Nucl - Phys.B v.604

Nuclear Physics B 604 (2001) 331
www.elsevier.nl/locate/npe
Resonant sneutrino production in supersymmetry

with R-parity violation at the LHC
G. Moreau a , E. Perez b , G. Polesello c
a Service de Physique Thorique, CEA-Saclay, F91191, Gif-sur-Yvette Cedex, France
b Service de Physique des Particules, DAPNIA, CEA-Saclay, F91191, Gif-sur-Yvette Cedex, France
c INFN, Sezione di Pavia, Via Bassi 6, Pavia, Italy
Received 7 March 2001; accepted 12 April 2001
Abstract
The resonant production of sneutrinos at the LHC via the R-parity violating couplings ij k
Li Qj Dkc is studied through its three-leptons signature. A detailed particle level study of signal
and background is performed using a fast simulation of the ATLAS detector. Through the full
reconstruction of the cascade decay, a model-independent and precise measurement of the masses
of the involved sparticles can be performed. Besides, this signature can be detected for a broad class
of supersymmetric models, and for a wide range of values of several ij k coupling constants. Within
the MSSM, the production of a 850 GeV sneutrino for 211 > 0.05, and of a 300 GeV sneutrino
for 211 > 0.01 can be observed within the first three years of LHC running. 2001 Published by
Elsevier Science B.V.
PACS: 12.60.Jv; 13.85.Hd; 13.85.Lg; 14.80.Ly
1. Introduction
The most general superpotential respecting the gauge symmetries of the Standard Model
(SM) contains bilinear and trilinear terms which are not taken into account in the Minimal
Supersymmetric Standard Model (MSSM). Restricting to the trilinear part, these additional
terms read as:

1
1 c c c
c

c
ij k Li Lj Ek + ij k Li Qj Dk + ij k Ui Dj Dk ,
W
(1)
2
2
i,j,k
where i, j, k are generation indices, L (Q) denote the left-handed leptons (quarks)
superfields, and E c , D c and U c are right-handed superfields for charged leptons, down
and up-type quarks, respectively.
E-mail address: gmoreau@spht.saclay.cea.fr (G. Moreau).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 8 6 - 9
G. Moreau et al. / Nuclear Physics B 604 (2001) 331
The first two terms in Eq. (1) lead to violation of the lepton number (L
/ ), while the last
one implies violation of the baryon number (B
/ ). Since the simultaneous presence of L
/ and
B
/ couplings could lead to a too fast proton decay, a discrete multiplicative symmetry which
forbids the above terms in the superpotential has been imposed by hand in the MSSM.
This symmetry, called R-parity (Rp ), is defined as Rp = (1)3B+L+2S , where B, L and
S, respectively, denote the baryon number, the fermion number and the spin, such that
Rp = 1 (Rp = 1) for all supersymmetric (SM) particles. However other solutions can
ensure the proton stability, e.g., if L only is violated, or if only Uic Djc Dkc interactions
are allowed and the proton is lighter than the Lightest Supersymmetric Particle (LSP).
Moreover, on the theoretical point of view, there is no clear preference, e.g., in models
/ p and Rp conservation [1]. It is thus
inspired by grand unified or string theories, between R
mandatory to search for SUSY in both scenarios.
/ p lies in the possibility for
On the experimental side, the main consequence of R
the LSP to decay into ordinary matter. This is in contrast to scenarios where Rp is
conserved, in which the LSP is stable and escapes detection, leading to the characteristic
search for missing energy signals in direct collider searches. Moreover, while in Rp
conserved models, the supersymmetric (SUSY) particles must be produced in pairs, R
/p
allows the single production of superpartners, thus enlarging the mass domain where
/ p couplings offer the opportunity to resonantly
SUSY could be discovered. In particular, R
/ p coupling constants are severely
produce supersymmetric particles [2,3]. Although the R
constrained by the low-energy experimental bounds [1,48], the superpartner resonant
production can have significant cross-sections both at leptonic [4] and hadronic [9]
colliders. This is this possibility which is exploited throughout this paper.
The resonant production of supersymmetric particles is attractive for another reason:
/ p coupling, this reaction would
since its rate is proportional to a power 2 of the relevant R
/ p couplings than the pair production. In fact in
allow an easier determination of the R
the latter case, the sensitivity on the R
/ p coupling is mainly provided by the displaced
vertex analysis for the LSP decay, which is difficult experimentally especially at hadronic
colliders.
In this paper, we focus on the resonant SUSY particle production at the Large Hadron
Collider (LHC) operating at a center of mass energy of 14 TeV with special reference
to the ATLAS detector. At the LHC due to the continuous distribution of the centre of
mass energy of the colliding partons, a partonparton resonance can be probed over a wide
mass domain. This is a distinct advantage over the situation at lepton colliders, where the
search for narrow resonances requires lengthy scans over the centre of mass energy of the
machine.
At hadronic colliders, either a slepton or a squark can be produced at the resonance
through a or a coupling constant, respectively. In the hypothesis of a single dominant
R
/ p coupling constant, the resonant SUSY particle could decay through the same R
/p
coupling as in the production, leading then to a two quarks final state for the hard
process [1014]. In the case where both and couplings are non-vanishing, the slepton
produced via can decay through giving rise to the same final state as in Drell
/p
Yan process, namely two leptons [13,1517]. However, for most of the values of the R
Fig. 1. Feynman diagrams for the single chargino production at hadronic colliders via the ij k
coupling (symbolised by a circle in the figure). The arrows denote the flow of the particle momentum.
coupling constants allowed by present indirect searches, the decays of the resonant SUSY
particle via gauge interactions are dominant if kinematically accessible [4]. In this favoured
situation, typically, the produced superpartner initiates a cascade decay ended by the R
/p
decay of the LSP. In case of a dominant coupling constant, due to the R
/ p decay of the
LSP into quarks, this cascade decay leads to multijet final states which have a large QCD
background [9,10]. Only if leptonic decays such as, for instance, 1+ li i 10 enter the
cascade clearer signatures can be investigated [18]. The situation is more favourable in the
hypothesis of a single coupling constant, where the LSP can decay into a charged lepton,
allowing then multileptonic final states to be easily obtained.
We will thus assume a dominant ij k coupling constant. At hadronic colliders, either
a i sneutrino or a li charged slepton can be produced at the resonance via ij k and the
initial states are dj dk and uj dk , respectively. The slepton produced at the resonance has
two possible gauge decays, either into a chargino or a neutralino. In both cases particularly
clean signatures can be observed. For example, the production of a neutralino together
with a charged lepton resulting from the resonant charged slepton production can lead to
the interesting like-sign dilepton topology [1921] since, due to its Majorana nature, the
neutralino decays via ij k into a lepton as 0 li uj dk and into an anti-lepton as 0
li u j dk with the same probability.
In this article, we consider the single lightest chargino production at LHC as induced
by the resonant sneutrino production pp i 1+ li . The single 1 production also
receives contributions from the t and u channel squark exchange diagrams shown in
Fig. 1. In many models, the 10 neutralino is the LSP for most of the SUSY parameter
space. In the hypothesis of a 10 LSP, the produced 1 chargino mainly decays into the
neutralino as 1 10 qp qp or as 1 10 lp p . The neutralino then decays via ij k as
10 li uj dk , li u j dk or as 10 i dj dk , i dj dk . We concentrate on the decays of both
the chargino and the neutralino into charged leptons, which lead to a three leptons final
state. This signature has a low Standard Model background, and allows the reconstruction
of the whole decay chain, thus providing a measurement of some parameters of the SUSY
model.
2. The signal
2.1. Theoretical framework
Our theoretical framework in Sections 2 and 3 will be the R
/ p extension of the Minimal
Supersymmetric Standard Model. In Section 4 we will also give results in the Minimal
Supergravity (mSUGRA) model. The MSSM parameters are the following. M1 , M2
and M3 are the soft-SUSY breaking mass terms for the bino, the wino and the gluino,
respectively. is the Higgs mass parameter. tan = Hu /Hd is the ratio of the vacuum
expectation values (vev) for the two-Higgs doublet fields. At , Ab and A are the third
generation soft-SUSY breaking trilinear couplings. In fact, since these trilinear couplings
are proportional to the fermion masses one can neglect the first two generations couplings
without any phenomenological consequence in this context. Finally, mq , ml and m are the
squark, slepton and sneutrino mass, respectively. The value of the squark mass enters our
study mainly in the determination of the relative branching ratios of the 0 into lepton or
neutrino and of the into 0 + quarks or 0 + leptons. The remaining three parameters
m2Hu , m2Hd and the soft-SUSY breaking bilinear coupling B are determined through
the electroweak symmetry breaking conditions which are two necessary minimisation
conditions of the Higgs potential.
We choose to study the case of a single dominant 2j k allowing the reactions pp
. In Section 3 the analysis will be performed explicitly for the 211 coupling, since
it corresponds to the hard subprocess d d 1 which offers the highest partonic
luminosity. We will take 211 = 0.09, the upper value allowed by indirect bound: 211 <
0.09(mdR /100 GeV) [1] for a squark mass of 100 GeV. A quantitative discussion will be
given below for the general case of a single dominant 2j k coupling constant. We will not
treat explicitly the 1j k couplings which are associated to the e production, since the
low-energy bounds on these couplings are rather more stringent than the constraints on 2j k
and 3j k [1]. However, the three-leptons analysis from sneutrino production should give
similar sensitivities on the 1j k and 2j k couplings since isolation cuts will be included in
the selection criteria for the leptons. We will not perform the analysis of the 3j k couplings
which correspond to the production. A technique for mass reconstruction in the
ATLAS detector using the hadronic decays of the has been demonstrated in [22]. The
detailed experimental analysis needed to extract a signal is beyond the scope of this work.
Besides, in this case the sneutrino and chargino mass reconstruction studied in Section 3.1
is spoiled by the neutrinos produced in the decay.
2.2. Single chargino production cross-section
In order to establish the set of models in which the analysis presented below can be
performed, we need to study the variations of the single chargino production rate (pp
) with the MSSM parameters.
In Fig. 2, we present the cross-sections for the 1 production through several

2j k couplings as a function of the parameter for the fixed values: tan = 10, M2 =
Fig. 2. Cross-sections for the

2j k couplings, for tan =
1 production as a function of the parameter through various

10, M2 = 200 GeV and m = 400 GeV. In the case of 211 the
cross-section for 2 is also shown as the dashed line. The values of the R
/ p couplings have been
chosen equal to: 211 = 0.09, 212 = 0.09, 213 = 0.09, 221 = 0.18, 222 = 0.18, 223 = 0.18,
231 = 0.22, 232 = 0.39 and 233 = 0.39, which correspond to the low-energy limits for a sfermion
mass of 100 GeV [1].
200 GeV, and m = 400 GeV. For this choice of parameters and independently of , the
chargino 1 is lighter than the .
In this case the contributions of squark exchange in
the t and u channels are negligible compared to the resonant process so that the 1
production cross-section does not depend on the squark mass. The values for the considered
R
/ p coupling constants have been conservatively taken equal to the low-energy limits for
a sfermion mass of 100 GeV [1]. The cross-sections scale as 2
2j k .
We see on this figure that the dependence of the rates on is smooth for || > M2 . This
is due to the weak dependence of the 1 mass on in this domain. In contrast, we observe
a strong decrease of the rate in the region || < M2 where the 1 chargino is mainly
composed by the higgsino. Most of the small || domain (|| smaller than 100 GeV for
tan = 1.41 and m0 = 500 GeV) is however excluded by the present LEP limits [32].
We also show as a dashed line on the plot the rate for the 2 production through
the 211 coupling. The decrease of the 2 production rate with increasing || is due to an
Fig. 3. Cross-section for 1 production as a function of m and m in the MSSM for the
1
choice of values = 500 GeV, tan = 10 and 211 = 0.09. The hatched region at the upper left
corresponds to m < m . The cross-hatched region at low m is excluded by the preliminary
1
1
LEP results at s = 196 GeV [32].
increase of the 2 mass. We will not consider the contribution to the three-leptons final
state from the 2 production since the rate becomes important only for a very limited
range of small || values not yet excluded by LEP data.
Fig. 2 also allows to compare the sensitivities that can be reached on various 2j k
couplings using the single chargino production. If we compare, for instance, the crosssections of the 1 production via 211 and 221 at = 500 GeV, we can see that for
equal values of the R
/ p couplings the ratios between the cross-sections associated to211

and 221 is 2.15. Therefore, the sensitivity that can be obtained on 221 is only 2.15
times weaker than the sensitivity on 211 , for a 400 GeV sneutrino. Note that the crosssection ratio, and hence the scaling to be applied, in order to infer from the reach on 211
the sensitivity on another coupling 2j k , depends on the sneutrino mass. The reason is that
the evolution of the parton densities with the x-Bjorken variable is different for sea quark
and valence quark and for different quark flavours.
In order to study the dependence of the cross-section on the masses of the involved
sparticles, the parameters m and M2 were varied, and the other model parameters affecting
the cross-section were fixed at the values: 211 = 0.09, = 500 GeV and tan = 10.
The cross-section for 1 production as a function of m and m is shown in Fig. 3.
1
Since the 1 mass is approximately equal to M2 as long as M2 < ||, and becomes
equal to || for M2 > ||, we studied masses between 100 and 950 GeV, and values
of M2 between 100 and 500 GeV. For increasing m the cross-section decreases due to
a reduction of the partonic luminosity. A decrease of the cross-section is also observed for
m approaching m , since the phase space factor of the decay 1 following the
1
resonant sneutrino production is then suppressed. In the region m > m , the chargino
1
production still receives contributions from the s channel exchange of a virtual sneutrino,
as well as from the t and u channels squark exchange which in that case also contribute
significantly. However, in this phase space domain where the resonant sneutrino production
is not accessible, the cross-section is considerably reduced.
Finally, the single chargino production rate depends weakly on the A trilinear couplings.
Indeed, only the t and u channels squark exchange, varying with the squark mass which
can be influenced by A, depends on these couplings. The dependence of the rate on the
tan parameter is also weak.
2.3. Three leptons branching ratio
We calculate the total three leptons rate by multiplying the single chargino cross-section
by the chargino branching ratio, since we neglect the width of the chargino. The threeleptons final state is generated by the cascade decay 1 10 lp p , 10 ud.
2.3.1. 1 branching ratio
If the LSP is the 10 and mW + m 0 , mH + m 0 , m 0 , ml, m , mq > m , the
1
main kinematically allowed decays of the lightest chargino are 1 10 lp p and 1

10 dp up , p being the generation index. These three-body decays occur either through
a virtual W -boson or virtual sfermions. The associated branching ratio values are typically
of order (for m < m 0 + mb + mt ) B(1 10 lp p ) 11% for each of the three lepton
1
generations (p = 1, 2, 3) and B(1 10 dp up ) 33% for the first two quark generations
(p = 1, 2) due to the colour factor 3. In the case where mH + m 0 , m 0 , ml, m , mq >
1
m > mW + m 0 , the two dominant chargino decays are once more 1 10 lp p and
1
1 10 dp up (with branching fractions of 3 11% and 2 33%, respectively)

since those take place through the two-body decay 1 W 10 . However, in the case
where mW + m 0 , m 0 , ml, m , mq > m > mH + m 0 , the dominant chargino decay
1
is the two-body decay 1 H 10 [23]. The branching ratios are also modified if one of
the sfermions becomes lighter than the chargino. For instance, in a scenario where m 0 ,
2
mW + m 0 , mH + m 0 , ml, m > m > mqL,R , the dominant chargino decay is 1

1
10 dp up since it takes place via the two-body decay 1 qL,R qp .

/ p modes
Due to the strong present experimental constraints on the couplings [1], the R
can compete with the gauge couplings, affecting then the 1 branching fractions, only
in particular kinematic configurations in which the 1 and 10 masses are close enough.
However, this does not happen as long as the chargino is sufficiently heavier than the
neutralino, as is the case for example in supergravity inspired models.
Therefore, within the framework of the mSUGRA model (in which the squarks are
typically heavier than the lightest chargino) and for m > m (which is the relevant
1
scenario in the present study of a resonant sneutrino decaying into the lightest chargino),
p
10
the branching fraction of the chargino decay into an electron or a muon is of order 22% in
most of the considered parameter space.
2.3.2. 10 branching ratio
When 10 is the LSP (as it occurs in most regions of the mSUGRA parameter space), it
/ p interactions. For instance, in the case of a single dominant R
/p
can only decay through R
coupling constant of type 211 , the 10 decays involving this R
/ p interaction are 10 ud
3 ), the 0 ud decay receives contributions
and 10 dd. If the 10 is a pure wino (W
1
from the exchanges of a left-handed smuon ( L ) and a left-handed up-squark (u L ), while
the 10 dd decay receives contributions from the exchanges of a left-handed muonsneutrino ( L ) and a left-handed down-squark (dL ). In such a scenario, the 10 ud and
3 L (W
3 dL d)
10 dd decays have nearly identical branching ratios since the W
coupling value is exactly the opposite of the W3 L (W3 u L u) coupling value

(opposite SU(2)L gauge group charges) [33]. The difference between these two branching
fractions is due to the mass differences m L m L , mdL mu L , m m and md mu . If
the 0 ud decay receives contributions from the exchanges
the 10 is a pure bino (B),
1
of a L , a dR and a u L , while the 10 dd decay receives contributions from the
exchanges of a L , a dR and a dL . In such a scenario, the 10 ud and 10 dd
dL d) coupling
L (B
decays have approximately equal branching ratios since the B

value is exactly the same as the B L (Bu L u) coupling value (same U (1)Y gauge
group hypercharges) [33]. The difference between these two branching fractions comes
once more from the mass differences m L m L , mdL mu L , m m and md mu .
Finally, in the higgsino domain, namely for || M2 , the 10 dd decay becomes
dominant, spoiling the three-leptons signature.
3. Experimental analysis
3.1. Mass reconstruction
The analysis strategy is based on the exploitation of the decay chain:
1+
|
10 W + e+ (+ )
|
q q
(2)
which presents a sequence of three decays which can be fully reconstructed. The strong
kinematic constraint provided by the masses of the three sparticles in the cascade is
sufficient to reduce the contribution of the different background sources well below the
signal rate.
The signal events were generated with a version of the SUSYGEN Monte Carlo [24]
modified to allow the generation of pp processes. The hard-subprocess q q is
first generated according to the full lowest order matrix elements corresponding to the
diagrams depicted in Fig. 1. Cascade decays of the s are performed according to the
11
relevant matrix elements. The parton showers approach [25] relying on the DGLAP [26]
evolution equations is used to simulate QCD radiations in the initial and final states, and
the non-perturbative part of the hadronisation is modeled using string fragmentation [25].
The events were then processed through the program ATLFAST [27], a parameterised
simulation of the ATLAS detector response.
In this section, the analysis will be performed for the R
/ p coupling 211 = 0.09 and for
the following MSSM point: M1 = 90 GeV, M2 = 180 GeV, = 360 GeV, tan = 10,
At = Ab = A = 0, mf = 300 GeV.
For this set of MSSM parameters, the masses of the relevant gauginos are:
m 0 = 89.9 GeV,
1
m = 173.7 GeV,
1
and the
decay into an on shell W has a branching ratio of order 96%. The total crosssection for the resonant sneutrino production pp is 37 pb. If we include the branching
fractions into the three leptons, the cross-section is 2.2 pb, corresponding to 66000
events for the standard integrated luminosity of 30 fb1 for the first three years of LHC
data taking.
The signal is characterised by the presence of three isolated leptons and two jets. For the
initial sample selection we require that:
Exactly three isolated leptons are found in the event, with pT1 > 20 GeV, pT2,3 >
10 GeV, where pT is the momentum component in the plane perpendicular to the
beam direction, and pseudorapidity || < 2.5.
At least two of the three leptons must be muons.
At least two jets with pT > 15 GeV are found.
The invariant mass of any + pair is larger than 10 GeV and outside 6.5 GeV of
the Z mass.
The isolation prescription on the leptons is necessary to reduce the background from the
semileptonic decays of heavy quarks, and consists in requiring an energy deposition of less
than 10 GeV not associated with the lepton in a pseudorapidity-azimuth ( ) cone of
opening 0R = 0.2 around the lepton direction.
The efficiency for these cuts, after the branching fractions have been taken into account,
is 25%, where half of the loss comes from requiring three isolated leptons, and the other
half is the loss of jets from 10 decay either because they are not reconstructed, or because
the two jets from the decay are reconstructed as a single jet. The cut on the + pair
invariant mass gives a 10% loss in statistics. In order to avoid the combinatorial background
from additional QCD events we further require that no third jet with pT > 15 GeV is
reconstructed in the event. The efficiency after this cut is 15%.
The reconstruction of the sparticle masses could be performed either starting from the 10
reconstruction and going up the decay chain, or trying to exploit the three mass constraints
at the same time. We choose the first approach which is not optimal, but allows a clearer
insight into the kinematics of the events.
The first step in reconstruction of the 10 jet jet is the choice of the correct
muon to attempt the reconstruction. The three leptons come in the following flavour-sign
configurations (+ charge conjugates):
12
(1) e+ + ;
(2) e+ ;
(3) + + ;
(4) + ;
where the first lepton comes from the , the second one from the W , and the third one from
the 10 decay, corresponding to three final state signatures: (1) two opposite sign muons
and an electron, 1 (2) two same-sign muons and an electron, (3), (4) three muons. The
configuration with three same-sign muons does not correspond to the required signature
and is rejected in the analysis. For signature (1) the muon produced in the 10 decay is
defined as the one which has the same sign as the electron. For configuration (2) both
muons must be tested to reconstruct the 10 . For configuration (3), (4), the 10 muon must
be one of the two same-sign ones.
In order to minimise the combinatorial background we start the reconstruction from
signature (1) where each lepton is unambiguously attributed to a step in the decay. The
distribution of the -jet-jet invariant mass is shown in the left plot of Fig. 4. A clear peak
is visible corresponding to the 10 mass superimposed to a combinatorial background of
events where one of the two jets from the 10 was lost and a jet from initial state radiation
was picked up. The combinatorial background can be evaluated using three-jet events,
where at least one jet is guaranteed to come from initial state radiation. The shape of the
combinatorial background estimated with this method is shown as the shaded histogram
superimposed to the signal peak. After background subtraction, an approximately Gaussian
peak with a width of 6 GeV, and a statistics of about 1100 50 events is reconstructed,
shown in the right of Fig. 4. The error on the number of events is the estimated systematic
uncertainty on the background subtraction procedure. If we consider a window of 15 GeV
around the peak, corresponding to 2.5 of the Gaussian, 1400 events are observed in
Fig. 4. -jet-jet invariant mass for events in configuration (1). (See text.) Left: exclusive two jet
events with superimposed (hatched) the combinatorial background. Right: 10 peak after background
subtraction.
1 Here and in the following, electron stands for both e+ and e .
13
the sample, and the combinatorial contamination is approximately 25%. A tail towards
low mass values is observed, corresponding to events where a fraction of the parton energy
is lost in the jet reconstruction. From this distribution the 10 mass can be measured with
a statistical error of 100 MeV. The measurement error will in this case be dominated by
the systematic error on the jet energy scale which in ATLAS is estimated to be at the level
of 1% [34].
The 25% combinatorial background is due to the soft kinematics of the chosen
example point, with a 10 which is both light and produced with a small boost. In order
to show the effect of the mass hierarchy of the involved sparticles, the shape of the 10
mass peak is shown in Fig. 5 for a sneutrino mass of 500 GeV and different choices
for the 10 mass. In all cases the 1 mass is twice the 10 mass, corresponding to the
gauge unification condition and to || values of the same order as M2 . The combinatorial
background is in general smaller than for a 300 GeV sneutrino, due to the higher boost
imparted to the 10 , and it decreases with increasing 10 masses, due to the higher efficiency
for reconstructing both jets from the 10 decay. For this analysis no attempt has been done
Fig. 5. Lepton-jet-jet invariant mass for exclusive two-jet events where the 10 lepton is uniquely
defined, for four different values of the 10 mass: m 0 = 80, 150, 200 and 250 GeV. In all cases the
1
sneutrino mass is set at 500 GeV, = M2 = 2M1 , and tan = 10, yielding a 1 mass twice the
10 mass. All the sfermion masses are set to 500 GeV. The normalisation is arbitrary.
14
for the recalibration of the jet energy. This results in the skewing of the distributions
towards low masses, and in the peak value being slightly displaced with respect to the
nominal mass value.
Once the position of the 10 mass peak is known, the reconstructed 10 statistics can
be increased by also considering signatures (2) and (3), (4). For events coming from
signatures (2) to (4), the 10 candidate is defined as the muon-jet-jet combination which
gives a mass nearest to the mass peak determined from signature (1) events. In all cases
the reconstructed mass is required to be within 15 GeV of the peak position to define
a 10 candidate. In 80% of the events containing at least a combination satisfying this
requirement, only one 10 candidate is found, and this sample can be used to improve the
statistical precision on the 10 mass measurement.
Using the above definition of the 10 , we can go further in the mass reconstruction of
the involved sparticles. Only configurations (1) and (2) are used, i.e., the events containing
two muons and an electron in order to avoid ambiguities in the choice of the lepton from
the W decay. The preliminary step for the reconstruction of the 1 is the reconstruction of
the W boson from its leptonic decay. The longitudinal momentum of the neutrino from the
W decay is calculated from the missing transverse momentum of the event (considered as
pT ) and the requirement that the electron-neutrino invariant mass gives the W mass. The
resulting neutrino longitudinal momentum, has a twofold ambiguity. We therefore build
the invariant W 10 mass candidate using both solutions for the W boson momentum.
The resulting spectrum is shown in Fig. 6, as the full line histogram. A clear peak is
seen, superimposed on a combinatorial background. If only the solution yielding the 1
mass nearest to the measured mass peak is retained, the mass spectrum corresponding to
the shaded histogram is obtained. The peak in the unbiased histogram can be fitted with
a Gaussian shape, with a width of 6 GeV.
The combination with the mass nearest to the measured peak is taken as 1 candidate,
provided that the reconstructed mass is within 15 GeV of the peak. For 80% of the e
events where a 10 candidate is found, a W 10 combination satisfying this requirement is
reconstructed.
Finally the 1 candidates are combined with the leftover muon, yielding the mass
spectrum shown in Fig. 7. The mass peak at this point presents very limited tails, and
has a width of 10 GeV. We define fully reconstructed events as those for which this
mass lies within 25 GeV of the measured peak. From the estimate of the combinatorial
under the 10 peak, we expect approximately 2 1100 = 2200 events where all the jets and
leptons are correctly assigned over a total of 2350 events observed in the peak. The small
difference between the two numbers is given by events for which one of the two jets used
for the 10 reconstruction comes from initial state radiation. These jets are typically soft,
and therefore the reconstructed 10 candidate very often has a momentum which both in
magnitude and direction is close to the momentum of the original 10 . Therefore for such
events the reconstructed 10 behaves in the further steps in the reconstruction as the real
one, only inducing some widening in the 1 and peaks.
The statistics available at the different steps in the analysis for an integrated luminosity
of 30 fb1 is given in the first column of Table 1. For the assumed value of the coupling,
15
Fig. 6. Invariant mass of the 10 with the W candidate. The full line histogram includes both solutions
for the neutrino longitudinal momentum, the grey one only includes the solution which gives the mass
nearest to the measured peak.
Fig. 7. Invariant mass of the third lepton in the event with the 1 candidate.
211 = 0.09, the uncertainty on the measurement of all the three masses involved will be
dominated by the 1% uncertainty on the jet energy scale.
The efficiency for the reconstruction of the full decay chain with the analysis described
above is 2.5%. A more sophisticated analysis using also the three-muons events should
approximately double this efficiency.
16
Table 1
Cross-sections and expected numbers of events after cuts for the signal and the different Standard
Model background contributions considered in the analysis. The Loose cuts are described at the
beginning of Section 3.1, and the Jet veto consists in adding the requirement that no third jet with
pT > 15 GeV is reconstructed in the event. The line labelled 10 gives the number of events from
signatures (1) to (4) (e and ) for which a 10 candidate is found. The line labelled 1
shows the number of events from signatures (1) and (2) (e) where a 1 candidate is found in
addition, and the last line indicates the number of fully reconstructed events. In the case of the signal,
we give the cross-section for the resonant sneutrino production multiplied by the branching ratios
into three leptons
Process
(pb)
Nev (30 fb1 )
Loose cuts
Jet veto
10
Signal
tt
WZ
W bb
Wt
Zb
2.2
6.6 104
18400
11050
6250
590
1.7 107
2800
1400
205
26
8 105
90
65
10
300
9 106
2.4
60
1.8 106
3.5
1650
5 107
650
440
46
2530
2350
16
2
1.5
0.5
2.5
2.5
From the observed number of events and the mass a measurement of the quantity
BR, where BR is the product of the branching ratios of the decays shown in
Eq. (2), is possible. The measurement of additional SUSY processes will be needed to
disentangle the two terms of this product.
2
211
3.2. Standard Model background

The requirement of three isolated leptons in the events strongly reduces the possible
background sources. The following processes were considered as a background:
tt production;
W Z production;
W t production;
W bb production;
Zb production.
These backgrounds were generated with the PYTHIA Monte Carlo [28], except W t and
W bb for which the ONETOP parton level generator [29] was used, interfaced to PYTHIA
for hadronisation and fragmentation.
A special treatment was needed for both the W Z and Zb backgrounds, because in both
cases PYTHIA only generates the resonant W and Z bosons. As discussed in [36], the
contribution from and Z/ interference can be important (a stands for a virtual
particle), even if the events with the 3+ 3 invariant mass inside the Z peak are rejected
by the analysis cuts. In order to evaluate this contribution, we have generated at parton
level the processes pp W () Z () / and pp bZ ()/ with the subsequent leptonic
decays of the gauge bosons using the COMPHEP program [35]. The numbers of events
17
surviving the requirements on leptons were evaluated at parton level both for the case with
only the resonant W and Z bosons and for the full set of contributions: W , W , Z, Z , ,
Z/ interference and Z / interference. From the ratio of the two cases we calculated
a correction factor which was then applied to the W Z and Zb backgrounds generated with
PYTHIA in order to take into account the non-resonant processes. We need to apply this
procedure since the two hadronic jets required by the analysis cuts come from the initial
state radiation of the incoming quarks in the W () Z () / and Z () / b productions. Now
this radiation is only present in shower Monte Carlos like PYTHIA, and not on parton-level
calculations.
The cross-sections for the various processes, and the number of total expected events
for an integrated luminosity of 30 fb1 are given in Table 1, according to the cross-section
numbers used in the ATLAS physics performance TDR [34]. In particular, even when the
cross-section is known at NLO, as in the case of the top, the Born cross-section is taken for
internal consistency of the study.
For each of the background processes a sample of events between one seventh and a few
times the expected statistics was generated and passed through the simplified simulation of
the ATLAS detector.
After the loose selection cuts described in Section 3.1, the background is dominated
by top production, as can be seen from the numbers shown in Table 1. The distribution
of the -jet-jet invariant mass for background events, obtained as in Section 3.1 and
corresponding to the 10 candidates selection, is shown as the hatched histogram in Fig. 8.
In this figure we have superimposed the same distribution for the signal. Already at this
level, the signal stands out very clearly from the background, and in the following steps of
Fig. 8. Invariant mass of the 10 candidates entering the kinematic analysis superimposed to the
Standard Model background (hatched).
18
the reconstruction the background becomes almost negligible. The numbers of background
and signal events expected at the various steps of the reconstruction can be compared in
Table 1. The full analysis was performed only for the tt, W Z and Zb backgrounds because
for the other channels the background is essentially negligible compared to top production,
and in most cases the Monte Carlo statistics after the initial selection was too low to allow
a detailed study. For the SUSY model considered and the chosen value of the coupling
constant, even the loose selection applied allows to efficiently separate the signal from the
background.
3.3. Sensitivity on
From these results, it is possible to evaluate the minimum value of the 211 coupling
for which it will be possible to discover the signal. The starting point in the analysis is the
observation of a peak in the muon-jet-jet invariant mass over an essentially flat background.
All of the further analysis steps of the cascade reconstruction rely on the possibility of
selecting the events with a mass around the 10 peak.
For the observation of the peak, the best signal/background ratio is obtained using the
three-muons sample (configurations (3) and (4) above). In the Standard Model, which
incorporates lepton universality, about one eight of the three-leptons events present a threemuons configuration, whereas about half of the signal events come in this configuration,
thereby granting an improvement of a factor 4 in signal over background, with respect
to the full sample. The three muons come either in the + + or in the + sign
configuration, because the two muons from the decay chain (
) 1 10
0
must have opposite sign, whereas the 1 can decay to muons of either sign. Therefore
the muon for the 10 reconstruction must be chosen between the two same-sign ones.
The distribution for the -jet-jet invariant mass, for events containing two jets and three
muons is shown in Fig. 9, scaled down by a factor 20, corresponding to a value of 0.02,
superimposed to the expected top background. In the distribution each event enters twice,
for each of the two same-sign muons which can be used to reconstruct the 10 . We expect,
however, that the combination with the wrong muon gives in most cases a reconstructed
mass outside of the 10 peak.
A statistical prescription is needed to define the fact that a peak structure is seen in
the signal + background distribution. Given the exploratory
nature of the work, we adopt

the naive approach of calculating the value for which S/ B = 5, where S and B are,
respectively, the number of signal and background candidates counted in an interval of
15 GeV around the measured 10 peak. In this interval, for the chosen point, for an
integrated luminosity of 30 fb1 , S = 475000 ()2 and B = 72 events. In the hypothesis
that the tt background can be precisely measured from data, the lower limit on 211 is:
211 > 0.009.
The pair production of SUSY particles through standard Rp -conserving processes is
another possible source of background, due to the possibility to obtain final states with
high lepton multiplicity, and the high production cross-sections. This background can
19
Fig. 9. Invariant mass of the 10 candidates from three-muon events scaled down by a factor 20,
corresponding to a value of 0.02, superimposed to the Standard Model background (hatched).
only be evaluated inside models providing predictions for the whole SUSY spectrum.
As a preliminary study, a sample of events were generated with the HERWIG 6.1 Monte
Carlo [31] by setting the slepton masses at 300 GeV, the masses of squarks and gluinos
at 1000 GeV and the chargino-neutralino spectrum as for the example model. The total
Rp -conserving cross-section is in this case 6 pb. A total of 30 SUSY background
events which satisfy the requirements used above to define S and B are observed. All the
events surviving the cuts are from direct chargino and neutralino production, with a small
contribution from DrellYan slepton production. Since the contributions from squark and
gluino decays are strongly suppressed by the jet veto requirements, this result can be
considered as a correct order of magnitude estimate, independently from the assumed
values for the squark and gluino masses. Moreover, the reconstructions of the chargino
and sneutrino masses can also be used in order to reduce the SUSY background. A more
thorough discussion of the SUSY background will be given below in the framework of the
mSUGRA model.
4. Analysis reach in various models

For the example case studied in Section 3.1 it was shown that the sneutrino production
signal can be easily separated from the background, and allows to perform precision
measurements of the masses of the sparticles involved in the decay chain.
The analysis can be generalised to investigate the range of SUSY parameters in which
this kind of analysis is possible, and to define the minimum value of the constant which
20
gives a detectable signal in a given SUSY scenario. The different model parameters enter
the definition of the detectability at different levels:
The sneutrino production cross-section is a function only of the sneutrino mass and
of the square of the R-parity violating coupling constant.
The branching fraction of the sneutrino decay into three leptons is a function of all
the SUSY parameters.
The analysis efficiency is a function of the masses of the three supersymmetric
particles involved in the decay.
The dependences of the cross-section and branching ratios on the SUSY parameters were
discussed in Section 2 for the MSSM, and are summarised in Figs. 2 and 3. We only need at
this point to parameterise the analysis efficiency as a function of the sparticle masses. The
number of signal events for each considered model will then be obtained by multiplying
the expected number of three-lepton events by the parameterised efficiency.
4.1. Efficiency of the three-muon analysis
According to the discussion presented in Section 3.3, we need to calculate the efficiency
for the signal process to satisfy the following requirements:
to pass the initial selection cuts described in Section 3.1 (loose cuts), including the
veto on the third jet (jet veto);
to contain three reconstructed muons, with one of the -jet-jet invariant masses within
15 GeV of the 10 mass.
Three sneutrino masses, m = 300, 500 and 900 GeV were considered, and for each of
these the evolution of the efficiency with the 1 mass was studied. The mass of the 10
was assumed to be half of the mass of the 1 , relation which is in general valid in SUGRA
inspired models and correspond to a choice of values for || of the same order as M2 .
The analysis efficiency is shown in Fig. 10 as a function of the mass and of the
1 mass. The efficiency values are calculated with respect to the number of events which
at generation level did contain the three leptons, therefore they only depend on the event
kinematics and not on the branching ratios. The loss of efficiency at the lower end of the
1 mass spectrum is due to the inefficiency for detecting two jets from the 10 decay,
either because the two jets are reconstructed as a single jet, or because one of the two jets
is below the detection threshold of 15 GeV. A significant additional loss in efficiency for
low m and high m is caused by the fact that all the 1 decay products are collimated
1
by the 1 boost, and one of the leptons fails the isolation requirement. The efficiency then
becomes approximately independent of the masses of the sneutrino and of the 1 , up to
the point where the and 1 masses become close enough to affect the efficiency for the
detection of the muon from the 1 decay; for m m < 10 GeV the analysis
1
efficiency rapidly drops to zero. The moderate decrease in efficiency at high 1 masses for
m = 900 GeV can be ascribed to the fact that one of two energetic jets from the 10 decay
radiates a hard gluon, three jets are reconstructed, and the event is rejected by the jet veto.
At this point all the ingredients are available to study the reach in the parameter space
for the analysis presented in Section 3 within different SUSY models.
21
Fig. 10. Efficiency for reconstructing a -jet-jet invariant mass within 15 GeV of the 10 mass
in three-muons events, as a function of the 1 mass. The shown points were generated with the
following parameters: = M2 = 2M1 , tan = 10. All the sfermion masses are set equal to the
sneutrino mass. Points for m = 300, 500 and 900 GeV are shown.
4.2. Analysis reach in the MSSM

The region in the m m plane for which the signal significance is greater than 5 ,
1
as defined in Section 3.3, and at least 10 signal events are observed for an integrated
luminosity of 30 fb1 is shown in Fig. 11 for different choices of the 211 constant.
The behaviours of the sensitivity curves in the m m plane are well explained by the
1
variations of the single chargino production cross-section shown in Fig. 3 in the same
plane of parameters. The SUSY background is not considered in the plot, as it depends on
all the model parameters, in particular on the sfermion mass spectrum and couplings. For
a few example cases all the SUSY processes were generated for various assumptions on
the squark masses. We found that for our analysis cuts the SUSY background is dominated
by direct production of chargino/neutralino pairs (and more precisely by the pp 1 1
and pp 1 20 processes), with subsequent decays of chargino and neutralino involving
charged leptons. This background has at least four jets in the final state, coming from the
decays of the two 10 at the end of the chains, and is therefore rejected by the veto on the
third jet. However, for 10 masses lower than 100 GeV, the jets from 10 decay often
fall below the detection threshold, and a small fraction of the events survives the cuts.
The SUSY background becomes negligible in the limit of high 10 and masses both
because the drop in cross-section, and because of the harder jets from 10 decay. The main
effect of taking into account this background would therefore be to somewhat reduce the
significance of the signal for 1 masses lower than 200 GeV.
22
Fig. 11. 5 reach in the m m plane for three different choices of the 211 coupling for an
1
integrated luminosity of 30 fb1 at the LHC. The chosen model parameters were: = 500 GeV,
tan = 10, mq = ml = 300 GeV, At = Ab = A = 0, M2 = 2M1 . The significance is defined only
considering the Standard Model background, and a signal of at least ten events is required. The
hatched region at the upper left corresponds to m < m . The cross-hatched region at low m is
1
1
excluded by the preliminary LEP results at s = 196 GeV [32].
From the curves in Fig. 11 we can conclude that within the MSSM, the production of
a 850 GeV sneutrino for 211 > 0.05, and of a 300 GeV sneutrino for 211 > 0.01 can be
observed within the first three years of LHC running, provided that the sneutrino is heavier
than the lightest chargino.
The sensitivity on an R
/ p coupling of type 2j k can be derived from the sensitivity

obtained for 211 , as explained in Section 2.2. For example, we have seen that the
sensitivity on 221 was 1.46 times weaker than the sensitivity on 211 , for tan =
10, M2 = 200 GeV, = 500 GeV and m = 400 GeV. This set of parameters leads
to a sensitivity on 211 of about 0.016 as can be seen in Fig. 11, and hence to a sensitivity
on 221 of 0.023. In Table 2, we present the sensitivity on any 2j k coupling estimated
using the same method and for the same MSSM parameters. Those sensitivities represent
an important improvement with respect to the low-energy limits of [1].
In the case of a single dominant 2j 3 coupling the neutralino decays as 10 uj b and
the semileptonic decay of the b-quark could affect the analysis efficiency. Hence in this
case, the precise sensitivity cannot be simply calculated by scaling the value obtained for
211 . The order of magnitude of the sensitivity which can be inferred from our analysis
should however be correct.
23
Table 2
Sensitivities on the 2j k coupling constants deduced from the sensitivity on 211 for tan = 10,
M1 = 100 GeV, M2 = 200 GeV, = 500 GeV, mq = ml = 300 GeV and m = 400 GeV
211
212
213
221
222
223
231
232
233
0.01
0.02
0.02
0.02
0.03
0.05
0.03
0.06
0.10
4.3. Analysis reach in mSUGRA

Our framework throughout this section will be the so-called minimal supergravity
model. In this model the parameters obey a set of boundary conditions at the Grand Unification Theory (GUT) scale Mx . These conditions appear to be natural in supergravity scenario since the supersymmetry breaking occurs in an hidden sector which communicates
with the visible sector only through gravitational interactions. First, mSUGRA contains the
gauge coupling unification at Mx , such an unification being suggested by the experimental results obtained at LEP I. One can view the gauge coupling unification assumption as
a fixing of the GUT scale Mx . Second, the gaugino (bino, wino and gluino) masses at Mx
are given by the universal mass m1/2 . The parameters m1/2 and Mi (i = 1, 2, 3) are thus
related by the solutions of the renormalization group equations (RGE). Besides, since the
gaugino masses and the gauge couplings are governed by the same RGE, one has the wellknown relation: M1 = 53 tan2 W M2 . Similarly, at Mx , the universal scalars mass is m0 and
the trilinear couplings are all equal to A0 . Finally, in mSUGRA the absolute value of the
higgsino mixing parameter || as well as the bilinear coupling B are determined by the
radiative electroweak symmetry breaking conditions. Therefore, mSUGRA contains only
the five following parameters: sign(), tan , A0 , m0 and m1/2 .
Due to the small dependence of the single chargino production rate on the parameter
for M2 || (see Section 2), the study of the mSUGRA model in which || is fixed by
the electroweak symmetry breaking condition provides information on a broader class of
models. The single chargino production rate depends mainly on the values of m0 and m1/2
(see Section 2). We will set A0 = 0, and study the detectability of the signal in the m0 m1/2
plane. We show in Fig. 12 the curves of equal mass for and 1 for tan = 10 calculated
with the ISASUSY [30] package which uses one-loop RGE to get the SUSY spectrum
from the mSUGRA parameters.
The signal reach can be easily evaluated from the sparticle mass spectrum and branching
fractions by using the parameterisation of the analysis efficiency shown in Fig. 10.
4.3.1. Supersymmetric background
In the case of a well constrained model as mSUGRA, the SUSY background can in
principle be evaluated in each considered point in the parameter space. For this evaluation
the full SUSY sample must be generated for each point, requiring the generation of a large
number of events.
The sparticle masses for the model studied in detail in Section 3 uniquely define a model
in the mSUGRA framework. Therefore, as a first approach to the problem, a full analysis
24
Fig. 12. Curves of equal mass for and 1 in the m0 m1/2 plane for tan = 10. The grey region
at the upper left indicates the domain where the 10 is not the LSP. The cross-hatched region for low
m1/2 gives the kinematic limit for the discovery of 1 or l by LEP running at s = 200 GeV. The
dotted line shows the region below which the 1 decays to a virtual W .
25
Fig. 13. Number of SUSY background events for an integrated luminosity of 30 fb1 in the m0 m1/2
plane with tan = 10 for a few selected test models. The hatched region at the upper left corresponds
to m < m . The cross-hatched region for low m1/2 gives the kinematic limit for the discovery of
1
1 or l by LEP running at s = 200 GeV.
was performed for this model, corresponding to the parameter values:

m0 = 265 GeV,
m1/2 = 220 GeV,
tan = 10,
< 0,
A0 = 0.
For this mSUGRA point, the mass scale for squarks/gluinos is in the proximity of
500 GeV, and the total cross-section for all the SUSY particles pair productions is
approximately 67.5 pb, yielding a signal of 2 106 events for the first three years of
data-taking at the LHC. A total of 350k events were generated and analysed. The number
of surviving events after cuts in the three-muons sample was 24 12 for an integrated
luminosity of 30 fb1 . All the background events come from direct chargino and neutralino
production, as it was the case for the MSSM point studied in Section 3.3. As a cross-check,
we generated for the same mSUGRA point only the processes of the type pp + X,
where denotes either 0 or , and X any SUSY particle. The cross-section is in this
case 4.3 pb, and the number of background events is 19 3 events, in good agreement
with the number evaluated generating all the SUSY processes.
Based on this result, we have performed a scan in the m0 m1/2 plane the fixed values
tan = 10, < 0. On a grid of points we generated event samples for the pp + X
processes with the HERWIG 6.1 Monte Carlo [31]. The number of SUSY events with
a -jet-jet combination with an invariant mass within 15 GeV of the 10 mass is shown
in Fig. 13 in the m0 m1/2 plane for an integrated luminosity of 30 fb1 . The background
26
is significant for a 1 mass of 175 GeV (m1/2 = 200 GeV), and becomes essentially
negligible for 1 mass of 260 GeV (m1/2 = 300 GeV). This behaviour is due to the
combination of two effects: the production cross-section decreases with increasing
mass, and the probability of losing two of the four jets from the decay of the two 10 in
the event becomes very small for a 1 mass of 220 GeV. Indeed, the suppression of the
SUSY background is mainly due to the jet veto cut.
Given the high SUSY cross-section, and the high lepton multiplicity from 10 decays,
a prominent signal should manifest itself through R-conserving sparticle pair production in
this scenario. Single resonant sneutrino production will then be used as a way of extracting
information on the value of the Rp -violating coupling constant, and of precisely measuring
the masses of , 1 , 10 . Moreover, thanks to the very high number of produced 10
expected from q/
g pair production, the 10 mass will be directly reconstructed from q and
g decays, as shown in [34]. So, for the present analysis it can be assumed that the 10 mass is
approximately known, and an attempt to reconstruct the 1 peak can be performed even if
the 10 reconstruction does not yield a significant peak over the SUSY + SM background. In
order to perform the full reconstruction, one just needs to observe a statistically significant
excess of events over what is expected from the Standard Model background in the mass
region corresponding to the known 10 mass. The full kinematic reconstruction described in
Section 3.1 above will then easily allow to separate the process of interest from the SUSY
background.
4.3.2. Results
Based
on the discussion in the previous section we calculate the signal significance as
S/ B, where for the signal S we only consider the resonant sneutrino production, and
for the background B we only consider the SM background. We show in Fig. 14 for
tan = 10 and for the two signs of the regions in the m0 m1/2 plane for which the
signal significance exceeds 5 and the number of signal events is larger than 10, for an
integrated luminosity of 30 fb1 . The reach is shown for three different choices of the 211
parameter: 211 = 0.01, 0.025, 0.05. Even for the lowest considered coupling the signal
can be detected in a significant fraction of the parameter space. The dotted line shows the
region below which the 1 decays to the 10 and a virtual W , thus making the kinematic
reconstruction of the decay chain described in Section 3.1 impossible. The reconstruction
of the 10 is however still possible, but the reconstruction efficiency drops rapidly due to
the difficulty to separate the two soft jets from the 10 decay. A detailed study involving
careful consideration of jet identification algorithms is needed to assert the LHC reach in
that region.
As observed in [37], the efficiencies quoted for this analysis rely on a branching ratio
of 100% for the decay 1 W 10 . This is in general true in SUGRA models as long
as the 1 is heavier than the 1 , corresponding to moderate values for tan . For high
tan , the decay 1 1 can become kinematically possible, and its branching ratio
can dominate the standard 1 W 10 . The stau in turns typically decays as 1 10 .
The three-leptons signature is in this case even enhanced, due to the higher branching
fraction into electrons and muons for the compared to the W , at the price of a softer
27
Fig. 14. 5 reach in the m0 m1/2 plane for tan = 10 and three different choices of the 211
coupling for an integrated luminosity of 30 fb1 at the LHC. The significance is defined only
dotted line shows the region below which the 1 decays to a virtual W .
28
lepton spectrum. The 10 reconstruction is still possible but the presence of three neutrinos
(two additional neutrinos come from the leptonic decay) renders the reconstruction of the
particles earlier in the decay chain difficult. The analysis efficiency is essentially unaffected
with respect to the low tan case as long as the mass difference between the 1 and the
10 is larger than 50 GeV. For 1 and 10 masses too much degenerate, the transverse
momentum of the charged lepton coming from the decay would often fall below the
analysis requirements, leading thus to a reduction of the signal efficiency. The reach in the
m0 m1/2 plane is shown in Fig. 15 for tan = 35 and three different choices of the 211
coupling. The branching fraction for the decay 1 1 is higher than 50% to the left of
the dotted line, and the region for which a reduced signal efficiency is expected is displayed
as a grey area. The reach for 10 detection is similar to the low tan case, but the region in
which the full reconstruction of the sneutrino decay chain is possible is severely restricted.
5. Conclusions
We have analysed the resonant sneutrino production at LHC in supersymmetric models
with R-parity violation. We have focused on the three-leptons signature which has a small
Standard Model background, and allows a model-independent mass reconstruction of the
full sneutrino decay chain.
A detailed study for an example MSSM point has shown that the mass reconstruction
analysis has an efficiency of a few percent, and that a precise measurement of the masses of
, 1 , 10 can be performed. Both the Standard Model background and the backgrounds
from other SUSY pair productions were studied in detail, and shown to be well below
/ p coupling 211 taken at the present
the expected signal for a value of the considered R
low-energy limit.
The trilepton signal from sneutrino production was then studied as a function of the
model parameters under different model assumptions, and sensitivity over a significant
part of the parameter space was found. Within the MSSM, the production of a 850 GeV
sneutrino for 211 > 0.05, and of a 300 GeV sneutrino for 211 > 0.01 can be observed in
the first three years of LHC running. In the framework of the mSUGRA model, the region
in the m0 m1/2 space accessible to the analysis was mapped as a function of the value
of the Rp -violating coupling for representative values of tan . A significant part of the
m0 m1/2 plane will be accessible for 211 > 0.01.
Although the detailed study was focused on the case of a single dominant 211 coupling,
we have found that the resonant sneutrino production analysis can bring interesting
/ p couplings of the type 2j k , compared to the low-energy
sensitivities on all the R
constraints. The resonant sneutrino production should also allow to test most of the 1j k
coupling constants.
In conclusion we have demonstrated that if minimal supersymmetry with R-parity
violation is realised in nature, the three-leptons signature from sneutrino decay will be
a privileged channel for the precision measurement of sparticle masses and for studying
the SUSY parameter space, over a broad spectrum of models. Analyses based on the
29
Fig. 15. 5 reach in the m0 m1/2 plane for tan = 35 and three different choices of the 211
coupling for an integrated luminosity of 30 fb1 at the LHC. The significance is defined only
dashed line shows the region below which the 1 decays to a virtual W . The region to the left of the
dotted lines has a branching ratio for 1 1 larger than 50%, and the grey area indicates the
region for which a low signal efficiency is expected.
30
study of events including three leptons, were often advocated in the literature [3851] as
a particularly sensitive way of attacking the search for SUSY at the LHC in the standard Rconserving scenario. The higher lepton multiplicity and the possibility to perform precise
measurements of the model parameters make this kind of analyses an even more attractive
possibility in the case of R-parity violation with dominant couplings.
Acknowledgements
We would like to acknowledge useful discussions with many members of the ATLAS
Collaboration about the details of the ATLAS experimental setup. This work was initiated
during a workshop held in Les Houches. We warmly thank Patrick Aurenche and all of the
organising team for the stimulating program, the excellent atmosphere and the outstanding
computing facilities. We are also deeply indebted to the HERWIG team for allowing us to
use a prerelease of the HERWIG 6.0 Monte Carlo for the initial part of this work.
References
[1] H. Dreiner, in: G.L. Kane (Ed.), Perspectives on Supersymmetry, World Scientific, 1998, hepph/9707435.
[2] S. Dimopoulos, L.J. Hall, Phys. Lett. B 207 (1987) 210.
[3] H. Dreiner, G.G. Ross, Nucl. Phys. B 365 (1991) 597.
[4] V. Barger, G.F. Giudice, T. Han, Phys. Rev. D 40 (1989) 2987.
[5] G. Bhattacharyya, Invited talk presented at Beyond the Desert, Castle Ringberg, Tegernsee,
Germany, 814 June, 1997; Susy 96, Nucl. Phys. B (Proc. Suppl.) 52A (1997) 83.
[6] R. Barbier et al., Report of the group on the R-parity violation, hep-ph/9810232.
[7] F. Ledroit, G. Sajot, GDR-S-008, ISN, Grenoble, 1998.
[8] B.C. Allanach, A. Dedes, H. Dreiner, Phys. Rev. D 60 (1999) 075014.
[9] S. Dimopoulos, R. Esmailzadeh, L.J. Hall, J. Merlo, G.D. Starkman, Phys. Rev. D 41 (1990)
2099.
[10] P. Bintruy et al., ECFA Large Hadron Collider (LHC) Workshop, Vol. II, Aachen, 1990.
[11] A. Datta, J.M. Yang, B.-L. Young, X. Zhang, Phys. Rev. D 56 (1997) 3107.
[12] R.J. Oakes, K. Whisnant, J.M. Yang, B.-L. Young, X. Zhang, Phys. Rev. D 57 (1998) 534.
[13] J.L. Hewett, T.G. Rizzo, Proceedings of the XXIX International Conference on High Energy
Physics, Vancouver, CA, 2329 July, 1998, hep-ph/9809525.
[14] P. Chiappetta et al., hep-ph/9910483.
[15] J. Kalinowski, R. Rckl, H. Spiesberger, P.M. Zerwas, Phys. Lett. B 414 (1997) 297.
[16] J. Kalinowski, Proceedings of Beyond the Desert 97 Accelerator and Non-accelerator
Approaches, Ringberg Castle, Germany, June 1997, hep-ph/9708490.
[17] S. Bar-Shalom, G. Eilam, A. Soni, Phys. Rev. D 59 (1999) 055012.
[18] E.L. Berger, B.W. Harris, Z. Sullivan, Phys. Rev. Lett. 83 (1999) 4472.
[19] H. Dreiner, P. Richardson, M.H. Seymour, in: Proceedings of the BTMSSM subgroup of the
Physics at run II Workshop on Supersymmetry/Higgs, in press, hep-ph/9903419.
[20] H. Dreiner, P. Richardson, M.H. Seymour, hep-ph/0001224.
[21] H. Dreiner, P. Richardson, M.H. Seymour, hep-ph/0007228.
[22] I. Hinchliffe, F. Paige, Phys. Rev. D 61 (2000) 095011.
[23] J.F. Gunion, H.E. Haber, Phys. Rev. D 37 (1988) 2515.
31
[24] N. Ghodbane, S. Katsanevas, P. Morawitz, E. Perez, SUSYGEN 3.0/06, lyoinfo.in2p3.fr

/susygen/susygen3.html;
N. Ghodbane, hep-ph/9909499.
[25] JETSET 7.3 and 7.4;
T. Sjstrand, Lund Univ. preprint LU-TP-95-20, August, 1995, p. 321;
T. Sjstrand, CERN preprint TH-7112-93, February, 1994, p. 305.
[26] V.N. Gribov, L.N. Lipatov, Sov. J. Nucl. Phys. 15 (1972) 78;
G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298;
Y.L. Dokshitzer, JETP 46 (1977) 641.
[27] E. Richter-Was, D. Froidevaux, L. Poggioli, ATLFAST 2.0: a fast simulation package for
ATLAS, ATLAS Internal Note ATL-PHYS-98-131, 1998.
[28] T. Sjstrand, Comput. Phys. Commun. 82 (1994) 74.
[29] D.O. Carlson, S. Mrenna, C.P. Yuan, private communication;
D.O. Carlson, C.P. Yuan, Phys. Lett. B 306 (1993) 386.
[30] H. Baer, F.E. Paige, S.D. Protopopescu, X. Tata, hep-ph/9305342;
H. Baer, F.E. Paige, S.D. Protopopescu, X. Tata, hep-ph/9804321.
[31] G. Corcella, I.G. Knowles, G. Marchesini, S. Moretti, K. Odagiri, P. Richardson, M.H. Seymour, B.R. Webber, JHEP 0101 (2001) 010;
G. Marchesini, B.R. Webber, G. Abbiendi, I.G. Knowles, M.H. Seymour, L. Stanco, Comput.
Phys. Commun. 67 (1992) 465.
[32] The ALEPH Collaboration, Internal Note ALEPH 99-078, CONF 99-050, 1999, contributed to
HEP-EPS 99.
[33] J.F. Gunion, H.E. Haber, Nucl. Phys. B 272 (1986) 1;
J.F. Gunion, H.E. Haber, Nucl. Phys. B 402 (1993) 567, Erratum.
[34] The ATLAS Collaboration, ATLAS Detector and Physics Performance Technical Design
Report, ATLAS TDR 15, CERN/LHCC/99-15.
[35] A. Pukhov, E. Boos, M. Dubinin, V. Edneral, V. Ilyin, D. Kovalenko, A. Kryukov, V. Savrin,
S. Shichanin, A. Semenov, CompHEP a package for evaluation of Feynman diagrams and
integration over multi-particle phase space, preprint INP MSU 98-41/542, hep-ph/9908288.
[36] J.M. Campbell, R.K. Ellis, Phys. Rev. D 60 (1999) 113006.
[37] F. Dliot et al., Phys. Lett. B 475 (2000) 184.
[38] P. Nath, R. Arnowitt, Mod. Phys. Lett. A 2 (1987) 331.
[39] R. Arnowitt, R. Barnett, P. Nath, F. Paige, Int. J. Mod. Phys. A 2 (1987) 1113.
[40] H. Baer, K. Hagiwara, X. Tata, Phys. Rev. D 35 (1987) 1598.
[41] R. Barbieri, F. Caravaglios, M. Frigeni, M. Mangano, Nucl. Phys. B 367 (1991) 28.
[42] T. Kamon, J.L. Lopez, P. McIntyre, J.T. White, Phys. Rev. D 50 (1994) 5676.
[43] H. Baer, C.-H. Chen, C. Kao, X. Tata, Phys. Rev. D 52 (1995) 1565.
[44] S. Mrenna, G. Kane, G. Kribs, J. Wells, Phys. Rev. D 53 (1996) 1168.
[45] H. Baer, C.-H. Chen, M. Drees, F. Paige, X. Tata, Phys. Rev. Lett. 79 (1997) 986.
[46] H. Baer, C.-H. Chen, M. Drees, F. Paige, X. Tata, Phys. Rev. D 58 (1998) 075008.
[47] V. Barger, C. Kao, T.-J. Li, Phys. Lett. B 433 (1998) 328.
[48] V. Barger, C. Kao, preprint FERMILAB-PUB-98/342-T, Phys. Rev. D 60 (1999) 115015.
[49] J. Lykken, K.T. Matchev, preprint FERMILAB-PUB-99/034-T, Phys. Rev. D 61 (2000) 015001.
[50] K.T. Matchev, D.M. Pierce, preprint FERMILAB-PUB-99/078-T, Phys. Rev. D 60 (1999)
075004.
[51] H. Baer, M. Drees, F. Paige, P. Quintana, X. Tata, Phys. Rev. D 61 (2000) 095007.

