3
B622 (2002) 393
B621 (2002) 3
B621 (2002) 3
B621 (2002) 179
B621 (2002) 3
B622 (2002) 279
B622 (2002) 227
B621 (2002) 3
B623 (2002) 287
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B622 (2002) 309
B623 (2002) 3
B621 (2002) 3
B621 (2002) 3
B623 (2002) 474
B621 (2002) 622
Raidal, M.
Rakow, P.E.L.
Rattazzi, R.
Reinhardt, H.
Restuccia, A.
Rico, J.
Riemann, P.
B621 (2002) 208

B623 (2002) 287
B622 (2002) 227
B621 (2002) 131
B622 (2002) 240
B621 (2002) 3
B621 (2002) 3
512
Riotto, A.
Riva, V.
Roda, C.
Romanino, A.
Rubbia, A.
Runkel, I.
B623 (2002) 97
B621 (2002) 571
B621 (2002) 3
B622 (2002) 73
B621 (2002) 3
B624 (2002) 452
Sahakian, V.
Salvatore, F.
Sarkar, S.
Satz, H.
Savc, M.
Schaefer, S.
Schfer, A.
Schahmaneche, K.
Schalm, K.
Schellekens, A.N.
Schierholz, G.
Schmidt, B.
Schmidt, T.
Schweigert, C.
Schweigert, C.
Sevior, M.
Shin, H.J.
Sierra, G.
Signer, A.
Sillou, D.
Soler, F.J.P.
Sommer, R.
Sommer, R.
Sorbo, L.
Sozzi, G.
Steele, D.
Stelle, K.S.
Stiegler, U.
Stipcevic, M.
Stirling, W.J.
Stolarczyk, Th.
Strumia, A.
Strumia, A.
Su, S.
Sugiyama, K.
Suyama, T.
B621 (2002) 62
B621 (2002) 3
B621 (2002) 495
B623 (2002) 493
B621 (2002) 479
B623 (2002) 287
B623 (2002) 287
B621 (2002) 3
B624 (2002) 219
B624 (2002) 219
B623 (2002) 287
B621 (2002) 3
B621 (2002) 3
B622 (2002) 269
B624 (2002) 452
B621 (2002) 3
B624 (2002) 431
B622 (2002) 593
B621 (2002) 257
B621 (2002) 3
B621 (2002) 3
B622 (2002) 328
B623 (2002) 271
B622 (2002) 393
B621 (2002) 3
B621 (2002) 3
B623 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 257
B621 (2002) 3
B622 (2002) 73
B623 (2002) 395
B624 (2002) 3
B622 (2002) 3
B621 (2002) 235
Takcs, G.
Takcs, G.
Tanii, Y.
Tareb-Reyes, M.
Taylor, G.N.
Tereshchenko, V.
Teschner, J.
Thormeier, M.
Tobe, K.
Toldr, R.
B622 (2002) 548

B622 (2002) 565
B624 (2002) 283
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B622 (2002) 309
B624 (2002) 163
B623 (2002) 395
B621 (2002) 495
Tonasse, M.D.
Tonasse, M.D.
Toropin, A.
Tth, G.Zs.
Touchard, A.-M.
Tovey, S.N.
Tran, M.-T.
Travaglini, G.
Trincherini, E.
Tsesmelis, E.
Tseytlin, A.A.
Tsujikawa, S.
B623 (2002) 316

B623 (2002) 325
B621 (2002) 3
B622 (2002) 548
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 101
B622 (2002) 227
B621 (2002) 3
B621 (2002) 179
B622 (2002) 393
Ulrichs, J.
B621 (2002)
Vacavant, L.
Valdata-Nappi, M.
Valuev, V.
Van Gulik, R.
Vannucci, F.
Varvell, K.E.
Veltri, M.
Vercesi, V.
Vermaseren, J.A.M.
Vidal-Sitjes, G.
Vieira, J.-M.
Vinogradova, T.
Vogt, A.
Voituriez, R.
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B623 (2002) 220
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 413
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 413
B621 (2002) 675
Wang, Y.-S.
Weber, F.V.
Weiglein, G.
Weisse, T.
Wesson, P.S.
Wilke, T.
Wilson, F.F.
Winter, W.
Winterhalder, A.
Winton, L.J.
Wu, X.-H.
Wu, Y.-S.
B622 (2002) 633

B621 (2002) 3
B624 (2002) 3
B621 (2002) 3
B621 (2002) 388
B624 (2002) 253
B621 (2002) 3
B622 (2002) 429
B621 (2002) 523
B621 (2002) 3
B624 (2002) 81
B622 (2002) 189
Yabsley, B.D.
Yamaguchi, S.
Yoshikawa, T.
B621 (2002) 3
B622 (2002) 3
B624 (2002) 377
Zaccone, H.
Zarembo, K.
Zhang, J.
Zuber, K.
Zuccon, P.
Zwirner, L.
B621 (2002) 3
B623 (2002) 474
B624 (2002) 360
B621 (2002) 3
B621 (2002) 3
B621 (2002) 337

Nucl - Phys.B v.624

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.624

Загружено:

Авторское право:

Доступные форматы

Nuclear Physics B 624 (2002) 344

Implications of the Higgs boson searches on

Received 28 June 2001; accepted 19 December 2001

E-mail addresses: sandro.ambrosanio@bancaroma.it (S. Ambrosanio), dedes@th.physik.uni-bonn.de

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

PACS: 12.60.Jr; 12.60.Fr; 14.80.Ly

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

2. The Higgs sector in soft SUSY-breaking scenarios

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

In our numerical procedure we employ a two-loop renormalization group analysis for

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

the observables depend on just 4 parameters and a sign,

is always the LSP in

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

2.5. Evaluation of predictions in the Higgs boson sector of the MSSM

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

mNLSP > 100 GeV (GMSB).

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

where Yt = h2t /(4) and

a 2 b2 ln(a 2 /b2 ) + b2 c2 ln(b 2 /c2 ) + c2 a 2 ln(c2 /a 2 )

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

boson searches in the mmax

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

1.5 tan 55,

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

As above, the two bounds quoted here refer to mt = 175 GeV.

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

Tevatron. At an LC with a CMS energy of s 1 TeV both the associated production of

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

[3] G. Altarelli, G. Isidori, Phys. Lett. B 337 (1994) 141;

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

[33] A. Dabelstein, Nucl. Phys. B 456 (1995) 25, hep-ph/9503443;

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

S. Ambrosanio et al. / Nuclear Physics B 624 (2002) 344

[91] J. Ellis, G. Ganis, D.V. Nanopoulos, K.A. Olive, hep-ph/0009355;

Nuclear Physics B 624 (2002) 4562

One-loop N-point superstring amplitudes with

brings down the term ( 9 w )2 , the first term in M does not

F (z, ) = 1 F (z , ) 2iz ,

9r 9s wr w s F (zr zs , )

Fig. 1. A diagrammatic representation of Z(0) and Z( R) as reflections about y = 0 and y = 0, with