One loop soft supersymmetry breaking terms

in superstring effective theories
Pierre Bintruy a , Mary K. Gaillard b , Brent D. Nelson b
a LPT, Universit Paris-Sud, Bat. 210 F-91405 Orsay Cedex, France
b Department of Physics, University of California, and Theoretical Physics Group,
50A-5101, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA

Received 15 November 2000; accepted 13 December 2000
Abstract
We perform a systematic analysis of soft supersymmetry breaking terms at the one loop level
in a large class of string effective field theories. This includes the so-called anomaly mediated
contributions. We illustrate our results for several classes of orbifold models. In particular, we
discuss a class of models where soft supersymmetry breaking terms are determined by quasi model
independent anomaly mediated contributions, with possibly non-vanishing scalar masses at the
one loop level. We show that the latter contribution depends on the detailed prescription of the
regularization process which is assumed to represent the Planck scale physics of the underlying
fundamental theory. The usual anomaly mediation case with vanishing scalar masses at one loop
is not found to be generic. However gaugino masses and A-terms always vanish at tree level if
supersymmetry breaking is moduli dominated with the moduli stabilized at self-dual points, whereas
the vanishing of the B-term depends on the origin of the -term in the underlying theory. We
also discuss the supersymmetric spectrum of O-I and O-II models, as well as a model of gaugino
condensation. For reference, explicit spectra corresponding to a Higgs mass of 114 GeV are given.
Finally, we address general strategies for distinguishing among these models. 2001 Elsevier
Science B.V. All rights reserved.
PACS: 12.60.Jv; 14.80.Ly; 11.25.Mj; 04.65.+e
1. Introduction
In any given supersymmetric theory, a consistent analysis of the soft terms is necessary
in order to make reliable predictions. Such a systematic analysis was performed at tree level
This work was supported in part by the Director, Office of Science, Office of Basic Energy Services, of the
U.S. Department of Energy under Contract DE-AC03-76SF00098 and in part by the National Science Foundation
under grants PHY-95-14797 and INT-9910077.
E-mail address: bdnelson@lbl.gov (B.D. Nelson).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 7 5 9 - 8
P. Bintruy et al. / Nuclear Physics B 604 (2001) 3274
33
by Brignole, Ibez and Muoz [1] some time ago for a large class of four-dimensional
string models. One of the nice features of this analysis was to make explicit the dependence
of the soft terms in the auxiliary field vacuum expectation values (vevs) and thus to relate
them directly to the supersymmetry breaking mechanism. In this respect, the auxiliary
fields FS and FT associated, respectively, with the string dilaton and the moduli fields are
expected to play a central role in these superstring models.
This analysis showed that, besides a universal contribution associated with the dilaton
field, soft terms generically receive from moduli fields a non-universal contribution which
may lead to a very different phenomenology from the standard one referred to as the
minimal supergravity model.
Recently, a new contribution to the soft supersymmetry breaking terms has been
discussed under the name of anomaly mediated terms [2,3] that arise at the quantum
level from the superconformal anomaly. They are truly supergravity contributions in the
sense that they involve the auxiliary fields of the supergravity multiplet, more precisely
the complex scalar auxiliary field M in the minimal formulation (see, e.g., [4] or [5]).
However if these contributions are included, then all one-loop contributions to the soft
terms should be taken into account. In what follows, we present the general form of these
contributions, expressed in terms of the auxiliary fields and we discuss them for several
classes of superstring models. We stress that some of the contributions depend on the way
the underlying theory regulates the low energy effective field theory. In particular we find a
model of anomaly mediation where the scalar masses might be non-vanishing at one loop.
2. General form of one loop supersymmetry breaking terms

In this section, we give the complete expressions for the soft supersymmetry breaking
terms. 1 Let us start by introducing our notations. We consider a set of chiral superfields
Z M (the associated scalar field will be denoted by zM ) which belong to two distinct classes:
the first class Z i denotes observable superfields charged under the gauge symmetries, the
second class Z n describes hidden sector fields, typically in the models that we will consider
the dilaton and T and U moduli fields. Their interactions are described by three functions:

M
the Khler potential K(Z M , Z
), the superpotential W (Z i , Z n ) and the gauge kinetic
a
n
functions f (Z ), one for each gauge group Ga .
The auxiliary fields are obtained by solving the corresponding equations of motion. They
read for the chiral superfields: 2

,
F M = eK/2K M N W
+ KN
W
N
(2.1)
1 We keep only the terms of leading order in m

3/2 /R , where m3/2 is the gravitino mass (typically less than
10 TeV) and R is the renormalization scale, taken to be the scale at which supersymmetry is broken (typically
1011 GeV or higher).
2 We follow the sign conventions of [5,6]. Let us note that the auxiliary fields differ by a sign from the ones
used by Brignole, Ibez and Muoz [1].
34
= W
/ Z
N and K M N is the inverse of the Khler metric
where, as is standard, W
N
N . The supergravity auxiliary field M simply reads:
KM N = 2 K/Z M Z
M = 3eK/2W.
(2.2)
As a sign of spontaneous breaking of supersymmetry, the gravitino

mass is directly
expressed in terms of its vev (in reduced Planck scale units MPl / 8 = 1 which we use
from now on):

1
.
m3/2 = M
= eK/2 W
3
In terms of these fields, the F -term part of the potential reads:
(2.3)

J 1 M M.
V = F I KI J F
(2.4)
3
Since in what follows we will assume vanishing D-terms we will only be interested in this
part of the scalar potential.
Finally, the holomorphic function fa (Z M ) is the coefficient of the gauge kinetic term in
superspace. Its vev yields the gauge coupling associated with the gauge group Ga :
Re fa =
1
.
ga2
(2.5)
In the weak coupling regime, the models that we consider have a simple gauge kinetic
function:

fa(0) Z n = ka S,
(2.6)
where S is the string dilaton and ka is the affine level. 3 In what follows, we will adopt
the description of the dilaton in terms of a chiral superfield, although all our results
were obtained in the linear superfield formulation as described in Appendix A. Quantum
corrections involve the moduli fields T . Of central importance at the perturbative level,
are the diagonal modular transformations:
T
aT ib
,
icT + d
ad bc = 1,
a, b, c, d Z,
(2.7)
that leave the classical effective supergravity theory invariant. At the quantum level there is
an anomaly [712] which is cancelled by a universal GreenSchwarz counterterm [12] and
model-dependent string threshold corrections [7,8]. In order to present the contributions of
these terms to the gaugino masses, we must be somewhat more explicit.
We take the standard form:
3

+ K T = k(S + S)
ln T + T ,
K(S, T ) = k(S + S)
(2.8)
=1
3 From now on, we will only consider affine level one nonabelian gauge groups, i.e., k = 1 (k = 5/3 for the
abelian group U (1)Y of the Standard Model).
35
for the moduli dependence of the Khler potential. We will assume for the simplicity of
the expressions which follow that the Khler metric for the matter fields has the form:

2
Ki j = i Z n i j + O Z i .
(2.9)
Indeed a matter field which transforms as

n
Z i icT + d i Z i
(2.10)
under the modular transformations (2.7) is said to have weight ni and has

n
T + T i .
i =
(2.11)
The superpotential transforms as

1
W W
.
icT + d
(2.12)
2.1. Gaugino masses

The tree level contribution to the masses of canonically normalized gaugino fields simply
reads: 4
ga2 n
F n fa .
(2.13)
2
The full one loop anomaly-induced contribution has been obtained recently [1315]. It is:

1
ga ()2 2ba0
1 i n
(1)
i
n
M
Ma an =
Ca F Kn
Ca F n ln i ,
Ca
2
3
8 2
4 2
Ma(0) =
(2.14)
Ca , Cai
where
are the quadratic Casimir operators for the gauge group Ga respectively in
the adjoint representation and in the representation of Z i , ba0 is the one loop coefficient of
the corresponding beta function:

1
0
i
ba =
(2.15)
Ca ,
3Ca
16 2
i
(Z n )
have been defined in (2.9). The first term is the one generally
and the functions i
quoted [2,3]: ba0 ga2 m3/2 using (2.3). It is often obtained by a spurion field computation [16]. It is a finite contribution related to the superconformal anomaly, rather than a
remnant of the ultraviolet divergences. The remaining terms have been obtained recently
[14,15] using a general supersymmetric expression for the anomaly-induced terms [17]
or PauliVillars regulators [18]. They reflect the Khler conformal and chiral anomalies
associated with ultraviolet divergences of the low energy effective field theory [9,10].
4 From now on, we will suppress the brackets indicating that all explicit expressions of soft terms are
given in terms of vevs of fields.
36
Other terms may appear in string models at one loop. The GreenSchwarz counterterm
has the following form

LGS = d 4 ELVGS ,
(2.16)
in a linear multiplet formalism [19,20] where L is a linear multiplet which includes the
degrees of freedom of the dilaton and of the antisymmetric tensor present among the
massless string modes. The real function VGS reads:

GS ni i 2
+ O 4 .
VGS =
(2.17)
T +T
ln T + T +
pi
2
24
The group-independent factor GS is simply equal to 3CE8 , where CE8 = 30 is the

Casimir operator of the group E8 in the adjoint representation, if there are no Wilson lines.
Otherwise, it can be smaller in magnitude. In the rest of this section, we will neglect 5 terms
of order 2 .
String threshold corrections may be interpreted as one loop corrections to the gauge
kinetic functions. They read:

GS
i
1
(1)
2
+ Ca
ln T
1 + 2ni Ca ,
fa =
(2.18)
3
16 2
i
where (T ) is the classical Dedekind function:

(T ) = e
T /12

1 e2nT ,
(2.19)
n=1
which transforms as
1/2

T
T icT + d
(2.20)
under the modular transformation (2.7). We will also use in the following the Riemann zeta
function:
1 d(T )
.
(T ) =
(2.21)
(T ) dT
Combining contributions from the GreenSchwarz counterterm and string threshold
corrections with the light loop contribution (2.14) yields a total one loop contribution [13]:

ga ()2 2 GS
1 i
1
(1)
0
1
+
3n
2(t
F
+
b
C
)
+
Ma =
a
a
i
2
3 16 2
8 2
t + t

2ba0
gs2
i
Ca
Ca F S .
+
(2.22)
M+
3
16 2
i
The last term involves the value of the string coupling at unification. In models with
dilaton stabilization through nonperturbative corrections to the Khler potential [21,22],
5 See Ref. [13] for formulas taking into account the terms of order 2 .
37
the value of the gauge coupling at the string scale (unification scale) Ms is related to gs2 =
2Ks by: 6

s + s
g 2 (Ms ) = gs2 1 + f gs2 /2 =
,
2
where the function f parameterizes nonperturbative string effects [23].
Let us note that the non-holomorphic Eisenstein function

2 (T , T ) 2 2 (T ) + 1/[T + T ] 2G2 (T , T ),
G
(2.23)
(2.24)
vanishes at the self-dual points T = 1 and T = ei/6 .

In the presence of the GS term (2.16), the scalar potential also receives some corrections.
In particular
= K + gs M VGS ,
KM N K
(2.25)
N
MN
MN
2
in (2.1) and (2.4). If pi = 0 in (2.17), the effect of (2.25) is to multiply the vev of F by
the numerical factor 1 1 gs2 GS /48 2 1.1 if gs2 = 0.5. Additional corrections are
given in Appendix A; they are unimportant if GS (t + t )1 |F WS /W |/48 2 |M/3|:
for example when supersymmetry breaking is dilaton dominated or if the superpotential is
independent of the dilaton. The domain of validity of this approximation is discussed in
Appendix A. We neglect all these corrections in the subsequent sections of the text, except
in Section 3.5 where pi = 0 in (2.17) is considered.
2.2. A-terms
A-terms are cubic terms in the scalar potential that generally arise when supersymmetry
is broken:
1
VA =
Aij k eK/2Wij k zi zj zk + h.c.
6
ij k
1
=
Aij k eK/2(i j k )1/2 Wij k z i z j z k + h.c.,
6
(2.26)
ij k
1/2
where z i = i
zi is a normalized scalar field, and Wij k = 3 W (zN )/zi zj zk . At tree
level we have

n

K
/Wij k .
A(0)
(2.27)
ij k = F n ln i j k e
The one loop contributions to A-terms (and to scalar masses and B-terms discussed
below) are considerably more sensitive to the details of Planck scale physics than the
gaugino masses considered in the preceding subsection. The most straightforward way
to regulate an effective theory is by introducing heavy fields known as PauliVillars
(PV) fields with masses of the order of the effective cut-off, and couplings to light
fields chosen so as to cancel ultraviolet divergences. The PV masses can be interpreted
6 In the linear multiplet formulation [19,20], g 2 = 25.
s
38
as parameterizing effects of the underlying theory. These masses are to some extent
constrained by supersymmetry. These constraints are much more powerful in determining
the loop-corrected gaugino masses than the other soft parameters, for the reasons that
follow.
All gauge-charged PV fields contribute to the vacuum polarization and to the gaugino
masses. Their gauge-charge weighted masses are constrained by finiteness and supersymmetry to give the result in (2.22). The superfield operator that corresponds to these terms
is the same one that contains the field theory chiral and conformal anomalies under Khler transformations of the type (2.7), and is therefore completely determined by the chiral
anomaly which is unambiguous. Specifically, the conformal and chiral anomalies are the
real and imaginary part of an F-term operator; the former is governed by the field dependence of the PV masses that act as an effective cut-off and are determined by supersymmetry from the latter [9].
On the other hand, only a subset of charged PV fields A contribute to the
renormalization of the Khler potential, which determines the matter wave function
renormalization and governs the loop corrections to soft parameters in the scalar
potential. Their PV masses are determined by the product of the inverse metrics of
these fields and of fields A to which they couple in the PV superpotential to generate
Planck scale supersymmetric masses, as well as by a priori unknown holomorphic
functions A (Z N ) of the light fields that appear in the PV superpotential. While the
Khler metrics of the A are determined by finiteness requirements, the metrics of
the A are arbitrary. In operator language, the conformal anomaly associated with the
renormalization of the Khler potential is a D-term; it is supersymmetric by itself and
there is no constraint, analogous to the conformal/chiral anomaly matching in the case
of gauge field renormalization with an F-term anomaly, on the effective cut-offs or
PV masses for this term. As a consequence the soft terms in the scalar potential
cannot be determined precisely in the absence of a detailed theory of Planck scale
physics.
The leading order A-term Lagrangian was given in [15]; from the definition (2.26) we
obtain for the one loop contribution:

1 a (0)
A(1)
i 2Ma ln |m
i ma |/2R + F n n ln(|m
i ma |)
ij k = i M +
3
a

(0)

lm
i Ailm ln |ml mm |/2R + F n n ln(|ml mm |)
+
lm
+ (i j ) + (j k),
(2.28)
where mi , ma are the PV masses of the supermultiplets i , a that regulate loop

contributions of the light supermultiplets, respectively Z i , Wa , and m
i is the PV mass
i

of a field , in the gauge group representation conjugate to that of i (and of Z i )
needed to complete the regularization of the gauge-dependent contribution to the one
39
loop Khler potential renormalization. 7 The parameters determine the chiral multiplet
wave function renormalization. In the supersymmetric gauge [24] the matter wave function
renormalization matrix is 8

i

1
j
j
2
2
K
j kl

4i
.
i =
ga Ta i e
Wikl W
(2.30)
32 2
a
kl
The matrix (2.30) is diagonal in the approximation in which generation mixing is neglected
in the Yukawa couplings; in practice only the T c Q3 Hu Yukawa coupling is important. We
have made this approximation in (2.28), and set
jk
j
j
i =
i +
ia ,
i i i ,
a
jk
ga2 2 i
T ,
8 2 a i
eK
(2.31)
(i j k )1 |Wij k |2 .
32 2
We are interested here in string-derived models, in which case the moduli dependence
of the function Wij k is fixed by modular invariance:
2(1+n +n +n )
i
j
k .
Wij k = wij k
(2.32)
T
ia =
jk
i =
Similarly, the quantum corrected theory should be perturbatively invariant under the
modular transformation (2.7). This can be achieved if the couplings of the relevant PV
i that contribute to the renormalization
fields are modular invariant. For the fields i , a ,
of the Khler potential, we have [18], for typical orbifold models,

n
T + T i ,
i : i = i =
i :
a :
i = i ,
a = ga2 eK
= ga2 ek

T + T
1
(2.33)
i , a ),
Setting for A = ( i ,

m
A : A = hA (S + S)
T + T A ,
(2.34)
the functions A (Z n ) and therefore the PV masses are fixed up to a constant by modular
covariance, and we obtain for the full A-term, using (2.28),

a lm
1
1 (0) 1
F
+ 2 t
i pia +
i plm
Aij k = Aij k i M
3
3
t + t
a
lm
7 Assuming a PV mass term of the form (Z N ) A A in the superpotential, we have explicitly:

A
1 1
m2A = eK A
|A |2 ,
A
and are defined in (2.33) and (2.34).
where A
A
8 We define the -function following the conventions of Cheng and Li [25].
(2.29)
40

a 2 lm 2
+F
ln ia +
i ln lm
s a i
lm

4 a (0) lm (0)

2
ln t + t t
i pia Ma +
i plm Ailm
S
+2
ia Ma(0) ln
lm
lm (0) 2

2
ia /2R +
i Ailm ln lm /2R
lm
+ cyclic(ij k),
(2.35)
with

1
ni + nj + mi + mj ,
2
2
ij = i j ek (hi hj )1/2 ,
pij = 1 +

1
1 + ma + m
i ni ,
2

1/2
2
ai = i a ek/2ga ha h i
,
pia
=
(2.36)
where i j , i a are constants. The tree level A-terms and gaugino masses are given
from (2.27) and (2.13), using (2.32), respectively, by

1
(0)

Aij k =
F ni + nj + nk + 1
+ 2 t
kS F S ,
t
+
t
Ma(0) =
ga2 () S
F = s ln ga2 F S .
2
(2.37)
For example, if the PV masses mi , ma , m

i in (2.29) are constant (as well as A ) 9 we
have from (2.29)
(A)
ni + mi = 1,
i = 1,
ni m
ma = 0,
(2.38)
= 0,
2ij and 2ai constants. A commonly (though often implicitly)
and thus pij = pia
made assumption in the literature is instead that A has the same Khler metric as A :
(B)
mi = ni ,
m
i = ni ,
ma = 1,
(2.39)
this gives 2ij constant, 2ia = ga2 , pij = 1 + ni + nj , pai

= ni . Distinguishing among
the possibilities from the theoretical point of view requires string-loop calculations similar
to those used to fix the moduli dependence of the gauge kinetic function [7,8]. We note
however that if supersymmetry breaking is moduli mediated (F S = 0) with the moduli
9 It was shown in [18] that the Khler potential for the untwisted sector from orbifold compactification can be
made modular invariant with the relevant masses constant. Since the tree level Khler potential for the twisted
sector is not known beyond quadratic order in twisted sector fields, the one loop corrections to it cannot be
calculated.
41
stabilized at self-dual points, as suggested by modular invariance, the tree level soft terms
(2.37) vanish, and the only one loop contribution is the standard anomaly mediated term
1
Aanom
(2.40)
ij k = M(i + j + k ).
3
Therefore if gaugino masses and/or A-terms are measured to be significantly larger than
the anomaly mediated values (see also (2.22)), in the string context of assumed modular
invariance this would quite generally suggest dilaton dominated supersymmetry breaking
and/or moduli vevs far from the self-dual points.
2.3. B-terms
B-terms are quadratic terms in zi and in z that appear in the scalar potential after
supersymmetry breaking if there are such quadratic terms in the superpotential and/or
Khler potential:
3
1 n i j
ij Z Z Z + O Z i ,
W Zi =
(2.41)
2
ij
i i 2 1 n n i j

=
Z Z + h.c. + O Z i 3 .
K Z ,Z
(2.42)
ij Z , Z
i Z +
2
i
ij
These terms give rise to masses for the chiral supermultiplets Z i :

1
2

eK/2 i ij j + h.c. + eK zi j |ij |2 ,
LM =
2
ij

1
n n ij .
ij = ij eK/2 Mij F
(2.43)
3
The Lagrangian (2.43) is globally supersymmetric although the mass term arising from
ij appears [26] only after local supersymmetry breaking: m3/2 = 0. The B-term potential
takes the form
1
VB =
Bij eK/2 ij zi zj + h.c.
2
ij
1
=
(2.44)
Bij eK/2 (i j )1/2 ij z i z j + h.c.
2
ij
At tree level we have

1

(0)
.
Bij = F n n ln i j eK /ij + M
(2.45)
3
The one loop contribution is easily extracted from the result for the leading order A-term
Lagrangian given in [15]; we obtain

1 a (0)
i 2Ma ln |m
i ma |/2R + F n n ln(|m
i ma |)
+
Bij(1) = i M
3
a

(0)

+
(2.46)
ilm Ailm ln |ml mm |/2R + F n n ln(|ml mm |) + (i j ).
lm
42
Using the assumptions and results of Section 2.2 we obtain for the full B-term in stringderived orbifold models

a lm
1 (0) 1
1
Bij = Bij i M
F
+ 2 t
i pia +
i plm
2
3
t + t
a
lm

2 lm 2
S
a
+F
ln ia +
i ln lm
s a i
lm

4
(0)
(0)
lm
2
ln t + t t
p M +
p A
i
+2
ia
lm
ilm
lm

lm (0) 2

ia Ma(0) ln 2ia /2R +
i Ailm ln lm /2R + i j,
(2.47)
lm
with the various parameters defined in (2.36). Because we have assumed modular
covariance for trilinear terms in the superpotential, 10 Eqs. (2.37) assure that the one loop
contribution to the B-term reduces to the anomaly mediated term
1
Bijanom = M(
i + j )
3
(2.48)
if supersymmetry breaking is moduli mediated (F S = 0) with the moduli stabilized at

self-dual points.
However tree level B-terms may not vanish in this case; they are sensitive to the origin
of the -term (2.43). A modular invariant Khler potential of the form (2.42) was
constructed [27] for (2, 2) orbifold compactifications of the heterotic string with both
T-moduli and U-moduli. Here we restrict the moduli to T-moduli in which case modular
) requires
invariance of the Khler potential K(Z i , Z

q 2k 2q
ij
ij ,
n = aij (S, S)
T + T ij T
T
ij Z n , Z
kij = qij + ni + nj ,
and modular covariance of the superpotential (2.41) requires
2w

ij
ij Z n = nij
T
, wij = 1 + ni + nj .
(2.49)
(2.50)
Bilinear terms in matter fields do not appear in the tree level superpotential in superstringderived models, but they can be generated from higher dimension terms when some fields
acquire vevs. Bilinear terms in the Khler potential could similarly be generated from
higher dimension terms. These will be modular invariant if only modular invariant fields
acquire vevs. For example D-term induced breaking of an anomalous U (1) above the scale
of supersymmetry breaking preserves modular invariance. On the other hand if Hu Hd = 0,
it is of the order of the electroweak scale: it presumably originates from the vev N of an
10 In fact we need only assume this for the dominant T c Q H term; in making the approximation (2.31) we
3 u
implicitly neglect the small Yukawa couplings that may themselves arise from higher dimension operators and/or
loop corrections.
43
electroweak singlet field N and there is no reason that modular invariance should still be
operative at such low energy scales. In any case, the corresponding B-term is generated by
an A-term in this instance.
To consider the case in which the -term is already present at the supersymmetry
breaking scale, we can parameterize ij , ij as in (2.49) and (2.50), but with the exponents
kij , qij , wij left a priori arbitrary; the case of modular invariance is recovered when the last
equalities in those equations are imposed. We also assume that Standard Model singlets N
whose vevs may generate quadratic terms in the superpotential or Khler potential do not
contribute to supersymmetry breaking: F N = 0.
If the -term (2.43) is generated by a superpotential term (2.41), we obtain for the tree
level B-term

(0)
1

Bij superpotential =
w
F 1 + ni + nj
+
2
t
ij
t + t
1
kS F S + M.
(2.51)
3
The coefficients of the moduli auxiliary fields vanish at the moduli self-dual points when
for
modular invariance (2.50) is imposed, but the B-term does not vanish: B (0) = 13 M
S
F = 0. Although it seems rather implausible that a hierarchically small value of ij TeV
would be generated at the supersymmetry breaking scale 1011 , it could conceivably arise
as a product of vevs in a superpotential term of very high dimension [28].
A more natural origin for a -term of the order of a TeV is a quadratic term in the Khler
potential as in (2.42). The expression for B (0) obtained from the general parameterization
(2.49) is rather complicated and does not in general vanish when F S = 0 and modular
invariance (2.49) is imposed. As an example, consider the simplifying assumptions that
= constant and q = W/t = 0, then for ij = 0
aij (S, S)
ij

2k
1
ij
ij = aij W t
,
F = t + t ,
3

(0)

1
1 + ni + nj
Bij Kahler potential =
F
+ 2 t kij
t + t
1
(kS + S ln W )F S + M.
(2.52)
3
In this case even the coefficients of the moduli auxiliary fields do not vanish at the moduli
self-dual points when modular invariance is imposed, and under the above conditions we
It is possible that a comparison of BHu Hd with A-terms might shed some
get Bij(0) = 23 M.
light on the origin of the -term (2.43).
2.4. Scalar masses
The expression soft scalar masses refers to mass terms in the scalar potential

2 2 i 2
Mi2 i zi =
Mi z ,
VM =
i
(2.53)
44
with no supersymmetric counterpart in the chiral fermion Lagrangian. The tree level soft
scalar masses are given by
2
1
m n m ln i .
(2.54)
F nF
= MM
9
Here and throughout the discussion of scalar masses, we drop terms proportional to the
vacuum energy, Eq. (2.4).
The one loop contribution [15] to soft masses is determined by the soft parameters of the
PV sector. The A-terms of the PV sector and the masses of i , a , i are determined by
the soft parameters of the light field tree Lagrangian. Denoting by NA the soft mass of A ,
the one loop scalar masses can be written in the form

(1)2
1 a 2 2 (0)2 (0) 2
Mi
=
Na + Ni Ma
Mi
2 a i

jk

2
2
(0)
(0)
+
i Nj2 + Nk2 + Mj
+ Mk
Mi(0)
jk

2
2
m

m + F n n ln |m
i ma |/2R
ia 3 Ma(0) Mi(0) + Ma(0) F

jk
jk

Mj(0)
2

2
2 1 (0) m

m + F n n
A
+ Mk(0) + A(0)
+
F
ij k
2 ij k

ln |mj mk |/2R

a (0) 1 j k (0)
1

+ M +M
i Ma +
i Aij k .
3
2
a
(2.55)
jk
For orbifold compactifications of string theory, with the Khler metrics given in (2.11),
we obtain for the tree level scalar masses

(0)2 1
+
n t + t 2 .
= MM
F F
Mi
(2.56)
i
9
+ t ) and M (0) = 0 in the no-scale case

Note that if W/t = 0, then F = 13 M(t
i

with ni = 1, as for the untwisted sector of orbifold models. The soft masses NA are
given by the standard formula (2.54) by just replacing i by A . The one loop contribution
is then given by

jk

(1)2 1
S
S
a
2
2
i F F
s s
Mi
= M M
i ln ia +
i ln j k
9
a
jk

jk

2
t + t
F F
ia pai
+
pjk
i
jk

a (0) 1 j k (0)
1

+ M +M
i Ma +
i Aij k
3
2
a
jk

F

a

1
+ 2 t
t + t
45

jk
1
(0)
ia pia
Ma(0) +
i pjk Aj k + h.c.
2
jk

j k (0)

1
+ FS
a M (0) ln 2ia +
i Aij k ln 2j k + h.c.
s a i a
2
jk

4
ln t + t t

2 (0) 2
3 Ma(0) Mi
ia pia
+
+

(0) 2
Mj
(0) 2
Mk
(0) 2
Aij k
jk

2
2

ia 3 Ma(0) Mi(0) ln 2ia /2R
jk
i pjk
jk

(0) 2
Mj

(0)2 (0) 2 2
ln j k /2R .
+ Mk
+ Aij k
(2.57)
jk
3. Orbifold models
Following [1], we will consider models where the supersymmetry breaking arises
through non-vanishing expectation values of the auxiliary fields F S , F and M and we
write:
1 1/2
iS
,
F S = MK
(3.1)
sin e
S
S
3
1 1/2
F = MK
(3.2)
cos ei ,

3

with 2 = 1. In the case where one considers a single common modulus T (the overall
radius of compactification), (3.2) simply reads:
1 1/2
F T = MK
cos eiT .
T T
3
(3.3)
Note that the vev of M is related to the gravitino mass through (2.3) and that these auxiliary
fields automatically satisfy the constraint that the potential V in (2.4) vanishes at the ground
state.
In contrast with Ref. [1], we have already included the effect of the GreenSchwarz term
on the scalar potential at tree level and thus the auxiliary fields considered here include to
a large extent the corresponding GreenSchwarz corrections. Additional corrections (see
(2.25) and following text) will be discussed in Appendix A.
46
In the orbifold models that we consider, i.e., with gauge kinetic function (2.6) and Khler
potential given by (2.8) and (2.11), the tree level soft terms have simple expressions:
g 2 1/2
Ma(0) = a Mk
sin eiS ,
s s
2 3

M
(0)
Aij k = cos
t + t G2 ni + nj + nk + 1 ei
3

ks
iS
,
1/2 sin e
ks s

1
sin
M
(0)
Bij = 1/2 [ks + s ln ij ]eiS
3
3 ks s

+ cos
ni + nj + 1 t ln ij ei ,
2
Mi(0)

MM
2
2
1+3
=
ni cos .
9
(3.4)
The one loop contributions to (3.4) are decidedly more cumbersome and the complete expressions are given in Appendix B. Below we consider the phenomenological implications
of some specific cases in which the soft supersymmetry breaking terms are simpler. In all
of the following G2 = (2 (T ) + 1/(T + T )), which is proportional to the Eisenstein
function (2.24).
3.1. Moduli domination at the self-dual point: the case for leading anomaly-induced
contributions
The analysis of the preceding sections indicates a very specific situation which turns
out to give quasi-model independent contributions. It is the case of moduli mediated
supersymmetry breaking (F S = 0 or = 0) where the moduli fields lie at a self-dual point
(t = 1 or ei/6 , and thus G2 = 0). Assuming (2.6), we have vanishing tree level gaugino
masses and A-terms and from (2.22), (2.35) and (2.48) we obtain:
Ma = ga ()2
ba0
M,
3
1
Aij k = M(
i + j + k ),
3
1
Bij = M(
i + j ).
3
(3.5)
(3.6)
(3.7)
Further assuming that 2 = 1/3 (as in the case of a single modulus T , see (3.3)), = 0

and ni = 1 (as in the untwisted sector), we have vanishing tree level scalar masses
and
Mi2

jk
1
a
= M M i
i pai
i pj k .
9
,a
47
(3.8)
,j k
i /9 whereas for the

For the choice (A) of PV weights (see (2.38)), one finds Mi2 = M M
choice (B) (see (2.39)), one obtains Mi2 = 0. This shows very clearly how dependent the
scalar masses are on the regularization scheme forced upon us by the underlying theory.
Case (B) corresponds to what is usually referred to as the anomaly mediated scenario in
which scalar squared masses arise at two loops but are negative for sleptons, thus implying
an unacceptable phenomenology without further ad hoc assumptions. As discussed in
Section 2.4, if the -term (2.43) has a low energy origin through the vev of a standard
model singlet in a superpotential term, we would expect that in this scenario the B-term
would also be dominated by the anomaly mediated contribution (2.48). On the other hand
if it arises from Planck-scale physics, we do not expect the tree level contribution to vanish.
Let us note moreover that any departure from our hypothesis (i.e., a small value for F S
or a departure from the self-dual point in moduli space) generates tree level values for the
soft terms which tend to overcome the one loop anomaly-induced contributions considered
here, as we will see in the next subsection.
When (3.8) represents the leading contribution to scalar masses we can see from (2.31)
that the positivity of scalar squared mass depends on the size of the Yukawa couplings
(which themselves are a function of the value of tan and of the scale UV at which the
and p of (2.36).
soft terms are determined) and the values of the high-scale parameters pia
ij
In the simple case of scenario (A) (2.38) mentioned above, the sign of the scalar squared
mass depends on the sign of the anomalous dimensions. Keeping all third generation
Yukawa couplings and taking the running masses of the third generation fermions at the
Z-mass to be {mt , mb , m } = {165, 4.1, 1.78} GeV, we investigated the range in tan
for which the scalar masses are positive for a GUT-inspired boundary scale of UV =
2 1016 GeV as well as an intermediate scale of UV = 1 1011 GeV. As can be seen from
Table 1 the problem of tachyonic scalar masses for the matter fields is eased considerably
in this scenario relative to the previously studied anomaly mediated scenario represented
by case (B) (2.39).
and p are varied
Let us now investigate the pattern of soft terms as the parameters pia
ij
= p p with p a constant. If the scale at which the soft terms
by assuming that pia
ij
emerge is taken to be UV = 1 1011 GeV then the spectrum of soft terms as a function
of p is displayed in Fig. 1. In general gaugino masses are an order of magnitude smaller
than scalar masses, except for values of p approaching the limiting case of p = 1 (which
is equivalent to scenario (B) given by (2.39)) where scalar masses go through zero. It
is important to note that all of the possibilities of Fig. 1 represent anomaly mediated
scenarios. However, it is only the extreme case of p = 1 that was studied previously in the
particular model of Randall and Sundrum [2].
One final aspect of these soft term patterns relevant to low energy phenomenology is the
relative size of the scalar masses and A-terms. It is well known that for any generation of
48
Table 1
Regions of positive squared mass in the anomaly dominated scenario. Range of tan for which scalar
squared masses are positive at the boundary scale UV using the PV scenario (A). The value of tan
was varied over the range for which the third generation Yukawa couplings remain perturbative up
to the scale UV . This corresponds to the range 1.3 tan 44 for UV = 1 1011 GeV and
1.6 tan 48 for UV = 2 1016 GeV
Scalar mass
UV = 1 1011 GeV
UV = 2 1016 GeV
2
MQ
1.4 tan 45
1.7 tan 44
1.8 tan 48
1.9 tan 44
1.3 tan 42
1.6 tan 41
1.3 tan 46
1.6 tan 44
1.3 tan 39
1.6 tan 41
always negative
1.3 tan 33
3.6 tan 33
1.6 tan 37
2
MU
3
2
MD
3
2
ML
3
2
ME
3
2
MH
u
2
MH
d
Fig. 1. Soft term spectrum for anomaly dominated scenario. Soft term magnitudes for third generation
scalars, Higgs fields and gaugino masses are given as a function of universal PV parameter p as a
fraction of gravitino mass m3/2 . Scalar particles are generally much heavier than gauginos except
for the limiting case of p 1.
matter with non-negligible Yukawa couplings the relation

|Aij k |2 3 Mi2 + Mj2 + Mk2 ,
(3.9)
evaluated at the scale of supersymmetry breaking, is a good indicator that the minimum of
the scalar potential will yield proper electroweak symmetry breaking: when the bound is
not satisfied it is typical to develop minima away from the electroweak symmetry breaking
point in a direction in which one of the scalars masses of a field carrying electric or color
charge becomes negative. Since the anomaly mediated A-term and the scalar squared
mass both have a single loop factor of 1/16 2 the condition (3.9) is generally satisfied.
For example, in scenario (A) discussed above
2

Mi + Mj2 + Mk2 = m3/2 Aij k ,
(3.10)
49
and since Aij k is loop-suppressed relative to the gravitino mass, as seen from (3.6), this
scenario is phenomenologically acceptable. Scenario (B) with its vanishing scalar masses
at one loop is problematic, however, and the two loop contributions are relevant to the
determination of its viability.
3.2. The O-II models
This class of orbifold models discussed in [1] has matter fields in the untwisted sector
with weights (n1i , n2i , n3i ) = (1, 0, 0) (0, 1, 0) or (0, 0, 1). Then, taking for simplicity
the same common value T for the T fields, 11 one obtains from (B.1):

2 ()
g
2
GS
2ba0
a
tot
0
+
b
Ma = M cos (t + t )G2
+
a
16 2
2 3
3
3

2

g
sin
Cai
.
+ 1/2 1 + s 2 Ca
(3.11)
16
k
s s
The above form suggests a closer investigation of the relative magnitude of the
contributions to gaugino masses arising from the dilaton sector (proportional to sin ), the
moduli sector (proportional to cos ) and the anomaly-induced piece (independent of the
goldstino angle). As mentioned in the previous section, any tree level contribution (from
the dilaton sector) will likely dominate the gaugino mass, particularly when the Green
Schwarz coefficient is smaller than 3CE8 . The anomaly-induced piece is typically quite
small and will only be relevant in the case of moduli domination (sin = 0) with moduli
stabilized very near their self-dual points and/or very small GreenSchwarz coefficient.
This behavior is demonstrated for the case of the U (1)Y gaugino mass M1 in Fig. 2. We
and set gs2 = 1/2.
have taken k = ln(S + S)
In Fig. 3 we look at the relative sizes of the three gaugino mass terms for the case of
moduli domination ( = 0) and a mixed case ( = /3) for real moduli vacuum values
Re t at a boundary scale of UV = 2 1016 GeV. Note that there is always a particular
value of the moduli vev such that a nearly degenerate gaugino mass spectrum is recovered.
As cos 0 this value gets ever larger as we approach the limiting case in which the
gaugino masses are independent of the value of Re t. At the GUT scale where g22 g12
1/2 the difference in SU(2) and U(1) gaugino masses is given by

m3/2
Re
t
+
2
7
1
+
cos
M2 M1
(3.12)
3
sin
,
3
40 2
where we have used the fact that for Re t > 1, (t) /12. For = 0 Eq. (3.12) indicates
that M1 = M2 at Re t 6/ while for = /3 this occurs when Re t 3.7.
When = 0 (3.12) implies that |M2 | |M1 | (the gaugino masses in this regime are
negative) whenever

3 2 3
tan
+
sec
+
1
.
Re t
(3.13)
7
11 All the expressions given in this and the following sections will assume zero phases = = 0.
S
T
50
Fig. 2. U (1)Y gaugino mass in the BIM O-II model. Contributions to the value of M1 from F T
(dashed) and F S (dotted) as well as total M1 (solid) are given as a function of goldstino angle for
two values of GS and four T-modulus vevs: Re t = 0.5 (a), Re t = 0.9 (b), Re t = 5 (c), and
Re t = 20 (d). All values are given as a fraction of the gravitino mass m3/2 .
In the case where = 0 so that there is no tree level contribution to gaugino masses we see
from Fig. 3 that |M1 | |M2 | in nearly all of the Re t parameter space. This relationship
between the boundary values of the SU(2) and U(1) gaugino masses is crucial to the low
energy phenomenology of the model in that it determines whether the lightest neutralino
is predominantly bino-like, predominantly wino-like or a mixed state. Thus a lightest
neutralino with a significant wino content need not necessarily imply that supersymmetry
breaking is due to pure anomaly mediation. We will return to this point when we investigate
sample spectra in the next section.
The scale at which the soft masses emerge is particularly important: the largest
contributions to gaugino masses generically arise from the tree level piece and the piece
proportional to the GreenSchwarz coefficient GS . These terms cancel in (3.12), however,
51
Fig. 3. Relative gaugino masses vs. Re t in the BIM O-II model with UV = 21016 GeV. Relative
sizes of the three gaugino masses M1 (dashed), M2 (dotted) and M3 (solid) are displayed as a
function of Re t for two values of the goldstino angle and three representative values of GS . The
vertical dotted line at Re t = 1 indicates the moduli self-dual point where gaugino masses become
independent of GS . When sin = 0 this point represents the case of leading anomaly contributions
discussed in Section 3.1. All masses are relative to the gravitino mass m3/2 .
when the difference is evaluated at the GUT scale. Thus the location of the crossover point
is independent of the choice of GS in Fig. 3.
An immediate consequence of the above is that measurement of the properties of
the lightest neutralinos may reveal information on the nature of the scale of ultraviolet
physics. In particular the region of parameter space for which the lightest neutralino
is predominantly wino-like becomes increasingly small as the scale of supersymmetry
breaking is lowered. This is illustrated in Fig. 4 where we plot the ratio of gaugino masses
M1 /M2 for two different boundary scales: UV = 2 1016 GeV and UV = 1 1011 GeV,
for which g22 (7/5)g12 . As the gauge couplings run farther apart the shaded areas in which
M1 /M2 1 (and hence where a wino-like lightest neutralino is possible) grow steadily
smaller. When GS = 0 the ratio M1 /M2 diminishes as the goldstino angle increases
until M2 begins to approach its vanishing value and the ratio passes through a discontinuity
before increasing rapidly as . When the GreenSchwarz coefficient GS is increased
the location of this discontinuity, as indicated in Fig. 4 by a heavy arrow, moves to smaller
values of .
52
Fig. 4. Ratio M1 /M2 in BIM O-II model. Contours of the absolute value of the ratio of U(1) to SU(2)
gaugino masses are given for boundary scales of UV = 2 1016 GeV for panels (a) and (b), and
UV = 1 1011 GeV for panels (c) and (d). The shaded area is the region of parameter space for
which |M1 | |M2 |. The arrow indicates the direction of smallest ratios as the discontinuity M2 = 0
is approached. The contour M1 = M2 is given by the heavy solid line.
The trilinear A-terms for these orbifold models are given by (B.3):

M
cos
i
(t + t )G2
ia pia +
ilm plm
Atot
ij k =
3
3
3
a
lm

g2
sin
ks
+ 1/2 +
ilm s ln 2lm +
ia s ln 2ia + a ln 2ia
3
2
ks s
a
lm

2 a

4
lm
ga i pia
ks i plm
ln (t + t )|(t)|
a
+ cyclic(ij k),
lm
(3.14)
53

where pia = pia
and plm = plm
. For scenario (A), as defined by (2.38), this
expression is particularly simple

sin

ks a ga2 2
i
M
tot
2
+
+ cyclic(ij k). (3.15)
ln ia /R
Aij k =
i
3
2
3 k 1/2
3
a
s s
It is worth noting that, with such a scenario for the PV metrics, this pattern for A-terms

goes beyond the BIM O-II model. Any of the following conditions: (i) (ni + nj +
nk + 1) = 0 with identical vacuum values for all T-moduli (as in the BIM O-II model),
(ii) cos = 0 (dilaton domination) or (iii) G2 = 0 (moduli stabilized at self-dual point),
yields the A-terms given by (3.15) above.
By contrast, for scenario (B) defined by (2.39) the A-terms take the form

sin

ks ga2 a 2
i
M
tot
i ln ga 1
Aij k = 1 + cos (t + t )G2 + 1/2 +
3
2
3
3
ks s
a

2 a
ln (t + t )|(t)|4
(3.16)
ga i
ks ilm
+ cyclic(ij k).
a
lm
This scenario also allows for the recovery of an anomaly mediated-like result of
A-terms proportional to anomalous dimensions in the moduli dominated limit (sin = 0).
Expression (3.16) differs from the situation in Section 3.1 in that for moduli domination
this scenario can accommodate proper electroweak symmetry breaking provided the
moduli are stabilized away from their self-dual points: in particular, using the fact that
for Re t > 1, (t) /12 we have (t + t )G2 1 for t 6/ 2 leading to A 0
from (3.16).
The expressions for the bilinear B-terms are similar, but with added model dependence at
the tree level involving the origin of bilinear terms in the Khler potential or superpotential.
For the case of scenario (A) the general form given in (B.4) yields

g2
sin
M
tot
ia a ln 2ia
Bij = 1/2 ks s ln ij +
2
3 ks s
a

1
M
M
+
(3.17)
cos 1
i
t ln ij +
+ (i j ),
6
3 2
while for case (B) the corresponding expression is

g2
sin

M
tot
Bij = 1/2 ks s ln ij +
ia a ln ga2 1
2
3 ks s
a

ln (t + t )|(t)|4
ia ga2
ilm ks
a
lm

1

M
M
i +
cos 1
+
t ln ij 2(t + t )G2 i
3 2
6
+ (i j ).
(3.18)
54
Finally, the scalar masses in the BIM O-II model are found from Eq. (B.6) of
Appendix B. Under the assumptions of scenario (A) this reduces to
2
Mitot

1 sin a 2 1 j k
1
i + 1/2
= |M|
i ga
i (ks + ks )
9
2
ks s 3 3 a
jk

jk

2
sin2
a
2
+
1
i ln ia + 2
i ln j k
9
a
jk

2
sin2
1 4 a 2 1 jk
+
,
g ln ia
i (ks ks + 2ks s ) ln j k
ks s
4 a a i
3

jk
(3.19)
and for scenario (B) the scalar masses are given by
2
Mitot

2 a 1 jk
1 sin
1 + cos (t + t )G2
= |M|
ga i
i (ks + ks )
2
3 3 ks1/2
a
jk
s

jk

a
sin2
4
1 + i + ln (t + t )|(t)|
i 2
i
+
9
a
jk

jk
2

a
2
i ln ga + 2
i ln j k
2
jk

5
sin2 a ga4
1 jk
i
i (ks ks + 2ks s ) ln 2j k
ln ga2 +
+
ks s
4
3
3
a
jk

a ga4

1 jk
4
i
i ks ks
.
+ ln (t + t )|(t)|
(3.20)
+
4
3
a
jk
The pattern of soft supersymmetry breaking terms that arise in this orbifold model
with uniform modular weights ni = 1 and with the same Khler metric for the A and
the A , as in scenario (2.39), will produce a low energy phenomenology very similar to
that of the recently proposed gaugino mediation scenario [29] if the GreenSchwarz
coefficient is sufficiently large, the supersymmetry breaking is moduli dominated and
the moduli are stabilized at Re t 2. Such a situation gives rise to exactly vanishing
scalar masses and nearly vanishing A-terms and the gaugino masses in such a regime
are very nearly universal, as can be seen from the lower panels of Fig. 3. However, as
the GreenSchwarz coefficient is reduced the gaugino masses become negligible at the
point Re t 2, eventually coming into conflict with direct search results at LEP and the
Tevatron. Specific spectra for the O-II model will be presented with spectra for orbifold
models with large threshold corrections, to which we now turn.
55
3.3. The O-I models

Models of this type were proposed with the goal of obtaining coupling constant
unification at the string scale, as opposed to the extrapolated unification scale of GUT
2 1016 GeV which is typically a factor of twenty or so lower than the string scale. This is
achieved through large string threshold corrections and the requirement of both particular
sets of modular weights for the massless fields and relatively large values of Re t far from
the self-dual points. Other solutions to this discrepancy of scales have been proposed since
but because the O-I models have often been discussed in the literature we include them in
the present discussion.
To investigate the phenomenological consequences of such models
we will assume a
common vacuum value for all three moduli and take = 1/ 3 as before. We shall
investigate two scenarios: (A) the original O-I scenario of Brignole et al. [1] with
modular weights nQ = nD = 1, nU = 2, nL = nE = 3, nHd , nHu = 4 and (B) a
Z3 Z6 compactification studied by Love and Stadler [31] with modular weights nQ =
nD = 0, nU = 2, nL = 4, nE = 1, nHd = nHu = 1. In what follows let us assume
that the soft terms emerge at a scale for which logarithms such as ln( 2ia ) and ln( 2j k )
are negligible and assume PV case (A) for simplicity. In this approximation the general
expressions of Appendix B take a simplified form. The gaugino masses, given by

M
1
GS
ba0 + cos (t + t )G2
+ ba0
Cai (1 + ni )
Matot = ga2 ()
3
16 2
8 2
i

2

3 sin
g
Cai
1 + s 2 Ca
,
+
(3.21)
1/2
16
2k
s s
are displayed in Fig. 5 with the value = 0 (where the impact of the differing modular
weights is the greatest) for three models: the BIM O-II case of Section 3.2, the original
BIM O-I case and the Love and Stadler case. The boundary scale is taken to be UV =
2 1016 GeV. 12 It is clear from Fig. 5 that the modular weights of the matter fields play
a crucial role in determining the gaugino mass spectrum provided the GreenSchwarz
coefficient is sufficiently small. As this parameter is increased it will quickly come to
dominate the other terms in (3.21).
However, looking at the tree level expressions for the scalar masses (3.4) it is apparent
that when cos = 1 any field with a modular weight such that ni < 1 will have a
negative tree level scalar squared mass, as was noted in [1]. Thus, to accommodate
these large threshold models proper electroweak symmetry breaking (i.e., positive scalar
squared masses) will generally require a goldstino angle such that sin is large and
the tree level terms in (3.21) are dominant. Models with a viable low energy vacuum
will therefore be models for which the impact of the matter fields modular weights
on the gaugino spectrum is considerably muted. This is displayed in Fig. 6 where
12 Though these models are designed to allow for unification of gauge couplings at the string scale
str
5 1017 GeV, we will investigate the pattern of soft supersymmetry-breaking terms at the GUT scale to allow
for comparison with other models.
56
Fig. 5. Relative gaugino masses in the BIM O-II and BIM O-I models with = 0. Relative sizes
of the three gaugino masses M1 (dashed), M2 (dotted) and M3 (solid) are displayed as a function
of Re t for two values of the GreenSchwarz coefficient GS and UV = 2 1016 GeV. The top
panels (a) represent the BIM O-II model from Section 3.2, the middle panels (b) represent the BIM
O-I model and the bottom panels (c) represent the Love and Stadler case from [31]. All masses are
relative to the gravitino mass m3/2 .
Fig. 6. Relative gaugino masses in the BIM O-I models with = /3 and GS = 0. Relative sizes of
the three gaugino masses are displayed as a function of Re t for the BIM O-I model (solid) and the
Love and Stadler model (dashed) at UV = 2 1016 GeV. All masses are relative to the gravitino
mass m3/2 .
57
gaugino masses in the BIM O-I model and the Love and Stadler model are displayed
for = /3 and GS = 0. We see that in these realistic cases the differences in
gaugino mass spectra between these models are small, making them hard to distinguish
experimentally.
The trilinear A-terms for scenario (A) are

M
cos
sin ks
(t
+
t
)G
=
+
(n
+
n
+
n
+
3)
Atot
i
2 i
j
k
ij k
3
3
3 k 1/2
s s
+ cyclic(ij k),
(3.22)
and the scalar masses are determined from

jk
tot2 |M|2
=
i (ni + nj + nk + 3) + ni cos2
Mi
(1 + i ) + cos (t + t )G2
9
jk
3 sin a 2 1 j k
+
(3.23)
i ga
i (ks + ks ) .
1/2
2
k
a
s s
jk
With these expressions we are now in a position to compare the typical spectra of these O-I
large threshold models with the models of Sections 3.1 and 3.2.
In Tables 2 and 3 we give some representative sample spectra for PauliVillars
scenario (A) defined by (2.38) and tan = 3 and tan = 10, respectively. The spectra
for scenario (B) are very similar and these values vary only minimally when UV is
varied. To obtain these spectra at the electroweak scale the renormalization group equations
(RGEs) were run from the boundary scale to the electroweak scale. All gauge and Yukawa
couplings as well as gaugino masses and A-terms were run with one loop RGEs while
scalar masses were run at two loops to capture the possible effects of heavy scalars on the
evolution of third generation squarks and sleptons. We chose to keep only the top, bottom
and tau Yukawas and the corresponding A-terms. The gravitino mass has been scaled in
each case to obtain a Higgs mass of 114 GeV, which can be considered either as a limiting
case or as an experimental requirement, depending on what happens next at LEP.
At the electroweak scale the one loop corrected effective potential V1-loop = Vtree +
JVrad is computed and the effective -term is calculated
2 =
(m2Hd + m2Hd ) (m2Hu + m2Hu ) tan2
1
MZ2 .
2
(3.24)
1
In Eq. (3.24) the quantities mHu and mHd are the second derivatives of the radiative
corrections JVrad with respect to the up-type and down-type Higgs scalar fields,
respectively. These corrections include the effects of all third generation particles. If the
right-hand side of Eq. (3.24) is positive then there exists some initial value of at the
supersymmetry breaking scale which results in correct electroweak symmetry breaking
with MZ = 91.187 GeV. 13
tan2
13 Note that for these tables we do not try to specify the origin of this -term (nor its associated B-term) and
merely leave them as free parameters in the theory ultimately determined by the requirement that the Z-boson
receive the correct mass.
58
Table 2
Sample spectra (in GeV) for typical models of Sections 3.1, 3.2 and 3.3. All cases are for PV
% and W
3 % represent the content of the
scenario (A), tan = 3 and UV = 2 1016 GeV (B
lightest neutralino in per cents). The first O-II case considered, while clearly ruled out experimentally,
is presented as an illustrative example
Model Anomaly (3.1)
GS
Re t
0
N/A
1
m3/2
mN
1
mN
2
%
B
3 %
W
m
1.9 104
51.81
168
0.01
99.7
51.83
mg
mh
mA
mtR
mtL
mb
R
mb
L
mR
mL
Atop
Abot
Atau
623
114
2237
860
1842
1805
1810
1191
1193
391
973
220
1617
BIM O-II (3.2)

0
0
6/
0
0
5
1.9 104 1.6 104

0.32
152.53
3.7
462
80.9
0.001
19.1
99.7
3.1
152.55
BIM O-I (3.3) L&S (3.3) [31]
0
0
20
0
90
6/
/3
90
6/
/3
90
16
/3
90
14.5
4500
248.97
759
0.001
99.7
249.00
1600
332
615
99.9
0.001
615
450
313
599
99.9
0.001
599
150
287
557
99.9
0.001
557
150
297
581
99.9
0.001
581
2245
114
1357
1597
1820
1769
1908
514
515
1423
1827
305
1281
2156
114
1447
1521
1804
1782
1883
329
330
1696
2819
466
1341
2164
114
1387
1610
1866
1847
1945
302
303
1541
2200
184
1302
2106
114
1810
1373
1709
1701
1881
198
281
560
405
7417
1577
2128
114
1568
1532
1793
1773
1871
290
301
1607
4650
2734
1297
3.6
114
2217
796
1810
1765
1770
1180
1182
71
463
273
1592
1468
114
1992
1142
1818
1802
1824
1076
1078
815
999
376
1501
Note that the gravitino mass varies greatly over the models considered in Tables 2 and 3.
For the anomaly case (which is equivalent to the BIM O-II model with sin = 0 and
Re t = 1) there is a large hierarchy between scalars and gauginos, as noted in Section 3.1,
which necessitates a large value of the gravitino mass to yield neutralinos with masses near
the current LEP limits. Having normalized our scales to yield Higgs masses of 114 GeV
we find chargino masses (for PV scenario (A) and thus p = 0 in Fig. 1) below the recently
reported bounds of m 86.1 GeV for the case of a chargino which is nearly degenerate
with a wino-like lightest neutralino [32]. As the PV scenario assumed is modified, however,
this relation between the chargino mass and Higgs mass varies. In particular as the value of
p approaches larger, positive values the gauginos steadily become heavier for a fixed Higgs
mass, eventually satisfying the experimental constraints. For the large threshold models,
by contrast, the large values of Re t necessary to ensure gauge coupling unification at the
59
Table 3
Sample spectra (in GeV) for typical models of Sections 3.1, 3.2 and 3.3. The same as in Table 2 but
for tan = 10. Neither of the large threshold models are viable at this value of tan
Model Anomaly (3.1)
BIM O-II (3.2)
GS
Re t
0
N/A
1
0
0
6/
m3/2
mN
1
mN
2
%
B
3 %
W
m
8000
20.20
70
0.08
98.0
20.21
8000
0.17
3.11
79.2
20.8
2.5
mg
mh
mA
mtR
mtL
mb
R
mb
L
mR
mL
Atop
Abot
Atau
280
114
797
449
797
739
763
493
503
190
398
83
578
BIM O-I (3.3) L&S (3.3) [31]

/3
90
6/
/3
90
16
/3
90
14.5
6500
1800
1200 200
62.11
98.72
139
129
187
301
260
244
0.002 1.9 107 99.3 99.1
97.8
97.4
0.001 0.002
62.14
98.75
260
244
N/A
N/A
0
0
5
0
0
20
0
90
6/
1.85 644
114 114
790 689
427 527
782 774
720 727
744 753
490 431
499 440
47 336
187 403
108 153
565 529
978
114
485
658
806
737
799
206
211
596
858
130
495
1020
114
560
663
849
792
838
147
156
796
1223
190
559
979
114
497
667
819
771
812
121
132
668
893
100
499
string scale make the gauginos typically heavier than the gravitino at the boundary scale
UV , due to the large value of (t + t )G2 , and have a smaller degree of hierarchy between
gauginos and scalars.
The O-II models can interpolate between these two extremes. When = 0 and GS = 0
the pattern of physical masses shows the anomaly mediated feature of a wino-like LSP.
As the value of Re t increases from Re t = 1 (the pure anomaly mediated case) it first
passes through the experimentally excluded values where Re t 6/ and the gaugino
masses are nearly zero. Thereafter the hierarchy between gauginos and scalars steadily
decreases until the spectra of masses is very similar to that of the more typical supergravity
spectra to the right of Table 2. However, as mentioned at the end of the previous section
the feature of a wino-like LSP persists. Once = 0 and/or GS = 0 the pattern of soft terms
immediately becomes relatively insensitive to the value of Re t and the LSP once again
becomes predominantly bino-like.
60
The models with large threshold corrections also tend to have very light staus. In fact, as
the value of tan increases the stau mass mR eventually becomes negative. The limiting
value of tan for which these models are phenomenologically viable depends slightly on
the value of GS : for = /3 the model of Love and Stadler requires tan < 9.1 when
GS = 90 and tan < 4.8 when GS = 0, while the original BIM O-I model requires
tan < 3.1 when GS = 90 and is not allowed at all for GS = 0. This is reflected in
the empty columns in Table 3. This problem is slightly ameliorated when the goldstino
angle is increased. For = 2/5, for example, the model of Love and Stadler requires
tan < 12.7 when GS = 90 and tan < 9.6 when GS = 0, while the original BIM O-I
model requires tan < 4.9 when GS = 90 and tan < 2.1 when GS = 0.
The pattern of masses exhibited in Tables 2 and 3 suggests that the hierarchy between
gauginos and scalars in any potential observation of supersymmetry will be a key to
understanding the nature of the underlying physics giving rise to supersymmetry breaking.
The observation of a lightest neutralino with significant wino content will not be enough
to distinguish between the pure anomaly mediated cases and the BIM O-II type models
but will indicate that supersymmetry breaking is moduli dominated within this class of
models. The presence of a large hierarchy between scalars and gauginos and large mixing
in the stop sector will point towards moduli stabilized at or near their self-dual values,
while the absence of such effects would suggest the moduli are stabilized far from these
values.
3.4. The BGW model
In this section we give the soft supersymmetry breaking parameters for the model
of Ref. [23], with an explicit mechanism for supersymmetry breaking through gaugino
condensation in a hidden sector, and dilaton stabilization by nonperturbative string
effects. An effective Lagrangian below the scale c of hidden gaugino condensation is
constructed [19,20] by replacing the linear multiplet L in (2.16) by a vector multiplet V
. The
whose components includes those of L and of a chiral multiplet U and its conjugate U

superfield U satisfies the same equations as the composite chiral superfield U = W W
constructed from the YangMills superfield strength, and is interpreted as the lightest
chiral superfield bound state of the effective theory below the condensation scale c =
|u|1/3 , where u = U | is the scalar component of the chiral supermultiplet U . An effective
potential for the gaugino condensates U , as well as matter condensates that are present
if there is elementary matter charged under the confined gauge group, is constructed by
field theory anomaly matching. Once the massive (m c ) composite degrees of freedom
are integrated out, this generates a potential for the dilaton and moduli.
The gaugino masses were given in [13]. In the notation adopted here they take the form 14
Ma =
ga2 () S
F + Ma(1) ,
2
(3.25)
14 As in the above subsections we set p = 0 in (2.17); modifications that occur when p = 0 are discussed in
i
i
the following subsection.
61
where Ma(1) is given in (2.22). The A-terms, squared soft scalar masses and B-terms are
given by (2.35), (2.56), (2.57) and (2.47), respectively, with (see Appendix A)

gs2 0
1 0
1 1
S
M = b+ u = 3m3/2,
F = KS S 1 + b+ u,
2
4
3
1
KS = gs2 ,
(3.26)
2
0
where gs2 is defined in (2.23) and b+
is the beta function coefficient, Eq. (2.15), of the
15
condensing gauge group G+ . The model of Ref. [23] is explicitly modular invariant,
so the moduli are stabilized at their self-dual points with F = 0, and supersymmetry
breaking is dilaton dominated. Then [23]
gs2 S
1
F ,
Mi(0) = |M| = |m3/2|.
2
3
Vanishing of the vacuum energy (2.4) now requires
A(0)
ij k =
0 )2
2 1
(2b+
KS S F S = M 2 ,
KS1
=
,
S
0 2
3
3 1 + 13 gs2 b+
S
0
F
2b+

M = 31 + 1 g 2 b0 .
3 s +
(3.27)
(3.28)
0 1 the tree level A-terms and gaugino masses are suppressed relative to the the
If b+
gravitino mass, whereas the scalar masses and B-terms, Bij(0) m3/2 , are not. Therefore
one loop corrections can be neglected for the latter, but may be important for the former. It
is clear from (2.22) and (2.35) that the dominant one loop corrections in this case are just
the anomaly mediated terms found in [2,3]:

0
b0

b+
0
= ga 2 m3/2
b
Ma Ma(0) + ga 2 a M
a ,
0
3
1 + 13 gs2 b+

0
gs2 b+
1
(0)
+ i + j + k .
Aij k Aij k M(i + j + k ) = m3/2
(3.29)
0
3
1 + 13 gs2 b+
This model was analyzed in detail in [30]. Over most of the allowed parameter space, 0.1
0
ba0 , the tree contributions dominate. However there is a small region of parameter
b+
0 that the gaugino masses and A-terms are similar
space with a sufficiently small value of b+
to those in an anomaly mediated scenario [2,3,16].
Using the expressions in Appendix B, together with (3.26) and (3.28) the pattern of soft
supersymmetry breaking terms can be obtained as a function of the condensing group beta
0 and the modular weights of the fields with Re t = 1 or Re t =
function coefficient b+
i/6
and sin = 1. The condensation scale in these models is typically of the order of
e
1 1014 GeV and we take this to be the boundary condition scale UV in what follows. In
0
as a fraction of the gravitino mass.
Fig. 7 the gaugino masses are displayed as a function b+
15 If there are several condensing gauge groups, the one with the largest value of b0 dominates supersymmetry
a
breaking.
62
Fig. 7. Gaugino masses in the BGW model. Gaugino masses M1 (dashed), M2 (dotted) and M3
(solid) are given at a scale UV = 1 1014 GeV as a function of the condensing group beta function
0 . The vertical dotted line in the left panel is the case of E condensation in the hidden
coefficient b+
6
sector with 9 27s of hidden sector matter studied in [30]. The right panel focuses on the region where
the three masses are approximately unified.
Fig. 8. Spectrum of soft supersymmetry breaking terms in BGW model. All values are given relative
to the gravitino mass m3/2 at a scale UV = 1 1014 GeV as a function of the condensing group
0.
beta function coefficient b+
63
In [30] it was shown that for weak coupling at the string scale (gs2 1/2) a reasonable scale
of supersymmetry breaking (i.e., gravitino masses less than 10 TeV) generally requires
0 0.085. The region with gravitino mass larger than 10 TeV is shaded in Fig. 7. Also
b+
indicated in Fig. 7 is a benchmark scenario consisting of an E6 gaugino condensate in
the hidden sector together with 9 27s of matter and having a beta function coefficient of
0
= 0.038.
b+
The spectrum of gaugino masses will typically be similar to that of the anomalymediated cases with M1 M2 and a lightest neutralino with substantial wino-like content
0 0.19. The location of the approximate unification of gaugino masses near
provided b+
0
is expanded in the right panel of Fig. 7.
this value of b+
In Fig. 8 we plot the relative sizes of all third generation scalar masses and A-terms,
Higgs masses and gaugino masses as a fraction of the gravitino mass for tan = 3,
assuming ni = 1 for all fields. As was the case in Sections 3.13.3, the gauginos are
typically an order of magnitude smaller than scalars (note the change in vertical scale in
Fig. 8). Despite this hierarchy, this model was shown in [30] to give rise to acceptable
low energy phenomenology provided tan was in the low to moderate range. Fig. 8
displays an important feature of the always-present one loop contributions arising from
the conformal anomaly: when tree level scalar masses are present and universal the nonuniversality arising from the anomaly pieces is negligible (here averaging less than a 1%
correction). However, the corrections to the gaugino masses may significantly alter the
gaugino spectrum provided the tree level contributions are absent or suppressed, as in the
BGW model considered here. Neglecting these one loop anomaly-induced contributions to
soft terms is an approximation whose validity needs to be assessed on a model-by-model
basis.
3.5. Matter couplings to the GreenSchwarz term
So far we have assumed the GreenSchwarz function VGS depends only on the moduli,
that is, we set pi = 0 in (2.17). Only the moduli couplings in this term are known from
string loop calculations [7,8] and they are proportional to the Khler potential for the
moduli. It is possible that the GS function is proportional to the full Klher potential,
in which case pi = p = GS /24 2 , or that it is proportional to the untwisted Khler
potential, i.e., to the logarithm of the determinant of the metric in the six-dimensional
compact space. In this last case we would have pi = p for untwisted matter and pi = 0 for
twisted matter. The presence of these terms modifies the soft parameters if F S = 0.
One effect of pi = 0 is a modification [33] of the effective matter Khler potential
(2.9):

1
i 1 + gs pi i .
(3.30)
2
The potential can still be written in the form given in (A.8) of the appendix with the
= K + 1 gs (VGS ) . However the effective metric is not
replacement KN N K
NN
NN
NN
2
Khler in this formulation. In addition F N does not take the usual form (2.1): F N =
), which reduces to (2.1) when W is holomorphic. This is not
N N (eK W W
eK/2 W 1 K
N
64
the case in the linear multiplet formulation for the dilaton that we are using here because
of the way the GS term enters in the dilaton potential, as described in Appendix A. For
these reasons Eqs. (2.27), (2.54) and (2.45) do not generally apply if F S = 0; the pi -terms
in these parameters depend on the specifics of the model for generating a potential for the
dilaton. The PV metrics (2.33) are similarly modified:

1
1
1
i 1 + gs pi i ,
(3.31)
i 1 + gs pi
i ,
2
2
as are the soft parameters in the PV potential. These give additional one loop contributions,
which can be important for gaugino masses which have no tree level contribution from
pi = 0.
Here we give the results only for the explicit dilaton dominated supersymmetry breaking
model of the previous subsection:
(0)
JAij k JAij k =
0)
pi (3 + gs2 b+

m3/2 + (i j ) + (j k),
0
1 + 12 gs pi
2b+
0
)
pi (3 + gs2 b+

m3/2 + (i j ),
1
0
2b+ 1 + 2 gs pi
0)
g 2 () Cai pi (3 + gs2 b+
JMa JMa(1) =

m3/2.
0 1 + 1 g2 p
8 2
2b+
2 s i
i
JBij JBij(0)
(3.32)
Note that in this special case the above results can in fact be obtained from the general
formulae (2.22), (2.27) and (2.45):
pi
(0)
JAij k = F S KS S
+ (i j ) + (j k),
1 + 12 gs pi
pi
JBij(0) = F S KS S
+ (i j ),
1 + 12 gs pi
pi C i
g 2 () S
a
JMa(1) =
(3.33)
F
K
,
S
S
1 2
8 2
1
+
g
2 s pi
i
since it follows from (3.30) that
F n n ln i F n n ln i F S KS S
pi
1 + 12 gs pi
(3.34)
where we used the relation

5
gs
=2
= KS S ,
(3.35)
s
x
given in Appendix A. However (2.54) does not apply even in this case; the tree level scalar
masses in this model have been given in [23]:

0
(0)

M = 1 3pi 2b+ m3/2 .
(3.36)
i

0
2
b+ 2 + pi gs
The results (3.32) then follow from (3.28). We see a considerable enhancement of all these
0 . Under the assumption that
parameters if pi = p b+
GS takes its maximum value of
65
90, the only viable scenario with some pi = p found in [30] is for pHu,d = 0 and pi = p
for all three generations of squarks and sleptons.
4. Conclusion
To conclude, let us first stress that even though we have been studying specific classes of
superstring models, the types of spectra that we obtained and discussed appear to be quite
generic. For example, scenarios from models with extra dimensions tend to give spectra
which can be related to one or another type considered here, whether it is the model of
Randall and Sundrum [2], or models of gaugino mediation [29].
In particular, soft terms that are proportional to beta function coefficients and anomalous
dimensions can be realized in a variety of ways in string-derived supergravity. The case that
is generally referred to as anomaly mediation is just one limiting value in a continuum of
such models. The importance of these anomaly-induced terms depends on the absence or
suppression of tree level contributions to the soft supersymmetry breaking parameters and
on the assumptions made regarding the underlying theory when regulating the effective
supergravity theory.
Once supersymmetry is discovered, the central issue will be to unravel the mechanism
of supersymmetry breaking. The search strategy will be of the most value if it is based on
large classes of different models, not just on a single minimal model. The models studied
above tend to show that a possible strategy could be based on three steps:
(i) identifying gaugino masses (the least model dependent aspect of these theories) and
the nature of the LSP,
(ii) identifying where (approximately) the bulk of the scalar masses lie and whether there
is an order of magnitude between gaugino and scalar masses,
(iii) then using the details of the scalar masses, in particular the mixing in the stop sector
and the degree of non-universality, to disentangle the possible scenarios.
Observation of non-universal supersymmetric parameters obeying the relations described in Sections 3.13.4 will likely shed light on the scale of supersymmetry breaking,
the nature of the fields responsible for this breaking and the origin of the -term, if not the
properties of the underlying superstring theory itself.
Acknowledgements
We thank Joel Giedt for discussions. P.B. thanks the Theory Group of LBNL for its
generous hospitality and the participants of the GDR Supersymmetry working group on
non-universalities, especially Laurent Duflot and Jean-Franois Grivaz, for discussions.
B.N. would like to thank the Laboratoire de Physique Thorique at the Universit Paris-Sud
where part of this work was completed. This work was supported in part by the Director,
Office of Science, Office of Basic Energy Services, of the U.S. Department of Energy
under Contract DE-AC03-76SF00098 and in part by the National Science Foundation
under grants PHY-95-14797 and INT-9910077.
66
Appendix A. The linear multiplet formalism for the dilaton

In this paper we have presented the soft supersymmetry breaking parameters in terms
of the various auxiliary fields of supergravity. In order to adhere as closely as possible
to the notation of [1], we used expressions of the form obtained in the standard chiral
formulation of supergravity. In the context of string theory, the dilaton 5 appears as
the scalar component of a linear multiplet L. The chiral multiplet formulation can be
recovered by a duality transformation, at least at the classical level. However the linear
multiplet formulation provides a simpler implementation of the GreenSchwarz anomaly
cancellation mechanism and a better framework for constructing an effective Lagrangian
for gaugino condensation. The effective theory of [23] made explicit use of the linear
multiplet formalism. In this appendix we show the correspondence between various terms
in the component Lagrangian of the linear formalism and of the expressions given in the
text. We also show how explicit cancellations among the light loop (anomaly) contribution,
the GS term and the string threshold corrections result in the final expression (2.22) for
the one loop gaugino mass. These cancellations are most readily displayed in the linear
multiplet formalism. Finally, we will display corrections to the soft parameters in the scalar
potential that are present if the dilaton and moduli sectors both contribute substantially to
supersymmetry breaking.
In the presence of a (nonperturbatively induced) potential for the dilaton, the tree level
scalar Lagrangian takes the form (dropping gauge charged matter)
Lscalar =
m t m t
(t + t )2
k (5)
5
m 5 m 5 m a m a V ,
45
k (5)
(A.1)
where the axion a is related to the two-form bmn of the linear multiplet by a duality
transformation:
1 mnpq
25
M
n bpq = m a.
2
k (5)
(A.2)
The potential V can be written in the form

V=
(t
1

+ 5 F 2 1 M M,
F F
+t )
k (5)
3
k (5) i
F=
f 5, t , z ,
45
(A.3)
where f (5, t , zi ) is a complex but nonholomorphic function of the scalar fields. For
example in the model of [23],

f 5, t , zi =
(A.4)
(1 + 5ba )u a (1 + 5b+ )u + ,
a
where u a (5, t , zi ) is the value of the gaugino condensate for a hidden gauge group Ga

with beta function coefficient ba = (Ca 13 i Cai )/8 2 = 2ba0 /3; the function (A.4) is
dominated by the condensate u + with the largest beta function coefficient b+ .
67
To cast this result in a form resembling the standard chiral formulation we introduce the
variable x(5) = 2g 2 (Ms ), which is twice the inverse squared gauge coupling (2.23). It is
related to the dilaton Khler potential k by the differential equation [5]
k (5) = 5x (5),
5 =
5
x,
k (5)
(A.5)
giving
k(x)
2 k(x)
5
5
5
= k (5)
= 5,
=
,
=
x
x
x k (5)
x 2
k (5)
5
1 2 k(x)
m 5 m 5 = m x m x =
m x m x.
45
4k (5)
4 x 2
(A.6)
Then setting
x = s + s ,
a = Im s,
(A.7)
(A.1) and (A.3) take the standard form (including gauge-charged chiral matter)

N + 1 M M,
Lscalar =
KN N m zN m z N + F N F
3
N

K = k(s + s ) + K t +
i |zi |2 ,
(A.8)
provided we identify F = F S and KS S = 5/k (5). When the fermion part of the Lagrangian
is included, one obtains for the gaugino masses
Ma(0) =
ga2
F,
2
(A.9)
in agreement with (2.13) with fa = s and F = F S .

The replacements (A.7) amount to a duality transformation to the chiral formulation for
the dilaton. When the GS term is included, after a two-form/scalar duality transformation,
Eqs. (A.1)(A.8) are modified by the replacements
2
2

b m Im t
,
t + t
(1 + b5) t + t
,
m a m a +
2 Re t
b = GS /24 2.
(A.10)
We may make a full superfield duality transformation by the additional replacements

x = s + s = s + s + b
k(s + s ) k s + s bK t , (A.11)
ln t + t ,
where s is the complex scalar component (Im s = a) of the dilaton chiral superfield. This
introduces mixing of the moduli [and of matter fields if pi = 0 in (2.17)] with the dilaton
in the Khler metric [1]. Working in the linear multiplet formalism for the dilaton, there is
no mixing of the dilaton with chiral fields; 16 in this case (A.1) and (A.3) are modified only
16 See, for example, the discussion of Eq. (4.20) in [9].
68
by (A.10). With this modification (A.3) is completely general; it includes the effects of
the GS term on the potential for 5 and t in the presence of a source of supersymmetry
breaking such as gaugino condensation. In fact the GS term coupling to the confined
hidden gauge sector, as in the model of Section 3.4, must be included to make the effective
supersymmetry breaking tree Lagrangian perturbatively modular invariant.
However it is inconsistent to include the GS term coupling to the unconfined
(observable) gauge sector without the corrections from the observable sector loops. Here
we illustrate the modular anomaly cancellation among the contributions to the gaugino
masses. In orbifold models the light loop contribution (2.14) takes the form

ga2 () 2ba0
5
(1)
i
Ca
Ca F
Ma an =
M+
2
3
8 2
i

1 i
2
0
+
(A.12)
F
Ca (1 + 3ni ) .
b
3(t + t ) a 8 2
The contribution of the GS term (2.16) is

GS
2
g 2 ()
Ma(1) GS = a
F
2
3(t + t ) 16 2
(A.13)
and the string threshold corrections (2.17) give a contribution

ga2 () 4 GS
1 i
(1)
0
Ma th =
F t
+ ba
Ca (1 + 3ni ) .
2
3
16 2
8 2
(A.14)
These combine to give the total contribution (2.22) with the substitutions
F F S,
5 gs2 /2 = KS ,
with the moduli t appearing only through the modular invariant expressions

1

F t + t
+ 2 t .
In the linear multiplet formulation for the dilaton, the tree level scalar potential takes the
form

F N F
N 1 M M,
Vtree =
K
NN
3
N

N N eK ww ,
M = 3eK/2w,
(A.15)
F N = w1 eK/2K
N
is defined in (2.25), and K
= K = 5/k (5). (A.15)
where the effective metric K
NN
SS
SS
reduces to the standard form if w is holomorphic. If a duality transformation to the
chiral formulation for the dilaton is always possible [19] in the effective theory below
the supersymmetry breaking scale, we must have

w = w s , t , zi ,
s=
x
+ ia,
2
1
s = s + VGS .
2
(A.16)
69
For example, in the BGW model we have 17

b+
v = eK/2 u,
w = W t , zi + v s , t ,
4
2(bb+ )/b+
K/2 s /b+
t
,
u = ce e
(A.17)
where c is a constant. In this case we have

1
S S [w + K w],
N w = wS N VGS ,
wS = 0,
F S = eK/2K
S
S
2

N N w + K w + 1 ( VGS )s ln w .
F N = eK/2K
N
N
2 N
(A.18)
Inserting these expressions in the potential (A.15) we obtain the following expressions for
the soft supersymmetry-breaking terms at tree level:

b
F s ln w + h.c. ,
+ t )

tree
(0)

b
F s ln w + h.c. ,
Bij superpotential = Bij superpotential
2(t + t )
tree
(0)
b
F s ln w,
Bij Kahler potential = Bij Kahler potential
2(t + t )
(0)
Atree
ij k = Aij k
2(t
(A.19)
where the expressions with index 0 are the tree level expressions given in the text with
W (Z N ) w(Z N , VGS ) and

b
t t w t + Kt w
F = eK/2 K
(A.20)
ln
w
.
s
2(t + t )
. If pi = 0
The scalar masses depend on the curvature of the effective scalar metric K
NN
they are complicated expressions in the general case; their values for the BGW model are
given in Section 3.5. If pi = 0, they reduce to the result given in Section 2.4, with the
substitutions W w and (A.20).
If pi = 0 the expressions in Section 2 receive no corrections if supersymmetry breaking
is dilaton mediated, F = 0. If there is no dilaton superpotential, ws = 0, the only
correction is the rescaling F (1 +b5)F . If a dilaton superpotential v is generated by
a single dominant gaugino condensate (and the associated matter condensates), the dilaton
dependence of v in (A.17) follows quite generally from anomaly matching, giving
s ln w = v/b+ w.
(A.21)
Since b+ b, the corrections in (A.19) can be significant if |v/w|, cos and 1/t are
all order one. The moduli dependence of v in (A.17) follows from perturbative modular
invariance. 18 To the extent that modular invariant condensation dominates supersymmetry
17 The full potential for the BGW model is given in (15) of [34]. The full expression for the field dependence of
the condensate u with zi = 0 is given in the second reference of [23], and reduces to (A.17) with the identification
of the axion as a = b+ in the notation of that paper.

18 Modular invariance could be broken in v if corrections to k(5) from string nonperturbative effects are moduli
dependent [35]. We have ignored this possibility throughout.
70
breaking, one gets essentially the BGW model with negligible contributions from F . On
the other hand if W is dominant, the corrections found in (A.19) again become negligible.
They are significant only if there are two comparable sources of supersymmetry breaking.
Even in this case they are unimportant in the large T limit if s ln w is not too large. Note
that the correction to the A-term does not vanish at the self-dual points for the moduli, so in
this case we would not get an anomaly mediated scenario at these points when F S = 0.
In the chiral formulation [1], there is mixing between dilaton and moduli F-terms. In that
language, the corrections to the results of Section 2, aside from the rescaling of F , arise
K + h.c. in the potential.
from terms proportional to F S F
ST
Appendix B. Soft supersymmetry breaking terms in orbifold models

In this appendix we collect the complete expressions (tree plus one loop correction)
for the soft supersymmetry breaking terms in orbifold models defined by (2.6), (2.8)
and (2.11) with supersymmetry breaking vevs parameterized by (3.1) and (3.2). We
neglect corrections proportional to GS /48 2 in the scalar potential that were discussed
in Appendix A.
The gaugino mass is determined from (2.37) and (2.22):

ga2 () 2
tot
cos
t + t G2
Ma = M
3
2 3

i
GS
1 i
0
+ ba
Ca 1 + 3ni e T
16 2
8 2
i

gs2
2ba0 sin
i
iS
Ca
.
+ + 1/2 1 +
Ca e
(B.1)
16 2
3
ks s
i
The trilinear A-terms are obtained from (2.35). The expression is simplified by utilizing
(2.31) to obtain the identities
F S s ia = ia Ma(0),
s ilm = ks ilm ,
(B.2)
where the last relation is true if i = i (s). This yields A-terms of the form:

1
i
M
tot
Aij k = + cos
t + t G2
ni + nj + nk + 1
3
3
3

2 a
lm
i plm ni + nl + nm + 1 ln lm
i pia
lm

4 lm
t + t G2
ln t + t t
i plm

ni + nl + nm + 1
lm
ei
71

2 ga2 2

ks a
+
ln
ln
ilm s ln 2lm
+
s
ia
ia
i
1/2
3
2
ks s
a
lm

4
2
a
lm
i
S

ln t + t t
g p ks
p
e
sin
a i
ia
lm
lm
+ cyclic(ij k).
(B.3)
The bilinear B-terms have a similar form

1
1
M
M
i + cos
ni + nj + 1 t ln ij
Bijtot =
3 2
2
3

lm
t + t G2
i pia +
i plm
lm

4 lm
t + t G2
ln t + t t
i plm

ni + nl + nm + 1

ilm

t +t
lm

sin
lm
G2

2 i
ni + nl + nm + 1 ln lm e

2

gas 2
ln
+
ln
ilm s ln 2lm
s
ia
ia
1/2
2
ks s
a
lm

4 a 2 lm

ln t + t t
i ga pia
i ks plm
ia
(ks + s ln ij ) eis
lm
+ (i j ).
(B.4)
The scalar masses arise from (2.56) and (2.57). Some degree of consolidation can be
obtained by employing the relation (B.2) as well as
S 2

F s s ln g 2 = M (0)2 ,
a
to allow the following identifications:

S s
F
ia Ma(0) ln 2ia
=
S s
F
=

!

S Ma(0) s ln 2ia 2 Ma(0) 2 ln 2ia ,
ia F
2
ilm A(0)
lm
ilm ln
lm

!

(0) S
S A(0) ln 2lm ks s F S 2 ln 2lm ,
ilm Ailm F
s ln 2lm + ks F
ilm

lm
(B.5)
72

S S
F F s s
=
ia ln
2
ia
ia
a
S S
F F s s
!

2 2

S s ln 2ia + h.c. ,
2 Ma(0) ln 2ia + F S s s ln 2ia Ma(0) F

ilm ln 2lm
lm
2 lm

!

= F S
i s s ln 2lm + (ks s + ks ks ) ln 2lm + ks s ln 2lm + h.c. .
lm
With these, the complete expression for the tree level plus one loop scalar masses is given
by

tot 2
Mi
4
1
1
(1 + i ) +
ln t + t t
9
9

jk
a
i pia 2
i pj k
= |M|2
jk
1
2 jk 2
ia ln 2ia +
i ln j k
9 a
9
jk

sin
1 a 2
1 j k iS
iS
+ 1/2
i ga cos S
i k s e
+ ks e
2
ks s 3 3 a
jk

cos
j
k
i ni + nj + nk + 1 cos
t + t G2
+
3 3
jk

jk
1
1
+ cos2
n 2
2
a p +
i pjk
3 i 3 a i ia
jk

1 a 2 2 1 j k
nj + nk 2 ln 2j k
i
ni ln ia +
i
3 a
3
jk

4
1 a 2
ln t + t t
p
ni
3 a i ia
1 j k

nj + nk 2
+
i pj k
3
jk

1 j k
t + t t + t G2 G2
+
i pj k
3

jk

i( )
ni + nj + nk + 1 ni + nj + nk + 1 e
73

1 j k
t + t t + t G2 G2
i pj k
3

jk

ni + nj + nk + 1 cos( )

1 j k
t + t t + t G2 G2
+ i
3

ni + nj + nk + 1 ni + nj + nk + 1 ln 2j k ei( )
+
sin cos
1/2

1
t + t G2
6
ks s
jk

i pjk ks ei(S ) + ks ei(S )

jk

1
ga2 ia pia
cos(S )
t + t G2
3
a
4 j k i( )

1
S + k e i( S )
+
ln t + t t
i pj k ks e
s
3
jk

t + t G2 ni + nj + nk + 1

1 j k
i
t + t G2 ni + nj + nk + 1
+
6
jk

2

s ln j k + ks ln 2j k ei(S )

1 j k
t + t G2 ni + nj + nk + 1
+
i
6
jk

2
2 i( )
S
s ln j k + ks ln j k e

1 4 a 2 1 a
sin2
+
ga i ln ia
i s s ln 2ia
ks s
4 a
3 a

1
+
ia ga2 s ln 2ia + s ln 2ia
3 a
jk 1
1

(ks ks + 2ks s ) ln 2j k + s s ln 2j k
i
3
3
jk

1
+ (ks s ln 2j k + ks s ln 2j k
2

4 1 4 a
1 jk

ln t + t t
g p +
i pj k ks ks
.
4 a a i ia 3
jk
(B.6)
74
References
[1] A. Brignole, C.E. Ibez, C. Muoz, Nucl. Phys. B 422 (1994) 125;
A. Brignole, C.E. Ibez, C. Muoz, Nucl. Phys. B 436 (1995) 747, Erratum, hep-ph/9707209;
A. Brignole, C.E. Ibez, C. Muoz, C. Scheich, Z. Phys. C 74 (1997) 157.
[2] L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 557, hep-th/9810155.
[3] G. Giudice, M. Luty, H. Murayama, R. Rattazzi, JHEP 9812 (1998) 027, hep-ph/9810442.
[4] J. Wess, J. Bagger, Supersymmetry and Supergravity, Princeton Univ. Press, 1983.
[5] P. Bintruy, G. Girardi, R. Grimm, Phys. Rep. 343 (2001) 255.
[6] P. Bintruy, G. Girardi, R. Grimm, M. Mller, Phys. Lett. B 189 (1987) 389;
P. Bintruy, G. Girardi, R. Grimm, preprint LAPP-TH-275/90.
[7] L.J. Dixon, V.S. Kaplunovsky, J. Louis, Nucl. Phys. B 355 (1991) 649.
[8] I. Antoniadis, K.S. Narain, T.R. Taylor, Phys. Lett. B 267 (1991) 37.
[9] M.K. Gaillard, T.R. Taylor, Nucl. Phys. B 381 (1992) 577.
[10] V.S. Kaplunovsky, J. Louis, Nucl. Phys. B 444 (1995) 191.
[11] P. Mayr, S. Stieberger, Nucl. Phys. B 412 (1994) 502.
[12] G. Lopes Cardoso, B.A. Ovrut, Nucl. Phys. B 369 (1993) 351;
J.-P. Derendinger, S. Ferrara, C. Kounnas, F. Zwirner, Nucl. Phys. B 372 (1992) 145.
[13] M.K. Gaillard, B. Nelson, Y.Y. Wu, Phys. Lett. B 459 (1999) 549.
[14] J. Bagger, T. Moroi, E. Poppitz, JHEP 0004 (2000) 009, hep-th/9911029.
[15] M.K. Gaillard, B. Nelson, Nucl. Phys. B 588 (2000) 197.
[16] A. Pomarol, R. Rattazzi, JHEP 9905 (1999) 013.
[17] G.L. Cardoso, B. Ovrut, Nucl. Phys. B 418 (1994) 535.
[18] M.K. Gaillard, Phys. Lett. B 342 (1995) 125;
M.K. Gaillard, Phys. Rev. D 58 (1998) 105027;
M.K. Gaillard, Phys. Rev. D 61 (2000) 084028.
[19] C.P. Burgess, J.P. Derendinger, F. Quevedo, M. Quiros, Phys. Lett. B 348 (1995) 428.
[20] P. Bintruy, M.K. Gaillard, T.R. Taylor, Nucl. Phys. B 455 (1995) 97.
[21] S.H. Shenker, in: Random Surfaces and Quantum Gravity, Proc. of the NATO Advanced
Study Institute, Cargese, France, 1990, O. Alvarez, E. Marinari, P. Windey (Eds.), NATO ASI
Series B, Vol. 262, Plenum, New York, 1990.
[22] T. Banks, M. Dine, Phys. Rev. D 50 (1994) 7454.
[23] P. Bintruy, M.K. Gaillard, Y.Y. Wu, Nucl. Phys. B 481 (1996) 109;
P. Bintruy, M.K. Gaillard, Y.Y. Wu, Nucl. Phys. B 493 (1997) 27;
P. Bintruy, M.K. Gaillard, Y.Y. Wu, Phys. Lett. B 412 (1997) 288.
[24] R. Barbieri, S. Ferrara, L. Maiani, F. Palumbo, C.A. Savoy, Phys. Lett. B 115 (1982) 212.
[25] T.P. Cheng, L.F. Li, Gauge Theories of Elementary Particle Physics, Clarendon Press, 1988.
[26] G.F. Giudice, A. Masiero, Phys. Lett. B 206 (1988) 480.
[27] I. Antoniadis, E. Gava, K.S. Narain, T.R. Taylor, Nucl. Phys. B 432 (1994) 187.
[28] J. Giedt, Nucl. Phys. B 595 (2001) 3.
[29] D. Kaplan, G. Kribs, M. Schmaltz, Phys. Rev. D 62 (2000) 035010;
M. Schmaltz, W. Skiba, Phys. Rev. D 62 (2000) 095005.
[30] M.K. Gaillard, B. Nelson, Nucl. Phys. B 571 (2000) 3.
[31] A. Love, P. Stadler, Nucl. Phys. B 515 (1998) 34.
[32] L3 Collaboration, L3 Note 2582 (2000), Presented at 30th International Conference on High
Energy Physics, Osaka, Japan, 27 July2 August 2000;
DELPHI Collaboration, CERN-EP/2000-033 (2000), to be published in Phys. Lett. B.
[33] P. Adamietz, P. Bintruy, G. Girardi, R. Grimm, Nucl. Phys. B 401 (1993) 257.
[34] M.K. Gaillard, D.H. Lyth, H. Murayama, Phys. Rev. D 58 (1998) 123505.
[35] E. Silverstein, Phys. Lett. B 396 (1997) 91.

Hidden supersymmetries in supersymmetric

quantum mechanics
J.A. de Azcrraga a , J.M. Izquierdo b , A.J. Macfarlane c
a Departamento de Fsica Terica, Universidad de Valencia and IFIC, Centro Mixto Universidad de
Valencia-CSIC, E-46100 Burjassot, Valencia, Spain

b Departamento de Fsica Terica, Universidad de Valladolid, E-47011 Valladolid, Spain
c Centre for Mathematical Sciences, D.A.M.T.P. Wilberforce Road, Cambridge CB3 0WA, UK
Received 15 January 2001; accepted 12 March 2001
Abstract
We discuss the appearance of additional, hidden supersymmetries for simple 0+1 Ad(G)-invariant
supersymmetric models and analyse some geometrical mechanisms that lead to them. It is shown that
their existence depends crucially on the availability of odd order invariant skewsymmetric tensors on
the (generic) compact Lie algebra G, and hence on the cohomology properties of the Lie algebra
considered. 2001 Elsevier Science B.V. All rights reserved.
PACS: 02.02.Qs; 11.30.-j; 11.30.Pb
1. Introduction
In supersymmetric quantum mechanics models with standard supersymmetry, the
supercharges Qa are related to the Hamiltonian H via {Qa , Qb } = H ab , a, b = 1, . . . , N .
[1,2],
In many of these models one can find additional or hidden supercharges Q
involving the structure constants of a Lie algebra, and perhaps a KillingYano tensor [3,4].
The appearance of the KillingYano tensor in this context is not surprising, since it also
plays a role in the existence of hidden symmetries [5,6].
The additional supercharges are required to satisfy

= 0,
Qa , Q
(1)
H ] = 0, so that the Qs
generate supersymmetries of the theory. We shall
hence [Q,
consider three models: one with bosonic superfields, and two with fermionic superfields
E-mail addresses: j.a.de.azcarraga@ific.uv.es (J.A. de Azcrraga), izquierd@fta.uva.es (J.M. Izquierdo),
a.j.macfarlane@damtp.cam.ac.uk (A.J. Macfarlane).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 0 9 4 - 3
76
J.A. de Azcrraga et al. / Nuclear Physics B 604 (2001) 7591
(with N = 1 and N = 2, respectively) for which the bosonic component variables are
auxiliary [7].
A typical example of the bosonic superfield case, in which the Lie algebra G is su(2),
is that of the non-relativistic motion of a spin- 21 particle in the background field of a Dirac
monopole [8] (see also [9]). The case G = SU(2) is, however, rather exceptional since
it is the only group for which the structure constants of its algebra coincide with the fully
antisymmetric tensor (ij k ) of a (dim G)-dimensional space. Thus, a natural question to ask
is what generalisations are possible when simple (and compact) algebras G of rank l > 1
are employed. Also, for l > 1 there exist other available skewsymmetric tensors (of odd
order > 3): they are provided by the higher order cocycles of the Lie algebra cohomology
of G. Thus, in order to investigate the appearance of hidden charges in a group theoretical
context, we look first at
3 = xi fij j 1 ifij k i j k ,
Q
(2)
3
where xi and i are the position and fermionic coordinates respectively, i = 1, . . . , dim G,
and fij is an antisymmetric second order KillingYano tensor [3,4] associated with the
structure constants fij k in such a way that (1) holds, and second we look at
5 = xi fij kl j k l 1 iij kpq i j k p q
Q
(3)
5
in various contexts. In (3), fij kl = f[ij kl] is a fourth order generalised KillingYano tensor
and ij kpq is a fifth order totally antisymmetric invariant tensor, associated with the third
order (RacahCasimir) invariant symmetric polynomial of such G as allow for one. This
exists for su(n), n 3, and su(3) will be good enough to illustrate the extent of most of
our results when using the fifth order cocycle.
In the fermionic superfield case, it is possible to construct models where the only
dynamical fields are fermionic, and whose Lagrangian includes an interaction term
constructed using the structure constants of the simple Lie algebra. Our aim is to construct,
in terms of the corresponding fermionic variables and the higher order cocycles, hidden
supercharges in the N = 1 and N = 2 cases. When N = 1 we shall restrict the Lie algebra
to su(n), whereas in the N = 2 case the simple Lie algebra G will be unrestricted. The
restriction reflects the fact that the discussion of the N = 1 case employs the identity
Cij k ij ks4 s2m1 = 0, which we believe holds for all G, but for which explicit detailed
proofs are available [10] only for su(n).
We consider only Ad(G)-invariant simple 0 + 1 supersymmetry models. In Section 2 we
describe the case of a non-relativistic system moving in a space that is the representation
space of the adjoint representation of the symmetry group G, coupled to a background
potential Ai . It is shown in Section 3 that in the free case there exist non-standard
supersymmetries associated with all the higher order cocycles. In the presence of the
background field Ai (Section 4), however, we find only one hidden supercharge for the
lowest order cocycle, i.e., for that given by the structure constants of G.
In Section 5, we consider a N = 1 simple purely fermionic model and show that one may
also construct hidden supercharges from each of the l su(n) algebra cohomology cocycles.
77
In Section 6 the case with N = 2 is considered. It is shown that we may construct two
hidden supercharges for each of the l cocycles of the Lie algebra G.
2. A particle model with bosonic and fermionic degrees of freedom

Let G be a compact, simple Lie group of algebra G, [Xi , Xj ] = ifij k Xk . We set out from
the superspace Lagrangian
1
i i Di + iqDi Ai () = K + L, i = 1, . . . , dim G,
(4)
2
where is a real Grassmann variable, the i = i (t, ) are scalar superfields, and the
covariant derivative D and the generator of supersymmetry Q are given by
L=
i (t, ) = xi (t) + i i (t),
D = i t ,
Q = + i t .
(5)
The Lagrangian is invariant under the (real, adjoint) action of G. Using

Di = i(i xi ),
Ai () = Ai (x) + i j j Ai (x),
(6)
the expansion (4) of L gives

1
K = xi i qi Ai ,
2
1
1
1
L = xi xi + ii i + q x i Ai iqFij i j ,
2
2
2
where
Fij = i Aj j Ai .
(7)
(8)
(9)
We easily find the momenta and canonical commutators

pi =
L
= xi + qAi ,
xi
[xi , pj ] = iij ,
{i , j } = ij ,
(10)
and compute
1
1
xi xi + iqFij i j .
(11)
2
2
The standard supersymmetry that leaves the action for (4) invariant is i = iQi , i.e.,
H=
xi = ii ,
i = x i .
(12)
Noethers theorem
iQ = xi pi + i
L
iK,
i
(13)
where the piece depending on K comes from the quasi-invariance of the Lagrangian, gives
for the conserved supercharge
Q = xi i .
(14)
78
It is easy to check that Q above generates (12) by means of the canonical formalism, and
that
Q2 = H,
(15)
reproduces the right-hand side of (11).
3. Hidden supersymmetries in the free case

3
3.1. The case of Q
Here we shall put q = 0 in expressions in Section 2. We intend first to seek an additional
3 } = 0 in the form
3 such that {Q, Q
supersymmetry Q
3 = xi fij j 1 ifij k i j k ,
Q
(16)
3
where fij is to be determined. Since it is easier to work classically, we use the Dirac bracket
formalism corresponding to (10), so that for F and G functions of dynamical variables
{F, G} =
F G
F G F G
+ i(1)F
.
xl pl pl xl
l l
Using Q = pi i from (14) and (16), we get

3 = l xi fij,l j i xl xi fil iflj k j k ,
Q, Q
where fij,l =
f .
x l ij
(17)
(18)
The second term in the r.h.s. of (18) is zero if
fij = fj i ,
(19)
and the other two cancel if

fi[j,l] = fij l .
(20)
We could even have written fij,k instead of fij k in (16), and then fij,k may effectively be
replaced by f[ij,k] in virtue of the i j k factor. Then (20) is
fi[j,l] = f[ij,l] ,
(21)
so that
(l fi)j = 0,
(22)
2fij,l = fli,j + fj l,i .
(23)
or
Eqs. (21) or (22) state that the derivative of the antisymmetric tensor (19) is also
skewsymmetric, and characterise fij as KillingYano tensor [3,4]. One way to satisfy this
condition sets
fij = fij k xk ,
(24)
79
giving
3 = Li i + 2 Si i ,
Q
3
where
Li = fij k xj xk ,
(25)
1
Si = ifij k j k .
2
(26)
Furthermore,
[Li , xj ] = ifij k xk ,

3 , xi = ifij k xj k ,
Q
[Si , j ] = ifij k k ,

3 , i = Li + 2Si .
Q
(27)
Both Li and Si are representations of the Lie algebra G since they obey
[Li , Lj ] = ifij k Lk ,
[Si , Sj ] = ifij k Sk .
(28)
3 supersymmetry does not close on the Hamiltonian, but instead we

We note that the Q
have

3 , Q
3 = J
2 + 1 S
2 ,
Q
(29)
Ji = Li + Si ,
3
where Ji is the conserved charge associated with the G-invariance of the action (4). The
result here parallels the result of [8] for particle motion in the background field of a Dirac
monopole.
3 looks similar to the Kostant fermionic operator K
We note in passing also that Q
[11,12]
i
(30)
fij k i j k ,
3! 2
where refers to some representation of G (cf. Li in (26)). In (30) the quantised fermion
operators i have been represented, i 1 i , by the Dirac matrices of a euclidean space
2
of dimension dim G which obey {i , j } = 2ij . However, (30) implies K 2 =
2 + 13 S
2 .
3 of (16), for any representation of G.
Clearly K is not proportional the supercharge Q
K = i i
3.2. Other hidden supersymmetries

We generalise now the previous paragraph to supercharges involving the higher order
5 as in (3)
cocycles of the Lie algebra G. Instead of (16), we consider Q
5 = pi fij kp j kp 1 iij kpq ij kpq ,
Q
5
where fij kp = fij kp (x),
fij kp = fi[j kp] ,
(31)
(32)
we have used the abbreviation ij k... = i j k , . . . , and ij kpq is by definition totally

antisymmetric in all its five indices. Instead of (32) we might consider the replacement of
ij kpq by
f[ij kp,q] ,
(33)
80
5 anticommutes with Q = pi i
where fij kp,q = fij kp /xq . We now demand that Q

5 = 0,
Q, Q
(34)
quantum mechanically for variety, and also because promotion of the classical calculation
in this context is not in this instance trivial. Thus

5 = pl l , Q
5 , pl l
5 + Q
Q, Q

= pl 3pi fij kl j k ilj kpq j kpq + ipl flj kp,q j kpq .
(35)
The first term of (35) can be eliminated by requiring that fij kl is antisymmetric in i and l,
and, hence, using (33) that
fij kp = f[ij kp] .
(36)
The other result needed to secure (34) is the vanishing of the part of
flj kp,q lj kpq ,
(37)
antisymmetric in jkpq. If we try the ansatz

flj kp = lj kpq xq ,
(38)
then this requirement is satisfied. Then

5 = pi ij kpq xq j kp 1 iij kpq +ij kpq
Q
5
(39)
7 ,
is a hidden, Ad(G)-invariant supercharge. This situation is one that applies to Q
involving 7 , etc. Actually, since the charges are constructed using the structure constants
and the higher-order cocycles, there will be a hidden supercharge for each cocycle in a
model with an arbitrary simple compact group. We may then conclude the following:
The Ad(G)-invariant free supersymmetric particle model described by L = 12 i Di
s , s = 1, . . . , l. These are determined by the l Lie algebra
admits l hidden supercharges Q
cohomology cocycles of the simple compact algebra G of rank l.
If we had written f[ij kp,q] in the second term of (31), then (37) corresponds to the
vanishing of the part of
fl[j kp],q = f[lj kp,q] ,
(40)
antisymmetric in jkpq. To extract the minimal condition, we write

4
1
fl[j kp,q] + f[j kpq],l .
5
5
Then (40) is equivalent to
f[lj kp,q] =
(41)
f[lj kp,q] = f[j kpq],l ,
(42)
which is an analogue of (23) and, together with the complete antisymmetry of fij kl , means
that fij kl is a KillingYano tensor of valence four. This is consistent with Tanimotos
analysis [6], although in a slightly different context. The previous result may be rephrased
in the following form:
81
The additional supersymmetry exists because each Lie algebra cocycle of order 2m 1
provides a KillingYano tensor of valence 2(m 1) by the analogous to (38) since then
fii1 i2m3 ,q + fqi1 i2m3 ,i = 0.
(43)
One might look for different solutions, but we did not find any. For example, the
condition (42) defeats the otherwise interesting ansatz fij kl = f[ij fkl] , where fij is a
second order KillingYano tensor. In fact, the KillingYano condition appears to be very
restrictive, and it is likely that (24) and (38) are the only possible solutions for the cases
considered.
3 in the presence of a background field
4. Q
We investigate now to what extent it is possible to reproduce the analysis of Section 3 in
3 by (16), where now pi = xi + qAi as in
the A = 0 case. When Q is given by (14) and Q
3 } = 0 requires
Section 2, we find classically that {Q, Q

3 = l xi fij,l j i x l xi fil iflj k j k + qflj Fil i j = 0,
Q, Q
(44)
so we have to impose (19), (23) and
fl[j Fi]l = 0,
(45)
where Fij is given by (9). It is sensible to try first the solution

fij = fij k xk ,
(46)
of (19) and (23). To satisfy (45) we may choose

Fij = xi yj xj yi ,
(47)
where yi = dij k xj xk , whenever there exists an invariant symmetric tensor dij k on the
algebra (which excludes su(2)). This is true because of the identities fij k xj xk = 0 =
fij k xj yk , the last one due to the invariance of the symmetric tensor dij k . However,
consistency with (9) demands that any ansatz for Fij , and (47) in particular, obeys
[i Fj k] = 0.
(48)
The simple ansatz Fij = a(Z)fij k xk where Z = xk xk , satisfies (48) when the algebra is
su(2), in which case a(Z) Z 3/2 , as is expected for the monopole [8]. But for su(n), e.g.,
for su(3), for which Z = xk xk and Y = dij k xi xj xk = xk yk , the ansatz Fij = a(Z, Y )fij k xk
fails. We can see this by contracting
0 = [i Fj k] = 2
a
a
x[i fj k]l xl + 3
y[i fj k]l xl + afij k ,
Z
Y
(49)
with yi , which gives

a
2 a
Y
fj kl xl + y 2
fj kl xl + afj kl yl = 0.
3 Z
Y
(50)
82
The independence of the tensors fij k xl and fj kl yl now gives a = 0. The method used here
can be extended to show the failure also of the more general ansatz
Fij = a(Z, Y )fij k xk + b(Z, Y )fij k yk .
(51)
The case of su(2) of course escapes such a failure because yi then does not exist.
The choice (47) can be used for su(n), n 3: it obeys (48) directly but also allows us to
write down suitable choices of Ai for which (9) reproduces (47). For example, we may use
1
Ai = Y xi .
3
More generally, any Ai of the form
Ai = (Z, Y )xi + (Z, Y )yi
gives Fij in suitable form:

3
Fij = 2
(xi yj xj yi ).
Z
Y
(52)
(53)
(54)
3 given by
Thus, for any G-invariant model (4) there exists an additional supersymmetry Q
3 = pi fij k j xk 1 ifij k i j k ,
Q
(55)
3
determined by the structure constants fij k of G. The contribution proportional to Ai
disappears from the first term of (54) again because of the identities fij k xj xk = 0 =
fij k xj yk . But this does not mean that we have recovered the free case, because Ai is present
satisfies the relations
in Q = (pi qAi )i . Q

3 , xi = ifij k xj k ,
3 , i = Li + 2Si ,
Q
Q

+ S
2 + 1 S
2 ,
3 = L
3 , Q
Q
(56)
3
where now Li = fij k xj pk .
5 for q = 0. If the KillingYano tensor is
We have not found, however, an analogue of Q
fij kl = ij klm xm and Fij is proportional to (47), the condition corresponding to (45),
fl[j mn Fi]l = 0,
(57)
is not satisfied because, in contrast with fij k xj yk = 0, ij klm xl ym = 0. It is possible that

the use of more general background fields (see [13]) opens the way to richer possibilities.
5. N = 1 fermion superfields
5.1. Basic formalism
We turn now to a different supersymmetric model in which hidden supersymmetries
related to higher order cocycles also occur. This model is a theory of fermions with all states
83
in one energy level, and without bosonic dynamical variables [7]. Consider the Ad(SU(n))invariant superspace Lagrangian given by
1
1
i Di + igfij k i j k ,
2
3!
where the i = i (t, ) are i = 1, . . . , dim G fermionic superfields,
L=
i (t, ) = i (t) + Bi (t),
Di = Bi i i .
(58)
(59)
The expansion of (58) may be written as L = K + L, where now

1
1
i Bi + igi j k ,
2
3!

1
L = i i i + Bi Bi + igfij k i j Bk .
(60)
2
As in Section 2, we may use the expression of K and L in Noethers theorem (see (13)) for
the variations
K=
i = Bi ,
Bi = i i ,
(61)
to obtain
1
Q = igfij k i j k .
(62)
6
All the bosonic components Bi are auxiliary. Using their EulerLagrange equations to
solve for Bi , one obtains the classical Lagrangian Lc = 12 ii i . The classical Hamiltonian
vanishes identically, 1 but the quantum Hamiltonian, defined by H = Q2 , and computed
using {i , j } = ij , is not zero but a constant,
Q2 =

1 2 2
g n n 1
48
(63)
for su(n).
5.2. Hidden supersymmetries in the fermionic model
for
As for the (4) model for q = 0, there exist in this case additional supercharges Q
(2m1)
every non-trivial cocycle of any su(n) Lie algebra. To see this, let i1 i2m1 be a non-trivial
cocycle corresponding to an invariant symmetric tensor t of order m. If t has components
tl1 lm , then the (2m 1)-order cocycle is given by 2
l
m1
i(2m1)
= f[ili1 i2 fi2m3
i2m2 ti2m1 ]li lm1 .
1 i2m1
(64)
Using it, we form

2m1 = (2m1) i1 i2m1 ,
Q
i1 i2m1
(65)
1 There is an S S part, which is proportional to f

i i
mij fmkl i j k l and which is zero classically by virtue
of the Jacobi identity and the fermionic character of the s.
2 For the relation among symmetric tensors and cocycles of generic compact simple Lie algebra G see
[14,10,15].
84
where
i1 i2m1 i1 i2m1 .
(66)
2m1 by expressing
We compute the quantum anticommutator of Q in (62) and Q
the products ii i2 i3 j1 j2m1 and j1 j2m1 ii i2 i3 as linear combinations of completely
antisymmetrised products of s. This can be done by repeated use of the identities
(deduced from {i , j } = ij )
1
p[ij1 j2 jp ] ,
2
1
j1 jp i = j1 jp i + p[j1 jp1 jp ]i .
(67)
2
Then we easily find that the terms with no s or an even number of them vanish identically
because the two contributions coming from the anticommutator cancel each other. So we
are left with

1
(2m1)

Q, Q2m1 = ig 3(2m 1)fki2 i3 kj
i2 i3 j2 j2m1
2 j2m1
6

1 (2m 1)!
(2m1)
fklm klmj4 j2m1 j4 j2m1 .
(68)
4 (2m 4)!
i j1 jp = ij1 jp +
The first term in (68) vanishes due to the Jacobi identity (since the indices i2 i3 j2 j2m1
are antisymmetrised due to the presence of the s) and the second also vanishes because
the maximal contraction of indices among the above two su(n) cocycles of different order
2m1 define new conserved fermionic charges of higher
gives zero [10]. Hence, the l Q
order. As in the case of Q in (62), they square to a constant. For example, let us consider
5 is given by
5 for su(n), n 3. The square of Q
the case of Q
25 = 5 5
Q
i1 i5 j1 j5 i1 i5 j1 j5 .
(69)
5 and I close
It is shown in [16] (see also [10]) that this square is a number. Hence, Q, Q
into a superalgebra.
6. N = 2 fermion superfields
6.1. Basic formalism
We now consider a purely fermionic model with two standard supersymmetries [7] (see
also [17]). The supersymmetry algebra in terms of the covariant derivatives for this model
is

D = i t ,
(70)
D = + i t ,
D, D = 2it .
The N = 1 superfields i are replaced by N = 2 superfields i = i (t, , ), i =
1, . . . , dim G, to which the chirality condition D i = 0 is imposed. This of course means
85
that i obeys Di = 0 and is antichiral. Solving as usual the chirality condition we obtain
the superfield expansions
i = ei
(i Bi ) = i Bi + i i ,
i = i Bi i i ,
(71)
where i are fermionic and Bi are bosonic. The following Ad(G)-invariant superspace
action has the property that the Bs are non-dynamical, and includes an interaction term:

1
S=
dt d d i i
2

1
1
+
(72)
dt d iCij k i j +k +
dt d iCij k i j k ,
6
6
where Cij k are the structure constants of the Lie algebra G. The Ad(G)-invariant
component Lagrangian is given by
1
1
1
1
i i i + i i + Bi Bi iCij k i j Bk iCij k i j Bk .
(73)
2
2
2
2
The EulerLagrange equations of the Bi , Bi are algebraic and can be used to eliminate
these variables. Then, by writing Ji = 12 iCij k j k , the Lagrangian becomes
L=

1
i i i + i i 2Ji Ji .
2
The canonical formalism yields the Dirac brackets

i , j = iij ,
{i , j } = 0,
i , j = 0,
L=
(74)
(75)
and classically we have

H = 2Ji Ji .
(76)
The non-zero supersymmetry variations of the fields i and Bi are given by

Bi = 2i i ,
i = Bi ,
i ,
Bi = 2i
i = Bi .
(77)
These variations correspond, via Noethers theorem, to the conserved charges

1
1
iCij k i j k ,
Q = iCij k i j k .
(78)
3
3
When we quantise the theory we shall regard the i as the creation operators. Hence, to
avoid confusion, the following replacements will be made from now on: i = i , i = ci ,
as in [17], so that = ci . Thus in the quantum theory, we have the anticommutation
relations
Q=
{ci , j } = ij .
Also the quantum mechanical Hamiltonian is defined via

Q, Q = 2Hq ,
(79)
(80)
86
which gives the result

1

Hq = Ji , Ji c2 ,
6
(81)
where c2 = Cij k Cij k (= n(n2 1) for su(n)), and where it should be noted that Ji and Ji
do not commute. One might expect that that Hq is closely related to the quadratic Casimir
operator X2 = Xi Xi of G, where Xi = iCij k cj k . It is simple to confirm this for the case
of su(n) by proving the following identities

Xi Xi = nN 2Ji Ji = n n2 1 N 2Ji Ji ,
(82)
where N = ci i is the total fermion number operator, and i ci = (n2 1 N). These
allow the commutator and anticommutator of Ji and Ji to be calculated, and give rise to
the result

1
Hq = n n2 1 Xi Xi .
(83)
3
The results (81) and (83), viewed together, seem a strangely related pair. However, their
agreement, as well as the correctness of (82), can easily be confirmed by considering
actions of the operators in question on each of the fermion number N = 0, 1, 2, 3, states
of the simple but non-trivial SU(2) version of the theory. Further, having set out from the
definition (80) of Hq , we know that all energy eigenvalues are non-negative.
In addition to
1
Q
Q
= iCij k ci cj ck ,
,
q03 = q30
=
(84)
2
3!
2
in which the first and second subscripts indicate the numbers of ci and i factors,
respectively, two further fermionic operators occur naturally:
q30 =
1
1
iCij k ci cj k ,
q12 = iCij k ci j k .
2
2
These operators each anticommute with each of q30 and q03 and obey
q21 =
(85)
1
2
2
Xi Xi ,
(86)
q21
= 0,
q12
= 0.
2
It follows that q21 and q12 commute with Xi Xi and with H .
We have found a second supersymmetry which anticommutes with the original one; its
closure does not give a new operator independent of H . However, in view of the results (81)
and (83), it is not appropriate to say that our theory has N = 4 supersymmetry. We have
simply found two additional supercharges naturally associated with the structure constants
of G, a consideration that is built on significantly in the next subsection.
{q21 , q12} =
6.2. Hidden supersymmetries

One obvious question asks whether it is possible to find new supercharges that generalize
those of Section 6.1. Consider the case of charges constructed using the five-cocycle i1 i5
87
rather than the three-cocycle Cij k . An analysis of the possibilities, q50 , q41 , q32 , q23 , q14 ,
q05 , leads one to conclude that only
1
1
ii1 i5 i1 i5 , q50 = ii1 i5 ci1 ci5 ,
(87)
5!
5!
are hidden conserved supercharges because only they anticommute with q21 and q12 .
Moreover, we have the following general result:
The l = rank G pairs of fermionic charges
q05 =
i
i i
i i2m1 ,
(2m 1)! 1 2m1 1
i
i i
q2m1,0 =
(88)
ci ci2m1 ,
(2m 1)! 1 2m1 1
determined by the (2m 1)-cocycles of the Lie algebra G, where the allowed values of
m depend on the specific G considered, also commute with q21 and q12 , and hence they
commute with Hq and are conserved supercharges.
q0,2m1 =
Proof. Let us restrict ourselves to q2m1,0 (the case q0,2m1 is completely analogous).
Consider first
{q2m1,0 , q12} i[i2 i2m1 Cj1 j2 ]i cj1 cj2 ci2 ci2m1 ,
(89)
where the antisymmetrization is forced by the presence of the cs. This expression vanishes
by the G-invariance of , since this implies
i[i1 i2m2 Ci2m1 ]ij = 0.
(90)
Hence, {q2m1,0, q12 } = 0. Now we have to check that the following anticommutator also
vanishes:
{q2m1,0 , q21} ii2 i2m1 Cij1 j2 cj1 ci2 ci2m1 j2
+ (m 1)ij i3 i2m1 Cijj1 cj1 ci3 ci2m1 .
(91)
The first term vanishes due to the antisymmetry in the indices j1 , i2 , . . . , i2m1 and the
invariance of . To show that the second term also vanishes, we have to prove that
D ij i1 i2m3 Ci2m2 ij ci1 ci2m2 ,
(92)
is equal to zero. Indeed, the invariance of allows us to write it in terms of the Cs

and the invariant symmetric tensor tl1 lm (see (64)) without having to involve i, j in the
antisymmetrization, so we have
D = Cii1 l1 Ci2 i3 l2 Ci2m4 i2m3 lm1 tl1 lm1 j Ci2m2 ij ci1 ci2m2 .
(93)
Now, using the Jacobi identity

Cii1 l1 Ci2m2 ij = Cii2m2 l1 Ci1 ij + Cii1 i2m2 Cl1 ij ,
(94)
we arrive at
D = Cii2m2 l1 Ci2 i3 l2 Ci2m4 i2m3 lm1 tl1 lm1 j Ci1 ij ci1 ci2m2
+ Cii1 i2m2 Ci2 i3 l2 Ci2m4 i2m3 lm1 tl1 lm1 j Cl1 ij ci1 ci2m2 .
(95)
88
The first term of this expression is equal to D due to the presence of ci1 and ci2m2 ,
and the second term vanishes because Cl1 ij is antisymmetric in l1 , j whereas tl1 lm 1j
is symmetric in these indices. So we have D = D, D = 0, {q2m1,0, q21 } = 0 and
[q2m1,0, Hq ] = 0.
Hence, the following result follows:
For every simple Lie algebra G of rank l, the model (72) has a series of 2l conserved
supercharges that are constructed from the primitive cocycles of G, which include the
supersymmetry generators.
7. Hidden supersymmetries as Noether charges

All the hidden supercharges discussed can be shown to be Noether charges associated
with actual supersymmetries of the actions of the models in question. One way to realise
this is to use the quantum commutator of the supercharge and the variables of the model
is the conserved supercharge and u is a generic
to extract the variations. Explicitly, if Q
component field in the model, its variation may be defined by

u,
u = Q,
(96)
where is the corresponding fermionic parameter.

is a symmetry of the classical action S = dt L, S = 0, for the constant
If Q
parameter . This means that, if we allow to become a function of t, its variation will be
of the form

,
S = dt i Q
(97)
is the conserved Noether charge for the symmetry (96).
ignoring boundary terms, where Q

and since for solutions of the Lagrange

Indeed, from (97) we get S = dt (i Q),
= 0 and Q

is the conserved charge.
equations S must be zero for any , it follows that Q
This procedure is particularly suitable when, as here, the complications addressed in [18]
do not arise.
We now give the variations obtained by using (96). In the bosonic case with Ai = 0, use
of (10) yields
xi = i i fij k xk j ,
i = xj fj ik xk i fj ki j k ,
(98)
3 (Eq. (25)). If now we put = (t) and ignore boundary

for the variation induced by Q
3 , recovering Q
3 as the Noether charge. The
terms in the integrand, we do find L = i Q
same applies to the other supercharges below.
5 (Eq. (31)). The corresponding formulae for the variations for the A = 0
Consider Q
model of Section 3.2 are,
xi = i ij kpq xq j kp ,
= 3 xj j kpiq xq kp i j kpqi j kpq .
(99)
Note that in the above variations i involves the derivative of x (mathematically, this
means that successive tangent spaces jet spaces are needed to define the action of
89
This is not the case for the two purely fermionic models for which the canonical
the Qs).
quantum commutators give
i = (2m 1) i1 i2m2 i i1 i2m2
(100)
2m1 (Eq. (65)), for the N = 1 model of Section 5.1, and

for Q
i
i i
i i2m2 ,
(2m 2)! 1 2m2 i1
i
i i
i =
i i i2m2 ,
(2m 2)! 1 2m2 1
i =
i = 0,
i = 0,
(101)
for the variations produced, respectively, by the supercharges q0,2m1 and q2m1,0
(Eq. (88)), of the N = 2 model of Section 6.1.
To proceed further, consider first the closure of , on xi , say, for the A = 0 model in
Section 4. First we find
[ , ]xi = 0,
(102)
3 } = 0 (Eq. (44)). For [ , ] we find

reflecting the fact that {Q, Q
[ , ]xi = 2i fij k Jj xk ,
(103)
where
1
ifij k j k
(104)
2
is the conserved charge (cf. (29)) associated with the adjoint transformations xi =
fij k aj xk , i = fij k aj k which leave the Lagrangian (8) invariant. This of course agrees
3 , Q
3 } expressed in the form (cf. again
with the variations on xi induced by the operator {Q
1
2
2
(29)) J + 3 S . A similar analysis can be performed for the corresponding actions upon
i , for which we get
Ji = fij k xj xk
[ , ]i = 2i fij k Jj k 2i fj lm fj ik xm xk l ,
(105)
in which we see the second term is zero on shell, using l = qFlj j where Flj is given
by (47).
8. Concluding remarks
We have shown in this paper that there exist hidden supersymmetric fermionic
charges associated with the Lie algebra cohomology cocycles of the symmetry group for
three simple Ad(G)-invariant supersymmetric models, one with bosonic and fermionic
coordinates and two (for N = 1 and N = 2) with only fermionic dynamical coordinates.
For the first one, and in the free case, there are l additional supersymmetries the existence
of which is tied to the KillingYano tensors of valence (2m 2) that may be constructed
from the (2m 1) Lie algebra cohomology cocycles. In the interacting A = 0 case the
same procedure seems to allow for only one additional supersymmetry, associated with the
90
structure constants fij k of the Lie algebra G considered. In the N = 1 fermionic model,
at least in the case of G = su(n) l additional symmetries may be constructed directly from
depend only on the cohomology of G
its cocycles. In general, these purely fermionic Qs
or, equivalently, on the topology of the corresponding compact group G. In this sense the
additional supercharges may be traced to the topology of G; however, they may be seen
to generate continuous symmetries of the system and may be obtained from Noethers
theorem. In the N = 2 case it is shown that the standard supercharges are in fact part of a
series of 2l conserved supercharges that can be constructed from the l cocycles of G.
Summarising, our analysis shows that the hidden symmetries appearing in Ad(G)invariant models are in fact supersymmetries because they stem from the existence of
the G-invariant odd skewsymmetric tensors associated with the cohomology of G. In this
sense, these additional fermionic charges could have been introduced directly, even before
having a supersymmetric model, as in the (G = su(2)) analysis of [19] of the results of [8].
However, the fact that they square to a Casimir has more to do with the structure of the
cocycles themselves [10], and hence with the structure of the generic symmetry group,
than with any other considerations of su(2). In fact, the expression (29) (see also (56)), that
3 } is given by a Casimir. Also, our
3 , Q
holds for any Lie algebra G explains why, e.g., {Q
analysis in Section 4 shows that the rotational symmetry of the model in [8] does not play
an essential role (cf. [19]), being just a result of the fact that the fully antisymmetric tensor
in 3 dimensions provides the structure constants.
Acknowledgements
This work was partly supported by the DGICYT, Spain (#PB 96-0756), the Junta de
Castilla y Len, Spain (#C02/199), and PPARC, UK.
References
[1] G.W. Gibbons, R.H. Rietdijk, J.W. van Holten, SUSY in the sky, Nucl. Phys. B 404 (1993)
4264.
[2] A.J. Macfarlane, A.J. Mountain, Hidden supersymmetries of particle motion in a WuYang
monopole field, Phys. Lett. B 373 (1996) 125129.
[3] K. Yano, Some remarks on tensor fields and curvature, Ann. Math. 55 (1952) 328347.
[4] W. Dietz, R. Rdiger, Spacetimes admitting KillingYano tensors I, Proc. R. Soc. London
A 375 (1981) 361378;
W. Dietz, R. Rdiger, Spacetimes admitting KillingYano tensors II, Proc. R. Soc. London
A 381 (1982) 315322.
[5] G.W. Gibbons, P.J. Ruback, The hidden symmetries of Taub-NUT and monopole scattering,
Phys. Lett. B 188 (1987) 226230;
G.W. Gibbons, P.J. Ruback, The hidden symmetries of multi-center metrics, Commun. Math.
Phys. 115 (1988) 267300.
[6] M. Tanimoto, The role of KillingYano tensors in supersymmetric mechanics on a curved
manifold, Nucl. Phys. B 442 (1995) 549.
91
[7] A.J. Macfarlane, Supersymmetric quantum mechanics without dynamical bosons, Phys. Lett.
B 394 (1997) 99104.
[8] F. De Jonghe, A.J. Macfarlane, K. Peeters, J.W. van Holten, New supersymmetry of the
monopole, Phys. Lett. B 359 (1995) 114117.
[9] M. Plyuschay, On the nature of fermionmonopole symmetry, Phys. Lett. B 485 (2000) 187
192.
[10] J.A. de Azcrraga, A.J. Macfarlane, Compilation of identities for the antisymmetric tensors of
the higher cocycles of su(n), math-ph/0006026, to appear in IJMPA.
[11] L. Brink, P. Ramond, Dirac equations light cone supersymmetry and superconformal algebras,
hep-th/9908208, contributed article to: M. Shifman (Ed.), Golfands Memorial Volume, World
Scientific, 1999.
[12] B. Kostant, A cubic Dirac operator and the emergence of the Euler number multiplets of
representations for equal rank subgroups, Duke Math. J. 100 (1999) 447501.
[13] J.W. van Holten, D = 1 supergravity and spinning particles, in: From Field Theory to Quantum
Groups, J. Lukierski Birthday Volume, World Scientific, 1966, p. 173.
[14] J.A. de Azcrraga, A.J. Macfarlane, A.J. Mountain, J.C. Prez Bueno, Invariant tensors for
simple groups, Nucl. Phys. B 510 (1998) 657687.
[15] J.A. de Azcrraga, A.J. Macfarlane, Optimally defined RacahCasimir operators for su(n) and
their eigenvalues for various classes of representations, math-ph/0006013, to appear in J. Math.
Phys.
[16] J.A. de Azcrraga, A.J. Macfarlane, Fermionic realisations of simple Lie algebras and their
invariant fermionic operators, hep-th/0003111, Nucl. Phys. B 581 (2000) 743760.
[17] C. Chryssomalakos, J.A. de Azcrraga, A.J. Macfarlane, J.C. Prez Bueno, Higher order BRST
and anti-BRST operators and cohomology for compact Lie algebras, hep-th/9810212, J. Math.
Phys. 40 (1999) 60096032.
[18] A.J. Macfarlane, A.J. Mountain, Construction of supercharges for the one-dimensional
supersymmetric nonlinear sigma model, JHEP 005 (1999) 9906.
[19] D. Spector, N = 0 supersymmetry and the non-relativistic monopole, Phys. Lett. B 474 (2000)
331335.

Systematics of one-loop YangMills diagrams from

bosonic string amplitudes
Alberto Frizzo a , Lorenzo Magnea a , Rodolfo Russo b ,
a Dipartimento di Fisica Teorica, Universit di Torino, and I.N.F.N., Sezione di Torino, Via P. Giuria 1,
I-10125 Torino, Italy

b Laboratoire de Physique Thorique de lEcole Normale Suprieure, 24 rue Lhomond,
F-75231 Paris Cedex 05, France

Received 15 January 2001; accepted 12 April 2001
Abstract
We present a general algorithm to compute off-shell, one-loop multigluon Green functions using
bosonic string amplitudes. We identify and parametrize the regions in the space of moduli of oneloop Riemann surfaces that contribute to the field theory limit of string amplitudes. Each of these
regions can be precisely associated with a Feynman-like scalar graph with cubic and quartic vertices,
whose lines represent the joint propagation of ghosts and gluons. We give a procedure to compute
the contribution of each graph to a gluon Green function, for arbitrarily polarized off-shell gluons,
reducible and irreducible diagrams, planar and non-planar topologies. Explicit examples are given
for up to six gluons. 2001 Published by Elsevier Science B.V.
PACS: 12.30.-t; 12.38.Bx
1. Introduction
The fact that string techniques can be used to compute on-shell scattering amplitudes in
YangMills theory at tree level [1,2] and one loop [35] has been known for many years,
and has inspired the developments of novel techniques also in field theory [6]. There are
several reasons, both from the point of view of strings and in view of further applications
to YangMills theory, that lead to consider the extension of this formalism to off-shell
correlation functions.
In the context of string theory, it is generally accepted that the framework developed
so far for the calculation of scattering amplitudes corresponds to a first-quantized theory,
and is properly suited only for on-shell quantities; a consistent extension off the mass-shell
of string-derived amplitudes in the field theory limit would provide a non-trivial testing
E-mail addresses: frizzo@to.infn.it (A. Frizzo), magnea@to.infn.it (L. Magnea).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 8 7 - 0
A. Frizzo et al. / Nuclear Physics B 604 (2001) 92120
93
ground for proposals to compute off-shell amplitudes in the full string theory [7]. Recently,
similar techniques have been applied also to a non-commutative generalization of field
theory [8].
In the context of particle theory, there are several points of interest in going off-shell.
First, only with off-shell gluons one can make a definite identification of the gauge
selected by string theory; this issue was addressed shortly after the first complete oneloop computations [9], and a combination of gauges that would reproduce the simplifying
features of the string calculation was identified. This identification was confirmed in
Refs. [10,11], where a consistent prescription to continue off-shell the string amplitude
was proposed, and it was shown that divergent parts of one-loop diagrams were consistent
with the identification made in [9]. A general argument detailing how the string amplitude
splits into particle graphs, and proving that all such off-shell graphs correspond to a definite
gauge, including finite contribution, was, however, still missing. A second motivation
is provided by the desire to extend the string formalism to diagrams with more loops:
although partial progress has been made, in the context of scalar amplitudes [1214],
and for the simplest gluon diagrams [15], tackling a general two-loop gluon amplitude
has proved difficult [16]. The main source of this difficulty is the fact that there are
several simplifying features of one-loop string calculations that are not preserved by a
two-loop generalization: at one loop, for example, it is possible to use partial integration
at the string level to eliminate completely the regions of integration corresponding to four
point vertices, and furthermore the topology of the graphs contributing non-trivially to the
amplitude is essentially unique. At two loops, it becomes necessary to understand in detail
how string moduli space degenerates into a set of graphs in the field theory limit, and
the study of scalar diagrams [13] has shown that the reconstruction of Feynman graphs
from contributions arising from different regions of moduli space is non-trivial. To solve
the problems that arise at two loops, it is then necessary to develop a more systematic
approach to the study of the field theory limit; in particular, it is necessary to identify
precisely the regions of integration corresponding to particle graphs of various topologies
and vertex structures. This can be done consistently only off the mass shell, with a precise
identification of the gauge choice, because in YangMills theory the contributions of
different graphs mix when the gauge is changed, and the differences only cancel when one
sums graphs to reconstruct a gauge-invariant quantity. As a final motivation, it should be
emphasized that in YangMills theory and in QCD there are many quantities of theoretical
and phenomenological interest that are defined off the mass shell; it would be interesting
to apply string techniques to these quantities as well.
In the present paper we will perform a systematic study, of the kind just described,
for one-loop YangMills amplitudes. Starting from the string master formula for one-loop
gluon amplitudes, we will give a precise characterization of all the regions of integration
contributing to the field theory limit. Each such region corresponds to a Feynman-like
graph, where, however, the color algebra, the Lorentz algebra, and the loop momentum
integration have already been performed; because of this, we will refer to such a graph as
a scalar graph. One-loop scalar graphs correspond to sums of Feynman diagrams, since
both gluons and ghosts circulate in the loop. Scalar graphs in general contain both cubic
94
and quartic vertices, and can be one-particle reducible or irreducible, as well as planar or
non-planar. We will give examples of each kind: in particular, we will consider in detail
the case of diagrams with quartic vertices, which arise from integration over finite regions
in string moduli space, and in some cases require a regularization of infrared singularities
arising from the propagation of bosonic string tachyons in intermediate channels.
The string master formula, when applied to off-mass-shell, non-transverse external
gluons, depends on a set of projective transformations (one for each external state),
defining a local coordinate system on the world-sheet around the point of insertion of each
external particle. At one loop, geometrical arguments lead to an essentially unique choice
of these projective transformation that preserves the global properties of the geometric
objects appearing in the string master formula. 1 We will show that, in each of the cases
examined, this choice leads to a well-defined off-shell continuation of the amplitude,
which corresponds to the combination of gauges described in Refs. [9] and [11]. The
correspondence is apparent at the level of the integrands when the amplitude is expressed
in terms of Schwinger or Feynman parameters, so it can be checked case by case without
having to perform any integration. Since the results are valid for external gluons with
arbitrary mass and polarization, polarization vectors could be dropped and the results could
be read directly as the values of the corresponding truncated Green functions, expressed as
functions of the external momenta.
We emphasize that, although a precise correspondence between scalar graphs arising
from string theory and sets of diagrams in field theory is established, it remains true that
the string method is vastly convenient from a practical point of view, essentially because
it generates automatically the answer for the sum of a subset of diagrams after loop
momentum integration, directly in terms of external momenta and polarizations, bypassing
all intermediate stages of the calculation, in which typically hundreds of thousands of
terms are generated. We emphasize also that, in its present form, the method lends itself
to be completely automated, in the sense that all expressions for the integrands of all
relevant scalar graphs can be machine-generated in a short time starting from the basic
string ingredients. The only obstacle remaining on the way to a complete calculation is the
computation of scalar integrals, something for which this approach can offer no help.
The paper is organized as follows. In Section 2, we introduce the necessary ingredients
of the string formalism, we explain the role of projective transformations, and give the
geometric prescription for the off-shell continuation. In Section 3 we describe the general
features of the procedure used to take the field theory limit, introducing the mapping
between string moduli and Schwinger parameters in field theory. In Section 4, we focus
on the regions in moduli space corresponding to diagrams with cubic vertices only; we
give examples of 3 and 4 point diagrams and specifically of a non-planar contribution.
In Section 5, we show how to treat quartic vertices, defining the necessary regularization
prescriptions. In Section 6 we turn our attention to reducible diagrams, and verify the nontrivial gauge choice performed by the string in this case. In Section 7 we summarize our
1 A choice of this kind was proposed in Ref. [7], where results similar to ours were obtained for the two- and
the three-point functions.
95
calculational method, which is now as general and systematic as conventional Feynman

rules, at least for one-loop correlation functions, and we give a concrete example of the
application of the method to a more complicated graph, computing a contribution to the
six-gluon one-loop amplitude. Finally, in Section 8, we draw our conclusions and discuss
possible developments and applications.
2. The one-loop off-shell master formula

Open bosonic string theory is the simplest string model containing a massless vector
state. Of course, the whole spectrum consists of an infinite tower of states with masses
proportional to the string tension T ; however, all the contributions to scattering amplitudes
coming from massive modes disappear in the low-energy limit, T , and the dynamics
of the surviving light modes can be effectively described by a field theory. As discussed in
Ref. [13], it is possible to reproduce different field theories in the infinite tension limit by
choosing different matching conditions for the coupling. Here we will be concerned with
the YangMills field theory limit (first discussed in Ref. [4]), which is obtained by allowing
the circulation of only massless states in the loop.
There are two potential technical problems in the computation of the YangMills field
theory limit: the fact that the ground state of the spectrum of the bosonic string is a
tachyon, and the fact that string amplitudes are computed in the critical dimension d = dc
= 26. Neither problem affects the result: it turns out that tachyonic contributions to gluon
amplitudes correspond to infrared divergences that can be easily identified and regulated by
hand, and furthermore one can continue the string amplitude to arbitrary values of d = 26,
obtaining in fact the correct result for the dimensionally regularized field theory of interest.
The continuation to arbitrary d can be justified form the point of view of string theory by
constructing a simple, though consistent, string model, having as low-energy limit the pure
d-dimensional YangMills theory [17]. It is sufficient to break the full 26-dimensional
Lorentz invariance by imposing different boundary conditions on the open string along
different directions. Since we want open strings to be free to move in d of the original 26
dimensions, we split our spacetime as M26 = Md M26d , where Md is d-dimensional
Minkowsky space; then we choose Neumann boundary conditions along Md and Dirichlet
boundary conditions along M26d ; in other words, we consider a bosonic D(d 1)-brane.
It is well known that the massless sector of such open strings also contains states which are
scalars from the d-dimensional point of view. In order to decouple these unwanted scalars,
one can choose M26d to be the orbifold R26d /Z2 , where the Z2 orbifold operation is
simply the reflection in R26d around the origin. The orbifolded theory must contain only
the open string states which are invariant under Z2 . By choosing a trivial action of this
Z2 operation on the ChanPaton factor, it is easy to see that all the massless scalar are
projected out by the orbifold operation, and one is left with a pure d-dimensional Yang
Mills theory (in the D-brane language, we are using bosonic fractional branes). The use
of this kind of D-branes, present in orbifolded theories, is a quite general technique for
generating pure YangMills theory also in the supersymmetric case.
96
Of course, the string master formula one derives in the orbifold model would be more
complicated than the one presented here, in Eq. (2.2). However, as long as one considers
scattering amplitudes only among vector states, the differences conspire precisely, in the
field theory limit, to change the effective value of the dimension from dc = 26 to d
[17]. Thus, computing the string amplitudes from Eq. (2.2) directly in d dimensions is,
at one loop and in the field theory limit, a shortcut perfectly equivalent to the computation
in the orbifolded model in the critical dimension. The gluon amplitude computed this
way automatically has the correct d dependence, and thus is given in dimensionally
regularized form. This also provides a simple criterion to identify tachyonic contributions:
they are in fact infrared divergent in the field theory limit for all values of d, i.e., they
are not regularized by dimensional continuation. Such contributions must be discarded,
just as the contributions of massive states, although they may leave finite remainders after
regularization, as we will see in Section 5.
A further technical problem of special relevance to the present paper is the fact that
string amplitudes must be computed between states satisfying the mass-shell condition,
to preserve conformal invariance on the string world-sheet. It turns out, however, that
the effects of off-shell continuation are parametrized in a simple way in the context of
the operator formalism [18]: the amplitude acquires a dependence on a set of projective
transformations defining local coordinate systems around the locations of the insertions
of external gluons. As we will see, there is a simple, geometrically motivated choice for
these projective transformations, which consistently yields the correct off-shell extension
for YangMills amplitudes.
Since our aim here is to provide a self-contained method to calculate gluon diagrams,
we will not emphasize technical details of the string formalism, which are available
elsewhere [18,19]. We will just give the general expression for the one-loop multigluon
string amplitude, and describe the geometric features necessary to define and perform the
field theory limit.
It should be kept in mind that string theory, although describing the propagation of
objects in a d-dimensional spacetime, has a two-dimensional structure describing the
world-sheet dynamics. Thus the amplitudes for the scattering of string states, and the
ingredients entering in their expression, must have a meaning also from the point of view
of the two-dimensional theory. String amplitudes must then be written in the language of
Riemann surfaces, in our case with boundaries and punctures representing the insertions
of external gluons. Using the operator formalism [18], Riemann surfaces turn out to
be represented in the Schottky parametrization [20], which is the one we will use. In
particular, one-loop open string amplitudes involve insertions of gluon vertex operators
on the boundaries of an annulus, which, in the Schottky parametrization, is represented as
in Fig. 1.
The boundaries of the annulus correspond to the segments AA and BB , whose
endpoints must be identified. External gluons will be inserted on the boundaries at
locations zi , often referred to as punctures. It is easy to see how the idea of adding loops
is implemented in this formalism. One starts from the upper half complex plane (equivalent
to the disk, representing the tree-level scattering amplitude), and adds two circles with
97
Fig. 1. The annulus in Schottky representation.
centers on the real axis, which must then be identified via a projective transformation.
Each loop is thus characterized by three real parameters: the positions of the two centers
on the real axis and the common radius of the circles, which fix, respectively, the positions
and the width of the holes added to the surface. The positions of the various circles are
related to the fixed points of the projective transformations under which the pairs of circles
are identified. Usually the fixed points are denoted by and , while the width of the
holes is determined by the third parameter characterizing the projective transformation, the
multiplier k. The fundamental building block entering the amplitudes is the Green function
of the two-dimensional Laplace operator defined on the string world-sheet, denoted by
G(zi , zj ): it is the correlator of two world-sheet bosons located at positions zi and zj and
of course depends on the parameters (, , k) defining the Riemann surface.
At this point we can write down the master formula for the one-loop multigluon string
amplitude, in terms of the Green function G. If one does not impose the mass-shell and
transversality conditions on the external states, the Green function must be evaluated in
terms of the projective transformations associated with the punctures, denoted by Vi (z),
which must satisfy
Vi (0) = zi ,
(2.1)
in order to define local coordinates around zi . The master formula is then given by
A(1) (1, . . . , M)

2M gSM
(2 )(Md2M2d)/4
= Tr a1 am Tr am+1 aM
(2)d

M
2d
dk d d
ln k d/2
i=1 dzi
1 kn

2
2
dVabc Vi (0) k ( )
2
n=1
2 pi pj

exp(G(Vi (z), Vj (y)))
Vi (0)
1

2 pi i

exp
2
Vi (0)
Vi (z) Vj (y)
i
i<j
z=y=0

exp
2 pj i z G Vi (z), Vj (y) z=y=0
i=j

1
+ i j z y G Vi (z), Vj (y) z=y=0
2
(2.2)
m.l.
where the subscript m.l. means that only terms multilinear in all polarization vectors
must be considered.
98
Eq. (2.2) gives the one-loop contribution to the color-ordered scattering amplitude of M
gluons characterized by momenta pi , polarization vectors i and color indices ai , in the
general case in which the mass-shell (pi2 = 0) and transversality (i pi = 0) conditions
have not been imposed. 2 Since this amplitude is color-ordered (one needs to sum over
all inequivalent orderings, to obtain the full amplitude), the color structure is factorized
and appears in the form of a ChanPaton factor. At one loop this factor contains at most
two traces of color matrices, corresponding to the insertion of gluons on both boundaries
of the annulus; in the planar case one has m = 0, and only one trace survives, multiplied
times a factor of N = Tr 1. The string coupling gS can be related to the gauge coupling
in d = 4 2 dimensions by matching the values of the tree-level three-point amplitude
derived from string theory with the corresponding YangMills Feynman rule. With our
choice of normalization for the SU(N) generators in the fundamental representation a ,

1
Tr a b = ab ,
(2.3)
2
the matching condition reads [11]
gd
gS = (2 )1d/4,
(2.4)
2
where gd = gYM , and , the Regge slope, is inversely proportional to the string
tension T . In the field theory limit one considers amplitudes where the typical energy
is much smaller than 1/2 , that is pi pj 1. This limit can be computed by formally
considering 0.
The KobaNielsen variables zi in Eq. (2.2) represent the locations of gluon insertions
on the boundary of the one-loop world-sheet. The symbol dVabc in the integration measure
reminds us that projective invariance of the string amplitude can be used to fix the
location of three integration variables, chosen among the punctures zi or the fixed points.
Multipliers, on the other hand, cannot be fixed; as we will see, they are related to Schwinger
parameters measuring the total length of the loops. In the present case, it is convenient to
fix = , = 0; next, we fix the position of one of the punctures, setting z1 = 1, as
indicated in Fig. 1.
The dependence of Eq. (2.2) on the projective transformations Vi (z) deserves further
comment. Local coordinates around the punctures must be introduced in the operator
formalism in order to perform the sewing procedure [18], which leads to the construction
of loop amplitudes from a tree-level multi-leg vertex operator (the N -Reggeon vertex). 3
If one imposes the mass-shell and transversality conditions, it can be easily verified that
the Vi dependence in Eq. (2.2) cancels, as a consequence of momentum conservation.
Furthermore, as we will see, even off-shell it is possible to reabsorb all the Vi dependence
in a simple redefinition of the Green function. This can be achieved by defining

1

1
G Vi (z), Vj (y) = G Vi (z), Vj (y) logVi (z) logVj (y).
(2.5)
2
2
2 Notice that we use the metric ( + + +) in string-derived expressions.
3 Roughly speaking, the sewing procedure generates a g-loop, m-point amplitude by inserting a propagator
between two external states of a (g 1)-loop, (m + 2)-point amplitude, and summing over the whole spectrum
of string states propagating in that channel.
99
Substituting Eqs. (2.4) and (2.5) into Eq. (2.2), and making use of Eq. (2.1), we recover
a simpler expression of the master formula, which was used in previous work [11], and
which has no explicit dependence on the projective transformations Vi (z). What is most
significant is the fact that the term proportional to pi i is canceled without imposing
the transversality condition to external gluons. The formula we find is then applicable to
unphysical as well as to physical polarizations. Focusing on the planar case, the practical
version of the master formula can be written as
A(1)
P (1, . . . , M)

= N Tr aM a1
gdM
(2 )(Md)/2
(4)d/2
1
0
1
1
dzM1
dzM
k
1
zM
dz2
z3
log k
2d
dk
1 kn
2
k
n=1
d/2

2 pi pj
exp G(zi , zj )
i<j

1
exp
2 pj i i G(zi , zj ) + i j i j G(zi , zj )
2
i=j
(2.6)
m.l.
To extract results from Eq. (2.6) in the case of off-shell gluons, one must of course make a
choice for the Vi (z). This choice can be guided by two-dimensional geometry. It is known
in fact that, for any Riemann surface, the Green function contains the logarithm of the
prime form, which is essentially a well defined analytic generalization of the monomial
zi zj to higher-genus surfaces. The prime form is not a function on the surface, but
rather a differential form of weight {1/2, 1/2}. One sees from Eq. (2.5) that one can
construct a well behaved Green function on the surface by choosing the Vi (z)s so that their
derivatives Vi (0) behave as differential forms of weight 1. The only globally defined oneforms on a genus g Riemann surface are the g Abelian differentials, so, at one loop, this
criterion leads to an essentially unique choice. One must choose

1
Vi (0) = (zi ) ,
(2.7)
where (z) is related to the unique Abelian differential on the torus, which for our choice of
fixed points ( = , = 0) is simply dz/z. We are lead uniquely to the choice Vi (0) = zi ,
already adopted in Refs. [11,13]. With this choice, the modified open string Green function
G(zi , zj ) defined in Eq. (2.5) is given by

zi
zj
1
zi 2

+
G(zi , zj ) = log
log
zj
zi
2 log k
zj

zj
zi
n
n
1 k zi 1 k zj
+ log
(2.8)
,
(1 k n )2
n=1
notice that G(zi , zj ) is actually a function only of the ratio ij = zi /zj , so that it is
translationally invariant on the torus parametrized by coordinates i log(zi ).
100
We conclude this section with a comment on the integration region over the moduli
given in Eq. (2.6). Notice that the zi s are ordered between k and z1 = 1. In fact, in the
Schottky parametrization, for a planar configuration, all punctures are constrained to lie
on the same boundary of the string world-sheet,
thus, having fixed
z1 = 1, all other zi
should be integrated over the interval B = k < zi < B = 1/ k, with the restriction
on the ordering implied by the color trace. This would complicate thecalculation of the
field theory
limit, since there would be contributions both from zi k 0 and from
zi 1/ k . It is possible to bypass this practical difficulty by making use of the fact
that the
in particular implies that the interval
string integrand is modular invariant, which

[1, 1/ k] can be mapped onto the interval [k, k ]. One can easily show that the effective
one-loop Green function in Eq. (2.5) satisfies
G(j i ; k) = G(ij ; k),
G(kj i ; k) = G(ij ; k),
(2.9)
where ij = zi /zj and with a slight abuse of notation we write G(ij ; k) G(zj , zi ).
Using these properties,
one can map all configurations with a subset of punctures in the
interval [1,
1/ k] to configurations in which those punctures have been moved to the
interval [k, k], preserving the ordering on the circle. This procedure yields the integration
region in Eq. (2.6).
3. General structure of the field theory limit

In the formal limit 0, Eq. (2.6) gives a vanishing or a divergent result, depending
on the number of external gluons M. It is clear that a finite result can be obtained only by
considering singular regions of integration, and parametrizing the singularities in order to
generate powers of , to compensate the explicit factor of (Md)/2 . In order to do this, it
will be necessary to change integration variables from the dimensionless quantities {k, zi }
to dimensionful parameters characterizing the size and shape of the string world-sheet in
units of .
The appropriate change of variables is once again suggested by the geometric structure
emerging from the operator formalism. The sewing procedure, in fact, indicates that when
the integrand of Eq. (2.6) is written out as a Laurent series expansion in powers of the
multiplier k, the coefficient of k n represents the contribution to the amplitude due to the
circulation in the loop of states belonging to mass level n + 1. Since the integrand has no
power-law divergence as k 0, except the explicit factor of k 2 , one can conclude that the
terms proportional to k 2 correspond to tachyon propagation (mass level n = 1), while
the contribution of gluons (mass level n = 0) will be given by terms proportional to k 1 .
This interpretation of the role played by k also rules out the possibility of expressing k
directly as the ratio of a dimensionful parameter and ; furthermore, after isolating the
correct power of k, the only singularities left in the integrand are logarithmic, and the
measure of integration for the multiplier is just d log k. This suggests [21] that one should
identify the logarithm of the multiplier with the length of the string loop, measured in units
of , in agreement with the fact that the limit k 0 corresponds to a very long and narrow
101
annulus, as seen explicitly in Fig. 1. It is then natural to think of the puncture coordinates
as associated to fractional lengths in a similar way. This leads to the choice
t
ti
log k = ,
(3.1)
log zi = ,
which has been employed in the past for both gluon [3,11] and scalar [13] amplitudes. It
turns out that the variables ti and t play the role of Schwinger parameters associated to
combinations of field theory propagators, as we will see explicitly below.
Once the change of variables in Eq. (3.1) has been effected, the field theory limit can be
reached by taking 0 with {t, ti } fixed. Thus one expands the integrand of Eq. (2.6) in
powers of k and zi , retains the coefficients of k 1 and zi1 , and keeps all terms that have
a finite limit as 0. As we will see, one must be careful to include all sources of
dependence, and all singular regions of integration.
This is a representative case of the general strategy that must be employed also for
amplitudes with more loops: the zero-slope limit has to be taken after introducing the
dimensionful field theory variables that have to be kept fixed. The exact form of the
change of integration variables identifies the particular corner of moduli space being
considered. To get the complete result, it is necessary to make sure that all singular corners
of integration have been taken into account. In particular, in the case of one-loop gluon
amplitudes, it is immediately clear that the region identified by Eq. (3.1) cannot give
the complete answer. Consider in fact the counting of powers of in Eq. (2.6) after
implementing Eq. (3.1): one finds an overall factor of ( /2)M/2 from the normalization
and the change of variables, and a further factor of m , with 0 m M/2, from the terms
multilinear in the polarization vectors arising from the expansion of the second exponential.
Specifically, only the terms containing no double derivatives of the Green function display
a complete cancellation of the overall power of , giving immediately a contribution to
the field theory limit. All terms containing double derivatives of G still have an overall
negative power of , and one must find further sources of positive powers of in order to
generate finite contributions from these terms. There are two such sources, which generate
classes of contributions which are easily identified in the field theory limit.
Positive powers of will arise from regions of integration in which certain Schwinger
parameters are small, in fact are themselves O( ). Notice that so long as, say, ti < tj ,
in the 0 limit the punctures are strongly ordered on the real axis, i.e., zj zi .
Taking ti tj = O( ), on the other hand, means that the punctures remain close to
each other as the field theory limit is taken. These integration regions must then, and
do, correspond to four-point vertices in field theory. From the point of view of string
theory such contributions only arise naturally in gluon amplitudes: if one takes a field
theory limit corresponding to scalars with a cubic coupling, these regions are always
suppressed by powers of . Notice also that the counting of powers of described
above correctly reproduces the counting of four-point vertices in gluon diagrams: in
an M-point one-loop amplitude there can be at most M/2 four-point vertices (or
(M 1)/2, for odd M), which is also the maximum number of powers of that
can be extracted with this method in Eq. (2.6) without getting a vanishing result. For
obvious dimensional reasons, diagrams with only four-point vertices have no powers
102
of momenta in the numerator, and thus give results in which all polarization vectors
are dotted into each other; this is also directly evident in Eq. (2.6).
A second source of positive powers of is the first exponential factor of Eq. (2.6). In
fact, upon substituting Eq. (3.1), one easily sees that this factor takes the form

2 pi pj

(3.2)
exp G(zi , zj )
= exp c0 (ti ) + c1 (ti ) .
i<j
Clearly, if the overall power of in Eq. (2.6) is negative, Eq. (3.2) must be expanded
as

2 pi pj

1
exp G(zi , zj )
= exp c0 (ti ) 1 + c1 (ti ) + 2 c1 (ti )2 + ,
2
i<j
(3.3)
and thus participates in the counting of . All terms arising from diagrams containing
only cubic vertices but containing factors of the form pi pj k l are generated
by string theory through this expansion.
The discussion developed so far is based upon the idea of expanding the integrand in
powers of k and zi , assuming that the punctures are not too close to each other, so that the
Green function is not dominated by its short distance singularity as zi zj . As a result,
one gets all 1PI contributions to the amplitude. As pointed out long ago [3], one can also
get one-particle-reducible contribution, by considering the opposite limit, taking zi zj
before the field theory limit, driven by k 0. This procedure is also a source of powers
of , since it generates dimensionless gluon propagators in the chosen channel, in the form
1/( pi pj ). This limit will be discussed in more detail in Section 6.
Summarizing, we have tentatively identified all the regions of integration in moduli
space contributing to the field theory limit. We see that in this limit the string amplitude
decomposes into a sum of contributions associated to graphs that, as we shall see, are
in straightforward correspondence with the Feynman graphs of the limiting field theory.
Specifically, the graphs for a given amplitude can be enumerated by counting the Feynman
graphs with the same number of external legs in a scalar theory with both cubic and quartic
interactions. Each string-derived graph corresponds to a sum of Feynman graphs in Yang
Mills theory, involving the propagation of both ghosts and gluons. Furthermore, the field
theory limit yields directly the Schwinger parameter integral, having bypassed all Lorentz
and color algebra, loop momentum integration, and the decomposition of tensor integrals
into scalar building blocks [23,24]. We will refer to the string-derived graphs as scalar
graphs. Scalar graphs are directly expressed in terms of scalar integrals, albeit with a
numerator structure reminiscent of the presence of momenta in YangMills vertices. At
least at one loop, string theory effectively succeeds in reducing the complexity of the
calculation of a YangMills amplitude to a level close to a scalar amplitude, even in the
general case of off-shell gluons with arbitrary polarizations.
In the next sections, we will describe in more detail how different scalar graphs must
be computed starting from Eq. (2.6). We will consider in turn irreducible graphs with
only cubic vertices, arising from the region in which the punctures are strongly ordered
103
(Section 4), then irreducible graphs with quartic vertices, arising when the strong ordering
condition is relaxed (Section 5), and finally reducible graphs (Section 6). In each case we
will give concrete examples. Because the graphs are computed with off-shell gluons, their
values are gauge-dependent in field theory. One is then able to make a precise identification
of the gauge choice made by string theory, and in fact our calculations confirm the
results of Refs. [9] and [11]: the string computes the amplitudes in a combination of the
background Feynman gauge (for irreducible diagrams) with the GervaisNeveu gauge (for
trees attached to the loop). It is tempting to identify the dependence of the off-shell string
amplitude on the choice of projective coordinates Vi (0) with a gauge dependence in the
field theory limit. However, we have so far been unable to make this identification precise.
We conclude this section by listing the explicit expressions that we will need for the
Green function and its derivatives in the various relevant limits. In the limit k 0, Eq. (2.8)
and its derivatives have the expansions

log2 ij
1
1
k ij +
G(ij ; k) = log(1 ij ) log ij +
2 + O k2 ,
2
2 log k
ij

log ij
1
1
1
1
k 2 + 1 + O k2 ,
+
i G(ij , k) =
zj
1 ij
2ij
ij log k
ij

log ij
1
zi
1
1
k 2 + 1 + O k2 ,
+
j G(ij ; k) = 2
1 ij
2ij
ij log k
zj
ij

ij
1
1
1
+
+
k
+
i j G(ij ; k) =
(3.4)
+ O k2 ,
ij
2
zi zj
log k (1 ij )
ij
where ij = zi /zj < 1, since the punctures are radially ordered on the boundary.
As we will see in Section 4, if the two punctures are on different boundaries, so that their
coordinates have opposite sign in Schottky parametrization, the Green function is slightly
modified and becomes

1
log2 |ij |
1

+ k |ij | +
+2
G(ij , k) = log 1 + |ij | log |ij | +
2
2 log k
|ij |

+ O k2 .
(3.5)
When considering reducible diagrams, we will also need the limit of the Green function as
1. In this limit, the logarithmic singularity dominates, and one can use simply
G(; k) = log(1 ) + O(1 ).
(3.6)
When the two punctures are on different boundaries, there is no logarithmic singularity in
the integration domain, at least when considering three-gluon vertices, so the corresponding limit for the non-planar Green function is not needed. We are now ready to turn to a
more detailed analysis of the field theory limit for different scalar graph topologies.
104
4. Irreducible diagrams with cubic vertices

Irreducible scalar diagrams with only cubic vertices are the most straightforward to
compute, since they arise naturally from the region of moduli space parametrized by
Eq. (3.1) in the limit 0. As pointed out above, in this region each gluon insertion
on the annulus is widely separated from the contiguous ones: thus, on both sides of each
external state there are loop propagators. Consider, for example, two contiguous punctures
zi and zi+1 . According to Eq. (3.1), in the field theory limit we set

zi = eti / ,
zi+1 = eti+1 / .
(4.1)
The existence of a loop propagator between the two insertions, requires that its proper time,
i.e., ti+1 ti , be kept finite as 0. Then
ti+1 ti
zi+1
(4.2)

0 zi+1 zi ,

zi
thus we conclude that the gluon labeled by i connects to the loop through a cubic vertex if
the KobaNielsen variables adjacent to it are strongly ordered, zi+1 zi zi1 . 4
To illustrate the technique, we will begin by computing the sum of the three diagrams
given in Fig. 2, using Eq. (2.6) for M = 3, and changing variables according to Eq. (3.1).
Concentrating on the color ordering (321), and retaining only the leading terms in the
expansion in powers of the multiplier, Eq. (2.6) for M = 3 reads

A(1)(1, 2, 3) = N Tr 3 2 1
gd3
(2 )2d/2
(4)d/2

1
1
1

log k d/2
dk
1
+
k(d
2)
dz
dz
3
2
2
k2
z3
0
k

exp 2 (p1 p2 G12 + p1 p3 G13 + p2 p3 G23 )

2 (1 p2 1 G21 + 1 p3 1 G31 )(2 p3 2 G32 + 2 p1 2 G12 )
(3 p1 3 G13 + 3 p2 3 G23 )

+ 1 2 1 2 G12 (3 p1 3 G13 + 3 p2 3 G23 ) + (cycl.) .
(4.3)
Here Gij denotes the expansion of G(ij , k) in powers of k, Eq. (3.4).

To proceed, we note that changing variables to Schwinger parameters the overall power
of becomes 1 . Thus for the terms in Eq. (4.3) that do not contain double derivatives of
the Green function all polynomial dependence on cancels and the field theory limit can
be taken directly. The terms proportional to double derivatives of G, however, need an extra
positive power of . This can be supplied by the term proportional to 1/ log k in Eq. (3.4),
but to get the complete result for these terms it is necessary to expand the exponential
as well, according to Eq. (3.3). Once this is done, the field theory limit can be taken by
4 We work in the configuration 0 k z z = 1. To state that z is widely separated from z = 1,
M
M
1
1
we may simply require it to be widely separated from k, since points k and 1 are identified on the annulus.
105
Fig. 2. Three-point diagrams: external gluons are background fields, marked by A. Momenta pi are
incoming.
picking the coefficient of 0 k 0 zi0 in the resulting polynomial, while the exponential factor
becomes simply

exp 2 (p1 p2 G12 + p1 p3 G13 + p2 p3 G23 )

t2
t3
exp p1 p2 t2 1
+ p1 p3 t3 1
t
t

t3 t2
(4.4)
.
+ p2 p3 (t3 t2 ) 1
t
To write our final answer in the most symmetric manner, it is convenient to introduce
the dimensionless Feynman parameters x1 = t2 /t, x2 = (t3 t2 )/t, x3 = 1 x1 x2 , and
to define xij as the sum of all Feynman parameters found between gluons i and j , when
moving clockwise around the loop (see Fig. 2), so that, for example, x13 = x1 + x2 . The
complete expression for Eq. (4.3) in the field theory limit becomes then
gd3
A (p1 , p2 , p3 ) = N Tr
(4)d/2
(1)
1
dx1 dx2 dx3 1
3 2 1

dt t 2d/2
0

xj exp t x1 x2 p22 + x2 x3 p32 + x3 x1 p12
j =1

1 pi 2 pj 3 pk (d 2)(x1i xi1 )(x2j xj 2 )(x3k xk3 )

i=2,3;j=1,3;
+ 8(x1i xi1 + x2j xj 2 + x3k xk3 )
k=1,2

2(d 2)
2
(xik xki ) + 16xki pi pj + 8pj
i j k pi
+
t
(i,j,k)=(1,2,3),(2,3,1),(3,1,2)
+
i j k pj
(i,j,k)=(1,2,3),(2,3,1),(3,1,2)
2(d 2)
(xj k xkj ) + 16xkj pi pj + 8pj2 8pk2
t

.
(4.5)
106
Fig. 3. Four-point diagram: external gluons are background fields, marked by an A. Momenta pi are
incoming.
In writing Eq. (4.5), we kept all terms proportional to longitudinal gluon polarizations, but
to get a symmetric expression we replaced, say, 1 p1 with 1 p2 1 p3 . The result
exactly matches the contribution of the corresponding color ordering to the three diagrams
in Fig. 2, computed using the background field method, with the Feynman rules given,
for example, in [22]. 5 We note in passing that the function appearing in the exponent of
Eq. (4.5) depends only on the topology of the diagram, and thus is the same that would
appear in a scalar theory.
Using symbolic algebraic manipulation programs, one can check at the level of
integrands that our prescription gives the correct results for scalar graphs also for
amplitudes with more gluons. We have tested the irreducible graph with only cubic vertices
in the case of the four gluon amplitude, but the complete result is too lengthy to report here.
It is more interesting to use the four-point amplitude as a testing ground for the computation
of non-planar contributions, since the first non-trivial example arises in this case. Consider
then the scalar graph containing the four-gluon diagram in Fig. 3, taking for simplicity
on-shell gluons.
With a growing number of external particles, it is convenient to fix helicities, which
helps to reduce the number of non-vanishing Lorentz invariant combinations of external
momenta and polarizations. Using standard helicity techniques, we may choose, for
example, the configuration + (p1 , p4 ), + (p2 , p4 ), (p3 , p1 ), (p4 , p1 ), where the first
argument is the external gluon momentum, while the second is the reference momentum,
as explained in [2]. In this case the only five non-vanishing scalar products are 2 3 ,
1 p3 = 1 p2 , 2 p1 = 2 p3 , 3 p4 = 3 p2 , and 4 p2 = 4 p3 . Summing
gluon and ghost loops, we find that the scalar graph, in the background field method, has
the expression
A(1) (p1 , p2 , p3 , p4 )

1
4
gd4
= 8iC
dt t 3d/2 dx1 dx2 dx3 dx4 1
xj
(4)d/2
0
j =1
5 The string result must be multiplied by a factor i, since the string rules give directly the T matrix.

exp t (s12 x2 x4 + s23 x1 x3 ) (1 p2 )(4 p3 )

2(2 p3 )(3 p2 ) 2x2 x4 + (d 2)x22 (x3 + x4 )(x1 + x4 )

1 2
2
2
+ (2 3 ) s23 (x1 + x3 + x4 ) + x2 + (d 2) x2 ,
t
where the color factor is

C = N Tr 1 2 3 4 + Tr 4 3 2 1

+ 2 Tr 1 2 Tr 3 4 + Tr 1 3 Tr 2 4 + Tr 1 4 Tr 2 3 .
107
(4.6)
(4.7)
Leading color terms can be obtained from the planar string amplitude, exactly as in the
previous example. Here we focus on subleading terms proportional to double traces,
which are reproduced by non-planar string amplitudes. They were absent in the three-point
amplitude, since Tr(i ) = 0.
Consider, for example, the term proportional to Tr(1 2 ) Tr(3 4 ). The string amplitude
with this ChanPaton factor is the one with z1 and z2 on one boundary (say, the positive
real axis in the Schottky representation, Section 2, with z1 = 1) while z3 and z4 are on
the other boundary. The non-planar master formula is similar to the planar one, with small
but significant differences. First of all, when a Green function connects two punctures
on different boundaries, its non-planar version, Eq. (3.5), must be used. Also, the correct
expression for the projective transformations is Vi (0) = |zi |. Finally, z2 and z3 are not
ordered, so that the resulting integration regions is the interval [1, k] for both punctures.
To compare non-planar and planar contributions, we change variables according to z3
z3 and z4 z4 ; all punctures are now integrated on the positive real axis, just as in the
planar case, however, now it is necessary to distinguish six different orderings of the four
punctures, namely,
(1) k z4 z3 z2 1,
(4) k z4 z2 z3 1,
(2) k z2 z3 z4 1,
(5) k z3 z4 z2 1,
(3) k z3 z2 z4 1,
(6) k z2 z4 z3 1.
Regions 1 and 2 correspond to the cyclic orderings (4321) and (1234), respectively.
The resulting integrands, expressed in terms of z2 , z3 and z4 , are equal, but, at the string
level, they differ from the planar amplitude. The difference, however, vanishes in the field
theory limit, so the two regions add up to give the relative factor of 2 between the single
trace planar contribution proportional to N Tr(1 2 3 4 ) and the double trace contribution
proportional to Tr(1 2 ) Tr(3 4 ) in Eq. (4.7).
Regions 3, 4, 5 and 6 correspond to the other cyclic orderings ((1324), (142
3), (1342), (1243), respectively), so they are to be identified with the subleading
terms of the other color-ordered field theory diagrams. This same mechanism has been
observed for scalar diagrams [13]. The fact that the differences between planar and nonplanar integrands vanish in the field theory limit is responsible for the relations between
single trace and double trace subamplitudes, noticed already in [25].
108
5. Irreducible diagrams with quartic vertices

Quartic vertices arise in the field theory limit when the strong ordering condition
between two punctures is relaxed. Changes of variables such as the one given in Eq. (4.1),
in fact, generate a certain number of explicit powers of in the integration measure. It
is clear, however, since Schwinger parameters are dimensionful, that there are potential
O( ) corrections arising from regions of integration where the Schwinger parameters
themselves, or their differences, are O( ). Such corrections are negligible in a scalar
theory, where all terms in the integrand have a uniform power of when expressed in
terms of the ti s; in a gauge theory amplitude, however, these terms do contribute, as
explained in Section 3.
Suppose we want two consecutive punctures, say zi and zi+1 , to attach to the loop
through a quartic vertex: the propagator between them would have a Schwinger parameter
proportional to ti+1 ti , so we must impose that this vanish in the zero-slope limit, setting
ti+1 ti = O( )
zi+1
y = O(1).
zi
(5.1)
The correct procedure to take into account this region of integration is to change variables
according to, say, zi = exp(ti / ), and zi+1 = yi zi , subsequently integrating over yi in a
neighborhood of yi = 1. Notice that we now have a different overall power of , since yi
is independent of , so we will be considering a different set of terms in the expansion of
the integrand in powers of .
To define precisely the yi integration, consider, for example, the integration region for
the puncture z2 ,
1
z3
1
dz2

z2
t3
dt2 .
(5.2)
If we relax the condition of strong ordering of the punctures, we must explicitly consider
regions in which t2 or t3 t2 are O( ). In these regions the counting of powers of is
clearly different, as can be seen by writing
1

t3
0
1
dt2 =
t3c
b
t3
1
1
dt2 +
dt2 +
dt2 ,
b
(5.3)
t3 c
where b and c are arbitrary O(1) constants. The contribution of diagrams with three-point
vertices, discussed in the previous section, arises from integrals such as the second term
on the RHS of Eq. (5.3), where we pick the terms in the integrand that cancel the explicit
factor of 1/ , and thus we may neglect the dependence in the limits of integration. The
other two integrals in Eq. (5.3) do not have an overall power of , because the size of the
integration region vanishes with . Defining t2 = 1 in the first integral, and t3 t2 =
2 in the second integral (corresponding to yi = exp(i )) we obtain
1

t3
b
dt2 =
1
d1 +
t3c
c
dt2 +
b
109
d2 .
(5.4)
The b and c dependence arising from the upper limits of integration of the integrals
must cancel the dependence arising from the t2 integral when the limits of integration are
expanded in powers of . The correct treatment of the i integrations is thus to perform
the integral exactly, but retain only the contribution from the lower limit of integration,
i = 0, corresponding to yi = 1. Notice that there is one (and only one) integration region
of this kind for every propagator in the loop, and the maximum number of regions that can
simultaneously contribute is correctly given by the overall counting of powers of in the
string master formula, Eq. (2.6). Notice also that the choice of integration variables is built
in the algorithm, since we are retaining only the contribution of one limit of integration:
a generic change of variables, in particular a shift by a constant, will in fact redistribute the
result of the integration between the two limits.
There is a class of terms contributing to scalar graphs with four-point vertices
that requires regularization, as was already pointed out in Refs. [11] and [15]. These
contributions arise when one takes zi+1 close to zi in a term containing a double derivative
of the Green function precisely with respect to these variables. Such terms diverge as
zi+1 zi , i.e., yi 1, because of the double pole in i i+1 G(zi , zi+1 ). To regularize
this divergence, one may retain also the y dependence arising from the exponential of the
same Green function, which is formally suppressed in the 0 limit. Its easy to see that
the resulting integral is of the form
1
I c, pi pi+1 =
dy(1 y)2+2 pi pi+1 y pi pi+1 ,
(5.5)
ec
where c is the arbitrary constant defining the boundary of the appropriate integration
region. Notice that there is some freedom concerning the inclusion of y-dependent terms
arising from the exponential in this regularization prescription: in Eq. (5.5) we included
exactly only the y dependence arising from G(zi , zi+1 ), while all other nonsingular terms
(arising from, say, G(zj , zi+1 ), with j = i) were dropped, to be treated according to the
general rules outlined at the beginning of this section. As we will see in the next section,
the inclusion of further y-dependent terms in Eq. (5.5) is equivalent to a reshuffling of
one-particle-reducible and irreducible contributions to the amplitude.
The integral I (c, pi pi+1 ) is linearly divergent at the upper limit of integration
when 0, and requires regularization. It is comforting to verify that three independent
regularization schemes give the same result in the field theory limit.
The simplest approach is the one that was adopted in Refs. [11] and [15]. One can set
0 ab initio in Eq. (5.5), expand the integrand in powers of y, and integrate the
series term by term, obtaining
1
I (c, 0) =
ec
dy
=
(1 y)2
1
ec
dy
n
n 1
ny =
y .
y
ec
n=1
n=1
(5.6)
110
The lower limit contribution depends on c, so it must be discarded, as explained

before. The remaining sum can be interpreted as a -function, yielding
n=1
1 = lim
s0
n=1
1
ns = (0) = .
2
(5.7)
This -function regularization was used in [15] also at the two-loop level, giving the
correct field theory limit.
Perhaps a more natural way to approach the regularization of Eq. (5.5) is to view
it as an incomplete B function. As we shall see, this viewpoint nicely ties together,
diagrams with four-point vertices with reducible diagrams. Since our answer must be
independent of c, we may turn the integral into an ordinary B function by taking the
c limit. Considering for simplicity a contribution to the two-point amplitude,
and thus replacing pi pi+1 with p2 , we find

I , p2 = B 1 2 p2 , 1 + p2 .
(5.8)
After analytic continuation we get I (, 0) = 1/2, exactly as in Eq. (5.7).
Finally, at one loop, these results are confirmed by integration by parts of double
derivatives of the Green function, at the level of the string master formula, as was
done in [3,11]. After integration by parts, the correct field theory limit is obtained by
considering only scalar graphs with cubic vertices, so that effectively the singularity
we are studying is removed. The regularization given by Eq. (5.7) turns out to be the
correct prescription to recover the same results without partial integration.
Our working prescription will then be

1
I c, p2 .
(5.9)
2
To illustrate the application of these rules, consider the diagram in Fig. 4, with off-shell
and non-transverse gluons. To verify the correctness of our prescription, Eq. (5.9), we can
focus on a specific kinematic structure, say 1 2 3 p4 4 p2 . 6 Setting z2 = y in the
Fig. 4. Four-point diagram with a quartic vertex: external gluons are background fields, marked by
an A. Momenta pi are incoming.
6 Notice that, in order to identify univocally the coefficients, we write p = p p p and
3 3
3 1
3 2
3 4
4 p4 = 4 p1 4 p2 4 p3 . The string master formula, in general, automatically generates the answer

with this convention, which yields a particularly symmetric expression.
111
master formula, Eq. (2.6) (recall that z1 = 1), we obtain

A(1) (p1 , p2 , p3 , p4 )

8N Tr 4 3 2 1
gd4
(1 2 )(3 p4 )(4 p2 )
(4)d/2

t
t4
d/2
dt4 dt3
dt t
0

s12 t3 (t t4 ) p32 t3 (t4 t3 ) p42 (t4 t3 )(t t4 )
exp
t

1
1
y
(2t4 t)(t 2t4 + 2t3 ) dy
+
y
+
(d
2)
,
4t 2
y
y
(1 y)2
(5.10)
where s12 = 2p1 p2 , and we wrote only the relevant limit of integration in y. The term
proportional to (1/y + y) integrates to zero, while for the term proportional to y/(1 y)2
we must use Eq. (5.9). The result for the contribution of the chosen kinematic structure to
the scalar graph containing Fig. 4 is
A(1) (p1 , p2 , p3 , p4 )

N Tr 4 3 2 1
gd4
(1 2 )(3 p4 )(4 p2 )
(4)d/2

1
3
d/2+2
dx1 dx2 dx3 1
xj
dt t
0
j =1

exp t s12 x1 x3 p32 x1 x2 p42 x2 x3 (2 d)(1 2x2 )(1 2x3) . (5.11)
The result is again in agreement with the computation in the background field Feynman
gauge, including the contribution of ghosts.
6. Reducible diagrams
Reducible diagrams (such as the one portrayed in Fig. 5) arise from the string master
formula when two adjacent punctures, say zi and zi+1 , are so close to each other on the
world-sheet that the corresponding correlator, G(zi , zi+1 ), is dominated by its logarithmic
short distance singularity even before the field theory limit is taken. In this integration
corner, often called pinching region [3], the string master formula effectively factorizes
into the product of a one-loop amplitude with an off-shell leg, times a tree-level amplitude.
All string states may flow through the propagator joining the two factors; the YangMills
field theory limit is reached by selecting the poles corresponding to the propagation of
massless states.
As we will see, reducible diagrams are the only diagrams where it is necessary to require
that a subset of the external legs (those forming the tree that attaches to the irreducible loop)
112
Fig. 5. Reducible four-point diagram: external gluons are background fields, marked by an A.
Momenta pi are incoming.
be on-shell. This however is also a feature of diagrammatic computations in field theory

when using the background field method [26]: there, reducible contributions to a scattering
amplitude only coincide with the ones obtained with conventional methods provided all
external legs on at least one side of the reducible propagator are put on shell.
The other relevant feature of the computation of reducible diagrams in the background
field method, namely the possibility of using different gauges for the tree and the oneloop subamplitudes, is also preserved in string theory. In fact, our analysis will confirm the
results obtained in [9] for on-shell diagrams and in [11] for the divergent part of off-shell
diagrams with transverse gluons: the 1PI part of the diagram is expressed in the background
field Feynman gauge, as expected from the previous sections; the tree part is computed by
the string in a different gauge, the GervaisNeveu gauge.
The GervaisNeveu gauge is a non-linear gauge, introduced to analyze the field theory
limit of open string theory [27]. The gauge-fixed action, ignoring ghosts, is:

1 2
GN
4
2 2
.
S = d x Tr F Tr A igA
(6.1)
2
The Feynman rules can be found, for example, in [5]. At tree level, the vertices are
simpler than those of the usual Feynman gauge, allowing efficient calculations. At one
loop, the string still exploits this advantage in the tree part of the diagrams, but avoids the
complications that ghost terms would generate if the gauge were used also for the one-loop
part of the diagrams.
Reducible contributions can be extracted from the string master formula by considering
the region of moduli space where zi zi+m k 0 [3,11]. To start with, we consider
the case i = m = 1, where, as in the previous section, z2 z1 = 1. In this limit the string
amplitude naturally factorizes in a sum over poles which correspond to the propagation of
different string states in a reducible leg connecting two separate parts of the diagram.
We have already studied the contribution of the first term of this sum, which is related to
the exchange the tachyonic ground state. With either of the two regularizations proposed,
(5.7) or (5.8), one recovers contributions to the irreducible diagram depicted in Fig. 4.
The basic idea of the pinching procedure adopted in [3,11] is to associate the
contribution due to the exchange of massless states to reducible diagrams. In order to select
the appropriate pole for the massless exchange one has to perform a Taylor expansions in
113
zi+j around zi of all string Green functions (3.4). However, for the pinched legs, which
are isolated on a reducible tree, we use an effective Green function determined only by
the short distance behavior of the string expression, which is equal to the tree-level Green
function,
GE (zi , zj ) = log |zi zj |.
(6.2)
As an example, let us consider the reducible diagram in Fig. 5, where we have to assume
that legs 1 and 2 are on shell, as explained above. Taking the limit z2 1 by means
of Eq. (6.2), we observe that the exponential involving G(z1 , z2 ) yields a factor of

(1 z2 ) s12 , while further factors of (1 z2 ) can be generated by expanding the other
Green functions and their derivatives around z2 = 1; for example,

1 z32
log z3
1 + z3
z2 1
+
+k
G32 G31 + (1 z2 )
,
2(1 z3 )
log k
z3

(1 + z32 )
1
1
z2 1
k
+
,
3 G32 3 G31 + (1 z2 )
(1 z3 )2 z3 log k
z32

2
1
z2 1
2k
,
2 3 G32 2 3 G32 z =1 + (1 z2 )
(6.3)
2
(1 z3 )3 z3 log k
where terms of order O((1 z2 )2 ) and O(k 2 ) have been neglected. Clearly, Taylorexpanding in powers of (1 z2 ) generates a series of poles in the s12 channel; in fact,
the general integral in z2 is of the form
1
dz2 (1 z2 ) s12 +a1 ,
(6.4)
z3
with integer a. Since z3 0 exponentially as 0, we can evaluate the integral

(possibly by analytic continuation in the external momenta) as
1
dz2 (1 z2 ) s12 +a1 = (a s12 )1 .
(6.5)
Different values of a correspond to different mass eigenstates propagating in the s12

channel. For example, a = 1 corresponds to tachyon propagation, which we now discard,
while the expected gluon pole 1/s12 is given by a = 0. Notice that, as announced, we
have to require p1 and p2 to be on shell in order to rewrite 2p1 p2 as s12 . Having
isolated the massless pole, one may apply the procedure illustrated in Sections 3 and 4
(only cubic vertices are connected to the loop in this case) and get the correct expression
for the diagram in Fig. 5. For example, selecting the term (1 2 )(3 p4 )(4 p2 ), one
finds
A(1) (p1 , p2 , p3 , p4 )

gd4 N Tr 4 3 2 1
2
(1 2 )(3 p4 )(4 p2 )
(4)d/2
114
dt t
d/2+2
1
dx1 dx2 dx3 1

xj
j =1

exp t s12 x1 x3 p32 x1 x2 p42 x2 x3

1
2(d 2)(1 2x2)t 1 8 p1 p3 + p1 p4 (1 2x2)
s12

+ (d 2)(1 2x2 )(1 2x3 ) p2 p3 (1 2x1 ) p2 p4 (1 2x3) .
(6.6)
The approach just described has a simple diagrammatic interpretation and, in particular,
it has the advantage of clearly separating reducible and irreducible contributions. However,
the use of an ad-hoc Green function, Eq. (6.2), may appear arbitrary and not completely
justified at the string level. Here we want to outline also an alternative method for
computing reducible diagrams, which is a natural extension of our rules for the calculation
of diagrams with four-point vertices. As we will show, in this case one does not have
to introduce any effective Green functions. On the contrary, it is sufficient to introduce
the parameter y = z2 /z1 , as in Eq. (5.1), and to carefully study the region y = 1. Using
this second method, diagrams with quartic vertices and reducible diagrams are computed
simultaneously.
The starting point of the alternative method is a natural generalization of the regularization (5.8) presented in Section 5. In particular, one has to retain all logarithmic dependence
on y in the exponential containing the sum of Green functions. In this way, one finds a
term of the form

t3
Y(y) = (1 y)2 p1 p2 y p1 p2 + p2 p3 (12 t
)+ p2 p4 (12 t4 )
t
(6.7)
which gives a natural regularization of the tachyonic divergence (1 y)2 . The ts above are
the Schwinger parameters related in the usual way to the string variables: t3 = log z3 ,
t4 = log z4 and t = log k. As explained in Section 5, for regular terms we perform
the y integral exactly, keeping only the contribution of the y = 1 limit of integration. In
this case, the factor (6.7) has no effect on the integration. For singular terms, we obtain
a generalization of Eq. (5.5). In particular, the tachyonic double pole at y 1 gives the
contribution
1
dy (1 y)
ec
1
Y(y)
dy (1 y)2 Y(y)

t3
= B 1 + 2 p1 p2 , 1 p1 p2 + p2 p3 1 2
t

t
4
+ p2 p4 1 2
t

1
1
t3
t4
=
p2 p3 1 2
+ p2 p4 1 2
+ O( ).
2 2p1 p2
t
t
(6.8)
115
Of course, Eq. (6.8) is just a factor of the amplitude; to recover the complete result one
has to include the appropriate measure of integration and the exponential factor appearing
in Eq. (5.10), which encode the contribution coming from the punctures not involved in
the pinching procedure. It is easy to see that the first term reproduces the prescription we
introduced in Eq. (5.9) to get the irreducible diagram with a quartic vertex, Fig. 4; the other
terms, containing a pole in 2p1 p2 , contribute to the reducible configuration, Fig. 5.
Simple poles in (1 y) must be treated in exactly the same way, obtaining again
1
dy(1 y)1 Y(y) =
ec
1
2 p1
p2

+ O ( )0 ,
(6.9)
which is another contribution to the reducible diagram. Notice the negative power of in
Eq. (6.9): it tells us that we need to consider also contributions of order , obtainable from
the expansion of the exponential, as outlined in Eq. (3.3); furthermore, one must expand
around y = 1 all the terms multiplying Eq. (6.7), to be sure to get all terms of the type (6.9).
The difference with the previous approach is that here we kept the exact dependence on
y = z2 in all exponentiated Green functions and no expansion in y was done in the terms
proportional to log(y), present, for instance, in G23 . Because of this, the contribution of
the simple pole is now different from the one of Eq. (6.6) and in particular the last term
proportional to (d 2) is absent. The complete result is recovered only once one takes into
account the reducible contribution coming from the new regularization (6.8).
The first method is maybe preferable from the computational point of view, because it
preserves the one-to-one correspondence between string regions and Feynman graphs. The
second method, on the other hand, is interesting because it ties together the prescription
for quartic vertices, Eq. (5.9), and the one for reducible diagrams.
It is not difficult to generalize the method we just presented to more complicated
situations where the pinching procedure involves two non-consecutive legs, like zi and
zi+m . In this case, the role of the variable y is taken by the combination zi+m /zi . Moreover,
one can introduce the variables l = (zi+l zi+m )/(zi zi+m ) for l = 1, 2, . . . , m 1,
which factorize the expression of the amplitude in two separate parts. The term depending
only on the s has the functional form of a tree-level amplitude and describes the legs
isolated in the reducible part, while the other term of the amplitude can be treated as usual
and gives the contribution of the irreducible part.
7. Systematics
We have completed the list of all regions of string moduli space contributing to the Yang
Mills field theory limit of the string master formula. Each region corresponds to a scalar
graph of a given topology, and we have given a procedure to compute the corresponding
contribution to the field theory amplitude in each case. We emphasize that all steps in
the computation may be automated, in a manner essentially independent of the number
of external gluons. In the special, but phenomenologically most relevant case of on-shell
116
scattering amplitudes, the steps of the calculation of a given one-loop multigluon amplitude
can be summarized as follows.
Use color and helicity decomposition techniques to reduce the computation of the full
amplitude to its basic building blocks, color-ordered fixed-helicity subamplitudes. For
each independent subamplitude, choose polarization vectors to minimize the number
of non-zero scalar products of polarizations and momenta.
List all scalar graphs contributing to the full amplitude, i.e., all one-loop graphs
that would contribute to the corresponding amplitude in a scalar theory with cubic
and quartic vertices. Because the answer will be given in dimensional regularization,
graphs that vanish in that scheme (such as tadpoles and bubbles on massless external
legs) need not be included.
Each subamplitude can now be written as a sum over scalar graphs,
(1)
A{i },C (1, . . . , M)

gdM
=
(4)2

dt t
0
ng

1
2+
0 i=1

dxi 1
ng

xj
j =1

( )
exp tSg (xi ; pi pj ) Rg t, xi ; pi , i i ,
(7.1)
where the subscript C denotes the color structure whose coefficient we are computing,
and ng is the number of propagators in the scalar graphs g contributing to the
amplitude. The dynamical information is encoded in the functions Sg and Rg , which
can be computed according to the rules given in the previous sections. Specifically,
Sg is the same function of external momenta and Schwinger parameters that would
appear in a scalar theory for a graph with the same topology; thus, it can be easily
computed in field theory as well as derived from string theory. The non-trivial
content of YangMills subamplitudes is given by the function Rg , which is a sum of
polynomials in the dimensionless Schwinger parameters xi , multiplied times integer
powers of t. As a consequence, the t integration can always be done at the outset,
yielding a function. Using helicity techniques from the beginning will, in general,
greatly simplify the function Rg , which in fact will vanish for many graphs for each
subamplitude.
Perform Schwinger parameter integrals. At this stage, string theory can offer no
further help; techniques have been developed [28], however, that are sufficient to
compute all one-loop scalar integrals that might be needed for YangMills amplitudes
in the foreseeable future.
The method just outlined allows, as is well known, to take advantage of the best
organization of the amplitude in terms of color and helicity decompositions. The
dimensional regularization scheme which naturally emerges in this approach is the socalled t HooftVeltman scheme [29]: polarizations of external, observed gluons are in four
dimensions, while polarizations of gluons circulating in the loop or otherwise unobserved
are in d = 4 2 dimensions, as indicated by the factor (1 + k(d 2)) of Eq. (4.3). It
should perhaps be further emphasized that a major practical advantage for amplitudes with
117
Fig. 6. Six-point diagram: external gluons are background fields, marked by an A. Momenta pi are
incoming.
several gluons is the fact that loop momentum integration has already been performed. To
illustrate this fact, consider the six-gluon diagram in Fig. 6 (with all gluons on-shell and
transverse).
Using Feynman rules and introducing Schwinger parameters, this diagram by itself
generates approximately 104 terms, each of which must still be integrated over the loop
momentum k. Suppose we focus on a particular kinematical structure, say 1 5 2 6 3
p5 4 p5 : 7 for most of the original 104 terms, this kinematical structure is generated only
upon loop momentum integration, so while we can postpone integration over Schwinger
parameters, we must integrate all terms in k. Methods to perform these integrals are wellknown [23], but they generate larger and larger expressions as the power of k grows; in the
present case, each contribution with 6 powers of k can generate up to 6 105 terms. When
all is said and done, the coefficient of the chosen kinematical structure must be picked out
of more than 4 106 terms, and that applies to just one diagram. Clearly, this is a time and
memory consuming task.
The string technique avoids all this. The change of variables identifying the field
theory limit for a given scalar graph is known a priori, and all subsequent computerized
manipulations involve only algebraic operations and series expansions. Using a symbolic
manipulation program, such as Mathematica, on a normal PC, the Feynman parameter
integrand for a given scalar graph, such as the one containing Fig. 6 and the corresponding
ghost loops, can be obtained in seconds. In this case, for example, computing the coefficient
of the chosen kinematic structure in the color-ordered amplitude multiplying Tr(6 1 )
yields
Sg (xi ; pi pj ) = 2p1 p2 x2 x6 + 2p2 p3 x1 x3 + 2p3 p4 x2 x4 + 2p4 p5 x3 x5
+ 2p5 p6 x4 x6 + 2p6 p1 x1 x5 + (p1 + p2 + p3 )2 x3 x6
+ (p2 + p3 + p4 )2 x1 x4 + (p3 + p4 + p5 )2 x2 x5 ,
(7.2)
7 Notice that for transverse gluons not all products p for fixed i, i = j , are independent. To pick
i
j
a basis, we choose to eliminate the products 3 p6 and 4 p6 using momentum conservation. The
kinematic structure considered is present, for example, in the six-gluon amplitudes with helicities fixed as
(p1 , p4 ), (p2 , p4 ), (p3 , p4 ), + (p4 , p3 ), + (p5 , p3 ), + (p6 , p3 ).
118
and
Rg (t, xi ; pi , i )

= 16t 3 x5 (d 2)x5 2t (p1 p3 + p1 p4 p2 p3 p2 p4 )

4tx5 (p1 p5 + p2 p6 ) 1 5 2 6 3 p5 4 p5 + .
(7.3)
Needless to say, the result in Eq. (7.3) agrees with the corresponding background field
method calculation.
We have completed a systematic analysis detailing how the one-loop string moduli space
splits into particle graphs in the field theory limit. For each class of graphs, a specific
-dependent change of variables has been introduced, which parametrizes the limiting
process to be implemented on the integrand of the string master formula. We have given a
geometric prescription for the choice of projective transformations associated with external
particles, so that our formulas apply to off-shell, arbitrarily polarized gluons, and can
be used directly to compute general gluon correlation functions, in a manner which is
completely alternative to, though much more efficient than conventional Feynman rules,
at least at one loop. Singularities arising at the boundaries of different regions of moduli
space have been consistently regulated, so as to avoid double counting. The method as
it stands can be implemented in computer code using available symbolic manipulation
programs. Such a program would be able to compute one-loop YangMills amplitudes and
gluon correlation functions, starting from the string master formula, provided the relevant
Schwinger parameter integrals are known in dimensionally regularized form.
An obvious application of the method is the computation of one-loop scattering
amplitudes with several gluons, which are relevant for next-to-leading order corrections
to multijet production at hadron colliders. Such amplitudes are known for up to five
gluons [30]; the six-gluon amplitude remains quite challenging even with the present
technology, but it is a natural and interesting testing ground; in particular, it could be used
to construct finite predictions for NLO partonic cross sections, since the seven-gluon treelevel amplitude is known in closed form [31].
Other possible applications arise from the fact that the method can be applied off shell.
At tree level, amplitudes with one off-shell gluon are related to currents [32], which can be
exploited to establish recursion relations for the calculation of on-shell matrix elements.
For certain classes of diagrams, these recursion relation have been generalized to one
loop [33], but the complete extension including diagrams with gluon loops is not known.
The present formalism would appear to be a natural framework to pursue this kind of
generalization. Other areas of interest include the study of the high-energy, collinear and
infrared limits of the amplitudes. For example, since one can now use string theory to
compute matrix elements uncontracted with polarization vectors, it would be interesting
to attempt a stringy derivation of the soft gluon current, recently derived at one loop by
conventional methods [34].
119
Finally, we believe that the present systematic study is a necessary step for the extension
of the method to more than one loop. As was shown in the case of scalar field theories,
at two loops the decomposition of string moduli space into particle graphs is much more
intricate than at one loop, even when considering only diagrams with cubic vertices [13].
A thorough understanding of the one-loop case on a graph by graph basis, including the
non-trivial case of diagrams with four point vertices, will be of considerable help in the
study of more complicated graph topologies.
Acknowledgements
We would like to thank C. Schubert for several useful observations. R.R. acknowledges
the support of the European Commission RTN programme HPRN-CT-2000-00131 and
of the Physics Department of the University of Neuchtel, where part of this work was
carried out. L.M. acknowledges the support of the European Commission TMR programme
FMRX-CT98-0194 (DG 12-MIHT). Work supported in part by the Italian Ministero
dellUniversit e della Ricerca Scientifica.
References
[1] A. Neveu, J. Scherk, Nucl. Phys. B 36 (1972) 155.
[2] M.L. Mangano, S.J. Parke, Phys. Rep. 200 (1991) 301.
[3] Z. Bern, D.A. Kosower, Phys. Rev. D 38 (1988) 1888;
Z. Bern, D.A. Kosower, Phys. Rev. Lett. 66 (1991) 1669;
Z. Bern, D.A. Kosower, Nucl. Phys. B 379 (1992) 451.
[4] Z. Bern, Phys. Lett. B 296 (1992) 85.
[5] Z. Bern, in: J. Harvey, J. Polchinski (Eds.), Proceedings of TASI 92, hep-ph/9304249;
Z. Bern, L. Dixon, D.A. Kosower, Ann. Rev. Nucl. Part. Sci. 46 (1996) 109, hep-ph/9602280.
[6] For a review, see, C. Schubert, Perturbative quantum field theory in the string-inspired
formalism, preprint LAPTH-761/99, to appear in Phys. Rep., and references therein.
[7] L. Cappiello, R. Marotta, R. Pettorino, F. Pezzella, Mod. Phys. Lett. A 13 (1998) 2433, hepth/9804032;
L. Cappiello, R. Marotta, R. Pettorino, F. Pezzella, Mod. Phys. Lett. A 13 (1998) 2845, hepth/9808164;
A. Liccardo, R. Marotta, F. Pezzella, Mod. Phys. Lett. A 14 (1999) 799, hep-th/9903027;
F. Cuomo, R. Marotta, F. Nicodemi, R. Pettorino, F. Pezzella, G. Sabella, hep-th/0011071.
[8] O. Andreev, H. Dorn, Nucl. Phys. B 583 (2000) 145, hep-th/0003113;
A. Bilal, C. Chu, R. Russo, Nucl. Phys. B 582 (2000) 65, hep-th/0003180;
C. Chu, R. Russo, S. Sciuto, Nucl. Phys. B 585 (2000) 193, hep-th/0004183.
[9] Z. Bern, D.C. Dunbar, Nucl. Phys. B 379 (1992) 562.
[10] P. Di Vecchia, A. Lerda, L. Magnea, R. Marotta, Phys. Lett. B 351 (1995) 445, hep-th/9502156.
[11] P. Di Vecchia, L. Magnea, A. Lerda, R. Russo, R. Marotta, Nucl. Phys. B 469 (1996) 235,
hep-th/9601143.
[12] P. Di Vecchia, L. Magnea, A. Lerda, R. Marotta, R. Russo, Phys. Lett. B 388 (1996) 65, hepth/9607141.
[13] A. Frizzo, L. Magnea, R. Russo, Nucl. Phys. B 579 (2000) 379, hep-th/9912183.
120
[14] R. Marotta, F. Pezzella, Phys. Rev. D 61 (2000) 106006, hep-th/9912158;

See also R. Marotta, F. Pezzella, hep-th/0003044.
[15] L. Magnea, R. Russo, in: J. Repond, D. Krakauer (Eds.), Proceedings of DIS 97, Chicago, USA,
1997, AIP Conf. Proc., Vol. 407, p. 913, hep-ph/9706396;
Also in: G. Eigen, P. Osland, B. Stugu (Eds.), Proceedings of Beyond the Standard Model V,
Balholm, Norway, 1997, AIP Conf. Proc., Vol. 415, p. 347, L. Magnea, R. Russo, hepph/9708471.
[16] B. Kors, M.G. Schmidt, hep-th/0003171.
[17] J. Gomis, M. Kleban, T. Mehen, M. Rangamani, S. Shenker, JHEP 0008 (2000) 011, hepth/0003215.
[18] See, for example, P. Di Vecchia, Multiloop amplitudes in string theory, Theor. Phys. (1992) 16,
and references therein.
[19] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, Cambridge University Press, 1987.
[20] P. Di Vecchia, F. Pezzella, M. Frau, K. Hornfeck, A. Lerda, S. Sciuto, Nucl. Phys. B 322 (1989)
317.
[21] K. Roland, Phys. Lett. B 289 (1992) 148.
[22] L.F. Abbott, Nucl. Phys. B 185 (1981) 189.
[23] G. Passarino, M. Veltman, Nucl. Phys. B 160 (1979) 151.
[24] W.L. van Neerven, J.A.M. Vermaseren, Phys. Lett. B 137 (1984) 241.
[25] Z. Bern, D.A. Kosower, Nucl. Phys. B 362 (1991) 389.
[26] L.F. Abbott, M.T. Grisaru, R.K. Schaefer, Nucl. Phys. B 229 (1983) 372.
[27] J.L. Gervais, A. Neveu, Nucl. Phys. B 46 (1972) 381.
[28] Z. Bern, L. Dixon, D.A. Kosower, Nucl. Phys. B 412 (1994) 751, hep-ph/9306240;
T. Binoth, J.Ph. Guillet, G. Heinrich, Nucl. Phys. B 572 (2000) 361, hep-ph/9911342.
[29] G. t Hooft, M. Veltman, Nucl. Phys. B 44 (1972) 189;
S. Catani, M.H. Seymour, Z. Trcsnyi, Phys. Rev. D 55 (1997) 6819, hep-ph/9610553.
[30] Z. Bern, L. Dixon, D.A. Kosower, Phys. Rev. Lett. 70 (1993) 2677, hep-ph/9302280.
[31] R. Kleiss, H. Kuijf, Nucl. Phys. B 312 (1989) 616.
[32] F.A. Berends, W. Giele, Nucl. Phys. B 306 (1988) 759.
[33] G. Mahlon, Phys. Rev. D 49 (1994) 4438, hep-ph/9312276;
G. Mahlon, in: Beyond the Standard Model 4, Lake Tahoe, 1994, p. 475, hep-ph/9412350.
[34] S. Catani, M. Grazzini, Nucl. Phys. B 591 (2000) 435, hep-ph/0007142.

Noncommutative gauge theory on fuzzy sphere

from matrix model
Satoshi Iso a,c , Yusuke Kimura a,b , Kanji Tanaka a,c ,
Kazunori Wakatsuki a,d
a High Energy Accelerator Research Organization (KEK), Tsukuba, Ibaraki 305-0801, Japan
b Department of Physics, Tokyo Institute of Technology, Oh-okayama, Meguro-ku, Tokyo 152, Japan
c Department of Particle and Nuclear Physics, The Graduate University for Advanced Studies,
Tsukuba, Ibaraki 305-0801, Japan

d Department of Physics, Tokyo Metropolitan University, Hachiohji, Tokyo 192-0397, Japan
Received 2 February 2001; accepted 4 April 2001
Abstract
We derive a noncommutative U (1) and U (n) gauge theory on the fuzzy sphere from a threedimensional matrix model by expanding the model around a classical solution of the fuzzy sphere.
ChernSimons term is added in the matrix model to make the fuzzy sphere as a classical solution
of the model. Majorana mass term is also added to make it supersymmetric. We consider two large
N limits, one corresponding to a gauge theory on a commutative sphere and the other to that on
a noncommutative plane. We also investigate stability of the fuzzy sphere by calculating one-loop
effective action around classical solutions. In the final part of this paper, we consider another matrix
model which gives a supersymmetric gauge theory on the fuzzy sphere. In this matrix model, only
ChernSimons term is added and supersymmetry transformation is modified. 2001 Published by
Elsevier Science B.V.
PACS: 02.10.Sp; 02.40.-k; 04.60.Nc; 04.62.+v
1. Introduction
To formulate nonperturbative aspects of string theory or M-theory, several kinds of
matrix models have been proposed [1,2]. These proposals are based on the developments
of D-brane physics [3,4]. IIB matrix model is one of these proposals [1]. It is a large N
reduced model [5] of ten-dimensional supersymmetric YangMills theory and the action
has a matrix regularized form of the GreenSchwartz action of IIB superstring. It is
postulated that it gives a constructive definition of type IIB superstring theory.
E-mail addresses: satoshi.iso@kek.jp (S. Iso), kimuray@post.kek.jp (Y. Kimura), kanji@post.kek.jp
(K. Tanaka), waka@post.kek.jp (K. Wakatsuki).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 7 3 - 0
122
S. Iso et al. / Nuclear Physics B 604 (2001) 121147
In the matrix model, matter and even spacetime are dynamically emerged out
of matrices [6,7]. Spacetime coordinates are represented by matrices and, therefore,
noncommutative geometry appears naturally. The idea of the noncommutative geometry is
to modify the microscopic structure of the spacetime. This modification is implemented by
replacing fields on the spacetime by matrices. It was shown [811] that noncommutative
YangMills theories in a flat background are obtained by expanding the matrix model
around a flat noncommutative background. The noncommutative background is a D-branelike background which is a solution of the equation of motion and preserves a part of
supersymmetry. Various properties of noncommutative YangMills have been studied from
the matrix model point of view [12]. In string theory, it is discussed that the world volume
theory on D-branes with NSNS two-form background is described by noncommutative
YangMills theory [13].
A different kind of noncommutative backgrounds, a noncommutative sphere, or a fuzzy
sphere is also studied in many contexts. In [14], it is discussed in the framework of matrix
regularization of a membrane. In the light-cone gauge, they gave a map between functions
on spherical membrane and hermitian matrices. In BFSS matrix model [2], membranes of
spherical topology are considered in [2325]. A noncommutative gauge theory on a fuzzy
sphere in string theory context is discussed in [26,27]. The approach to construct a gauge
theory on the fuzzy sphere were pursued in [1522].
The fuzzy sphere [15] can be constructed by introducing a cut off parameter N for
angular momentum of the spherical harmonics. The number of independent functions is
N
2
l=0 (2l + 1) = (N + 1) . Therefore, we can replace the functions by (N + 1) (N + 1)
hermitian matrices on the fuzzy sphere. Thus, the algebra on the fuzzy sphere becomes
noncommutative.
In this paper, to construct a noncommutative gauge theory on the fuzzy sphere, we first
consider a three-dimensional supersymmetric reduced model. We add ChernSimons term
and Majorana spinor mass term to the action of original supersymmetric matrix model.
An action with both of YangMills and ChernSimons terms are also considered in [27].
Although an ordinary matrix model has only a flat background as a classical solution, our
matrix model can describe a curved background owing to these terms.
We also study another type of action containing ChernSimons term but not Majorana
mass term. Although this action is not invariant under the original supersymmetry, it is
invariant under a modified supersymmetry. It is also invariant under a constant shift of the
fermion and hence it has N = 2 supersymmetry.
This paper is organized as follows. In Section 2, we study a matrix model which gives
a noncommutative gauge theory on a fuzzy sphere. We derive a noncommutative gauge
theory by expanding the model around a classical solution which represents a fuzzy sphere.
It is shown that in a commutative limit, it gives a standard theory on a commutative sphere.
In the latter part of this section, another large N limit corresponding to a sphere with
a large radius but the noncommutativity fixed is considered. By taking the limit, we can
obtain a noncommutative gauge theory in a flat noncommutative background, which is
considered in [8]. In Section 3, some properties of Dirac operator and chirality operator in
this model are considered. These operators become the correct operators on a commutative
123
sphere in a commutative limit. In Section 4, stability of two clssical solutions, diagonal

matrices and the fuzzy sphere, is examined. One-loop effective action is calculated for the
classical solutions. For diagonal matrices, all supersymmetry is preserved. Hence, oneloop effective action vanishes and eigenvalues can move freely. On the other hand, all
supersymmetry is broken for the fuzzy sphere. Therefore, one-loop effective action does
not vanish. It is shown that the value of the action including the one-loop corrections of
the fuzzy sphere is lower than that of the commuting matrices when N is sufficiently
large for fixed gYM . Another matrix model is analyzed by the same manner as the case
of a former model in Section 5. While in the former model the fuzzy sphere does not
preserve supersymmetry, in this model it is a BPS solution. Therefore we can obtain
a supersymmetric noncommutative gauge theory on a fuzzy sphere by expanding this
model around a fuzzy sphere. Section 6 is devoted to conclusions and discussions. In
Appendix A, a noncommutative product on a fuzzy sphere is constructed by following [14].
In Appendix B, by projecting a fuzzy sphere to a fuzzy complex plane stereographically
using a coherent state approach developed by [3133], a noncommutative product which
corresponds to the normal ordered product is considered. We also give mapping rules from
matrix models to field theories on the projected plane. This construction leads to field
theories with the Berezin type noncommutative product instead of the Moyal type.
2. Noncommutative gauge theory on fuzzy sphere

We start with a three-dimensional reduced model which is defined by the following
action,

2

1
1
1
i Ai , + . (1)
S = 2 Tr Ai , Aj Ai , Aj + iij k Ai Aj Ak +
g
4
3
2
Reduced models are obtained by reducing the spacetime volume to a single point [5].
We have added ChernSimons term and Majorana mass term to the reduced model of
supersymmetric YangMills [1]. Ai and are (N + 1) (N + 1) hermitian matrices, and
is a three-dimensional Majorana spinor field. i (i = 1, 2, 3) denote Pauli matrices. is
a dimensionfull parameter which depends on N .
This model possesses SO(3) symmetry and the following translation symmetry
Ai Ai + c1.
(2)
Gauge symmetry of this model is expressed by the unitary transformations,

Ai U Ai U ,
U U .
(3)
In addition to the above symmetries, this model has N = 1 supersymmetry:

Ai = i i ,

i
Ai , Aj ij .
2
Because of the fermionic mass term, translation symmetry of is not present.
(4)
124
We next consider the equation of motion of (1). When = 0, it is

Ai , [Ai , Aj ] = ij kl [Ak , Al ].
(5)
The simplest solution is realized by commuting diagonal matrices:

Ai = diag xi(N+1) , xi(N) , . . . , xi(1) .
(6)
Another solution represents an algebra of the fuzzy sphere:

xi , xj = iij k xk .
(7)
We impose the following condition for xi ,

x1 x1 + x2 x2 + x3 x3 = 2 .
(8)
In the 0 limit, xi becomes commutative coordinates xi :

x1 = sin cos ,
x2 = sin sin ,
x3 = cos ,
(9)
where denotes the radius of the sphere. Although the commuting matrices preserve the
supersymmetry, the solution representing the fuzzy sphere breaks it. The noncommutative
coordinates of (7) can be constructed by the generators of the (N + 1)-dimensional
irreducible representation of SU(2)
xi = L i ,
where
(10)

L i , L j = iij k L k .
(11)
and are related by the following relation

N(N + 2)
,
(12)
4
where we have used the fact that the quadratic Casimir of SU(2) in the (N +1)-dimensional
irreducible representation is given by N(N + 2)/4. The Plank constant, which represents
the area occupied by the unit quantum on the fuzzy sphere, is given by
2 = 2
N(N + 2) 2
4 2
=
.
N +1
N +1
Commutative limit is realized by
= fixed,
( 0).
(13)
(14)
From now on, we set = fixed.

Now we show that, expanding the model around the classical backgrounds (7) by the
similar procedure as in [8,9] it leads to a noncommutative YangMills on the fuzzy sphere.
We first consider U (1) noncommutative gauge theory on the fuzzy sphere. We expand the
bosonic matrices around the classical solution (7):

Li
+ a i .
Ai = xi + a i =
(15)
125
A correspondence between matrices and functions on a sphere is given as follow. Ordinary

functions on the sphere can be expanded by the spherical harmonics,
a() =

l

() =
alm Ylm (),
l=0 m=l
where
Ylm = l

l

lm Ylm (),
(16)
l=0 m=l
fa(lm)
x a1 x al ,
1 ,a2 ,...,al
(17)
is a spherical harmonics and fa1 ,a2 ,...,al is a traceless and symmetric tensor. The traceless
condition comes from xi xi = 2 . The normalization of the spherical harmonics is fixed by

d
1
Y Ylm =
4 l m
4
sin d Yl m Ylm = l l m m .
(18)
Matrices on the fuzzy sphere, on the other hand, can be expanded by the noncommutative
lm as
spherical harmonics Y
a =
N
l

lm ,
alm Y
l=0 m=l
N
l

lm .
lm Y
(19)
l=0 m=l
lm is a (N + 1) (N + 1) matrix and defined by

Y

lm = l
Y
falm
x a1 x al ,
1 ,a2 ,...,al
(20)
where the same coefficients as (17) are used. Angular momentum l is bounded at l = N
lm s form a complete basis of (N + 1) (N + 1) hermitian matrices. From

and these Y
the symmetry of the indices, the ordering of x corresponds to the Weyl type ordering.
=a
A hermiticity condition requires that alm
lm . Normalization of the noncommutative
spherical harmonics is given by

1
lm = l l m m .
Tr Y
m
l
N +1
A map from matrices to functions is given by
a =
N
l

lm a() =
alm Y
l=0 m=l
N
l

(21)
alm Ylm (),
(22)
l=0 m=l
and correspondingly a product of matrices is mapped to the star product on the fuzzy
sphere:
a b a $ b.
(23)
Detailed structures of the star product are summarized in Appendix A.

An action of Ad(L 3 ) is obtained by

lm =
lm .
Ad L 3 a =
alm L 3 , Y
alm mY
lm
lm
(24)
126
This property and SO(3) symmetry gives the following correspondence:

1
Ad L i Li ij k xj k .
i
The laplacian on the fuzzy sphere is given by

l(l + 1)
2
1
1
lm =
lm .
Ad L a = 2
alm L i , L i , Y
alm Y
2
2
lm
(25)
(26)
lm
Tr over matrices can be mapped to the integration over functions as

d
1
Tr
.
(27)
N +1
4
Let us expand the action (1) around the classical solution (7) and apply these mapping
rules. The bosonic part of the action (1) becomes
4 4
ij F
ij )
Tr(F
4g 2

i
1
1
i 4 3 ij k
m
Li , a j a k + a i a j , a k
ij m a a k
2 Tr
2g
3
2

2
2 Tr x i xi
6g

4 4
d
(Fij Fij )$
= 2 (N + 1)
4
4g

i
d 1
1
i
(Li aj )ak + ai [aj , ak ]
ij m a m ak
2 4 3 ij k (N + 1)
4
3
2
2g
$
SB =
4
N(N + 1)(N + 2),
24g 2
(28)
ij is defined as follows
where F
ij =
F
=

1
[Ai , Aj ] iij k Ak
2 2
1

1
1
Li , a j L j , a i + a i , a j iij k a k ,
(29)
and the corresponding function Fij () becomes

Fij () =

1
1
1
Li aj () Lj ai () + ai (), aj () $ iij k ak ().
(30)
This quantity is gauge covariant and becomes zero when the fluctuation is set to zero.
2 = 4g 2 /(N + 1) 4 2 . ( ) means that the
The YangMills coupling is found to be gYM
$
products should be taken as the star product. Hence commutators do not vanish even in the
case of U (1) gauge group.
The fermionic part of the action becomes

i L i + a i , + 2
SF = 2 Tr
2g
(N + 1)
2g 2

d 1
2
i [ai , ] +
i Li +
127
(31)
We next focus on the gauge symmetry of this action. The action (1) is invariant under
1 + i
the unitary transformation (3). For an infinitesimal transformation U = exp(i )
lm , the fluctuation around the fixed background transforms as

in (3) where = lm lm Y

i
Li , + i , a i .
After mapping to functions, we have local gauge symmetry

a i a i
i
Li + i[, ai ]$ .
Let us discuss a scalar field which is defined by

1

1
Ai Ai xi xi = xi a i + a i x i + a i a i .

2
2
It transforms covariantly as an adjoint representation

+ i , .
ai ai
(32)
(33)
(34)
(35)
Since the scalar field should become the radial component of a i in the commutative limit,
a naive choice is 0 = (xi a i + a i xi )/2. For small fluctuations this field is the correct
component of the field a i but large fluctuations of a i deform the shape of the sphere and
0 can be no longer interpreted as the radial component of a i . This is a manifestation of
the fact that matrix models or noncommutative gauge theories naturally unify spacetime
and matter on the same footing. An addition of the non-linear term a i a i makes transform
correctly as the scalar field in the adjoint representation.
We have so far discussed the U (1) noncommutative gauge theory on the fuzzy sphere.
A generalization to U (m) gauge group is realized by the following replacement:
xi xi 1m .
(36)
a is also replaced as follows:

2
m

a a T a ,
(37)
a=1
where T a (a = 1, . . . , m2 ) denote the generators of U (m). Then we obtain U (m)

noncommutative gauge theory by the same procedure as the U (1) case:

4 4
d
(Fij Fij )$
SB = 2 (N + 1) tr
4g
4

i 4 3 ij k
d 1
1
i
m
ij m a ak
2 (N + 1) tr
2g
4
3
2
$
4
mN(N + 1)(N + 2),
24g 2

d 1 i
2
i
SF = 2 (N + 1) tr
Li + [ai , ] + ,
2g
4
(38)
128
where trace is taken over m m matrices.

Let us consider a commutative limit. It is realized by the large N limit with fixed . The
following relations are satisfied in the commutative limit:
xi
ai (x) = Kia (x)ba (x) + ,
(39)
where i = 1, 2, 3, a = , and ba is a gauge field on the sphere. Two fields ba are defined
on the local coordinates , . Kia are two Killing vectors and defined by
Li = iKia (x)a .
(40)
The explicit forms of these Killing vectors are given as follows
K1 = sin ,
K1 = cot cos ,
K2 = cos ,
K2 = cot sin ,
K3 = 0,
K3 = 1.
(41)
From these Killing vectors, we obtain the metric tensor on S 2 as

g ab = Kia Kib .
(42)
R3
Three fields ai are defined in three-dimensional space

and contain a gauge field ba on
S 2 and a scalar field as well. In the commutative limit, we can separate the gauge field
ba from ai using the Killing vectors as in (39). The bosonic part of the action is rewritten
as

1
xk
SB = 2
d Kia Kjb Kic Kjd Fab Fcd + i2Kia Kjb Fab ij k
4gYM 2

+ 2Kia Kib (Da )(Db ) 2 2

1
xk
2 2 d iij k Kia Kjb Fab 2
2gYM

a
1
2i ab
2
ab
2
= 2
d x g Fab F + Fab + (Da ) D 2 , (43)
g
2gYM 2
where Fab = 1i a bb 1i b ba + [ba , bb ], Da = 1i a + [ba , ] and = = 1. The
fermionic part is also rewritten as

a
1
3 [, ] + 2
,
Da +
SF = 2
(44)
d
3
2
2gYM
where 3 = i xi / (see (64)) and a = i Kia . We thus obtained the action of a field theory
with a gauge field, an adjoint scalar and a gaugino field on a sphere.
In the commutative limit, the gauge transformation becomes
ba ba a ,
(45)
for U (1) gauge group and

ba ba a + i[, ba ],
(46)
129
for U (m) gauge group.

In the latter part of this section, we investigate the relation to a noncommutative gauge
theory in a flat background by taking limit while fixing . By virtue of the SO(3)
symmetry, we may consider the theory around the north pole without loss of generality.
Around the north pole, L 3 can be approximated as L 3 N/2. By defining L i = 2/N L i ,

the commutation relation (11) becomes

L 1 , L 2 i.
(47)
By further defining xi = L i and p i = 1 ij L j (i, j = 1, 2), we have

p1 , p 2 = i 2 ,
xi , p j = iij .
x1 , x2 = i 2 ,
(48)
We take the following limit to decompactify the sphere,

1 (N 1),
= fixed,
2
where
given by
= xi xi
2 2
N
N+2 2
2
N 2
2 .
(49)
In the coordinates of
xi ,
the Plank constant is
4 2
2 2 .
(50)
N +1
a $ b which is defined in (23) becomes the Moyal product a $M b because of (47) and the
Weyl type ordering property in (20). The following replacements hold in this limit,

d
d 2x
1
d 2x
Tr
=
(51)
=
,
2
N +1
4
4
4 2

1
1
Ad p i = ij Ad L j i
(i = 1, 2).
(52)
We can regard a 3 as the scalar field around the north pole. For simplicity, only U (1) case
is treated in the present discussions. The bosonic part of the action (1) becomes
2

4
4
SB = 2 Tr L 1 + a 1 , L 2 + a 2 + 2 Tr L i + a i , L i + a i ,
2g
2g
4

2i
+ 2 Tr L 1 + a 1 , L 2 + a 2
g
4

2

4
Tr p 2 a 2 , p1 + a 1
= 2
+ Tr pi + a i , pi + a i ,
2g

4i
+ Tr p 2 a 2 , p 1 + a 1 ,
(53)
where we have defined a i = 2/N a j j i (i, j = 1, 2) and = 2/N . This action can
be mapped to the following action,

2
6 N 2
2
SB =
d
x
F
(x)
$
F
(x)
+
d 2 x (x) Ad(D) (x)
12
12
2
16g

4i
+ d 2 x F12 (x) $ (x) ,
(54)
130
where Ad(Di ) = [ 1i i + ai , ] and F12 = 1i 1 a2 1i 2 a1 + [a1 , a2 ]$ . It is found that the

2 = 4g 2 /N 2 6 . Similarly, the fermionic part of the action (1)
YangMills coupling is gYM
becomes
i

L i + a i , + 2
SF = 2 Tr
2g

3 2

, + ,
= 2 Tr i p i + a i , +
(55)
2g
where we have defined j = j i i (i, j = 1, 2). This is also mapped as follows

SF =

(N + 1)
2g 2

d 2x 1
i i (x) + (x)
i [ai (x), (x)]$
(x)
4 2 i

2
3
(x), (x) $ + (x)(x)

+ (x)
.
The gauge transformation (32) becomes

ai ai i + i , ai $ .
(56)
(57)
We have seen that in the large radius and = fixed limit (49), the noncommutative
YangMills theory in the flat backgrounds can be obtained. The last term in (54) and the
last term in (56) become small in the . To discuss the meaning of these terms, we
integrate out the scalar field . The action (54) becomes

2
1
2
S =
d
x
F
(x)
$
F
(x)
+
d 2 x (x) Ad(D) (x)
12
12
2
4gYM

4i
+ d 2 x F12 (x) $ (x)

1
4
1
2
2
(x) + 2 d 2 x F12 (x) 2 F12 (x) + O a 3
d 2 x F12
4gYM

3
1
4
2 2
2 2
=
d x F12 (x) + 2 d x d y F12 (x)(x y)F12 (y) + O a
2
4gYM

1
4
2
=
(58)
p
1
+
d
F12 (p)F12 (p) + O a 3 ,
2

2
2
p
4gYM
where 1/ 2 = (x y) =
be
Mgauge =
2
.

d 2 k eik(x y ) /k 2 . The mass of the gauge field is found to
(59)
We found that the gauge field has acquired the mass by absorbing the degree of freedom
of the scalar field. From (56), the mass of the gaugino field is 2/ which is degenerate
with the mass of the gauge field. In the limit, the gauge field and the gaugino field
become massless.
131
3. Some properties of Dirac operator

In this section, we investigate some properties of Dirac operator. Dirac operator in our
model is given by

= 1 i Ad(L i ) + 1 .
D
(60)
Other Dirac operators are also constructed in [17,19,20]. In the commutative limit, this
operator becomes
D=
1
(i Li + 1),
(61)
which is the standard massless Dirac operator on the sphere [28]. The second term of this
operator comes from the contribution of the spin connection on the sphere. Around the
north pole, the operator behaves as

2
1
i pi + ,
D=
(62)
N
and in the limit, it approaches the massless Dirac operator in the flat background.
We consider the following chirality operator
3 =
1
i xi ,
(63)
which becomes, in the commutative limit,

3 =
1
i xi .
(64)
It is the standard chirality operator on the sphere [28]. This chirality operator obeys 32 =
1 23 / N(N + 2) and does not satisfy the usual condition for the chirality operator
before taking the commutative limit. A chirality operator which satisfies the condition
32 = 1 and becomes (64) in the commutative limit is given [17] by

1
i xi +
,
3 =
(65)
M
2
where M = (N + 1)/2 is a normalization factor. Anticommutation relation between the
Dirac operator and the chirality operator is

= 2 xi Ad L i
3 + 3 D
D
2
(66)
which vanishes in the commutative limit as can be seen from (25). Another type of a Dirac
operator which anticommutes with (65) is constructed in [17].
We next investigate the spectrum of the Dirac operator. We consider an eigenstate
of J 2 and J3 , satisfying J 2 j m = j (j + 1)j m and J3 j m = (m + 1/2)j m , where
132
j m , square of the
Ji = Li + i /2 and 1/2 j N + 1/2 and l m l. Acting on
Dirac operator becomes

j m = i L i ,
j m
j m + L i , L i ,
j m +
2D

2
2 3

j m
j m +
j m + Ad(L) 2
= Ad(J ) Ad(L)
4

1 2
j m .
= j+
(67)
2
We have used the relation J 2 = L2 + L + 3/4. This spectrum is identical with the
spectrum in the commutative limit.
4. Stability of classical solutions

We have argued the noncommutative gauge theories on the fuzzy sphere. This section is
devoted to a discussion of stability of the classical solutions. Let us first evaluate classical
values of the action for the solutions of (6) and (7).
The action vanishes
S = 0,
(68)
for commuting matrices (6), and becomes

S=
1
N +1
2
,
N(N + 1)(N + 2) 4 = 2 4
2
24g
3g
N(N + 2)
(69)
for the fuzzy sphere 1 (7). The solution of the fuzzy sphere has lower energy than the
solution of commuting matrices, and, therefore, the fuzzy sphere is more stable than the
commuting matrices at the classical level.
We next investigate one-loop effective action [1,6] around the classical solutions. We
decompose the matrices Ai and into the classical backgrounds and fluctuations:
Ai = Xi + A i ,
= + .
(70)
We add the following gauge-fixing term [1],

1
1
Sg.f. = 2 Tr [Xi , Ai ]2 + [Xi , b][Ai , c] ,
(71)
g
2
where b and c are ghost and anti-ghost, respectively. We expand the action up to the second
order of the fluctuations and set = 0:

2

1
1
S = 2 Tr
Xi , A j + Xi , Xj A i , A j iij k Xi A j , A k
g
2

1 i
Xi , + [Xi , b][Xi , c] .
(72)
2
1 The classical value of the action depends on the quadratic Casimir. That is, it depends on the representation.
We can construct an alternative representation in (10). The value of the action is larger than the irreducible
representation. Therefore, the solutions of reducible representation are unstable [25].
Then the one-loop effective action is calculated as

W = log d A d db dc eS .
133
(73)
We first consider the one-loop effective action for the commuting matrices. The contributions from A and are calculated as follows

(i)

1
4 2
(j ) 2
WB =
log x x
,
+ log 1 (i)
2
(x x (j ) )2
i=j

(i)
1
4 2
(j ) 2
WF =
(74)
+ log 1 (i)
log x x
.
2
(x x (j ) )2
i=j
Contributions from ghosts are included in WB . We find that the one-loop effective action
for commuting matrices vanishes due to the cancellation between bosonic and fermionic
contributions as expected from supersymmetry. Therefore, there are no net forces between
the eigenvalues if we neglect the effect of the fermionic zero modes [6]. For the fuzzy
sphere, one-loop effective action can be calculated as

2
1
Tr log Ad(L) ,
2

2 1
4 + 3 i Ad(Li )
1
WF = Tr log Ad(L) Tr log 1 +
.
2
4
(Ad(L))2
WB =
(75)
In this case, the one-loop effective action does not vanish because this background
preserves no supersymmetry. One-loop quantum corrections to the classical action (69)
are

1
4 + 3 i Ad(Li )
W = Tr log 1 +
(76)
.
4
(Ad(L))2
Let us evaluate this quantity. This expression can be rewritten using a replacement
i Ad(Li ) j (j + 1) l(l + 1) 3/4 = s(2l + 1) 1/2, where s = 1/2:
W =

N
3s(2l + 1) + 52
1
(2l + 1) log 1 +
4
l(l + 1)
l
1
2
s=1/2
N

(2l + 1) log
(l + 2)(l 1)
.
l(l + 1)
(77)
In the large N limit, we have

W 2 log N.
(78)
The value of the effective action including both the classical and the one-loop quantum
corrections becomes
Seff =
2 4 N +1
+ 2 log N =
N(N + 2) + 2 log N.
2
2
3g 2 N(N + 2)
6 gYM
(79)
134
For large enough N and gYM = fixed, the fuzzy sphere is more stable than the commuting
matrices. 2 On the other hand, there is a region of Seff > 0 if we take gYM sufficiently large
so that the second term dominates the first one. In the region, the commuting matrices is
more stable than the fuzzy sphere.
5. Supersymmetric noncommutative gauge theory on fuzzy sphere

In this section, we consider another action which gives a supersymmetric noncommutative gauge theory on the fuzzy sphere. Supersymmetric gauge theories on a fuzzy sphere is
also considered in [22]. The action is defined by

i j 2

1
1
1 i
i j k
S = 2 Tr Ai , Aj A , A + iij k A A A + Ai , .
(80)
g
4
3
2
It has has the following N = 2 supersymmetry:

i
(1) =
(1) Ai = i i ,
(81)
Ai , Aj iij k Ak ij ,
2
and
(2) Ai = 0,
(2) = .
(82)
The transformation for the gaugino in (81) is modified in comparison to the transformation
in the usual YangMills reduced models. 3 Let us check that these two transformations
form the N = 2 supersymmetry algebra. Algebra is also modified due to the modification
of supersymmetry. We have the following relations:

i
(1)
(1) (1)
(1)
= i[, ] + ii ,
1 2
2 1
2

(1) (1)
(1) (1)
1 2 2 1 Ai = i[Ai , ] + ij k j Ak ,
(83)
where = 2i(2 j 1 )Aj and j = 2i(2 j 1 ). The second term in the right hand side is
a new term corresponding to SO(3) rotation. Other commutation relations are calculated
as follows,

(1) (2) (2)(1) Ai = i i ,
(1) (2) (2) (1) = 0,
(84)
and

(2) (2)
(2) (2)
1 2 2 1 = 0,

(2) (2)
(2) (2)
1 2 2 1 Ai = 0.
(85)
2 As studied in papers [29], the partition function of d = 3 supersymmetric reduced models is not convergent
against integral over bosonic variables. It implies that eigenvalues are more likely to be separated from each other
and the fuzzy sphere solution is unstable nonperturbatively, in contrast to our perturbative analysis. From this
point of view, it is more interesting if we can construct nontrivial curved backgrounds from convergent higher
dimensional supersymmetric reduced models. We thank Staudacher for informing us of their analysis on reduced
model integrals.
3 In [30], there is a discussion about the supersymmetry transformation on the fuzzy sphere. In the context of
string theory, the spherical D2-brane can be a supersymmetric configuration. The fact that our model (80) has the
modified supersymmetry (81) which vanishes for the fuzzy sphere is consistent with the comment given in [30].
We thank K. Hashimoto and K. Krasnov for their stimulating comments.
135
If we take a linear combination of (1) and (2) as

(1) = (1) + (2) ,

(2) = i (1) (2) ,
(86)
we can obtain the following commutation relations up to a gauge symmetry and SO(3)
symmetry,

(1)(1) (1) (1) = 0,
(2) (2)

(2)
(2) = 0,
(1) (2)

(2)
(1) = 0,
(1) (1)

(1) (1) Ai = 2i i ,
(2) (2)

(2)
(2) Ai = 2i i ,
(1) (2)

(2)
(1) Ai = 0.
(87)
We find that these commutation relations indeed show the N = 2 supersymmetry algebra.
A new feature is the appearance of SO(3) rotation. This model has the same classical
solutions as the previous model, commuting diagonal matrices and the fuzzy sphere. In this
model, the fuzzy sphere preserves half of the N = 2 supersymmetries since (81) vanishes
for the fuzzy sphere while (82) does not, and this solution corresponds to a 1/2 BPS
background. Looking at the algebra (83), the remaining supersymmetry on the fuzzy sphere
generates SO(3) rotation instead of a constant shift of Ai . It is natural since translation on
a sphere is generated by SO(3) rotation. On the other hand, commuting matrices break all
the supersymmetry.
By expanding the bosonic matrices around the fuzzy sphere solution (7) as in (15)
and applying the mapping rule which is given in Section 2, we can obtain an N = 1
supersymmetric U (1) or U (n) noncommutative gauge theory on the fuzzy sphere. For
simplicity only U (1) gauge theory is considered in the following. The bosonic part of the
action is
4 4

Tr Fij Fij
4g 2

i
1
1
i
Li , a j a k + a i a j , a k
ij m a m a k
2 4 3 ij k Tr
3
2
2g
2

2 Tr x i xi
6g

4 4
d
(Fij Fij )$
= 2 (N + 1)
4
4g

i
d 1
1
i
ij m a m ak
2 4 3 ij k (N + 1)
2g
4
3
2
$
SB =
4
N(N + 1)(N + 2),
24g 2
where Fij is defined in (30).
(88)
136
The fermionic part is

i L i + a i ,
SF = 2 Tr
2g

d 1
i Li +
i [ai , ] .
= 2 (N + 1)
4
2g
$
(89)
This supersymmetric noncommutative gauge theory has the following N = 1 supersymmetry:

i
i 2 2
i ,
Fij ij .
=
(90)
2
Applying this transformation twice generates translation on the sphere, that is the rotation.
Let us consider a commutative limit. The action in the commutative limit is obtained by
the same procedure as in Section 2. In terms of the gauge field ba and the scalar field ,
the bosonic part becomes

a
1
2i ab
2
ab
2
SB = 2
d x g Fab F + Fab + Da D 2 . (91)
g
2gYM 2
ai =
This is the same as (43). The fermionic part becomes

a
1
3 [, ] .
Da +
d
SF = 2
2
2gYM
(92)
We have rescaled as 3/2 . The difference between the action obtained in this section
and the action obtained in Section 2 is that the former is supersymmetric while the latter is
not. The N = 1 supersymmetry transformation (90) is rewritten in terms of the gauge field
and the scalar field as
= i 3 ,
ba = i a
(93)
and

i
Fab ab 2 3 + Da 3 , a ,
(94)
2
where ba g ab bb , 3 = xi i /, a = Kia i and = 1/2 .
We next consider stability of the two classical solutions against quantum corrections
in the same manner as discussed in Section 4. One-loop corrections to the commuting
matrices are calculated as

2

1
4 2
WB =
log x (i) x (j ) + log 1 (i)
,
2
(x x (j ) )2
=
i=j
2
1 (i)
WF =
log x x (j ) .
2
(95)
i=j
Then, the value of the effective action including both classical and one-loop quantum
effects is given by

1
4 2
CM
Seff =
(96)
.
log 1 (i)
2
(x x (j ) )2
i=j
137
From this expression, we can read off that the eigenvalues tend to collapse into a single
ball whose size is of order . This shows that there exist a minimum length scale set by
CM is negative and of
in this theory. Since there are N(N + 1)/2 pairs of eigenvalues, Seff
2
order N .
For the fuzzy sphere, one-loop correction can be calculated as

2
1
Tr log Ad(L) ,
2

2
1
WF = Tr log Ad(L) ,
(97)
2
and the effective action for the fuzzy sphere does not receive quantum corrections
perturbatively as expected. Therefore, the value of the effective action is
WB =
FS
Seff
=
2 4 N +1
=
N(N + 2).
2
2
2
3g
N(N + 2)
6 gYM
(98)
In this case, since both of (96) and (98) are negative valued and of the same order N 2 , it is
difficult to conclude perturbative stability of the fuzzy sphere at this level.
6. Summary
In this paper, we have studied noncommutative gauge theories on the fuzzy sphere.
We considered two different types of supersymmetric three-dimensional matrix model
actions with a ChernSimons term. These models have a classical solution which
represents a fuzzy sphere. By expanding the models around this solution, we have obtained
noncommutative gauge theories on the fuzzy sphere. The gauge field acquires a mass due
to the ChernSimons term.
We have discussed two large N limits. One corresponds to a commutative limit. By
taking this limit, we obtained a gauge theory on a commutative sphere. Another limit
corresponds to a decompactifying limit. A noncommutative gauge theory on the fuzzy
sphere becomes a noncommutative gauge theory on a noncommutative plane.
Two types of three-dimensional supersymmetric matrix model actions we have considered in this paper have different supersymmetry properties. The first type contains a Majorana mass term in order to preserve the original supersymmetry in the flat space. The
second one does not have any other term than the YangMills and the ChernSimons terms
but it is invariant under a modified supersymmetry. The fuzzy sphere preserves supersymmetry in the second type while it does not in the first one. Therefore, the second one will be
more natural from the fuzzy sphere point of view. On the other hand, as for the commuting
matrices, the situation is opposite. A solution with commuting matrices preserves the supersymmetry in the first type and eigenvalues can move freely at least perturbatively. In the
second type, eigenvalues collapse into a small ball whose size is of the noncommutative
scale.
We have also investigated the stability of the fuzzy sphere against quantum fluctuations.
By calculating the one-loop effective action, we showed that in the first model the fuzzy
138
sphere is stable for fixed gYM in the large N limit. However, if we take gYM sufficiently
large with the large N limit, the fuzzy sphere becomes unstable and decays into a set
of diagonal eigenvalues. On the other hand, the fuzzy sphere is stable against quantum
corrections because of the nature of a BPS state in the second model.
Finally let us comment on ambiguities of operator orderings and corresponding freedom
for mapping from matrices to functions. Matrices on the fuzzy sphere are expanded by
a noncommutative spherical harmonics as discussed in Section 2. The ordering of the
noncommutative coordinates in the noncommutative spherical harmonics corresponds to
the Weyl type ordering. Therefore, a product of the noncommutative spherical harmonics
becomes Moyal product in the decompactifying limit. On the other hand, a stereographic
projection from a fuzzy sphere to a noncommutative complex plane as discussed in
Appendix B enables us to construct a normal ordered type basis. After mapping from
matrices to functions, a product of functions are written by the so called Berezin product.
Since such noncommutative products are known for general Khler manifolds, it would be
an interesting problem to obtain noncommutative gauge theories on more general curved
backgrounds from matrix models.
Appendix A. Star product on fuzzy sphere

In this section we summarize the noncommutative products on fuzzy sphere. We mainly
follow [14]. We have expanded functions and matrices as (16) and (19):
a() =
l=N

l

alm Ylm (),
(A.1)
l=0 m=l
a =
l=N

l

lm .
alm Y
(A.2)
l=0 m=l
Normalization is fixed as in (18) and (21). Combining (A.1) and (A.2), we can obtain a map
between the field and the function:
1

Tr Ylm a Ylm ().
a() =
(A.3)
N +1
lm
lm and Y
l m . We have required
Let us consider the product of the two spherical harmonics, Y
that maximum value of l is N . This product is expanded by the spherical harmonics and it
l+l . We assume that N is large such that l + l does not exceed N . We define the
contains Y
noncommutative product on the fuzzy sphere as
1

a $ b()
Tr a bYlm Ylm ()
N +1
lm

1
=
d d Yl m ( )Yl m ( )a( )b( )
N +1

lm l m l m

l m Y
l m Y
Ylm ().
(A.4)
Tr Y
lm
139
Components of the spherical harmonics can be given using WignerEckart theorem:

N

N N
N
l N2
s
2

2
Ylm ss = 2 , s Ylm 2 , s = (1)
(2l + 1)(N + 1), (A.5)
s m s
where s, s = N/2, N/2 + 1, . . . , 0, . . . , N/2.
Then we can calculate the trace of three spherical harmonics:

1
l1 m1 Y
l2 m2 Y
l3 m3
Tr Y
N +1

l1 l2
l
l
l
1
2
3
= (1)N+l1 +l2 +l3
m1 m2 m3 N N
2
2

(2l1 + 1)(2l2 + 1)(2l3 + 1)(N + 1),
l3
N
2
(A.6)
where ( ) and { } are Wigners 3j -symbol and 6j -symbol respectively. { } behaves as

N 3/2 for N [14]. ( ) becomes zero only when m1 + m2 + m3 = 0. By substituting
this quantity into (A.4), we get the explicit expression of the noncommutative product on
the fuzzy sphere.
Commutator of two matrices is

l1 m1 , Y
l2 m2
al1 m1 bl2 m2 Y
a,
b =
l1 m1 l2 m2

l3 m3 .
al1 m1 bl2 m2 fl13m13l2 m2 Y
l m
(A.7)
l1 m1 l2 m2 l3 m3
l m
fl13m13l2 m2 is zero when l1 + l2 + l3 = even and is given as follows when l1 + l2 + l3 = odd:

l m
fl13m13l2 m2 = 2 (2l1 + 1)(2l2 + 1)(2l3 + 1) N + 1 (1)N1
l1 l2 l3

l1
l2
l3
(A.8)
.
m1 m2 m3 N N N
2
2
2
This quantity behaves as N 1 when N .
Appendix B. Wick type star product

Here we explain the star product on the fuzzy sphere which corresponds to a normal
ordered operator product [3133] 4 and derive mapping rules from matrix models to field
theories with the normal ordered products la Berezin [31].
We first rescale xi as yi = xi / where xi s are defined by (7). Then

yi , yj = iij k yk ,
(B.1)
4 See [34,35] about Weyl ordered star product on Khler manifold.
140
where = /. Thus yi2 = 1 and 2 = 4/N(N + 2). We define the stereographic

projection of yi as
1
z = y ,
1
z = y+ ,
2
(B.2)
where y = y1 i y2 / 2 , = 2(1 y3 )1 . Note that the complex coordinate projected

from the sphere of radius is w = z . By the above definition, we obtain the commutation
relation

1
2
2
,
z , z = 1 + |z| 1 + |z|
(B.3)
2
2
and by using yi2 = 1 we can solve in terms of z and z :

= 2 + 1 4 1 + 2 2 ,
(B.4)
where |z|2 = z z and = 1 + |z|2 . For sufficiently large N , 1 + |z|2 and the
commutation relation is simplified as
1
2
1 + |z|2 .
z , z
N
(B.5)
As discussed in [32,33,36], the operator z can be written in terms of an annihilation

operator of a usual harmonic oscillator as
z = f (n + 1)a,
where

and

a,
a = 1,
(B.6)
n = a a,
N n + 1
1
f (n)
=
.
N/2(N/2 + 1) + N/2 n
N + 1 n
(B.7)
(B.8)
Eq. (B.8) is derived from a correspondence [33]

|N/2; m |n,
(B.9)
respectively. Thus the eigenvalue of

where |N/2; m and |n are eigenstates of L 3 and n,
L3 is given by m = n N/2 where 0 n N . To construct a normal ordered star product,

we define the following coherent state |z. Requiring z |z = z|z , we obtain
N
1
|z = M |z|2 2
n=0
zn
|n,
[ (n)]!
where M(|z|2 ) is a normalization factor, (n) n f (n) and

(n) (n 1) (1) (n = 0),
(n) ! =
1
(n = 0).
(B.10)
141
Eq. (B.10) satisfies the condition of the coherent state only for sufficiently large N and
when we can neglect the state of |N + 1. The normalization factor is determined from
z|z = 1 as
N
N

N
|z|2n
|z|2n
M |z|2 =
= 1 + |z|2 .
[N
+
1
n]!
[ (n)2 ]!
n!
n=0
(B.11)
n=0
Also the completeness condition

|nn| = (N + 1) d (z, z )|zz|,
1=
(B.12)
determines the measure d as

d (z, z )
i dz d z
.
2(1 + |z|2 )2
(B.13)
(See [33] for details.)

Normal ordered operators can be expanded in terms of |nm|
a m
(z )n
z m
a )n
nm = |nm| = (
Z
(B.14)
|00| =
|00|
,
[ (m)]!
n!
m! [ (n)]!

where |00| = k ck (z )k z k and ck is determined by M(x)1 = k ck x k . The corresponding orthonormalized basis of functions are defined by
nm |z =
Znm (z, z ) = z|Z
z n zm
.
[ (n)]! [ (m)]! M(|z|2)
Thus any operators

nm ,
a nm Z
a =
(B.15)
(B.16)
nm
are mapped to corresponding functions as

a a(z, z ) = z|a|z
=
a nm Znm (z, z ).
(B.17)
nm
The trace of an operator is mapped to an integral over the projected plane

1
Tr d(z, z ).
N +1
The analytic continuation of this function is defined by
a(z, )
=
|a|z
,
|z
(B.18)
(B.19)
where |z = M(||2) 2 M(|z|2) 2 M(z).

Using this definition, a product of matrices is
mapped to the star product as
1
a $ b(z, z ) = z|a b|z

= (N + 1)

d (, )
z|a|
|b|z
||z|2
z|
|z
142

M(z)M(
z)
b(z, )
= (N + 1) d(, )
a(, z )
2
M(|| )M(|z|2)

1
i d d
+1
a(, z )
2(1 + ||2 )2

1
(1 + z)(1
+ z)
b(z, ),
(1 + ||2 )(1 + |z|2 )
(B.20)
where = 1/N . This star product is nothing but the Berezin product on the sphere [31].
nm , the corresponding function satisfies the following simple

From the definition of Z
relations
Znm $ Zrs = mr Zns ,
and
(B.21)
nm $ Zn m = nn mm .
d(z, z )Z
(B.22)
Next we write down functions corresponding to the operators y s. y , y3 are rewritten

by z , z and n 5 as
y3 = (n N/2),
(B.23)
2 y+ = (1 y3 )z ,
2 y = z (1 y3 ).
Thus, the corresponding functions can be expanded by Znm as
y3 =
N

(n N/2)Znn ,
n=0
y+ = (1 y3 ) $ z =
N1

yn Zn+1 n ,
n=0
y = z $ (1 y3 ) =
N1

yn Zn n+1 ,
(B.24)

yn = 1 (n + 1 N/2) (n + 1) = n + 1 N n.
(B.25)
n=0
where
Also the functions corresponding to the

z=
N1

n=0
(n + 1)Zn n+1 ,
operators z , z
z =
N1

can be expanded as
(n + 1)Zn+1 n .
(B.26)
n=0
Now we will derive differential operators corresponding to the adjoint operators Ad(yi )
which are necessary to rewrite the matrix model action in terms of field theory on the
projected plane. We first define differential operators
kz =
1 1
z
1
M z M
+ z ,
N
1 + |z|2 N
5 n can be written in terms of z , z . However, it is not necessary here.
kz =
1 1
z
1
M z M
+ z .
2
N
1 + |z|
N
143
(B.27)
Note that the derivatives above act not only on M but on functions on the right of M. We
can easily obtain the following relations
zkz Znm = m Znm ,
m
Zn m1 ,
kz Znm =
(m)
zZnm = (m + 1)Zn m+1 ,
z kz Znm = n Znm ,
n
Zn1 m ,
kz Znm =
(n)
z Znm = (n + 1)Zn+1 m ,
where n = n/N , m = m/N . By using these relations, we have

nm (n m)Znm = N(zkz zkz )Znm (zz zz )Znm

y3 , Z
and hence the adjoint operator in matrix models Ad(y3) becomes the following differential
operator after mapping matrix models to field theories:
Ad(y3 ) = [y3 , ] (zz zz ).
We can also obtain similar rules for adjoint operators:

Ad 2 y+ z + z 2 z ,

Ad 2 y z + z2 z ,
2

2
Ad yi 2 1 + |z|2 z z .
(B.28)
(B.29)
(B.30)
(B.31)
Using mapping rules (B.17), (B.18), (B.20) and (B.29) (B.31), the matrix model actions
can be written in terms of field theories on the projected plane.
In the commutative limit, the action is simplified by writing the vector fields in terms of
projected ones as (39). Killing vectors on the projected plane are given
i
z
= ,
K+
2
i 2
z
K = z ,
2
K3z = iz,
i
z
K+
= z 2 ,
2
i
z
K = ,
2
K3z = i z ,
which can be seen from (B.29) (B.31). Thus the field strength of U (1) gauge field in the
commutative limit can be written in terms of z, z as

i a b
i
F+ = 2 K+
K (a bb b ba ) = 2 1 |z|2 1 + |z|2 Fzz ,
2

i a b
i
F+3 = 2 K+ K3 (a bb b ba ) =
z 1 + |z|2 Fzz ,
2 2

i a b
i
K3 (a bb b ba ) =
z 1 + |z|2 Fzz ,
F3 = 2 K
(B.32)
2 2
where
Fzz = z bz z bz .
(B.33)
144
We have set the scalar field = 0 for simplicity. Also we get

2
2
2
2 2(N + 1) d (z, z ) F12

+ F23
+ F31
Tr F

2
= 2(N + 1) d (z, z ) F+
+ 2F+3 F3

4 2

2(N + 1)
=
d (z, z ) 1 + |z|2 Fz
z
4

16(N + 1)
=
d (z, z )Fab F ab ,
4
(B.34)
where
g zz = g z z = 0,
g zz = g z z =
(1 + |z|2 )2
.
4
(B.35)
This is the well-known result.

Finally we consider the flat limit (49). Here, it should not be discussed around the north
pole but the south pole because of the definition (B.2). So y3 can be approximated as
y3 N/2 1,
(B.36)
x
and a rescaled coordinate corresponding to is

2
2

w=i
z = i z.
w =i
N
N
Thus we can rewrite the star product (B.20) in the flat case:

N

M |z|2 1 + |z|2

|w |2 N
=
1 2

2 N 2
|w |2 |w |2 2 |w |
=
1 2
N
2
|w |2
i dz d z
2(1 + |z|2 )2

|w |2 2 i dw d w
=
1 2
2 2

i dw d w

,

2
2
(B.37)
(B.38)
d(z, z )
a $ b(w , w ) =
i d d 1 (w + w w w )
a , w e
b w , ,
2
(B.39)
(B.40)
where = 2 /(N + 1) 2 /2 . Thus the star product becomes the Berezin product
on the flat plane.
145
References
[1] N. Ishibashi, H. Kawai, Y. Kitazawa, A. Tsuchiya, A large N reduced model as superstring,
Nucl. Phys. B 498 (1997) 467, hep-th/9612115.
[2] T. Banks, W. Fischler, S. Shenker, L. Susskind, M-theory as a matrix model: a conjecture, Phys.
Rev. D 55 (1997) 5112, hep-th/9610043.
[3] J. Polchinski, TASI Lectures on D-Branes, hep-th/9611050.
[4] W. Taylor, Lectures on D-branes, gauge theory and M(atrices), hep-th/9801182.
[5] T. Eguchi, H. Kawai, Reduction of dynamical degrees of freedom in the large N gauge theory,
Phys. Rev. Lett. 48 (1982) 1063;
G. Parisi, A simple expression for planar field theories, Phys. Lett. B 112 (1982) 463;
D. Gross, Y. Kitazawa, A quenched momentum prescription for large N theories, Nucl. Phys.
B 206 (1982) 440;
G. Bhanot, U. Heller, H. Neuberger, The quenched EguchiKawai model, Phys. Lett. B 113
(1982) 47;
S. Das, S. Wadia, Translation invariance and a reduced model for summing planar diagrams in
QCD, Phys. Lett. B 117 (1982) 228;
A. Gonzales-Arroya, M. Okawa, The twisted EguchiKawai model: a reduced model for large
N lattice gauge theory, Phys. Rev. D 27 (1983) 2397.
[6] H. Aoki, S. Iso, H. Kawai, Y. Kitazawa, T. Tada, Spacetime structures from IIB matrix model,
Prog. Theor. Phys. 99 (1998) 713, hep-th/9802085.
[7] H. Aoki, S. Iso, H. Kawai, Y. Kitazawa, T. Tada, A. Tsuchiya, IIB matrix model, Prog. Theor.
Phys. Suppl. 134 (1999) 47, hep-th/9908038.
[8] H. Aoki, N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, T. Tada, Noncommutative YangMills in
IIB matrix model, Nucl. Phys. B 565 (2000) 176, hep-th/9908141.
[9] M. Li, Strings from IIB matrices, Nucl. Phys. B 499 (1997) 149, hep-th/9612222.
[10] I. Bars, D. Minic, Non-commutative geometry on a discrete periodic lattice and gauge theory,
hep-th/9910091.
[11] J. Ambjorn, Y.M. Makeenko, J. Nishimura, R.J. Szabo, Finite N matrix models of noncommutative gauge theory, JHEP 9911 (1999) 029, hep-th/9911041.
[12] N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, Wilson loops in noncommutative YangMills, Nucl.
Phys. B 573 (2000) 573, hep-th/9910004;
Y. Kimura, Y. Kitazawa, IIB matrix model with D1D5 backgrounds, Nucl. Phys. B 581 (2000)
295, hep-th/9912258;
S. Iso, H. Kawai, Y. Kitazawa, Bi-local fields in noncommutative field theory, Nucl. Phys. B 576
(2000) 375, hep-th/0001027;
N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, String scale in noncommutative YangMills, Nucl.
Phys. B 583 (2000) 159, hep-th/0004038;
J. Ambjorn, Y.M. Makeenko, J. Nishimura, R.J. Szabo, Lattice gauge fields and discrete
noncommutative YangMills theory, JHEP 0005 (2000) 023, hep-th/0004147;
Y. Kimura, Y. Kitazawa, Supercurrent interactions in noncommutative YangMills and IIB
matrix model, Nucl. Phys. B 598 (2001) 73, hep-th/0011038.
[13] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032,
hep-th/9908056.
[14] J. Hoppe, Quantum theory of a massless relativistic surface and a two-dimensional bound state
problem, MIT, Ph.D. Thesis, 1982;
B. de Wit, J. Hoppe, H. Nicolai, On the quantum mechanics of supermembranes, Nucl. Phys.
B 305 (1988) 545;
J. Hoppe, Diffeomorphism groups, quantization, and SU(), Int. J. Mod. Phys. A 4 (1989)
5235.
[15] J. Madore, The fuzzy sphere, Class. Quantum Grav. 9 (1992) 69.
146
[16] H. Grosse, C. Klimcik, P. Presnajder, Towards finite quantum field theory in non-commutative
geometry, Int. J. Theor. Phys. 35 (1996) 231, hep-th/9505175;
H. Grosse, C. Klimcik, P. Presnajder, Field theory on a supersymmetric lattice, Commun. Math.
Phys. 185 (1997) 155, hep-th/9507074;
H. Grosse, C. Klimcik, P. Presnajder, Topologically nontrivial field configurations in noncommutative geometry, Commun. Math. Phys. 178 (1996) 507, hep-th/9510083.
[17] U. Carow-Watamura, S. Watamura, Noncommutative geometry and gauge theory on fuzzy
sphere, Commun. Math. Phys. 212 (2000) 395, hep-th/9801195;
U. Carow-Watamura, S. Watamura, Differential calculus on fuzzy sphere and scalar field, Int.
J. Mod. Phys. A 13 (1998) 3235, q-alg/9710034;
U. Carow-Watamura, S. Watamura, Chirality and dirac operator on noncommutative sphere,
Commun. Math. Phys. 183 (1997) 365, hep-th/9605003.
[18] C. Klimcik, Gauge theories on the noncommutative sphere, Commun. Math. Phys. 199 (1998)
257, hep-th/9710153.
[19] H. Grosse, J. Madore, A noncommutative version of the Schwinger model, Phys. Lett. B 283
(1992) 218.
[20] H. Grosse, P. Presnajder, The Dirac operators on the fuzzy sphere, Lett. Math. Phys. 33 (1995)
171.
[21] S. Baez, A.P. Balachandran, S. Vaidya, B. Ydri, Monopoles and solitons in fuzzy physics,
Commun. Math. Phys. 208 (2000) 787, hep-th/9811169;
A.P. Balachandran, S. Vaidya, Instantons and chiral anomaly in fuzzy physics, hep-th/9910129;
A.P. Balachandran, T.R. Govindarajan, B. Ydri, The fermion doubling problem and noncommutative geometry, Mod. Phys. Lett. A 15 (2000) 1279, hep-th/9911087;
A.P. Balachandran, X. Martin, D. OConnor, Fuzzy actions and their continuum limits, hepth/0007030.
[22] C. Klimcik, A nonperturbative regularization of the supersymmetric Schwinger model,
Commun. Math. Phys. 206 (1999) 567, hep-th/9903112.
[23] D. Kabat, W. Taylor, Spherical membranes in matrix theory, Adv. Theor. Math. Phys. 2 (1998)
181, hep-th/9711078.
[24] S.-J. Rey, Gravitating M(atrix) Q-balls, hep-th/9711081.
[25] R.C. Myers, Dielectric branes, JHEP 9912 (1999) 022, hep-th/9910053.
[26] A.Yu. Alekseev, A. Recknagel, V. Schomerus, Non-commutative world-volume geometries:
branes on SU(2) and fuzzy spheres, JHEP 9909 (1999) 023, hep-th/9908040.
[27] A.Yu. Alekseev, A. Recknagel, V. Schomerus, Brane dynamics in background fluxes and noncommutative geometry, JHEP 0005 (2000) 010, hep-th/0003187.
[28] C. Jayewardena, Schwinger model on S 2 , Helv. Phys. Acta 61 (1988) 636.
[29] W. Krauth, M. Staudacher, Finite YangMills integrals, Phys. Lett. B 435 (1998) 350, hepth/9804199;
W. Krauth, M. Staudacher, Eigenvalue distributions in YangMills integrals, Phys. Lett. B 453
(1999) 253, hep-th/9902113.
[30] K. Hashimoto, K. Krasnov, D-brane solutions in non-commutative gauge theory on fuzzy
sphere, hep-th/0101145.
[31] F.A. Berezin, General concept of quantization, Commun. Math. Phys. 40 (1975) 153.
[32] V.I. Manko, G. Marmo, E.C.G. Sudarshan, F. Zaccaria, Wigners problem and alternative
commutation relations for quantum mechanics, quant-ph/9612007.
[33] G. Alexanian, A. Pinzul, A. Stern, Generalized coherent state approach to star products and
applications to the fuzzy sphere, hep-th/0010187.
[34] S. Saito, K. Wakatsuki, Symmetrization of Berezin star product and path-integral quantization,
Prog. Theor. Phys. 104 (2000) 893, hep-th/9912265;
K. Wakatsuki, Symmetrization of Berezin quantization, hep-th/0006096.
147
[35] T. Masuda, Normalized Weyl-type $-product on Khler manifolds, Mod. Phys. Lett. A 15
(2000) 2177, hep-th/0010009.
[36] T. Holstein, H. Primakoff, Field dependence of the intrinsic domain magnetization of
ferromagnet, Phys. Rev. 58 (1940) 1098.

Nonabelian noncommutative gauge theory via

noncommutative extra dimensions
Branislav Jurco, Peter Schupp, Julius Wess
Universitt Mnchen, Sektion Physik, Theresienstr. 37, 80333 Mnchen, Germany
Abstract
The concept of covariant coordinates on noncommutative spaces leads directly to gauge theories
with generalized noncommutative gauge fields of the type that arises in string theory with background
B-fields. The theory is naturally expressed in terms of cochains in an appropriate cohomology; we
discuss how it fits into the framework of projective modules. The equivalence of star products that
arise from the background field with and without fluctuations and Kontsevichs formality theorem
allow an explicitly construction of a map that relates ordinary gauge theory and noncommutative
gauge theory (SeibergWitten map). As application we show the exact equality of the Dirac
BornInfeld action with B-field in the commutative setting and its semi-noncommutative cousin
in the intermediate picture. Using noncommutative extra dimensions the construction is extended
to noncommutative nonabelian gauge theory for arbitrary gauge groups; an explicit map between
abelian and nonabelian gauge fields is given. All constructions are also valid for non-constant B-field,
Poisson structure and metric. 2001 Published by Elsevier Science B.V.
1. Introduction
A natural approach to gauge theory on noncommutative spaces can be based on the
simple observation that multiplication of a field by a (coordinate) function is not a covariant
concept if that function does not commute with gauge transformations [1]. This can
be cured by adding appropriate noncommutative gauge potentials and thus introducing
covariant functions in complete analogy to the covariant derivatives of ordinary gauge
theory. 1 This construction is of particular interest because of its apparent relevance for
the description of open strings in a background B-field [24], where the D-brane world
volume can be interpreted as a noncommutative space whose fluctuations are governed
E-mail addresses: jurco@theorie.physik.uni-muenchen.de (B. Jurco),
schupp@theorie.physik.uni-muenchen.de (P. Schupp), wess@theorie.physik.uni-muenchen.de (J. Wess).
1 From the phase space point of view ordinary gauge theory is in fact a special case of this construction with
gauge potentials for only half of the coordinates (momenta).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 9 1 - 2
B. Jurco et al. / Nuclear Physics B 604 (2001) 148180
149
by a noncommutative version of YangMills theory [59]. It has been noticed (at least
in the case of a constant B-field) that there can be equivalent description of the effective
theory both in terms of noncommutative gauge theory and ordinary gauge theory. From
the physics perspective the two pictures are related by a choice of regularization [9,10]
which suggests a somewhat miraculous field redefinition that is usually called Seiberg
Witten map [9]. The inverse B-field, or more generally the antisymmetric part of the
inverse sum of B-field and metric, defines a Poisson structure whose quantization
gives rise to the noncommutativity on the D-brane world volume. Classically, the field
strength f describes fluctuations of the B-field. The SeibergWitten map expresses the
noncommutative gauge parameter and noncommutative
noncommutative potential A,
field strength in terms of their classical counter parts a, , f as formal power series
in , such that noncommutative gauge transformations are induced by ordinary gauge
transformations :
+ a),
A(a)
+ A(a)
= A(a
(1)
where is a function both of a and . In previous work [11,12] we have focused on the
rank one case and have explicitly constructed the maps A(a)
and (a,
) to all orders in
theta for the general case of an arbitrary Poisson manifolds which are relevant for the case
of non-constant background fields. 2 The corresponding star products can be computed
with Kontsevichs formula [13]; this formula continues to make sense even for non-closed
B-field although the corresponding star product will no longer be associative (see also [14])
but the non-associativity is still under control by the formality (90). For a noncommutative
gauge theory the rank one case does already include some information about the nonabelian
case, since it is always possible to include a matrix factor in the definition of the underlying
noncommutative space. In this article we shall make this more precise and will extend
our previous results to non-abelian gauge theories for any Lie group. (A brief description
of the construction was given in [15].) The noncommutative gauge potential and field
strength in general take values in the universal enveloping algebra, nevertheless thanks
to the existence of the SeibergWitten map the theory can be consistently formulated in
terms of only a finite number of fields; this important observation has been discussed
in [16]. A prerequisite for all this is an appropriate formulation of gauge theory on a more
or less arbitrary noncommutative space. (Here we are interested in the general case of
an arbitrary associative algebra of non-commuting variables, important special examples
with constant, linear and quadratic commutation relations have been discussed in [1].)
Particularly well-suited is the approach based on the notion of covariant coordinates that we
mentioned above because it finds a natural interpretation in the framework of deformation
quantization [13,17,18]. This is the natural setting since we are dealing with associative
algebras and formal power series it also allows rigorous statements by postponing, or
rather circumventing difficult questions related to convergence and representation theory.
Deformation quantization for non-constant and possibly degenerate Poisson structures
2 This is of course neither restricted to magnetic fields B 0i need not be zero nor to even-dimensional
manifolds.
150
goes far beyond the basic WeylMoyal product and the problem has only recently found
a general solution [13]. To construct a SeibergWitten map we do in fact need the
even more general formality theorem of Kontsevich [13]. A link between Kontsevich
quantization/formality and quantum field theory is given by the path integral approach [19]
which relates the graphs that determine the terms in the formality map to Feynman
diagrams. The relevant action - a Poisson sigma model was originally studied in [20,
21]; see also [22,23]. Our discussion is entirely tree-level. Aspects of the quantization of
nonabelian noncommutative gauge theories have been discussed by several authors [24
26] (and references therein). Closest to the present discussion is the perturbative study of
-expanded noncommutative gauge theories [27,28].
1.1. Noncommutative gauge theory in string theory
Let us briefly recall how star products and noncommutative gauge theory arise in string
theory [3,4,9]: consider an open string -model with background term

1
SB =
(2)
Bij ab a x i b x j ,
2i
D
where the integral is over the string world-sheet (disk) and B is constant, nondegenerate
and dB = 0. The correlation functions on the boundary of the disc in the decoupling limit
(g 0, 0) are

f1 x(1 ) fn x(n ) B = dx f1 fn (1 < < n )
(3)
with the WeylMoyal star product
i h ij

(f g)(x) = e 2

i j

f (x)g(x )x x ,
(4)
which is the deformation quantization of the Poisson structure = B 1 . More generally a

star product is an associative, [[h ]]-bilinear product
f g = fg +

n=1
(i h )n Bn (f, g),

(5)
bilinear
which is the deformation of a Poisson structure :

[f , g] = i h {f, g} + O h 2 , {f, g} = ij (x)i f j g.
(6)
We now perturb the constant B field by adding a gauge potential ai (x): B B + da,
SB SB + Sa , with

Sa = i d ai x( ) x i ( ).
(7)
D
Classically we have the naive gauge invariance

ai = i ,
(8)
151
but in the quantum theory this depends on the choice of regularization. For PauliVillars (8)
remains a symmetry but if one expands exp Sa and employes a point-splitting regularization
then the functional integral is invariant under noncommutative gauge transformations 3
A i = i + i A i i A i .
(9)
Since a sensible quantum theory should be independent of the choice of regularization there
should be field redefinitions A(a),

(a,
) (SeibergWitten map) that relate (8) and (9):
+ a).
= A(a
A(a)
+ A(a)
(10)
It is instructive to study the effect of the extra factor exp Sa in the correlation function (3)
in more detail: It effectively shifts the coordinates 4
x i x i + ij A j =: Dx i .
(11)
More generally, for a function f ,

f f + fA =: Df.
(12)
The mapping A : f fA plays the role of a generalized gauge potential; it maps a

function to a new function that depends on the gauge potential. The shifted coordinates
and functions are covariant under noncommutative gauge transformations:

(Df
) = i , Df .
Dx i = i , Dx i ,
(13)
The first expression implies (9) (for constant and nondegenerate).
The covariant coordinates (11) are the background independent operators of [9,32]; they
and the covariant functions (12) can also be introduced abstractly in the general case of an
arbitrary noncommutative space as we shall discuss in the next section.
2. Gauge theory on noncommutative spaces

2.1. Covariant functions, covariant coordinates
Take a more or less arbitrary noncommutative space, i.e., an associative unital algebra
Ax of noncommuting variables with multiplication and consider (matter) fields on this
space. The fields can be taken to be elements of Ax , or, more generally, a left module of it.
The notion of a gauge transformation is introduced as usual 5
,
(14)
where is an invertible element of Ax . In analogy to commutative geometry where

a manifold can be described by the commutative space of functions over it, we shall
3 In this form this formula is only valid for the MoyalWeyl star product.
4 Notation: D should not be confused with a covariant derivative (but it is related).
5 We shall often use the infinitesimal version = i of (14) this is purely for notational clarity. Other
transformations, like, e.g., or 1 can also be considered.
152
refer to the elements of Ax also as functions. Later we shall focus on the case where
the noncommutative multiplication is a star product; the elements of Ax are then in
fact ordinary functions in the usual sense of the word. The left-multiplication of a field
with a function f Ax does in general not result in a covariant object because of the
noncommutativity of Ax :
f f = (f ).
(15)
(As in ordinary gauge theory the gauge transformation only acts on the fields, i.e., on the
elements of the left-module of Ax and not on the elements of Ax itself.) To cure (15) we
introduce covariant functions
Df = f + fA ,
(16)
that transform under gauge transformations by conjugation

Df Df 1 ,
by adding gauge potentials fA with appropriate transformation property 6

fA f , 1 + fA 1 .
(17)
(18)
Further covariant objects can be constructed from covariant functions; the 2-tensor

F (f, g) = [Df , Dg] D [f , g] ,
(19)
for instance plays the role of covariant noncommutative field strength.
2.1.1. Canonical structure (constant ) and noncommutative YangMills
Before we continue let us illustrate all this in the particular simple case of an algebra Ax
generated by coordinates x i with canonical commutation relations
i j
x , x = i ij , ij C.
(20)
This algebra arises in the decoupling limit of open strings in the presence of a constant
B-field. It can be viewed as the quantization of a Poisson structure with Poisson tensor ij
and the multiplication is then the WeylMoyal star product
i
f g = f e2

ij
i
j
g.
(21)
(This formula holds only in the present example, where ij is constant and we shall also
assume that it is non-degenerate. In the rest of the paper we drop both restrictions.) Let us
focus on the coordinate functions x i . The corresponding covariant coordinates are
Dx i = x i + xAi = x i + ij A j ,
(22)
where we have used to lower the index on A j . Using (20), we see that the transformation (18) of the noncommutative gauge potential A j is

A j i j 1 + A j 1 ,
(23)
6 Notation: [a , b] a b b a [a, b] .

153
or, infinitesimally

A j = j + i , A j .
The noncommutative field strength

kl = k A l l A k i A k , A l
F
(24)
(25)
transforms covariantly
kl F
kl 1 .
F
We have again used to lower indices to get (25) from the definition (19)

kl ik j l F x i , x j = xAi , x j + x i , x j + xAi , x j .
iF
A
A

(26)
(27)
Note, that we should in general be more careful when using to lower indices as in (22) or
(27) because this may spoil the covariance when is not constant as it was in this particular
example. Relations (23), (25) and (26) define what is usually called Noncommutative
YangMills theory (NCYM) in the narrow sense: ordinary YangMills with all matrix
products replaced by star products. This simple rule, however, only really works well for
the MoyalWeyl product, i.e., constant . In the general case it is wise to stick with the
manifestly covariant and coordinate-independent 7 objects defined in (16) and (19). The
fundamental objects are really the mappings (differential operators) D and F in these
equations. The transformation of A = D id : f fA under gauge transformations is
exactly so that (16) transforms by conjugation. The mappings A Hom(Ax , Ax ) and
F Hom(Ax Ax , Ax ) play the role of generalized noncommutative gauge potential
and noncommutative field strength. There are several reasons, why one needs A and D
and not just Ai A(x i ) (or A i , for constant): if we perform a general coordinate

transformation x i x i (x j ) and transform Ai (or A i ) naively as its index structure
suggests, then we would obtain objects that are no longer covariant under noncommutative

gauge transformations. The correct transformation, A(x i ) A (x i ), is more complicated
and will be discussed in Section 3.3. Furthermore, we may be interested in covariant
versions of scalar fields (x). These are given by the corresponding covariant function
D((x)).
In the next section we will propose an abstract definition of the type of noncommutative
gauge theory that is of present interest. Then we shall proceed to give an interpretation in
the framework of deformation quantization and will construct a particular important class
of these operators.
2.2. Abstract formulation of noncommutative gauge theory
Finite projective modules take the place of fiber bundles in the noncommutative
realm [29]. This is also the case here, as we shall explain below, but may not have
been apparent since we have been working with component fields as is customary in the
7 We would like to thank Anton Alekseev for stressing the importance of this point.
154
physics literature. We have argued in the previous section that A C 1 , F C 2 with C p =

p
Hom(Ax , Ax ), C 0 Ax . These p-cochains take the place of forms on a noncommutative
space Ax , which for now is still an arbitrary associative algebra over a field k with
multiplication . It is actually more convenient to start with the Hochschild complex of Ax ,
p
C p (Ax , Ax ) = Homk (Ax , Ax ), with values in Ax considered as a left module of Ax . We
have a coboundary operator d : C p C p+1 , d2 = 0, d 1 = 0,
d C = [C, ]G ,
(28)
where [ , ]G is the Gerstenhaber bracket (A.4),

(d C)(f1 , . . . , fp+1 ) = f1 C(f2 , . . . , fp+1 ) C(f1 f2 , f3 , . . . , fp+1 )
+ C(f1 , f2 f3 , . . . , fp+1 )
+ ()p C(f1 , f2 , . . . , fp fp+1 )
+ ()p+1 C(f1 , . . . , fp ) fp+1 ,
(29)
and we have a cup product : C p1 C p2 C p1 +p2 ,

(C1 C2 )(f1 , . . . , fp1 +p2 ) = C1 (f1 , . . . , fp1 ) C2 (fp1 +1 , . . . , fp1 +p2 ).
(30)
For a function C 0 Ax the coboundary operator is defined as

(d )(f ) = f f
(31)
and the cup product reduces to the multiplication of Ax in the obvious way. For
this reason and since there seems to be little chance of confusion we have used the
same symbol for the cup product and the multiplication. Let us apply the Hochschild
formalism to the gauge transformation dependent map D C 1 that we introduced in the
definition of covariant functions (16) in the previous section. In view of the way that
Ai ij A j appeared in the definition of covariant coordinates (22) we define an abstract
noncommutative gauge potential A C 1
A D id.
(32)
Applying A to coordinate functions in the setting of (22) we indeed recover Ai :

A x i = D x i x i = Ai .
Let us compute the behavior of A under a gauge transformation. Using (17) and the
definitions of d and the cup product we find
A d 1 + A 1 ,
(33)
which gives (18) when evaluated on a function. The corresponding infinitesimal version is
A = i(d + A A ).
(34)
Next we introduce the Hochschild field strength FH C 2

FH d A + A A
(35)
155
and compute
FH (f, g) = Df Dg D(f g)
(36)
and find the Bianchi identity

d FH + A FH FH A = 0.
Evaluated on three functions f , g, h the latter reads

D (f g) h D f (g h) + (Df Dg) Dh Df (Dg Dh),
(37)
(38)
which is zero by associativity of Ax . FH transforms covariantly under gauge transformations

FH FH 1 ;
(39)
infinitesimally
FH = i( FH FH ).
(40)
When we compare (19) and (36), we see (as expected) that our noncommutative field
strength F of the previous section is an antisymmetric version of FH . We can obtain F
directly by taking the antisymmetrized version of Hochschild, where one considers Ax as
a Lie algebra with bracket [a , b] = a b b a; this is the Chevalley complex of Ax
p
with values in Ax : C p = Homk (Ax , Ax ). We find the relevant formulas in this setting
by using corresponding antisymmetrized formulas for the coboundary operator d and the
cup product which we then denote by . The action of Lie Ax on the module Ax is given
by -multiplication as before. We now see that Eq. (19),

F (f, g) = [Df , Dg] D [f , g] ,
(41)
can be written
F d A + A A.
(42)
The remaining equations also do not change in form (as compared to the Hochschild case):
(42) implies
d F + A F F A = 0
(43)
and the behavior under (infinitesimal) gauge transformations is

A = i(d + A A ),
(44)
F = i( F F ).
(45)
Eqs. (42), (43), (44) and (45) are reminiscent of the corresponding equations of ordinary
(nonabelian) gauge theory. The correspondence is given by the following dictionary: oneforms become linear operators on Ax which take one function as argument and yield
a new function, two-forms become bilinear operators on Ax which take two functions
as arguments and return one new function, and the Lie bracket is replaced by the
156
antisymmetrized cup product . 8 As in ordinary gauge theory, it may not be possible

to use one globally defined gauge potential A; we may need to introduce several D and
corresponding gauge potentials A for functions defined on different patches. We shall
come back to this later.
Remark. One reason for going through the slightly more general Hochschild construction
first is that the symmetric part of FH may also contain interesting information as we will
see in Section 3.5. For invertible D there is still another interesting object:
D1 F
F
(46)
measures noncommutativity:
(f, g) = [f , g] [f , g],
F
(47)
where the (associative) product is defined by f g = D1 (Df Dg). This field

strength satisfies the CartanMaurer equation
= [F,
F
]G .
d F
(48)
2.2.1. Projective modules

We shall now discuss how our formulae fit into the framework of finite projective
modules: the calculus of p-cochains in C p with the coboundary operator d uses only
the algebraic structure of Ax ; it is related to the standard universal calculus and one can
obtain other calculi by projection. Consider a (finite) projective right Ax -module E. We
introduce a connection on E as a linear map : E Ax C p E Ax C p+1 for p N0
which satisfies the Leibniz rule
() = ()
+ ()p d
= ()p d 1,
for all E Ax C p , C r , and where

d (a ) = (d a) + ()q a d
(49)
(50)
for all a C q , and d 1 is the identity operator on Ax . (The transformation of matter fields
= i leads to a slight complication here; for fields that transform in the adjoint (by
d and not and d .)

star-commutator) we would only need ,

Let (a ) be a generating family for E; any E can then be written as = a a with
a Ax (with only a finite number of terms different from zero). For a free module the
a are unique, but we shall not assume that. Let the generalized gauge potential be defined
a = b Aba . In the following
by the action of on the elements of the generating family:
= .A, etc. We compute
we shall suppress indices and simply write = .,

= (.) = . A + d = .(D ).
(51)
8 More educated: n-forms become n-cochains. An even closer match with the usual physics conventions is
achieved by multiplying our d and F by i (Section 3 and later: multiply by i h).
157
Evaluated on a function f Ax the component D yields a covariant function times the

matter field, (D )(f ) = (Df ) , so in this framework covariant functions are related
to the covariant derivative d + A:

(52)
d + A (f ) = f + A(f ) .
The square of the connection gives
2 = .(A A + d A). = .F .
(53)
with the field strength

F = d A + A A.
(54)
3. Ordinary versus noncommutative gauge theory

We are particularly interested in the case where the algebra of our noncommutative
space Ax is given by a star product (via a quantization map). A star product on a smooth
C -manifold M is an associative C[[h ]]-bilinear product

i h n
Bn (f, g), f, g C (M),
f g = fg +
(55)
2
n=1
where Bn are bilinear operators and h is the formal deformation parameter; it is a

deformation quantization of the Poisson structure
{f, g} ij (x)i f j g = B1 (f, g) B1 (g, f ).
(56)
Equivalent star products can be constructed with the help of invertible operators D
D(f g) = Df Dg,
Df f +
h n Dn (f ),
(57)
n=1
where Dn are linear operators. This operation clearly does not spoil associativity. There
are also inner automorphisms for each invertible element and their infinitesimal version,
inner derivations,
f f 1 ,
f = [i , f ];
(58)
these operations do not change the star product.

The striking similarity between Eqs. (16), (17) and Eqs. (57), (58) suggests the
following interpretation of noncommutative gauge theory in the star product formalism:
the covariance maps D = id + A are gauge equivalence maps D for the underlying star
product, combined with a change of coordinates
D = D ,
D(f g) = Df Dg,
gauge transformations are inner automorphisms of the star products.
(59)
158
3.1. Motivation from string theory

A Poisson tensor enters the discussion of Seiberg and Witten [9] via a background
B-field in the open string picture. In this setting

ij
1
ij

= 2
(60)
g + 2 B A
appears in the propagator at boundary points of the string world sheet. (g is the closed string
metric and A denotes the antisymmetric part of a matrix.) The 2-form 12 Bij dx i dx j
is a symplectic form, provided B is nondegenerate and d = 0, which is, e.g., obviously
the case if B is constant (but we shall not require it to be constant). In the zero slope limit
(or in the intermediate picture with = B [9], see Section 3.4)
ij

ij = B 1
(61)
defines then a Poisson structure. It has been discussed by several authors how the Moyal
Weyl star product enters the picture as a quantization of this Poisson structure in the
constant case [3,4,9]. A direct approach that is most suitable for our purposes and that also
works for non-constant is given by Cattaneo and Felder [19] in their QFT realization of
Kontsevichs star product.
In Section 1.1 we have discussed what happens if we add fluctuations f (with df = 0,
i.e., locally f = da) to the background B field: the action is then naively invariant under
ordinary gauge transformations a = d, but the invariance of the quantum theory depends
on the choice of regularization. A point splitting prescription [9,10] leads in fact to a
noncommutative gauge invariance. Since in a consistent quantum theory the choice of
regularization should not matter, Seiberg and Witten argued that there should exist maps
that relate ordinary and noncommutative gauge theory such that (1) holds. A more abstract
argument that leads to the same conclusion, but gives the SeibergWitten maps more
directly and also works for non-constant can be based on a quantum version of Mosers
lemma [11,12]. Here is briefly the idea, in the next section we will review the details: the
addition of f = da to B defines a new Poisson structure
=
1
B
=
1
,
B +f
(62)
which, according to Mosers lemma [30], is related to the original one by a change of
coordinates given by a flow a that depends on the gauge potential a. After quantization
and give rise to equivalent star products and . The equivalence map Da , the full
quantum flow Da = Da a and the noncommutative gauge potential
Aa = Da id
(63)
are also functions of a. An additional infinitesimal gauge transformation

a = d
(64)
does not change the Poisson structure (since f = 0), but it still induces an infinitesimal
canonical transformation. After quantization that transformation becomes an inner deriva-
159
tion of the star product and thus a noncommutative gauge transformation with =
(,
a), such that
Aa+d = Aa + Aa .
(65)
3.2. Quantum version of Mosers lemma: SeibergWitten map

Consider an abelian gauge theory on a manifold that also carries a Poisson structure .
The gauge potential, field strength and infinitesimal gauge transformations are
1
f = fij dx i dx j = da,
2
a = d.
fij = i aj j ai ,
a = ai dx i ,
(66)
We will first construct a semiclassical version of the SeibergWitten map, where all star
commutators are replaced by Poisson brackets. The construction is essentially a formal
generalization of Mosers lemma to Poisson manifolds.
3.2.1. Semi-classical construction
Let us consider the nilpotent coboundary operator of the Poisson cohomology (see
[45]) the semiclassical limit of (28):
d = [ , ]S,
(67)
where [, ]S is the SchoutenNijenhuis bracket (A.1) and = 12 ij i j is the Poisson

bivector. Acting with d on a function f gives the Hamiltonian vector field corresponding
to f
d f = { , f } = ij (j f )i .
(68)
It is natural to introduce a vector field

a = ai d x i = j i ai j
(69)
corresponding to the abelian gauge potential a and a bivector field

1
f = d a = ik fkl lj i j
2
(70)
corresponding to the abelian field strength f = da. We have d f = 0, due to d2

[, ]S = 0 (Jacobi identity).
We are now ready to perturb the Poisson structure by introducing a one-parameter
deformation t with t [0, 1]: 9
t t = ft
(71)
with initial condition 0 = . In local coordinates:

ij
t t = (t f t )ij ,
ij
0 = ij ,
(72)
9 In this notation the equations resemble those of Mosers original lemma, which deals with the symplectic
2-form , the inverse of (provided it exists). There, e.g., t t = f for t = + tf .
160
with formal solution given by the geometric series

t = tf + t 2 f f t 3 f f f =
1
,
1 + tf
(73)
if f is not explicitly -dependent. (The differential equations (71), (72) and the rest of the
construction do make sense even if f or a are -dependent.) t is a Poisson tensor for all t
because [t , t ]S = 0 at t = 0 and
t [t , t ]S = 2dt ft [t , t ]S .
The evolution (71) of t is generated by the vector field a :
t t = dt at = [at , t ]S .
This Lie derivative can be integrated to a flow (see Appendix A.3)

a = exp(at + t ) exp(t )t =0 ,
(74)
(75)
that relates the Poisson structures = 1 and = 0 . In analogy to (32) we define a semiclassical (semi-noncommutative) generalized gauge potential
Aa = a id.
(76)
Under an infinitesimal gauge transformation a a + d the vector field (69) changes by

a Hamiltonian vector field d = ij (j )i :
a a + d .
(77)
Let us compute the effect of this gauge transformation on the flow (75). After some
computation (see Appendix A.4) we find (infinitesimally: to first order in )

a , i.e., a+d
= (id + d )
(f ) = a (f ) + a (f ),
a+d
(78)
and
Aa+d = Aa + d + {Aa , },
(79)
with
(, a) =

(at + t )n ()
.
(n + 1)! t =0
(80)
n=0
Eqs. (80) and (76) with (75) are explicit semi-classical versions of the SeibergWitten
map. The semi-classical (semi-noncommutative) generalized field strength evaluated on
two functions (e.g., coordinates) f , g is

Fa (f, g) = f, g {f, g} = {f, g} {f, g} .
(81)
Abstractly as 2-cochain:
1
1
Fa = ( )ik i k = (f )j l ij kl i k
2
2
(82)
161
with f = f , or
f =
1
f,
1+f
(83)
which we recognize as the noncommutative field strength (with lower indices) for
constant f , [9]. The general result for non-constant f , is thus simply obtained by
the application of the covariantizing map (after raising indices with s).
The SeibergWitten map in the semiclassical regime for constant has previously been
discussed in [33,35], where it was understood as a coordinate redefinition that eliminates
fluctuations around a constant background.
We will now use Kontsevichs formality theorem to quantize everything. The goal is to
obtain (1) in the form (65) of which (79) is the semi-classical limit.
3.2.2. Kontsevich formality map
Kontsevichs formality map is a collection of skew-symmetric multilinear maps Un for
n = 0, . . . , that map tensor products of n polyvector fields to differential operators.
More precisely Un maps the tensor product of n ki -vector fields to an m-differential
operator, where m is determined by the matching condition
m = 2 2n +
n

ki .
(84)
i=1
U1 in particular is the natural map from a k-vector field to a k-differential operator

U1 (1 k )(f1 , . . . , fk ) =
k

1
sgn( )
i (fi ),
k!
k
(85)
i=1
and U0 is defined to be the ordinary multiplication of functions:

U0 (f, g) = fg.
(86)
The Un , n 1, satisfy the formality condition [13]

1
d Un (1 , . . . , n ) +
U|I | (I ), U|J | (J ) G
2
=
I J =(1,...,n)
I,J =

Un1 [i , j ]S , 1 , . . . ,
i , . . . ,
j , . . . , n ,
(87)
i<j
where d C [C, ]G , with the commutative multiplication (f, g) = f g of functions;

the hat marks an omitted vector field. See [13,19] for explicit constructions and more
details and [13,37] for the definition of the signs in this equation. In the following we
collect the three special cases that we actually use in this paper.
Consider the formal series (see also [38])
() =

(i h )n
n=0
n!
Un+1 (, , . . . , ).
(88)
162
According to the matching condition (84), Un+1 (, , . . . , ) is a tridifferential operator

for every n if is a trivector field, it is a bidifferential operator if is a bivector field, it is
a differential operator if is a vector field and it is a function if is a function; in all cases
is assumed to be a bidifferential operator.
Star products from Poisson tensors. A Poisson bivector gives rise to a star product
via the formality map: according to the matching condition (84), Un (, . . . , ) is a
bidifferential operator for every n if is a bivector field. This can be used to define a
product
f g =

(i h )n
n=0
n!
Un (, . . . , )(f, g) = fg +
i h ij
i f j g + .
2
(89)
The formality condition implies

d = (i h )2 (d ),
(90)
or, [, ]G =

S ), i.e., associativity of , if is Poisson. (If is not Poisson,
i.e., has non-vanishing SchoutenNijenhuis bracket [, ]S , then the product is not
associative, but the non-associativity is nevertheless under control via the formality
condition by (90).)
(i h)2 ([, ]
Differential operators from vector fields. We can define a linear differential operator 10
(i h )2
U3 (, , ) +
2
for every vector field . For Poisson the formality condition gives
( ) = +
(91)
d ( ) = i h (d ) = i h d + .
(92)
Vector fields that preserve the Poisson bracket, d = [, ]S = 0, give rise to

derivations of the star product (89): from (92) and the definition (A.7), (29) of d

0 = d ( ) (f, g) = ( ) (f g) + f ( ) (g) + ( ) (f ) g.
(93)
Inner derivations from Hamiltonian vector fields. Hamiltonian vector fields d f give rise
to inner derivations of the star product (89): we can define a new function 10
(i h )2
U3 (f, , ) +
f (f ) = f +
2
for every function f . For Poisson the formality condition gives
d f = i h (d f ).
Evaluated on a function g, this reads

1
(d f ) (g) = [g , f ].
i h
The Hamiltonian vector field d f is thus mapped to the inner derivation hi [f , ].
10 U (, ) = 0 and U (f, ) = 0 by explicit computation of Kontsevichs formulas.
2
2
(94)
(95)
(96)
163
3.2.3. Quantum construction

The construction mirrors the semiclassical one, the exact correspondence is given by the
formality maps Un that are skew-symmetric multilinear maps that take n polyvector fields
into a polydifferential operator. We start with the differential operator
a =

(i h )n
n!
n=0
Un+1 (a , , . . . , ),
(97)
which is the image of the vector field a under the formality map (91); then we use the
coboundary operator d (28) to define a bidifferential operator
f = d a .
(98)
This is the image of f = d a under the formality map:

f =

(i h )n+1
n!
n=0
Un+1 (f , , . . . , ).
(99)
A t-dependent Poisson structure (71) induces a t-dependent star product via (89)
g t h =

(i h )n
n=0
n!
Un (t , . . . , t )(g, h).
(100)
The t-derivative of this equation is

t (g t h) =

(i h )n+1
n=0
n!
Un+1 (ft , t , . . . , t )(g, h),
(101)
where we have used (71) and the skew-symmetry and multi-linearity of Un . Comparing
with (99) we find
t (g t h) = ft (g, h),
(102)
or, shorter, as an operator equation: t (t ) = ft . But ft = dt at = [at , t ]G , so the
t-evolution is generated by the differential operator at and can be integrated to a flow (see
Appendix A.3)

Da = exp(at + t ) exp(t )t =0 ,
(103)
that relates the star products = 1 and = 0 , and that defines the generalized
noncommutative gauge potential
Aa = Da id.
(104)
The transformation of a under an infinitesimal gauge transformation a a + d can be

computed from (77) with the help of (95), see (31):
a a +
1
d .
i h
(105)
164
The effect of this transformation on the quantum flow and on the noncommutative gauge
potential are (see Appendix A.4)

1
i
Da+d = id + d Da , i.e., Da+d(f ) = Da f + , Da f ,
(106)
h
i h

1
d A + A
Aa+d = Aa +
(107)
i h
with
a) =
(,

(at + t )n ()
.
(n + 1)! t =0
(108)
n=0
Eqs. (104) with (103) and (108) are explicit versions of the abelian SeibergWitten
map to all orders in h . They are unique up to (noncommutative) gauge transformations.
Perhaps more importantly this construction provides us with an explicit version of the
covariantizer Da (the equivalence map that sends coordinates and functions to their
covariant analogs) in terms of a finite number of (classical) fields ai . The noncommutative
gauge parameter (108) also satisfies the consistency condition
a) =
(,
a) (,
a) , (,
a) ,
(,
h
(109)
with (ai ) = i , () = 0, that follows from computing the commutator of abelian

gauge transformations on a covariant field [16].
The generalized noncommutative field strength evaluated on two functions (or coordinates) f , g is

Fa (f, g) = Da [f , g] [f , g] .
(110)
Up to order 2 the series for Aa and agree with the semiclassical results. In
components:

1
Aa x i = ij aj + kl al k ij aj ij fj k + ,
2

1
1
= + ij aj i + kl al k ij aj i ij fj k i + .
2
6
(111)
(112)
There are three major strategies for the computation of the SeibergWitten map:
(i) From the gauge equivalence condition (1) one can directly obtain recursion
relations for the terms in the SeibergWitten map. For constant these can be
cast in the form of differential equations [9]. Terms of low order in the gauge fields
but all orders in can be expressed in terms of n -products [41,42]. 11
(ii) A path integral approach can be based on the relationship between open Wilson
lines in the commutative and noncommutative picture [31,44].
11 Some motivation for the latter was provided in [40], and [43] provided some more concrete understanding of
the relationship of the generalized star product and SeibergWitten map.
165
(iii) The equivalence of the star products corresponding to the perturbed and unperturbed Poisson structures leads to our formulation in the framework of deformation
quantization. This allows a closed formula for the SeibergWitten map to all orders
in the gauge fields and in .
3.3. Covariance and (non)uniqueness
The objects Da and are not unique if all we ask is that they satisfy the generalized
SeibergWitten condition (106) with star product and have the correct classical limit
where D2 is an -algeD = a + , = + . The pair D2 Da D1 , D2 (),
bra automorphisms and D1 is an equivalence map (possibly combined with a change
of coordinates of the form id + o( 2 )) is an equally valid solution. If we allow also a
transformation to a new (but equivalent) star product then we can relax the condition
on D2 : it may be any fixed equivalence map possibly combined with a change of
coordinates. The maps (differential operators) D1 and D2 may depend on the gauge
potential ai via fij ; it is important, however, that they are gauge-invariant. The freedom
in the choice of D1 and D2 represents the freedom in the choice of coordinates and/or
quantization scheme in our construction; different Da , related by D1 , D2 should be
regarded as being equivalent.
Fig. 1 illustrates how the semi-classical and quantum constructions are affected by a
change of coordinates : the quantization of and in the new coordinates leads to star
products and that are related to the star products and in the old coordinates by
Fig. 1. Two nested commutative diagrams that illustrate the covariance of the semiclassical and
quantum constructions under a change of coordinates given by . The dashed lines indicate
Kontsevich quantization. and are the equivalence maps (including ) that relate the star
products that were computed in the new coordinates with those computed in the old coordinates. The
top and bottom trapezia illustrate the construction of the covariantizing equivalence maps in the old
and new coordinates.
166
equivalence maps and , respectively. (We have included in the definition of these
maps.) Note that in general = . The covariantizing equivalence map, generalized
gauge potential and field strength in the new coordinates and old coordinates are related
by:
Da = 1 Da ,
Aa =
( ) +
(113)
1
Aa .
(114)
Explicit (but complicated) expressions for and in terms of , , and the gauge
potential a can be computed with methods similar to the ones that we have used to compute
Da in the previous section.
3.4. DiracBornInfeld action in the intermediate picture
Seiberg and Witten have argued that the open string theory effective action in the
presence of a background B-field can be expressed either in terms of ordinary gauge theory
written in terms of the combination B + F or in terms of noncommutative gauge theory
, where the B-dependence appears only via the -dependence of the star
with gauge field F
product and the open string metric G and effective coupling Gs . (Here we implicitly need
to assume that is Poisson, which is of course the case for constant .) There is also an
,
+ F
intermediate picture with an effective noncommutative action which is a function of

where is a covariant version of some antisymmetric matrix , with a -dependent star
product and effective metric G and string coupling Gs . The proposed relations between
the new quantities and the background field B, the given closed string metric g and the
coupling gs are 12
1
1
=
,
G+
g+B
det1/2 (g + B) det1/2 (G + )
=
.
gs
Gs
The first relation can also be written more symmetrically:

1 + (G + ) 1 (g + B) = 1,
Gs = gs det1/2 1 (g + B) .
(115)
(116)
Given g and B we can pick essentially any antisymmetric matrix in particular one that
satisfies the Jacobi identity and find G and as symmetric and antisymmetric parts of
the following expression
G+ =
1
(g + B).
1 (g + B)
(117)
For = 1/B (as in the zero slope limit): G = Bg 1 B, = B, Gs = gs det1/2 (Bg 1 ).

For slowly varying but not necessarily small fields on a D-brane the effective theory
is given by the DiracBornInfeld (DBI) action. In the following we will show that
the ordinary DBI action is exactly equal to the semi-noncommutative DBI action in the
12 To avoid confusion with the matrices we will use bold face letters for tensors and forms in this section
(e.g., = 12 ij i j , = 12 ij dx i dx j , = 1 ); for simplicity we shall assume that all matrices are
nondegenerate when needed.
167
intermediate picture. There are no derivative corrections. By semi-noncommutative we

mean the semiclassical limit of a noncommutative theory with star commutators, e.g., in
the noncommutative transformation law, replaced by Poisson brackets as in Section 3.2.1.
Using (116) we can derive the following identity for scalar densities

1
1
1/2
1/2
(118)
det (g + B + F ) =
det
det1/2 (G + + F ),
gs
Gs

where
1
1
,
F =
F
(119)
1 + F
1 + F
(understood as formal power series). Raising indices on F with we get
{x i , x j } {x i , x j } which we recognize as the semiclassical version of

ij = x i , x j x i , x j ,
F
(120)
=
compare Eq. (47). The semi-noncommutative field strength changes by canonical transformation under gauge transformations (78); it is obtained from the invariant by action
of the covariantizing (see also (19), (81)):

( ) = x i , x j x i , x j .
(121)
The corresponding object with lower indices is
= (F ).
F
(122)

change of coordinates :
The Poisson structures and are related by the

matrices , are consequently related to the Jacobian det( (x)/x) of :

(x)
= 1.
det1/2 det

x
= . The
(123)
Using this we can derive the following exact equality for the DBI action with background
field B and the semi-noncommutative DBI action without B (but with ) in the intermediate picture,

1
1 det1/2 ( ) 1/2
), (124)
+ F
d p x det1/2 (g + B + F ) = d p x
det (G +
s det1/2
gs
G
G,
= F that transform semis Gs , G
, F
with covariant G
classically under gauge transformations (78). The only object without a , det1/2 ,
is important since it ensures that the semi-noncommutative action is invariant under gauge
transformations, i.e., canonical transformations. The factor det1/2 ( )/ det1/2 ( ) can be
absorbed in a redefinition of Gs ; it is equal to one in the case of constant , since does
not change a constant;

1
1
+
+ F
) ( const.). (125)
d p x det1/2 (g + B + F ) = d p x
det1/2 (G
s
gs
G
General actions invariant under the semiclassical SeibergWitten map, which includes the
BornInfeld action as well as some actions with derivative terms, have been discussed
in [34] in the case of constant = B 1 and constant metric g.
168
One can consider a fully noncommutative version of the DBI action with replaced
by the equivalence map D and with star products in the appropriate places. That action no
longer exactly equals its commutative cousin but differs only by derivative terms as can be
seen [39] by using the explicit form of the equivalence map (103) ordering ambiguities
in the definition of the action also contribute only derivative terms. The equivalence up
to derivative terms of the commutative and noncommutative DBI actions was previously
shown by direct computation [9] in the case of constant ; in this case an alternative
derivation that is closer to our present discussion and is based on a conjectured formula
for the SeibergWitten map was given in [42]. Requiring equivalence of the commutative
and noncommutative descriptions one can compute derivative corrections to the DBI
action [36].
The semi-noncommutative actions have the general form

1
S = d p x
(126)
L(Gs , G, , , ) ,
(,a)
1/2
det
where L is a gauge invariant scalar function. The gauge potential enters in two places: in
via the gauge invariant field strength f and in (,a)

. It is interesting to note that even the
metric and the coupling constants will in general transform under gauge transformations
since they depend on the gauge potential via . Under gauge transformations L
transforms canonically:

L = L, .
(127)
Due to the special scalar density det1/2 , the action S is gauge invariant and covariant
under general coordinate transformations.
In the background independent gauge = B 1 , = B [32] the action (124)
becomes simply

1
p
1
1/2

det (1 + g ) ,
SDBI = d x
(128)
gs
det1/2
with = (B + F )1 . Expanding the determinant to lowest nontrivial order we find the
following semi-noncommutative YangMills action

1
1
p
j k
li
gij gkl
d x
4gs
det1/2

1
1
= dpx
(129)
gij x j , x k g kl x l , x i ,
1/2 4g
det
s
with covariant coupling constant gs = gs , metric g ij = gij and coordinates x i =
x i . An analogous fully noncommutative version can be written with the help of the
covariantizing equivalence map D

1
p
1/2
j k
li
SNC = d x det D
(130)
gij F gkl F ,
4gs
with an appropriate scalar density det1/2 that ensures upon integration a cyclic trace (this
is as given in
is important for the gauge invariance of the action). In the zero-slope limit F
j
ij
i

= [x , x ]. In the latter case
(120), in the background independent gauge F

SNC =
d p x det1/2
j k
l i
1
,X
gkl X
,X
.
gij X
4gs
169
(131)
This has the form of a matrix model potential (albeit with nonconstant gs , g) with covariant
i = Dx i as dynamical variables.
coordinates X
3.5. Some notes on symmetric tensors
The matrix

ij
1
B +g
ij
= ij + Gij
(132)
plays a central role for strings in a background B-field (with = 0): its symmetric
ij
ij
part S is the effective open string metric and its antisymmetric part A provides the
Poisson structure (provided it is indeed Poisson) that leads upon quantization to the
noncommutativity felt by the open strings. One may now ask out of pure curiosity whether
it is possible to quantize directly, i.e., whether it is possible to find an associative star
product which in lowest order in h is given by (which is not antisymmetric):

i h ij
(133)
i f j g + o h 2 .
2
This is indeed possible, provided A is Poisson (i.e., satisfies the Jacobi identity), since the
f g = fg +
symmetric part of the star product can be gauged away by an equivalence map
(f g) = (f ) (g)
(134)
where

i h ij
(135)
A i f j g + o h 2 .
2
An equivalence map that does the job can be given explicitly in terms of the symmetric
part of :

i h ij
= exp S i j .
(136)
4
f g = fg +
It is enough to check terms up to order h

i h ij
i h ij
i f j g S i j (fg)
2
4

i h ij
i h ij
= fg + A i f j g S (i j f )g + f (i j g) .
2
4
To quantize we can thus proceed as follows: first one quantizes A , e.g., with Kontsevichs formula and then one uses to get from . In the previous sections we saw that the
full information about the noncommutative gauge fields is encoded in the equivalence map
Da . Here we can similarly reconstruct the metric field from by evaluating its Hochschild
field strength
fg +

i h

FH x i , x j = x i x j x i x j = Gij + o h 2 ,
2
(137)
170
or more directly: Gij = 2ih ( (x i x j ) x i x j ). Two more questions come up naturally: when
1
is = (B + g)1
A Poisson? Why is the relevant star product not a quantization of B
as in the zero-slope limit? The answer to the second question is of course that this is determined by the open string propagator in the presence of a background B and that happens
to have an antisymmetric part given by ij and only in the zero-slope limit this is equal to
B 1 . Nevertheless the two questions turn out to be related the star products based on
and are equivalent provided that the 2-forms
1 1
1
(138)
dx i dx j ,
= (gg)ij dx i dx j
ij
2
2
are closed, = d for some 1-form , and B(t) = B + t is nondegenerate for t [0, 1];
the closeness conditions on B and ensure in particular that and are Poisson. According to Mosers lemma the symplectic structures B(1) and B(0) are related by a change
of coordinates generated by the vector field = ij j i and, moreover, according to
Kontsevich the star products resulting from quantization of B(0) and B(1) are equivalent.
Since B(0) = 1 and B(1) = 1 we have demonstrated our claim. Let us remark that
B and are closed if g and B are derived from a Khler metric, since then B is closed
and B, gB 1 g are proportional. Instead of using Mosers lemma we could also drop some
assumptions and work directly with Poisson structures as in previous sections. We would
then require that is Poisson and gg is closed and introduce a 1-parameter deformation
(t), t [0, 1], with
B=
(0) = ,
t (t) = (t) (gg) (t)
(139)
and solution
(t) = + tgg + t 2 gggg + =
1
.
1 t (g )2
(140)
This is the antisymmetric part of

(t) =
1
.
B + t 1/2 g
(141)
The symmetric part is t 1/2 Gij (t) = t 1/2 [ (t)g ]ij with G(0) = B 1 gB 1 and
G(1) = G, while (0) = B 1 and (1) = with G and as given in (132). This suggest that = gg represents metric fluctuations around the background B that can
be gauged away by an equivalence transformation that curiously leads to the zero-slope
values G(0), (0) of the metric and Poisson structure.
4. SeibergWitten map for nonabelian gauge fields

We will now extend the discussion to nonabelian gauge theories, i.e., Lie algebravalued gauge potentials and gauge fields. We will argue that a SeibergWitten map
can be explicitly constructed for any gauge group by treating both the spacetime
noncommutativity and the noncommutativity of the nonabelian gauge group on equal
footing. Both structures are obtained from appropriate Poisson structures by deformation
171
quantization. This construction generalizes to fairly arbitrary noncommutative internal

spaces.
Let us mention that it is possible to absorb a matrix factor (e.g., GL(n) or U(n) in the
defining representation) directly into the definition of the noncommutative space Ax [46]
and then work with the abelian results of the previous sections, however, for other gauge
groups it is not a priori clear how to do this consistently. In any case even for GL(n) and
U(n) that approach would not give a very detailed description of the nonabelian Seiberg
Witten map.
4.1. Nonabelian setting
In this section we shall establish notation and will give a precise definition of the problem
that we would like to solve. Consider a manifold (noncommutative) spacetime with
a noncommutative structure provided by a star product that is derived from a Poisson
structure . On this space consider a nonabelian gauge theory with gauge group G,
field strength F , that can be locally expressed as
F = A A i[A , A ]
(142)
with nonabelian gauge potential A = Ab

where
Lie(G) are generators with
commutation relations i[T a , T b ] = Ccab T c , and nonabelian gauge transformations
Tb
Tb
A = + i[, A ].
(143)
Our main goal is to find a noncommutative gauge potential A = A(A ) and a noncommu , ) such that a nonabelian gauge transformation of A
tative gauge parameter (A
induces a noncommutative gauge transformation of A:
) + A(A
).
+ A ) = A(A
A(A
(144)
) should be a universal enveloping algebra-valued formal power series in ,

A(A
starting with A , that contains polynomials of A and its derivatives. Similarly
, ) should be a universal enveloping algebra-valued formal power series in ,
(A
starting with , that contains polynomials of A , and their derivatives. The product in
the definition of the noncommutative gauge transformation is a combination of the star
product on spacetime and the matrix product of the T a . We expect that it should be
possible to find expressions, where the structure constants Ccab do not appear explicitly,
except via commutators of the Lie algebra-valued A , .
A secondary goal is to find a construction that stays as close as possible to the method
that we used in the abelian case. There, we used a generalization of Mosers lemma to relate
Poisson structures and (and, after quantization, star products and ). The motivation
for this and some of the complications of the nonabelian case can be most easily understood
in the special case of invertible, i.e., symplectic, Poisson structures. The inverses of and
define closed 2-forms B and B that differ by the addition of a (closed) gauge field f
(62). Physically B is the background B-field plus fluctuations f . In the nonabelian case
we would like to keep this picture but with f replaced by the nonabelian field strength F :
B = B + F ,
172
where B = 1 . The trouble with this is that dF = A A = 0 in the nonabelian case so

B and B cannot both be closed 2-forms, which they should be if we want to interpret their
inverses as Poisson structures. Ignoring this, we could then look for a vector field that
generates a coordinate transformation that relates B and B . A natural generalization from
the abelian case (69) is
= A D ,
where we have replaced the abelian gauge potential by the nonabelian one and have also
switched to a covariant derivative D = + i[A , ]. The trouble here is that it is not
clear how to act with the matrix-valued on coordinates. is certainly no vector field;
it is not even a derivation. (Its action turns out to involve complete symmetrization over
the constituent matrices T a .)
The solution to both problems is to consider a larger space that is spanned by the space
time coordinates x and by symbols t a for the generators T a of the Lie Group. is
then the projection onto spacetime of a true vector field and and are the space
time components of true Poisson structures on the enlarged space. We obtain the desired
nonabelian noncommutative gauge theory by quantizing both the external and internal part
of the enlarged space at the same time. To use the method of the previous sections we
need to encode the nonabelian data in an abelian gauge theory on the enlarged space. This
program is successful, if the commutative nonabelian gauge theory can be recovered at
an intermediate step.
4.2. Abelian data
Notation: Greek indices , , , . . . belong to the external space, indices from the
beginning of the alphabet a, b, c, . . . belong to the internal space and indices i, j, k, . . .
run over the whole space (internal and external). We shall use capital letters (A , F ,
) for things related to the nonabelian theory on the external space and small letters (ai ,
fij , ij ) for objects related to the abelian theory on the enlarged space or the internal space
(ab , bc ).
The t a are commutating coordinate functions on the internal space (Lie algebra) just
like the x are commuting coordinate functions on the external space (spacetime). We
later recover the Lie algebra in the form of star-commutators on the internal space and the
matrices T a by taking a representation of that algebra. The star product on the internal
space is a quantization of its natural Poisson structure
a b
t , t Lie = Ccab t c =: ab .
(145)
In the new language
F = A A + {A , A }Lie
(146)
with A (x, t) = Ab (x)t b and

A = + {A , }Lie
(147)
173
with (x, t) = b (x)t b . (Lie algebra-valued translates into linear in t.) We equip the
enlarged space with a Poisson structure ij which is the direct sum of the external
and internal ab Poisson structures

0
=
(148)
.
0
Only for t = 0 is the Poisson structure block-diagonal. (t) and in particular = (1)
acquire off-diagonal terms through the t-evolution
(0) = ,
t (t) = (t)f (t), i.e.,
(t) = tf (t) + t 2 f (t)f (t)
(149)
generated by an abelian gauge field f that is itself not block-diagonal (but whose internal
components fab are zero as we shall argue below). The spacetime components of (t)
can be re-summed in a series in and we miraculously obtain an expression that looks
like the series for (t) but with f replaced by F (t) f tfa ab fb , which at t = 1
(and ab = 0, see below) becomes the nonabelian field strength F F (1) = a a +
{a , a }Lie :
(t) (t) = t i fij j + t 2 j fj k kl flm m

= t f tfa ab fb + .
(150)
To all orders in :
(t) = tF (t) + t 2 F (t)F (t) =
1
.
1 + tF (t)
(151)
In the case of invertible , (with (1)) we have

1
1
(152)
+ F.
=

This resembles the relation B = B + F that one would have naively expected, but
we should note that , are not necessarily Poisson (they are just the spacetime
components of the Poisson structures , ) and F is not exactly a non-abelian field
strength (it is in fact a gauge-invariant expression in abelian gauge fields that coincides
with the nonabelian field strength in the special gauge ab = 0; see next section). The other
components of (t) are computed similarly (again using fab = 0):
b (t) = b (t) = t (t)fa ab ,

ab
ab (t) = ab + t 2 f (t)f .
(153)
(t) is not the only object that acquires a non-abelian look at t = 1; this is also the case
for Mosers vector field
a = ( )ij aj i = ( ) A D + ab ab a ,
(154)
where A = a fa ab ab and D = fa ab b are the first terms in the expansions

for the nonabelian gauge potential and covariant derivative, valid around the special gauge
ab = 0 (see next section).
174
Now we need to identify appropriate abelian gauge fields and gauge transformations
on the enlarged space that upon quantization give the desired nonabelian noncommutative
gauge fields and noncommutative gauge transformations. For this we consider the terms
of lowest order in of the SeibergWitten condition, where we expect to see a purely
nonabelian gauge transformation. Up to this order it is in fact enough to work with the
semiclassical condition (79)
Aa+d = Aa + d + {Aa , }.
Evaluating this on x and collecting terms of order we get

A (a + d) = A (a) + + A (a), Lie ,
(155)
with A (a) and (, a) defined by

Aa x = A (a) + o 2 ,
(156)
= (, a) + o().
An abelian gauge transformation ai = i thus results in a nonabelian gauge transformation of A with gauge parameter . Since we would like to identify A and with
the gauge potential and parameter of the ordinary nonabelian gauge theory that we started
with, they should both be linear in the coordinates t a of the internal space. Studying gauge
transformations and the explicit expression (75), (76) for Aa (x ) (see next section) we
find that this implies that the internal components of the abelian gauge potential are independent of the t a , while the external components are linear in the t a . This is preserved by
abelian gauge transformations with gauge parameters that are linear in the t a :
a = ab (x)t b ,
a = i( b )t b ,
ab = ab (x),
ab = ib .
= b (x)t b ,
(157)
The gauge invariant characterization of the desired abelian gauge fields is fab = 0, fb =
fb independent of the t a , f linear in the t a . By a gauge transformation with parameter
ab (x)t b we can always go to a special gauge with vanishing internal gauge potential
ab = 0 (and a = a (ab )t b ).
We can now apply the method that we developed for the abelian case in the previous
sections to obtain the desired nonabelian noncommutative gauge fields in terms of the
abelian data (157). These are -expanded noncommutative gauge fields that become
ordinary nonabelian gauge fields at lowest nontrivial order in (but all orders in ). We
claim that a re-summation of the -series gives in fact -expanded noncommutative gauge
fields in terms of nonabelian gauge fields. This is a much stronger statement and can be
checked by inspection using the special gauge ab = 0. A rigorous formal proof is however
missing.
4.3. Mini SeibergWitten map
Nonabelian gauge theory is of course also a type of noncommutative gauge theory and
one may thus wonder whether a SeibergWitten map exists from abelian to nonabelian
175
gauge fields. Computing A (a) and (, a) (156) using the results from the previous
sections does in fact provide such maps:
a

e 1
A (a) = ea (a ) +
(158)
( ),
a

a
e 1
()
(, a) =
(159)
a
with (x, t) = ab (x)t b , the parameter of the gauge transformation that gives ab = b
starting from the special gauge ab = 0. Note that a = ab ab a = {, }Lie . In components
A (a) =
b
1
t a M n a (ab nfb ),
(n + 1)!
n=0
(, a) =
n=0
b
1
t a M n a b ,
(n + 1)!
(160)
(161)
with the matrix Mab = Cabc ac and ab t b = a , b t b = . Under an abelian gauge

transformation ai = i ,

A (a) = (, a) + A (a), (, a) Lie .
(162)
In the special gauge of vanishing internal gauge potential ab = 0 the maps becomes simply
A (a) = a , (, a) = .
Acknowledgements
We would like to thank P. Aschieri, J. Madore, S. Schraml and A. Sitarz for helpful
discussions.
Appendix A. Brackets, evolution and parameters

A.1. SchoutenNijenhuis bracket 13
The SchoutenNijenhuis bracket of two polyvector fields is defined by
[1 k , 1 l ]S

=
()i+j [i , j ] 1 i j l ,
i,j
[f, 1 k ]S =

()i1 i (f )1 i k ,
(A.1)
if all s and s are vector fields and f is a function. The hat marks omitted vector fields.
A Poison tensor is a bivector field = 12 ij i j that satisfies the Jacobi identity
0 = [, ]S

1 il j k
l
+ j l l ki + kl l ij i j k .
3
13 A good reference for the material in this section and the next is [45].
(A.2)
176
In terms of the coboundary operator

d = [, ]S ,
(A.3)
this can also be expressed as d = 0 or d2 = 0.

A.2. Gerstenhaber bracket
The Gerstenhaber bracket is given by
[C1 , C2 ]G = C1 C2 ()(p2 +1)(p1 +1) C2 C1 ,
(A.4)
where composition for C1 C p1 and C2 C p2 is defined as

(C1 C2 )(f1 , f2 , . . . , fp1 +p2 1 )

= C1 C2 (f1 , . . . , fp2 ), fp2 +1 , . . . , fp2 +p1 1

()p2 C1 f1 , C2 (f2 , . . . , fp2 +1 ), fp2 +2 , . . . , fp2 +p1 1

+ ()(p2 +1)(p1+1) C1 f1 , . . . , fp1 1 , C2 (fp1 , . . . , fp1 +p2 1 ) ,
p
(A.5)
p
C p may be either Homk (Ax , Ax ) or the space of p-differential operators Dpoly . In

analogy to (A.2) we can express the associativity of a product C 2 as

[, ]G (f, g, h) 2 (f g) h f (g h) .
[, ]G = 0,
(A.6)
In terms of the coboundary operator (see also (29))
d : C p C p+1 ,
d C = [C, ]G ,
(A.7)
this can also be written as d = 0 or d2 = 0.

A.3. t-evolution
Consider a t-dependent function f (t) whose t-evolution is governed by

t + A(t) f (t) = 0,
(A.8)
where A(t) is an operator (vector field or differential operator of arbitrary degree) whose
t-dependence is given. We are interested to relate f (1) to f (0). There is a simple way to
integrate (A.8) without having to resort to t-ordered exponentials: by Taylor expansion
et f (t) = f (t 1).
(A.9)
Due to (A.8) we can insert exp(t + A(t)) without changing anything

et et +A(t ) f (t) = f (t 1).
(A.10)
The trick hereby is that due to the BakerCampbellHausdorff formula all t are saturated
in the product of the exponentials; there are no free t-derivatives acting on f (t), so we can
evaluate at t = 1 and get

et et +A(t ) t =1 f (1) = f (0),
(A.11)
or, slightly rearranged

et +A(t ) et t =0 f (1) = f (0).
177
(A.12)
+ .
The first few terms in the expansion of the exponentials are 1 + A + 12 (A2 + A)
A.4. Semi-classical and quantum gauge parameters
Let A and B be two operators (vector fields or differential operators) and
B0 B,
Bn+1 = [A, Bn ],
(A.13)
then
eA+B eA =

n=0

Bn
eA + o 2 .
(n + 1)!
(A.14)
Semi-classical
We would like to proof (75) and (80)

a+d
a = (d ) a + o 2 ,
(, a) =

(at + t )n ()|t =0
.
(n + 1)!
(A.15)
n=0
It is helpful to first evaluate

[t + at , dt ] = dt (at + t )() .
(A.16)
(Note that both dt and dt at () are Hamiltonian vector fields.)

Proof. (We suppress the t-subscripts and the explicit t-dependence of .)

[t + a , d ](f ) = t {f, } + a {f, } a (f ),

= f (f, ) (d a )(f, ) + f, a ()

= d a () (f ).
(A.17)
We have repeatedly used the definition of d (68) and, in the last step, (70). Now we can
use (A.14) to evaluate

a = et +at +dt et +at et t =0

a+d

dt (at + t )n () t +a t
te
e
=
+ o 2 .
(A.18)

(n + 1)!
t =0
n=0
Quantum
We would like to proof (106) and (108)

1
Da+d Da =
d Da + o 2 ,
i h
(at + t )n ( )|t =0
.
(, a) =
(n + 1)!
n=0
(A.19)
178
First we evaluate

.
= dt (at + t )()
[t + at , dt ]
(A.20)
are inner derivations. Also note that the proof of this

(Note that both dt and dt at ()
equation is in principle much harder than its semi-classical counterpart, since we are now
dealing with differential operators of arbitrary degree.)
Proof. (We suppress the t-subscripts and the explicit t-dependence of .)
[t + a , d ](f
)

= t [f, ] + a d ( )f d ( ) a (f )

= f (f, ) f ( , f ) + a [f, ] a (f ),
f (,
f ) (d a )(f, )
+ [f, a ]
f)
+ (d a )(,
= f (f, )

= d a ( ) (f ).
The desired result (A.19) follows now from (A.14) as in the semi-classical case.
(A.21)
References
[1] J. Madore, S. Schraml, P. Schupp, J. Wess, Gauge theory on noncommutative spaces, Eur.
Phys. J. C 16 (2000) 161, hep-th/0001203.
[2] C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, String loop corrections to beta functions,
Nucl. Phys. B 288 (1987) 525;
A. Abouelsaood, C.G. Callan, C.R. Nappi, S.A. Yost, Open strings in background gauge fields,
Nucl. Phys. B 280 (1987) 599.
[3] C. Chu, P. Ho, Noncommutative open string and D-brane, Nucl. Phys. B 550 (1999) 151, hep-th/
9812219;
C. Chu, P. Ho, Constrained quantization of open string in background B field and noncommutative D-brane, Nucl. Phys. B 568 (2000) 447, hep-th/9906192.
[4] V. Schomerus, D-branes and deformation quantization, JHEP 9906 (1999) 030, hep-th/
9903205.
[5] A. Connes, M.R. Douglas, A. Schwarz, Noncommutative geometry and matrix theory:
Compactification on tori, JHEP 9802 (1998) 003, hep-th/9711162.
[6] M.R. Douglas, C. Hull, D-branes and the noncommutative torus, JHEP 9802 (1998) 008, hepth/9711165.
[7] B. Morariu, B. Zumino, Super YangMills on the noncommutative torus, in: Relativity, Particle
Physics and Cosmology, World Scientific, Singapore, 1998, hep-th/9807198.
[8] W.I. Taylor, D-brane field theory on compact spaces, Phys. Lett. B 394 (1997) 283, hep-th/
9611042.
[9] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032,
hep-th/9908142.
[10] O. Andreev, H. Dorn, On open string sigma-model and noncommutative gauge fields, Phys.
Lett. B 476 (2000) 402, hep-th/9912070.
[11] B. Jurco, P. Schupp, Noncommutative YangMills from equivalence of star products, Eur.
Phys. J. C 14 (2000) 367, hep-th/0001032.
[12] B. Jurco, P. Schupp, J. Wess, Noncommutative gauge theory for Poisson manifolds, Nucl.
Phys. B 584 (2000) 784, hep-th/0005005.
179
[13] M. Kontsevich, Deformation quantization of Poisson manifolds, I, q-alg/9709040.

[14] L. Cornalba, R. Schiappa, Nonassociative star product deformations for D-brane worldvolumes
in curved backgrounds, hep-th/0101219.
[15] B. Jurco, P. Schupp, J. Wess, Nonabelian noncommutative gauge fields and SeibergWitten
map, hep-th/0012225.
[16] B. Jurco, S. Schraml, P. Schupp, J. Wess, Enveloping algebra valued gauge transformations for
non-Abelian gauge groups on non-commutative spaces, Eur. Phys. J. C 17 (2000) 521, hep-th/
0006246.
[17] F. Bayen, M. Flato, C. Fronsdal, A. Lichnerowicz, D. Sternheimer, Deformation theory and
quantization. I. Deformations of symplectic structures, Ann. Phys. 111 (1978) 61.
[18] D. Sternheimer, Deformation quantization: Twenty years after, math.qa/9809056.
[19] A.S. Cattaneo, G. Felder, A path integral approach to the Kontsevich quantization formula,
Commun. Math. Phys. 212 (2000) 591, math.qa/9902090.
[20] P. Schaller, T. Strobl, Poisson structure induced (topological) field theories, Mod. Phys.
Lett. A 9 (1994) 3129, hep-th/9405110.
[21] N. Ikeda, Two-dimensional gravity and nonlinear gauge theory, Ann. Phys. 235 (1994) 435,
hep-th/9312059.
[22] P. Schaller, T. Strobl, Poisson sigma models: A generalization of 2-d gravity YangMills
systems, hep-th/9411163;
P. Schaller, T. Strobl, Introduction to Poisson sigma-models, hep-th/9507020.
[23] A.Y. Alekseev, P. Schaller, T. Strobl, The Topological G/G WZW model in the generalized
momentum representation, Phys. Rev. D 52 (1995) 7146, hep-th/9505012.
[24] L. Bonora, M. Salizzoni, Renormalization of noncommutative U(N) gauge theories, hep-th/
0011088.
[25] A. Armoni, Comments on perturbative dynamics of non-commutative YangMills theory, Nucl.
Phys. B 593 (2001) 229, hep-th/0005208.
[26] A. Armoni, R. Minasian, S. Theisen, On non-commutative N = 2 super YangMills, hep-th/
0102007.
[27] A.A. Bichl, J.M. Grimstrup, L. Popp, M. Schweda, R. Wulkenhaar, Perturbative analysis of the
SeibergWitten map, hep-th/0102044.
[28] A.A. Bichl, J.M. Grimstrup, L. Popp, M. Schweda, R. Wulkenhaar, Deformed QED via
SeibergWitten Map, hep-th/0102103.
[29] A. Connes, Noncommutative Geometry, Academic Press, 1994;
A. Connes, Noncommutative geometry: Year 2000, math.qa/0011193.
[30] J. Moser, On the volume elements on a manifold, Trans. Amer. Math. Soc. 120 (1965) 286.
[31] K. Okuyama, A path integral representation of the map between commutative and noncommutative gauge fields, JHEP 0003 (2000) 016, hep-th/9910138.
[32] N. Seiberg, A note on background independence in noncommutative gauge theories, matrix
model and tachyon condensation, JHEP 0009 (2000) 003, hep-th/0008013.
[33] L. Cornalba, D-brane physics and noncommutative YangMills theory, Adv. Theor. Math.
Phys., in press, hep-th/9909081.
[34] L. Cornalba, Corrections to the Abelian BornInfeld action arising from noncommutative
geometry, JHEP 0009 (2000) 017, hep-th/9912293.
[35] N. Ishibashi, A relation between commutative and noncommutative descriptions of D-branes,
hep-th/9909176.
[36] L. Cornalba, On the general structure of the non-Abelian BornInfeld action, hep-th/0006018.
[37] D. Arnal, D. Manchon, M. Masmoudi, Choix des signes pour la formalite de M. Kontsevich,
math.QA/0003003.
[38] D. Manchon, Poisson bracket, deformed bracket and gauge group actions in Kontsevich
deformation quantization, math.QA/0003004.
[39] P. Aschieri, private communication.
180
[40] S.R. Das, S.-J. Rey, Open Wilson lines in noncommutative gauge theory and tomography of
holographic dual supergravity, Nucl. Phys. B 590 (2000) 453, hep-th/0008042.
[41] T. Mehen, M.B. Wise, Generalized *-products, Wilson lines and the solution of the Seiberg
Witten equations, JHEP 0012 (2000) 008, hep-th/0010204.
[42] H. Liu, -trek II: n operations, open Wilson lines and the SeibergWitten map, hep-th/
0011125.
[43] S.R. Das, S.P. Trivedi, Supergravity couplings to noncommutative branes, open Wilson lines
and generalized star products, hep-th/0011131.
[44] K. Okuyama, Comments on open Wilson lines and generalized star products, hep-th/0101177.
[45] A. Cannas du Silvia, A. Weinstein, Geometric Models for Noncommutative Algebras, AMS,
1999.
[46] J. Madore, Modification of KaluzaKlein theory, Phys. Rev. D 41 (1990) 3709.

Collapsing D-branes in CalabiYau moduli space I

B.R. Greene, C.I. Lazaroiu
Departments of Physics and Mathematics, Columbia University, New York, NY 10027, USA
Abstract
We study the quantum volume of D-branes wrapped around various cycles in CalabiYau
manifolds, as the manifolds moduli are varied. In particular, we focus on the behaviour of these Dbranes near phase transitions between distinct low energy physical descriptions of the resulting string
theory. Whereas previous studies have solely considered quantum volumes in the context of twocycles in perturbative string theory or D-branes in the specific example of the quintic hypersurface,
we work more generally and find qualitatively new features. On the mathematical side, as we briefly
note, our work has some interesting implications for certain issues in arithmetic. 2001 Published
by Elsevier Science B.V.
PACS: 11.25; 11.25.M; 11.15
Keywords: D-branes; Compactification; Mirror symmetry; Moduli space
Introduction
Over the years, studies in quantum geometry have revealed both quantitative and
qualitative deviations from expectations based on classical geometry. Such distinctions,
in turn, have helped us understand physical properties that are intrinsically stringy. From
the earliest studies, based solely on perturbative string theory, we learned that a quantum
CalabiYau is in some sense a classical CalabiYau manifold together with all of its
rational curves [43]. More recently, with the advent of D-branes as non-perturbative probes,
we realized that a quantum CalabiYau manifold is sensitive to all of its supersymmetric
cycles since, regardless of the cycles dimension, there is some D-brane that can wrap
around it (thinking collectively in type IIA and type IIB string theory).
In this paper, we focus on the notion of volume from a string theoretic perspective. 1
There are two papers that provide the groundwork for this undertaking. The first is [34]
which made use of mirror symmetry to calculate the instanton action associated with
E-mail addresses: greene@phys.columbia.edu (B.R. Greene), lazaroiu@phys.columbia.edu (C.I. Lazaroiu).
1 Other approaches to quantum gravity most notably, loop quantum gravity have found interesting
quantum mechanical features of volume [6] so perhaps this area may be a fruitful point of intersection.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 5 4 - 7
182
B.R. Greene, C.I. Lazaroiu / Nuclear Physics B 604 (2001) 181255
strings whose worldsheet is wrapped around a holomorphic two-cycle. The second is [36],
in which another approach to measuring quantum volumes based on the masses of
wrapped D-branes was proposed.
Our goal in this paper is to begin a systematic union of the analyses of those papers in
the following sense: in [34] phase boundaries in the N = 2 moduli space were studied with
only the crudest of tools perturbative string dynamics. This implies that [34] only had
access to the quantum volume of two-cycles. A natural question that immediately follows
is: what can be learned from extending this analysis by probing phase boundaries with
higher-dimensional D-branes? In [36] we learned, in one example (the quintic hypersurface
of [43]), of the kind of surprises that such a study may reveal: it was shown there that the
six-cycle acquires zero quantum volume at the mirror of the conifold point, though the
quantum volumes of two and four cycles remain nonzero. 2 This is a significant deviation
from classical expectations since on a manifold with only one Khler modulus a single
breathing mode classical geometry requires that vanishing of a six-cycles volume
entails the vanishing of all sub-cycle volumes.
We anticipate that some D-brane mass becomes zero at real codimension one along
any phase boundary (since such boundaries intersect the discriminant locus of the mirror
CalabiYau space), and the lesson of [36] is that classical geometric expectations regarding
such degenerations can be completely misleading. We will see that this lesson is quite
general by finding disagreement between the classical and quantum pictures in a variety
of one and two-parameter models. As an interesting aside we point out that the vanishing
of certain D-brane masses at special boundary points in N = 2 moduli space can be used
as a tool for generating arithmetic identities which appear to be independent of known
relations. While we have little insight beyond that afforded by the physics of D-branes, this
mathematical question seems to be ripe for further study.
The present work can be thought of as complimentary to the approach of [46]. The
authors of that paper use conformal field theory methods [49,50] to construct D-brane
boundary states at the LandauGinzburg point in the quintic moduli space, and then study
the behaviour of those states as one perturbs away from this limit. Their goal, which
they largely accomplish, is to find the geometric interpretation of such states in terms of
wrapped D-branes and extract quantum geometrical implications from a conformal field
theory analysis. In our work, we start at the other end of the moduli-space (the large
radius limit), in which the classical geometry is manifest. We then move toward other
phases (including but not limited to the LanduaGinzburg phase) and study what happens
to wrapped D-branes. In time, these two approaches should shed substantial light on the
non-perturbative structure of the string theory moduli space.
In Section 1 we briefly review the results of [34] and [36] in the light of modern
understanding of D-branes. We also give a short discussion of what has (and has not)
been achieved in the literature. Section 2 introduces some old but largely overlooked 3
2 This was also found independently by [13,14].
3 An example where Meijer functions were used to perform the analytic continuation of periods can be found
in [1]. We thank E. Zaslow for bringing this reference to our attention. Issues of analytic continuation were also
183
mathematical methods that will greatly assist the calculations we perform. In Section 3
we discuss some relevant issues regarding mirror symmetry and integral structures. In
Sections 4 through 7 we apply these methods to a number of compact and non-compact
examples, investigating the quantum volumes of two, four, and six-cycles at various
interesting points in their moduli space. Finally, Section 8 discusses some open problems
and presents our conclusions. The two appendices give a brief primer on the notion of
monodromy weight filtration, as well as a list of useful identities.
1. Quantum volumes
In this section we review some ideas relevant for the analysis of quantum volumes in
string theory, and then outline our strategy for implementing them.
1.1. Perturbative quantum volumes
String theory on CalabiYau manifolds provides a rich arena for investigating quantum
geometry. When a CalabiYau space X is large, observables in the corresponding
string model can be expressed in terms of geometrical properties of the compactification
manifold. As this space decreases in size, the values of many observables continuously
shift from their large radius limits, thereby requiring shifted or deformed geometrical
constructs for their interpretation. Collectively, these constructs which reduce to
classical geometric counterparts in the limit of large scales yield what we mean by
quantum geometry. 4
A well known example of this idea arises in perturbative string theory, and provides the
first approach to discussing quantum volumes. If we consider the three-point correlation
function of chiral primary operators Oi associated with classes Ai in H 1,1 (X, C), we find

q
Oi Oj Ok = Ai Aj Ak +
(1)
N (Ai , Aj , Ak ),
1 q
X
where runs over the homology classes of rational curves in X, q is a monomial in

the variables ql = e2itl , with tl the coordinates of the complexified Khler class B +

iJ = l tl El in a basis El of H 2 (X, Z) and N (Ai , Aj , Ak ) are the GromovWitten
invariants [3,4] of X (see [27] for a review). The sum over GromovWitten invariants in
the right-hand side appears through an instanton expansion in the nonlinear sigma model,
with the powers of ql characterizing the contributions from instantons wrapping various
holomorphic 2-cycles. In the limit that the volume of X and all of its rational curves goes
recently discussed in [2], in connection with the Konsevitch conjecture [17].
4 We adhere to common terminology which does not distinguish between stringy geometry and quantum
geometry. More precisely, stringy geometry is concerned with the behaviour of compactified string theory for
small string coupling constant, i.e., with the moduli space of closed conformal field theory defined on the Riemann
sphere. In this paper, we go beyond this framework by including D-branes, so strictly speaking we study the
moduli space of open conformal field theory at genus zero, i.e., the moduli space of boundary conformal field
theory on a sphere.
184
to infinity, the instanton corrections are suppressed and the correlator can be interpreted in
terms of the triple intersection form of X:

Oi Oj Ok Ai Aj Ak .
(2)
X
But as we move away from this limit, the instanton corrections are generally nonzero,
yielding the quantum triple intersection the quantum cohomology ring on
H 1,1(X, C).
The right-hand side of (1) is calculationally unwieldy, but as first shown in [57] and
subsequently evaluated to successful conclusion in [43], if X has a mirror Y then (1) can
be evaluated exactly by a variation of Hodge structure calculation on Y . This procedure
has been employed in a variety of circumstances, one of the most well-known being the
enumeration of rational curves on the quintic hypersurface carried out in [43]. Our interest
here, though, is not in enumerating rational curves. Rather, as implicitly used in [43] and
developed in [34], we note that at a given point in the conformal field theory moduli
space we can use the instanton expansion on the right-hand side to define the quantum
volume of the rational curves in X. Namely, the curves are assigned whatever volume is
necessary to make the expansion correct. More precisely, the variables tl appearing in the
instanton
expansion are tuned to those values which ensure that equality (1) holds. Since

tl = el (B + iJ ) (with (el ) the basis of H2 (X, Z) dual to (El )), we interpret the absolute
value of tl as the quantum volume of the holomorphic two-cycles in the class el .
As strongly motivated in [43] and subsequently proved in [24], this constraint is met if
tl are related to the complex structure of the mirror Y through the mirror map

(z)
tl = Bl + iJl = l
(3)
,
0 (z)
where z are coordinates on the complex structure moduli space of Y , l H3 (Y, Z) are
suitably chosen three-cycles in Y (to be discussed in more detail shortly), and (z) is the
holomorphic three-form of Y at the point z.
Given a mirror pair (X, Y ), it follows that we can map out the quantum volumes of
rational curves in X as its Khler structure is varied by calculating appropriate ratios
of periods of , as functions of the complex structure of Y . Classically, Bl + iJl can
take on any value such that Jl 0, but the approach just described generally yields a
different conclusion. For instance, in some examples, Jl rl > 0, implying a minimum
size for certain rational curves. This departure from the classically allowed two-cycle
volumes
motivates the use of the term quantum volume even though in this case Bl +
iJl = el (B + iJ ) is a classical formula. The quantum aspect arises when this formula is
examined over the space of physically realizable models.
This approach for studying the quantum volumes of two-cycles was developed and
applied in a number of different contexts in [34] and [35]. For example, the quantum
volume of two-cycles involved in the flop transitions of [33] were shown to attain zero
value at the center of the flop (the conifold point of the transition), in keeping with classical
expectations. But in another example in which one moves from a smooth CalabiYau phase
185
to a phase in which the CalabiYau manifold acquires a Z3 orbifold singularity, two-cycle

volumes were shown to be bounded from below, in conflict with geometric expectations
(at an abelian quotient singularity of this type, an exceptional divisor has been blown down
to a point, which entails vanishing of the geometric volume of all holomorphic two-cycles
embedded in the divisor). Thus, there can be significant differences between the quantum
and classical properties of two-cycle volumes.
While revealing these interesting features of quantum volumes, the approach of [34] is
deficient for two main reasons: first, it is limited to the study of two-cycle volumes and
second (relatedly) it only uses perturbative string probes. D-branes, as emphasized in [36]
provide a natural tool for going further.
1.2. Nonperturbative quantum volumes
Since BPS saturated D-brane states wrapped around cycles in a CalabiYau space have
a mass that depends on the size of the cycles they wrap, we can use these masses to
define the quantum volume of cycles. Since type IIA string theory has even-dimensional
branes while type IIB string theory has odd-dimensional branes, considering both theories
compactified on the same CalabiYau manifold X allows us to define a notion of quantum
volume for cycles of all possible dimensions.
More specifically, in type IIB string theory on X, BPS saturated D3-branes wrap threedimensional special Lagrangian cycles of X [45] (these are the type A branes of [44]).
While it is a difficult mathematical problem to identify the special Lagrangian 3-cycles
of a CalabiYau space, if C is such a cycle then the mass of a D3-brane wrapped on C
depends only on the homology class = [C] H3 (X) of C, and identifying elements
of H3 (X, Z) is a much easier problem. If, for example, we choose an integral basis i of
H3 (X, Z), then we can express = q i i with q i Z (note that [C] is an integral class
since C is a submanifold of X) and write the BPS mass as [18,36]:

|q i i |
| |
=
.
m( ) =
(4)
1/2 ( )
1/2
( )
X
The integers q i can be interpreted as gauge charges with respect to the U (1)-gauge fields
resulting from the KaluzaKlein reduction of the RamondRamond four-form along the

cycles i , while the periods i are related to the vacuum expectation values of the scalar
fields which parameterize the complex structure moduli space. Hence (4) takes the standard
form of charges times vevs. Since C is special Lagrangian, 5 we have m( ) = vol(C) with
the proper normalization of . Hence the mass of an A-type D-brane is proportional to
the volume of the cycle it wraps, a relation which is quantum-mechanically exact. Thus,
modulo issues of marginal stability, no interesting quantum effects contribute to the masses
of A-type D-branes.
5 We assume that C is special Lagrangian with respect to (z). This condition depends on the point z on the
moduli space, as discussed in some detail in [9] such dependence is related to issues of marginal stability,
which we ignore in this paper.
186
In the type IIA theory on X, BPS saturated D-branes wrap holomorphic cycles (these are
the type B branes of [44]). If we choose an integral basis hi of Heven (X, Z) = H0 (X, Z)
H2 (X, Z) H4 (X, Z) H6 (X, Z), then a D-brane wrapping a holomorphic cycle in the
class h = q i hi has mass of the form |q i i | where i are the vevs of the scalar fields
associated with the Khler structure of X. In the large radius limit the mass of the Dbrane is proportional to the classical volume of the cycle it wraps, but this simple relation
is destroyed by quantum corrections as one moves away from this limit.
As observed in [36], one can can turn this argument around and define the quantum
volume of a holomorphic cycle of X to be proportional to the mass of the type B brane
wrapping it. This is a natural generalization of the geometric notion of volume inasmuch
as its reduces to the former in the large radius limit, where the full content of the string
theory compactified on X (at weak string coupling) can be described in terms of standard
geometric concepts.

Unfortunately, whereas the integrals i give a simple, closed form for the relevant
vevs for the odd-dimensional branes on X (associated with the complex structure of X)
there is no simple, closed form expression for the i . Indeed, the semiclassical formula
relating the mass of such a brane to the volume of the associated holomorphic cycle (and
the values of the B-field and worldvolume gauge field) is known to receive important
quantum corrections, which are difficult to analyze directly on X. The situation is in many
ways similar to that encountered for the action of string instantons wrapped over rational
curves and encoded by relations (1) and (3). However, we can follow [36] and circumvent
this obstacle by using mirror symmetry.
For this, consider the type IIB theory compactified on the mirror Y of X. Conformal
and effective field theory arguments [44] show that under mirror symmetry D2k -branes in
the IIA theory on X are mapped to D3-branes in the IIB-theory on Y . 6 This identifies the
mass of a type B D-brane with the mass of its mirror type A brane, which is amenable
to direct computation via Eq. (4). Hence invoking both X and its mirror allows us reduce
the problem of computing masses of BPS saturated D-branes to computing periods of
holomorphic three-forms. 7
1.3. Strategy of calculation
In view of (4), our strategy for measuring quantum volumes of holomorphic
submanifolds of X is in principle as follows:
6 More precisely, they are mapped to D3-branes wrapping special Lagrangian cycles whose homology classes
belong to the component Wk of the reduced monodromy weight filtration associated with the large complex
structure point of Y (see Appendix A).
7 It is believed [19] that the moduli space of type B branes admits a physically correct compactification related
to certain moduli spaces of semi-stable simple sheaves on X, which should form the natural framework for
finding an expression of the associated BPS charge directly in terms of type B D-brane data. Such an analysis
will necessarily encode important information about moduli spaces of stable sheaves on X. By using mirror
symmetry, we are somehow summing up this information and re-expressing it in terms of the data of Y
its complex structure and the collection of its special Lagrangian 3-cycles. It should be possible to follow this
approach in the other direction, in order to extract information about moduli spaces of sheaves on X.
187
(1) Choose an integral basis of three-cycles i in H3 (Y, Z).

(2) Calculate the periods i with respect to this basis for all points in the complex
structure moduli space of Y .

(3) Find the explicit mirror map between j H2j (X, Z) and H3 (Y, Z).

(4) Determine the quantum volume of any even-dimensional cycle in j H2j (X, Z) by
equating it (up to an overall normalization) with the mass of a D3-brane wrapped on the
mirror three-cycle in Y .
In practice, one encounters two 8 difficulties in carrying out this strategy. First, it is
generally a significant challenge to find explicitly a basis of integer three-cycles, i.e., it is
difficult to carry out step (1). Second, while the mirror map between H2 (X, C) and (a part 9
of) H3 (Y, C) is well understood, and while we certainly know that this map extends to

an isomorphism between
j H2j (X, Z) and H3 (Y, Z), realizing this map explicitly is
generally beyond what as yet has been worked out, i.e., step (3) nontrivial. How, then,
do we proceed?
Let us deal with the second problem first. In [43] the authors showed that integral twocycles on X are mirror to integral three-cycles on Y with prescribed monodromy properties
around a large complex structure point in the moduli space. Concretely, in a one-parameter
example, the three-cycle l in the numerator of (3) must satisfy l l + 0 upon parallel
transport about a large complex
structure point. When translated into the language of

periods, this implies that l has a logarithmic dependence on the coordinate z (chosen
such that z = 0 is the large complex structure point). In [12,15,22,43,61], this notion was
formalized and generalized to the statement that holomorphic cycles of real dimension
2j on X are mirror to three-cycles on Y whose periods have leading logj z behavior
near z = 0. Thus, finding a complete set of periods of and classifying their leading
logarithmic behavior gives us a means of identifying the dimension of their even-cycle
counterpart on X.
Although this is part of the approach we will follow, it is certainly deficient since it
says nothing about the subleading logarithmic dependence of the periods. For instance, is
the six-cycle on X mirror to a three-cycle on Y whose period is purely proportional to
log3 (z) or is there some nontrivial dependence on log2 (z) and log(z)? This is a question
that, at present, we do not know how to answer completely it awaits a more thorough
understanding of the action of mirror symmetry on the integral structure of a mirror pair.
Thus, in this paper when we speak about a cycle on X of real dimension 2j , that will often
refer to a cycle on X with 2j being the maximal-dimensional component, but with the
8 In fact, there is a third and much more serious problem involved in the whole subject, which belies
many difficulties of interpretation and various issues of marginal stability touched upon in [46]. This rests on the
following question: given an integral class H3 (Y, Z), does this class support a special Lagrangian 3-cycle? If
so, what is the moduli space of cycles in this class? We will mostly ignore this issue in the present paper, though
we plan to discuss some aspects of this problem elsewhere. For recent work on this subject we refer the reader
to [9] (see also [10] for a partial physical translation of some of those results). A complete understanding of this
issue will most probably require a systematic development of the theory of moduli spaces of special Lagrangian
submanifolds along the lines of [7,8].
9 This is the component W H (Y ) of the large complex structure reduced monodromy weight filtration.
1
3
The behaviour of the elements of W1 under mirror symmetry is clearly discussed in [30,62].
188
identity of the admixture of lower cycles left unspecified. In Section 3 we will quantify
how close we can currently get toward erasing this ambiguity. Gaining greater precision in
this regard is an important problem for future work.
As for the first problem mentioned above that of ensuring that the three cycles with
which we work are integral we propose the following. First, our main focus in this
paper is to examine D-branes wrapping cycles whose mass vanishes at special points
in the moduli space. For this purpose we can weaken the requirement that the i are
integral classes, and demand only that they are proportional to such classes. That is, if
= H3 (Y, Z) is the integral lattice of 3-cycles in H3 (Y, C), then all we require is that each
of the i we study is proportional to an element in , with a common proportionality factor.
In
this case, given a cycle which is an integral linear combination of i , the vanishing of
at some point in the moduli space implies the existence of an integral cycle (of class
proportional to ) having the same property. We call cycles that are proportional to integral
cycles weakly integral. Second, given a three-cycle which is weakly integral, we can
generate other weakly integral cycles (j ) in H3 (Y, C) by acting on it with monodromy
transformations. A cycle H3 (Y, C) with the property that the set of weakly integral
cycles thus produced is a system of generators for the complex vector space H3 (Y, C)
will be called cyclic; indeed, it is a cyclic vector for the monodromy representation of the
fundamental group of the moduli space by automorphisms of H3 (Y, C). If is both cyclic
and weakly integral, then there exists some complex constant such that 1 is integral.
Since the monodromy operators preserve the lattice , it follows that the cycles 1 (j )
are integral as well. Then one can easily construct the maximal lattice 0 H 3 (Y, C)
having the property that it contains all (j ) , and it is an easy exercise to show that :=
1 0 is a full sublattice 10 of , though it may fail to coincide with itself.
For example, in compact one parameter models, three-cycles undergo monodromy
transformations not only about the large complex structure point z = 0 but also about the
points z = 1 and z = (with our choice of coordinate on the moduli space). There always
exists a fundamental cycle 0 H3 (Y, C) with the property that it is left invariant by all
monodromy transformations about the large complex structure point (in the framework
of [15], this is the cycle mirror to a 0-cycle, namely the homology class of the fiber of the
special Lagrangian T 3 -fibration of Y ). However, 0 will not generally be invariant under
T [] and T [1], and it is often the case that it is a cyclic vector for the action of these
monodromy operators. In fact, a particular version of this procedure (involving only the
monodromy operator T [] and applicable for the case when z = is a LandauGinzburg
point) underlies the method used in [43] for producing a basis of periods.
We can now recast our strategy
for measuring quantum volumes into the following form:

(1) Calculate the period 0 for a weakly integral three-cycle 0 .
(2) Calculate the monodromy matrices T [0], T [1], and T [], and use them to produce
a weakly integral basis of three-cycles, and their corresponding periods
(3) Search for places in moduli space where a period associated to a weakly integral
three-cycle vanishes.
10 That is, a maximal rank sublattice of , i.e., a sublattice such that the / is a finite group.
0
189
(4) Interpret the vanishing period of a weakly integral three-cycle on Y with leading
logj (z) behavior around z = 0 as the vanishing quantum volume of a holomorphic cycle
on X of real dimension 2j .
As mentioned before, a more precise version of step (4) would involve identifying the
admixture of lower-dimensional cycles on X mirror to the vanishing cycle on Y . This
requires understanding mirror symmetry at the level of integral structures and in Section 3
we discuss aspects of this issue. In the next section we consider the problem of efficiently
calculating a complete set of periods over a basis of three-cycles, together with procedures
for identifying the monodromy matrices that is, techniques for carrying out steps (1)
through (3). We will then use these results to carry out the four steps above in various
examples.
Before proceeding with this analysis, let us make a few brief remarks on the status
of results which can be found in the literature. Most previous work which involved the
calculation of periods was concerned with deepening and generalizing the computation of
GromovWitten invariants first performed in [43]. Since these invariants can be extracted
from the mirror map (3), the literature is mainly focused on the computation of the
fundamental and log1 periods. Ironically, though, some of the earliest works on applying
mirror symmetry to enumerative geometry did consider the higher periods (e.g., [22,37,
38,43]), but their physical relevance in those pre-D-brane studies was, of course, not
considered. In this work we present a method for analyzing all of the periods, throughout
the moduli space, in a scheme that is calculationally easier than previous methods and
which allows their physical content to be directly extracted.
2. Techniques
Although we shall consider an example with h1,1 > 1, we will follow [34] and focus
on judiciously chosen one-dimensional subspaces in the relevant moduli space. Hence, in
this section we start by discussing a calculationally efficient technique for analyzing oneparameter systems. The approach we describe is based on classical mathematical results
due to C.S. Meijer and uses the so-called G-functions (or Meijer functions). As we show
below, these functions are especially well-suited for a systematic computation of a basis
of periods and its monodromies. Our use of G-functions grew out of our effort to find a
more effective approach to such computations, and hopefully represents an improvement
over the methods used in [43]. From an abstract point of view, the approach taken there
consists of the following four steps. First, one computes the fundamental period 0 near a
large complex structure point by explicitly integrating the holomorphic 3-form along a
carefully chosen 3-cycle. Second, one analytically continues this period to a point of finiteorder monodromy (assuming that such a point exists, which is not always the case). Third,
assuming that the order of the monodromy action at that point is sufficiently high and
that the fundamental period is a cyclic vector for the corresponding monodromy operator,
one can produce a basis of periods j by repeatedly acting with this operator on 0 . In
practice, this is carried out by using an explicit realization of the finite-order monodromy
190
as a local orbifold action on a branched multiple cover of the moduli space; this procedure
then amounts to rotating the branch cut used in the expansion of 0 near the small radius
point in order to obtain the higher periods j . Finally, one can in principle analytically
continue back from the finite-monodromy point to the large complex structure point by
using carefully chosen integral representations compatible with the branch-cuts introduced
in defining j . In practice, finding convenient integral representations of j for this
purpose is not always straightforward.
This approach depends on various assumptions and can be difficult to carry out in
practice, due to the computational complexity involved and the less than algorithmic nature
of some of the steps. The alternative we propose is calculationally simpler and more
algorithmic. It consists of the following steps.
First, we use the theory of Meijer functions to systematically (and rather easily)
produce a special basis of periods adapted to the large complex structure monodromyweight filtration. (The reader unfamiliar with this concept can find a brief description in
Appendix A.) These periods (which we will call Meijer periods) are expressed in terms
of Meijer functions, which are defined by certain integral representations of Mellin type.
Analytic continuation of the Meijer periods is thus straightforward to perform, with the
choice of branch-cuts automatically fixed by the condition of convergence of the associated
contour integral. This systematically produces the expansions of the periods in all regions
of the moduli space, without requiring the computation of monodromies. If knowledge of
the monodromy matrices is required, then they can be computed systematically from the
expansions of the Meijer periods. Once this has been achieved, a cyclic basis of the type
considered in [43] can be easily produced, and can be used to provide some (but generally
not complete) information about the integral lattice H 3 (X, Z) H 3 (X).
We begin with a short review of Meijer functions, with the main purpose of fixing our
notation and conventions. Then we illustrate the technique outlined above in a class of oneparameter examples which includes the mirror quintic as well as other models considered
in the literature. Finally, we consider a certain degeneration of this class of models, which
in particular contains a model we will encounter again when studying a two-parameter
example in Section 6.
2.1. Meijer functions
We start by reviewing a few basic facts about a class of higher transcendental functions
introduced by C.S. Meijer. These functions can be used to generate a fundamental system
of solutions of a generalized hypergeometric equation, and therein lies their utility. In a
sense, they form the natural class of functions associated with hypergeometric systems.
As the periods of the holomorphic three-form satisfy a generalized hypergeometric
equation, we will be able to use these methods to calculate quantum volumes. (We refer
the interested reader to [52] for details about the general theory of such functions and to
the excellent paper [55] for a review of their applications to hypergeometric equations.)
The modern definition of Meijer functions proceeds via MellinBarnes integrals. To
simplify subsequent formulae, we introduce the notation
191
Fig. 1. The defining contour for a MellinBarnes integral. A-type poles are represented by circles
and B-type poles by short vertical lines.
x1 , . . . , xn

y1 , . . . , ym
:=
(x1 ) (xn )
(y1 ) (yn )
(5)
for any complex numbers xi , yj . Consider the MellinBarnes integral in the complex plane

a1 , . . . , aA b1 , . . . , bB
(z)

c1 , . . . , cC d 1 , . . . , d D

1
a1 s, . . . , aA s b1 + s, . . . , bB + s
=
(6)
ds
(z),
c1 s, . . . , cC s d1 + s, . . . , dD + s
2i
with complex ai , bj , ck , dl and z. For generic a, b, c, d, the integrand has two series of
poles
(A) ai s = n,
(B)
bj + s = n
with n a nonnegative integer. The contour is taken to extend from i to +i (in this
order) in such a way as to separate the A-type poles from the B-type poles (see Fig. 1).
The conditions for convergence of (6) are discussed [52], to which we refer the reader for
details. Here we only mention that an essential requirement is that |arg(z)| < , which
is necessary in order to assure rapid decrease of the integrand at imaginary infinity. This
is an important point implicit in the following sections: when considering the analytic
continuation of periods from large to small radius for one-parameter models, the choice of
branch-cuts is automatically enforced by the use of MellinBarnes representations.
192
The Meijer functions are defined by

1 , . . . , r 1 , . . . , r
G
(z)
1 , . . . , s 1 , . . . , s

1 , . . . , s 1 1 , . . . , 1 r
=I
(z)
1 , . . . , r 1 1 , . . . , 1 s
(7)
whenever the Mellin integral appearing on the right-hand side converges. We will
frequently use I rather than G since the former displays the structure of the MellinBarnes
integrand more clearly. The functions I and G satisfy certain differential equations which
can be obtained by considering the finite difference identities obeyed by the associated
integrands. This is easiest to write down for G. Indeed, it is not hard to see that the function

r r+1 ,...,R
u(z) = G 11,...,
,...,s s+1 ,...,S (z) satisfies

D
z
1 , . . . , R
1 , . . . , S

u = 0,
(8)
with the operator

1 , . . . , R
Dz
( i ) (1) z
=
1 , . . . , S
i=1,...,S
( j + 1),
(9)
j =1,...,R
d
where = z dz
and = R r s (mod 2) = r s (mod 2). While u depends
strongly on the partitions (1 , . . . , R ) = (1 , . . . , r ; 1 , . . . , r ) and (1 , . . . , S ) =
(1 , . . . , s ; 1 , . . . , s ) used for its definition, the differential operator (9) depends on
these partitions only through the number Z2 . Thus, by considering different partitions
of (1 , . . . , R ) and (1 , . . . , S ) as arguments in G, together with the observation that

+1 1 ,...,R
R
Dz 11,...,
,...,S = Dz
1 ,...,S , we learn that the functions

a1 , . . . , aA b1 , . . . , bB
(z),
c1 , . . . , cC d 1 , . . . , d D

a1 , . . . , 1 dl , . . . , aA
b1 , . . . , bB
I
(z)
c1 , . . . , cC
d1 , . . . , dl , . . . , dD

and

I
a1 , . . . , aA
c1 , . . . , 1 b j , . . . , cC
b1 , . . . , bj , . . . , bB
d1 , . . . , dD

(z)
(where a hat indicates removal of the corresponding element from the list) satisfy the
same differential equation. Therefore, given a Meijer function I, one can produce other
solutions of the same equation by the simple procedure of lifting/lowering the parameters
diagonally in the symbol of I, together with replacing each parameter thus moved by
1 and performing a change of sign of the argument z for each such operation. This
simple observation is the basis of our procedure for obtaining all logj -monodromy periods
starting from the fundamental period, as we explain in more detail below.
193
The Meijer functions are relevant to our problem for the following reason. The periods
of the holomorphic three-form on a CalabiYau manifold are closely related to generalized
hypergeometric functions, 11

1 + n, . . . , p + n zn
1 , . . . , p
1 , . . . , q

(z) =
p Fq
1 , . . . , q
1 , . . . , p
1 + n, . . . , q + n n!
n=0

(1 )n (p )n zn
n=0
(1 )n (q )n n!
(with p = q + 1), which are solutions to the generalized hypergeometric equation:

( + i 1) z
( + j ) u = 0.
i=1,...,q
(10)
j =1,...,p
The generalized hypergeometric functions, in turn, are particular cases of Meijer functions: 12

1 , . . . , p
(z)
p Fq
1 , . . . , q

1 , . . . , q
1 1 , . . . , 1 p
=
G
(z)
1 , . . . , p
0
1 1 , . . . , 1 q

1 , . . . , q
0 1 , . . . , p
=
I
(z)
1 , . . . , p
1 , . . . , q
(where indicates that the corresponding group of parameters is missing). Indeed, the
hypergeometric operator

1 , . . . , p
( + i 1) z
( + j )
=
dz
(11)
1 , . . . , q
i=1,...,q
j =1,...,p
is related to the Meijer operator by

1 , . . . , p
1 1 , . . . , 1 p
dz
= D1z
.
1 , . . . , q
0, 1 1 , . . . , 1 q
(12)
In our CalabiYau applications, the hypergeometric equations arise as typical Picard

Fucks equations obeyed by the periods of a one-parameter model. In such a model, we
will always choose the coordinate z on the moduli space such that z = 0 corresponds
to the large complex structure (or large radius) point, z = to the small radius limit
and z = 1 to the other regular singular point (the analogue of the conifold point of [43]).
This assures that the PicardFucks equation will always have the standard hypergeometric
form (10). We will sometimes call this coordinate on the moduli space the standard or
hypergeometric coordinate. Then the parameters i , j will always be such that z = 0 is
a point of maximally unipotent monodromy, which implies that the only solutions of (10)
11 Here (x) := x(x + 1) (x + n 1), (x) := 1 denotes the Pochhammer symbol.
n
0
12 Note that convergence of the integral on the right-hand side requires |arg(z)| < .
194

p
which are regular at the origin are multiples of p Fq 1 ,...,
(z) (the hypergeometric
1 ,...,q
function itself is characterized among such solutions by the fact that it has value 1 at z = 0).
The hypergeometric solution thus corresponds to the fundamental period of [40], and is
trivial to write down in practice. Writing this solution as a Meijer function and making
simple operations on its symbol as above will allow us to produce a basis of solutions
of (10), i.e., a basis of periods for the model. Moreover, the Meijer periods thus produced
are automatically adapted to the monodromy weight filtration (see Appendix A for an
explanation of this concept) associated with the large complex structure point. In fact,
these periods display a rather universal behaviour around z = 0, as we will see in detail
below.
2.2. Monodromies
The strategy outlined in Section 1 requires that we understand the behavior of the
Meijer periods under monodromy transformations. In this section we set up some general
formalism that will allow us to accomplish this goal. Our remarks will clarify the
connection with the general theory of Fuchsian systems as explained, for example, in [53]
(see also [54]).
2.2.1. General remarks
Let us start with a discussion of monodromies for the Meijer equation (8). For simplicity
(and direct relevance to this paper) we consider only the case S = R := p, which includes
the hypergeometric equation. The logarithmic form of the Meijer operator (9) is

p1

1
1 , . . . , p
p
=
+
Bk (z) k ,
D
z
1 , . . . , p
1 + (1)+1 z
(13)
k=0
where we defined
Bk (z) :=
Qk + (1)+1 zRk
,
1 + (1)+1 z
(14)
with
Qk := (1)pk
i1 ipk ,
(15)
(i1 1) (ipk 1).
(16)
1i1 <<ipk p
Rk := (1)
pk
1i1 <<ipk p
The coefficient functions Bk (z) interpolate between Bk (0) = Qk and Bk () = Rk .

The Meijer equation (8) can be rewritten as the first order system
w(z) = A(z)w(z),
where
(17)
195
u(z)
u(z)
w(z) :=

p1 u(z)
and we set
0
1
0
0
A(z) :=
0
0
B0 (z) B1 (z)
0
1
0
0
0
0
0
1
Bp2 (z) Bp1 (z)
(18)
Given an arbitrary basis {Uj (z)}j =0,...,p1 of solutions of (8), we define its monodromy
matrices T [0] and T [] around z = 0 and z = by

2i
T [0]U (z),
for |z| 1,
U e z =
T []U (z), for |z| 1,
where
U0 (z)
U1 (z)
U (z) :=
.
Up1 (z)
If (z) is the associated fundamental matrix of (17), i.e., the matrix having
Uj (z)
Uj (z)
wj =
p1 Uj (z)
(j = 0, . . . , p 1) as its columns, then the differential version of the nilpotent orbit
theorem states that there exists a regular 13 matrix-valued function S (which, following
modern terminology, we call the nilpotent orbit of ) and a matrix R such that 14
(z) = S(z)zR .
(19)
t
This immediately gives T = e2iR and implies that S(z) satisfies the matrix differential
equation
S(z) = A(z)S(z) S(z)R.
(20)
(n) n
Expanding S(z) = n=0 S z shows that the matrix S (0) := S(0) (which plays the role
of initial condition for (20), and thus must be invertible if (z) is to be a fundamental
matrix for (17)) must satisfy
S(0)1 A(0)S(0) = R.
13 Regular in the vicinity of z = 0.
14 This decomposition has a standard geometric interpretation, which we recall in Appendix A.
(21)
196
Under a change of basis U (z) MU (z), we have (z) (z)M t and

R M t RM t ,
S(z) S(z)M
(22)
(23)
(where M t := (M 1 )t ). In particular, we have S(0) S(0)M t . Choosing M := S(0)t

gives a distinguished fundamental system can (z) with the properties
0
1
0
0
0
0
0
1
0
0

Rcan [0] := A(0) =
(24)
,
0
0
0
0
1
Q0 Q1 Qp2 Qp1
Scan (0) := I,
(25)
whose associated basis Ucan will be called the canonical basis.
To understand the behaviour at z = , note that 1/z = z , which gives

(1)+p+1 1 1 , . . . , 1 p
1 , . . . , p
D1/z
Dz
=
.
1 , . . . , p
1 1 , . . . , 1 p
z
(26)
This is again a Meijer operator, with symbols j := 1 j and i := 1 i . The canonical

monodromy is again of the form (24), but with Qk replaced by Qk := (1)pk Rk :
0
1
0
0
0
0
0
1
0
0
Rcan [] :=
(27)
.
0
0
0
0
1
Q0 Q1 Qp2 Qp1
The minus sign in front of this expression keeps track of the fact that the transformation
z e2i z corresponds to 1z e2i 1z ; that is, inversion in the complex plane reverses
the orientation of a contour which surrounds the origin. This amounts to inversion of the
associated monodromy matrix, an observation which will be used again in Section 4.3.
2.2.2. The hypergeometric case
Let us now apply these results to the hypergeometric case, which is the case of interest
for us. In view of (11), this is defined by R = S = p = q + 1 and
i := 1 i
1 := 0,
(i = 1, . . . , p),
j +1 := 1 j
(j = 1, . . . , q).
According to the discussion above, if u(z) satisfies the generalized hypergeometric equa

q+1
tion with (hypergeometric) symbol 1 ,...,
, then u(t)
:= u(1/t) satisfies the Meijer
,...,q
1
1 ,...,q
equation Dt 1,
u = 0 (note that this equation is not generally of hypergeometric
1 ,...,q+1
type). Applying the general results derived above shows that the canonical forms of the
hypergeometric monodromies around z = 0 and z = are
0
1
0
0
0
0
1
0
Rcan [0] : =
0
0
0
0
0 Q1 Qq1
0
1
0
0
0
1
Rcan [] : =
0
0
0
(1)q R0 (1)q1R1
with
Qk : =
0
0

,
1
Qq
0
Rq1
(i1 1) (iqk+1 1),
197
(28)
0
0
,
1
Rq
(29)
(30)
1i1 <<iqk+1 q
Rk : =
i1 iqk+1 ,
(31)
1i1 <<iqk+1 q+1
where k = 0, . . . , q. In this paper we only consider models for which all i are equal to 1.
Then all Qk are zero and the matrix Rcan [0] is in Jordan form. On the other hand, the
Jordan form of the matrix Rcan [] depends on the relative values of the parameters k , as
we will illustrate in detail below.
Before proceeding to apply the theory of Meijer functions to various examples, let us
mention that the remaining regular singular point of (10) (namely the point z = 1) is
somewhat special in the following sense. While one can change variables (for example,
by the transformation z (1 z)/z) in such a way as to move the point z = 1 to the origin
of the complex plane, the resulting equation is not of Meijer type. This is the root of the
technical difficulties encountered when studying the behaviour of a system of solutions
of (10) around z = 1, and the main reason why a completely uniform treatment of this
problem is not yet available. We refer the interested reader to [55] for a thorough discussion
of what is known about this subject.
On the other hand, the monodromies around z = 1 can be easily computed for a
convenient basis of solutions around that point, by making use of the general theory of [53].
However, this will not be relevant for us, for the following reason. The canonical bases
associated with the points 0 and (as well as the more general standard basis associated
with the point z = 1) generally correspond to different fundamental systems of solutions
of (10). What one is interested in when studying one-parameter models is to compute the
monodromies of a given basis around these three points; in other words, one considers the
analytic continuation of a fundamental system throughout the entire complex plane and
one asks for the monodromies of the basis of solutions thus obtained. If T [0] and T []
are the monodromies of such a system about 0 and , then the monodromy of that system
about z = 1 is easily obtained as T [1] = T [0]1 T [], an equation which follows from the
structure of the fundamental group of the thrice-punctured Riemann sphere P1 {0, 1, }.
This relation, as well as our convention for the monodromy matrix T [1] are depicted in
Fig. 2. In practice, it will thus suffice to compute T [0] and T [] for an analytically
198
Fig. 2. Generators of the fundamental group 1 (P1 {0, 1, }, p) of the thrice-punctured sphere
(using the base point p = 1/2). The monodromy operators T [0], T [1], T [] used in this paper
correspond to the paths 0 , 1 , , all of which are oriented in counterclockwise manner. The
relation 0 1 = induces the equality T [1] = T [0]1 T [].
continued fundamental system. As we illustrate in detail below, the most convenient choice
is to consider the analytic continuation of the Meijer basis, which is easily obtained from
its MellinBarnes representation through a computation of residues. This basis has the
added advantage that it allows for a very systematic computation of the monodromies
about 0 and . The utility of the canonical monodromies computed above is that they will
appear as intermediate steps in this procedure, which essentially consists in computing the
transition matrices from the Meijer basis to the two canonical bases associated with the
points 0 and .
3. The mirror map and integral structure
Identifying the B-type D-brane state associated with a collapsing holomorphic cycle
requires a deeper understanding of the map induced by mirror symmetry between H3 (Y, Z)
and Heven (X, Z). In this section we discuss what is known about this issue and identify the
obstacle that as yet needs to be surmounted. While the we believe that the essential point we
make below has not been widely appreciated, most of what follows involves manipulating
and reinterpreting ideas of [5] and [43] to which we refer the reader for background and
notation.

In deriving the mirror map tl = l / 0 for the quintic mirror-quintic pair (X, Y ),
the authors of [43] proceeded in the following way. Choose an integral symplectic basis
(Aa , B b ) of H3 (Y, Z) with dual cohomology basis ( a , b ), where a, b = 1, 2. Then write
the holomorphic three-form of Y as

a

b ,
=
(32)
Aa
Bb
and the prepotential G as

1
G=
.
2
Aa
Ba
Defining coordinates za in terms of the periods through za =

= >j gj ,
199
(33)

Aa
allows us to write
(34)
with >t = (z1 G, z2 G, z1 , z2 ) and g = (1 , 2 , 1 , 2 ).

One can follow a similar procedure for X. Choose an integral basis of cycles
(E1 , E2 , F 1 , F 2 ) for H0 (X, Z), H2 (X, Z), H4 (X, Z), and H6 (X, Z), respectively. Let
( 1 , 2 , 1 , 2 ) be dual integral cohomology classes, and following [5], let and be anticommuting constant Grassman parameters so that ( 1 , 2 , 1 , 2 ) := ( 1 , 2 , 1 , 2 )

has a symplectic intersection form. Then, introduce k H k,k (X, Z), the mirror to ,
expressed as

a
=
(35)

b ,
Ea
Fb
and the mirror prepotential on X

1
F=
.
2
Ea
(36)
Fa
The latter expression is purely formal as we know of no a priori means of calculating the
periods of , without recourse to mirror symmetry. However, following [5] and [43], we
can write down an expression for F which is valid up to worldsheet instanton corrections,
by recalling that Yukawa couplings are given by third derivatives of the prepotential.
1
2 for w1 C, we
Indeed, writing the complexified Khler form on X as B + iJ = w
w2
have

1 2 3 (w1 )3 1 1 2
1 2
1
2
F =
(37)
+ a w + bw1 w2 + c w2 + O ew /w ,
2
3!
w
2
2
X
where
is an auxiallary homogeneous coordinate on the moduli space and t = w1 /w2 .
Now, if we could actually evaluate (37) directly (and exactly), then we could write
w2
= j fj ,
(38)
with t = (w1 F , w2 F , w1 , w2 ) and f = (1 , 2 , 1 , 2 ). Then, the basis-independent

equality = (an equality that follows since in the underlying conformal field theory
and correspond to one and the same operator) can be expressed using (34) and (38) as
i = Nji >j ,
(39)
gj = Nji fi ,
(40)
where N is a matrix relating the bases f and g.
200
In practice, we cannot carry out this program as written because we do not know how to
evaluate (37) exactly. But as noticed and used in [43], if we call F pert the approximation
to F with all instantons corrections dropped, equating and in the large complex
structure/large radius limit via
pert i = Nji >j
(41)
is enough to determine N . This statement relies on the conjecture of Aspinwall and

Lutken [48] which asserts that N should be an integral symplectic matrix. As such, it
can be determined exactly even if we only impose (41) in the large radius limit.
Once we have determined N , we can use it to relate the integral cohomologies of X and
Y via (40), or the integral homologies via its dual relation
t
t

1 2
F , F , E1 , E2 = N B 1 , B 2 , A 1 , A 2 .
(42)
In principle, this solves the problem of identifying the map between H3 (Y, Z)

and
k H2k (X, Z). Although we have recounted this discussion at the level of the
quintic/mirror quintic example, it is of course more general. Nevertheless, there is an
important outstanding issue: even if we work in the large radius limit and thereby drop
instanton corrections to F , from (37) we see that there are three numbers, a, b, c which are
as yet undetermined. As pointed out in [43], the real part of these numbers do not contribute
to the Yukawa couplings or the metric on the moduli space, and the only imaginary parts
they can contain (which do affect the metric) arise at four loops in the sigma model, thereby
only contributing to c. In the case of the quintic, a, b, c enter into N in the following
manner
1 + 2a b a 0
2b
c b 1
N =
(43)
2
0 1
0
0
1
0
0
25
25i

with a = a + 11
2 , b = b 12 , c = c + (3). In [43], the authors arbitrarily chose

a = b = c = 0, noting that other choices differ by an integral symplectic change of basis
that would not affect the calculations of their paper. In our case, though, such a change of
basis affects the mirror map on integral homology and hence does affect our conclusions.
For example, in the case of the quintic with the choice a = b = c = 0, we have the map
A2 F 2 . A2 is the cycle whose period vanishes at the conifold point of the mirror quintic
while F 2 is its mirror six-cycle on the quintic itself. Hence one is tempted to conclude that
the pure six-cycle collapses on the quintic when its Khler form is tuned to be mirror to the
conifold point [36]. But, for arbitrary a = b = c , A2 is mapped to a linear combination of
the six, four, two, and zero cycles. The fact that the coefficient of the six-cycle is nonzero
for any choice of a = b = c tells us that a six-brane is collapsing, but the lower order
contributions dictate a particular binding of lower brane charges to the six-brane.
Can we calculate a = b = c from first principles? At the level of the sigma model its
hard to see how, as their real parts seem to have no impact on sigma model observables. At
the level of D-brane field theory, perhaps there is a way to calculate them, but as yet this is
201
an unsolved problem. Hence, as discussed in the introduction, we are forced in this paper
to discuss the masses of wrapped 2j -branes, leaving their lower brane content unspecified.
Let us add a few remarks about the connection with recent work on the relation between
D-branes and K-theory. It is now well-known [11] that the charges of type B D-branes
on X should generally be considered to take values in certain K-groups K(X). While
this was initially discovered [11] through anomaly-cancellation arguments, it is in fact
not surprising given the relation between D-brane moduli spaces and moduli spaces
of holomorphic vector bundles proposed in [19]. From this point of view, the lattice
H even (X) appears as a coarser version of the space of charges, via the natural embedding
K(X) H (X). Above, we considered the problem of identifying the integral structure
from a perturbative (conformal field-theoretic) point of view, but a probably more powerful
approach to this question is to analyze the issue form the point of view of the Strominger
YauZaslow conjecture [15,16]. In that set-up, mirror symmetry is expected to map special
Lagrangian submanifolds C of Y (together with a flat connection on C) to holomorphic
vector bundles on X (or, rather, a certain type of gerbe generalization of this concept). The
trace of such a map in some cohomology theory should then yield the mirror map between
(a version of) H (Y ) and K(X). Finally, the map H 3 (Y ) H even (X) would be induced
from this through the embedding K(X) H (X). Some expectations of these type are
summarized by the homological mirror symmetry conjecture of Kontsevich (see [17]), but
it is difficult to be more precise without first gaining of a proper understanding of the many
physical and mathematical issues involved in such an approach.
We should perhaps also mention that much recent work on D-brane geometry identifies
D-brane charges by following [36] in conventionally fixing the ambiguity discussed in
detail above. We will not always point out this issue in the following, but the reader should
bear it in mind.
4. Examples I: the generic family of compact one-parameter models

We are now ready to apply the methods discussed above, and we begin with the
hypergeometric equation satisfied by the periods of a class of one-parameter examples:

4
z( + 1 )( + 2 )( + 3 )( + 4 ) u = 0.
(44)
1 ,2 ,3 ,4
This is associated with the hypergeometric symbol
, where we take j to
1,1,1
be rational numbers. 15 In particular, the mirror quintic of [43] is of this type. We first
illustrate our method for computing a fundamental system of periods and the associated
monodromies for this class of models.
To simplify subsequent discussion, we assume that:
(1) None of the differences k l (with k = l) is an integer.
(2) None of the number k is an integer.
(3) k > 0 for all k = 1, . . . , 4.
15 This seems to always be the case in practice, and thus is not a significant reduction in generality.
202
These conditions will allow us to avoid certain exceptional cases in the discussion of
analytic continuation below.
Before proceeding with a detailed analysis, let us make some brief comments on the
role and range of applicability of the results we derive in this section. The models we
consider are generic 16 from the point of view of their hypergeometric equation, inasmuch
as conditions (1)(3) are satisfied for generic systems of rational parameters 1 , . . . , 4 .
This class of examples exhibits most of the qualitative features familiar from the study of
the mirror quintic performed in [43]; in particular, such models admit a LanduaGinzburg
point, due to the fact that the expansion of the periods near the small radius limit does
not contain logarithmic factors. 17 Of the three conditions above, only (1) is essential for
assuring this behaviour; imposing the other conditions plays mostly the role of avoiding
formulae which would be too complicated to treat in a general fashion.
On the other hand, there exist many one-parameter examples which do not satisfy
condition (1) and hence do not entirely fit into this scheme. Even in such cases, however,
the expansions and monodromies we derive near the large complex structure point can
often be applied. Such models can be thought of as degenerations of the generic family
considered here. Hence this class can be viewed as a generating or universal family
which contains other models as various limits. This leads to a hierarchy of one-parameter
models (discussed in more detail in Section 5), having the generic family at its top.
4.1. The Meijer periods

2 ,3 ,4
(z) and applying the
Starting with the fundamental period U0 (z) = 4 F3 1 ,1,1,1
procedure discussed in the previous subsection gives the following fundamental system
of solutions:

(j +1)

1, 1, 1
0
1 , 2 , 3 , 4
(1)j +1 z ,
Uj (z) =
(45)
I
(3j
)
1 , 2 , 3 , 4
1
where j = 0, . . . , 3 and the notation x (k) indicates that the symbol x is repeated k times.
The expansions of (45) around the points z = 0 (large complex structure) and z =
(LandauGinzburg) can be obtained easily by a computation of residues. Indeed, let
us introduce the following notation for the integrand of the associated MellinBarnes
representations (see Fig. 3):

s
(s)(j +1) s + 1 , . . . , s + 4
1, 1, 1

(1)j +1 z
j (s) :=
(s + 1)(3j )
1 , 2 , 3 , 4

(s)j +1 i=1,...,4 (s + i )
s
1
(1)j +1 z .
=
(46)
3j
(s + 1)
i=1,...,4 (i )
16 This is not to say that such models are generic (meaning abundant) as a class of one-parameter examples. In
fact, quite the contrary seems to be the case, at least based on a limited investigation of the models constructed in
the literature.
17 This is enforced by condition (1), which, as can be checked immediately by using the results of Section 2.2,
and will be confirmed by explicit computation below, assures that the monodromy around the point z = is of
finite order.
203
Fig. 3. The defining contour for the MellinBarnes representation of the Meijer periods.
With this notation, we have

1
Uj (z) =
ds j (s).
2i
(47)
The integrand j (s) has poles at

(A) s = n,
(B)
s = i n,
for each non-negative integer n.

Consider first the expansions for |z| < 1. In this case, the asymptotic behaviour of the
integrand in the complex plane allows us to close the contour to the right at + (see
Fig. 4).
Only A-type poles contribute to the associated Cauchy expansion, and these poles are
multiple of order j + 1. The computation of the associated residues proceeds in the
following standard manner. In order to isolate the poles of the integrand, notice that the
identity
(s) =
(1)n+1 (s + n + 1)
s n (s n + 1)n
allows us to write
fj (s)
(1)(j +1)(n+1)
,
j (s) :=
j +1
i=1,...,4 (i ) (s n)
with
fj (s) =
(s + n + 1)j +1
(s + 1)
3j
(48)
[(s
s
i=1,...,4 (s + i )
(1)j +1 z .
j
+1
n + 1)n ]
(49)
204
Fig. 4. Contour for the expansion around the large complex structure point.
(j )
Computing the residues is thereby reduced to calculating the derivatives fj (n) for j =
0, . . . , 3. One can achieve this efficiently by first computing the derivatives of gj (s) =
(j )
log(fj (s)), with the result that fj (n) = fj (n)j (n), where
0 = 1,
1 (n, z) = g1 (n, z) = 1 (n) + log(z),

2

2
2 (n, z) = g2 (n, z) + g2 (n, z) = 2 (n) + 2 (n) + log(z) ,
3 (n, z) = g3 (n, z) + 3g3 (n, z)g3 (n, z) + g3 (n, z)3

3
= 3 (n) + 33 (n) 3 (n) + log z + 3 (n) + log z ,
(50)
with
(i)
j (n) =
4
(i) (n + k ) (3 j ) (i) (n + 1)
k=1

n

1
(1)i (j + 1) (i) (1) + i!
.
l i+1
(51)
l=1
This gives the expansion of the Meijer periods for |z| < 1:
Uj (z) =
(1)j (1 )n (2 )n (3 )n (4 )n
j (n, z)zn .
j!
(n!)4
(52)
n=0
If |z| > 1 then one can close the contour to the left (Fig. 5). In this case, we have
contributions only from the (B)-type poles of the integrand, which are all simple
Resnk j (s)
=
j +1
nk
(1)n (n + k )
l=1,...,4 (l k n)
(1)j +1 z)

3j
n!
(1 n k )
k=1,...,4 (k )
205
Fig. 5. Contour for the expansion around the LandauGinzburg point.

=
sin(k )
3j
(k )4

i=1,...,4 (i )
(l k )
(1)j +1 z
l=1,...,4
]4
[(k )n
zn ,
n! l=1,...,4 (1 + k l )n

where l=1,...,4 { } := l=1,...,4,l=k { } and the second equality follows upon using the
completion formula:
.
(x)(1 x) =
(53)
sin(x)

Putting everything together, we obtain the expansion of the Meijer periods for |z| > 1:
Uj (z) =

4

sin(k ) 3j
i=1,...,4 (i ) k=1

(l k ) (1)j +1 z k
l=1,...,4

4 F3
(k )4
k , k , k , k
. . . , 1 + k 4
1 + k 1 , . . . , 1,

(1/z),
(54)
where 1 indicates absence of 1 = 1 + k k in the hypergeometric symbol.

4.2. Monodromies of the Meijer periods
The canonical form of the monodromies around the points z = 0 and z = is easily
determined by making use of the general relations (28). In our case, one has Sk = 0 (k =

1, . . . , 3) and Rk := 1i1 <<i4k 4 i1 i4k . In particular, it follows that the matrix
Rcan [0] is in Jordan form
206
0 1 0 0
0 0 1 0
Rcan [0] =
0 0 0 1,
0 0 0 0
while the Jordan form of Rcan [] is given by
0
0
0
1
0
0
0
2
.
RJ [] =
0
0
0
3
(55)
(56)
0
0
0
4
Since i are all rational, it immediately follows that TJ [] (and thus Tcan []) has finite
order, so that z = can be interpreted as a LandauGinzburg point. Following the
program outlined above, we proceed to compute the monodromies of the Meijer basis
around the points z = 0, 1, .
4.2.1. Meijer monodromies around z = 0
t
The Meijer monodromy T [0] = e2iR[0] around z = 0 can be determined as follows.
Let U (z) be the column vector with entries U0 (z), . . . , U3 (z), (z) be the associated
fundamental matrix of the first order system (17), which has entries ij (z) := i Uj (z),
and S(z) the associated nilpotent orbit, which satisfies (19) and (20). We can separate the
logarithmic factors in U (z) by writing
U (z)t = Z(z)q(z),
(57)
i.e.,
Uj (z) =
3
qsj (z)(log z)s ,
(58)
s=0
where Z(z) is the row vector

Z(z) := 1 log z (log z)2
(log z)3
(59)
and q(z) plays the role of the nilpotent orbit of U (z) as discussed in Section 2.2.
Under a change of basis U (z) U (z) = M 1 U (z) (with M an invertible matrix),
we have U (z)t = Z(z)q (z) with q (z) = q(z)M t and (z) = (z)M t , S (z) =
S(z)M t , R = M t RM t . As discussed earlier, choosing M such that S (0) = I , we have
R (0) = A(0) = Rcan [0]. In this case, (z) = S (z)zRcan [0] , and since Rcan [0] is in Jordan
form, we immediately obtain:
1 log z 12 (log z)2 61 (log z)3

1
2
0
1
log z
2 (log z) ,
zRcan [0] =
(60)

0
(zRcan [0] )
(log z)j i /(j
i)!. This allows us to write U (z)t = Z(z)q (z) with

i.e.,
ij =

q (z)sj := S0,j s (z)/s!. In particular, we obtain:
q (0) = diag(1, 1, 1/2, 1/6),
(61)
207
which allows us to determine M = q(0)t q (0)1 . Once M t is known, the monodromy

of the Meijer basis follows from the relation T [0] = MTcan [0]M 1 , where Tcan [0]t =
e2iRcan [0] is given by
1 2i 12 (2i)2 16 (2i)3
1
2
0
1
2i
2 (2i) .
Tcan [0]t =
(62)

0
0
0
1
In our class of examples, q(0) can be easily determined from the expansions given above.
Indeed, we can write
j (n) =
j
vsj (n)(log z)s ,
(63)
s=0
with
v00 := 1,
v01 (n) := 1 (n),

2
v02 (n) := 2 (n) + 2 (n) + i ,
v11 := 1,

v12 (n) := 2 2 (n) + i ,
v03 (n) := 3 (n) + 33 (n)3 (n) + 3 (n)3 ,
v23 (n) := 33 (n),
j
This gives Uj (z) = s=0 qsj (z)(log z)s , with
qsj (z) :=
v22 := 1,

v13 (n) := 3 3 (n) + 3 (n)2 ,
v33 := 1.
(64)
(1)j (1 )n (2 )n (3 )n (4 )n
vsj (n)zn .
j!
(n!)4
(65)
n=0
(i)
In particular, we have qsj (0) = (1)

j ! vsj (0). Substituting the formulae for j given above,
we obtain a complicated expression for q(0) in terms of polygamma functions which
we will not reproduce here. This expression can be simplified by the following indirect

procedure. Let us introduce the variable w = z, where = exp( 4k=1 (k ) 4(1))
(in practice, this plays the role of the natural coordinate on the moduli space dictated by the
monomial-divisor mirror map of [29]). One can then consider isolating the factors (log w)s
instead of (log z)s in the expansion of Uj (z) around z = 0. Doing so amounts to writing
Uj (w) =
j
qsj (w)(log w)s ,
(66)
s=0
where the functions qsj (w) are now given by
n
(1)j (1 )n (2 )n (3 )n (4 )n
w
qsj (w) :=
vsj (n)
,
j!
(n!)4
(67)
n=0
as can be seen upon expanding

s (n, w) =
j

s=0
vsj (n)(log w)s ,
(68)
208
with
v00 := 1,
v01 (n) := 1 (n),

2
v02 (n) := 2 (n) + 2 (n) + i ,
v11 := 1,

v12 := 2 2 (n) + i ,
v03 (n) := 3 (n) + 33 (n)3 (n) + 3 (n)3 ,
v23 (n) := 33 (n),
v22 := 1,

v13 (n) := 3 3 (n) + 3 (n)2 ,
v33 := 1.
(69)
Here (n) are defined through

j (n) + log (1)j +1 z = j (n) + log (1)j +1 w .

Since log(w) = log(z) + log() = log(z) + 4k=1 (k ) 4(1), we have
j (n) :=
(70)
4
n

1
.
(n + k ) (k ) + (3 j ) (1) (n + 1) (j + 1)
l
k=1
l=1
(71)
The new sequences j (n) have the convenient property j (0) = 0, which allows us to
obtain a simple expression for the matrix q(0):
1 0
1
2
( (0) ) 1 (0)
0
q(0)
:=
0
0
In this expression:
1
0
0
6 3
i
1
2
12 3 (0)
.
0
16
(72)
2 (0) = (1 ) + (2 ) + (3 ) + (4 ) + 1/3 2 ,
3 (0) = (1 ) + (2 ) + (3 ) + (4 ) + 2/3 2 ,
3 (0) = (1 ) + (2 ) + (3 ) + (4 ) + 8(3).
(73)
Using log(w) = log() + log(z), it is easy to see that

q(0) = cq(0),
(74)
p
with c an upper triangular matrix whose nonzero entries are given by csp := s (log )ps
for all s p. Moreover, it is not hard to see that the matrix q (0)ct q (0)1 commutes with
t , so that we can use M
:= q(0)
Rcan
t q (0)1 instead of M. In other words, we have
1 ,
can [0]M
T [0] = MT
(75)
which allows for the computation of T [0] upon using (62):
1
0
0
0
2i
1
0
0
.
T [0] =
(76)
4 2 2i
1
0
0
0
2i 1
Remarkably, this expression is independent of k , which underscores the universal
behaviour of the Meijer periods in the large complex structure limit, as mentioned in
209
the introduction. In particular, these periods are adapted to the large complex structure
monodromy weight filtration, as can be seen form the form of T [0] or directly from the
expansions (52).
4.2.2. Meijer monodromies around z =
The monodromy of the Meijer basis around z = is straightforward to extract. For this,
first notice that

(1)j +1 z k = zk j,odd + j,even eik ,
(77)
where j,odd equals 1 if j is odd, and 0 if j is even (and a similar definition holds for
j,even ). This allows us to write
Uj (z) =
3
aj k uk (z),
k=0
where
uk1 (z) := zk 4 F3
k , k , k , k
. . . , 1 + k 4
1 + k 1 , . . . , 1,

(1/z)
and the transition matrix A := (aj k )j,k=0,...,3 is given by

A=
i=1,...,4 (k )
CD,
(78)
where

sin(k+1 ) 3j
Cj k :=
j,odd + j,even eik+1 ,
D := diag(d0 , d1 , d2 , d3 ),
(l k+1 ).
dk := (k + 1)4

l=1,...,4,l=k+1
The monodromy of the basis u has the simple form:

Tu [] = diag e2i1 e2i4 ,
(79)
which coincides with TJ = e2iRJ [] , where RJ [] is the Jordan form of the matrix
Rcan []. This allows us to compute the monodromy of the Meijer basis via the relation
T [] = ATu []A1 . Since both D and Tu [] are diagonal, we have DTu []D 1 = Tu ,
which allows us to reduce this relation to
t
T [] = CTu []C 1 .
(80)
The last equality is more convenient for computing T [] since it involves only the
matrix C.
210
4.3. The mirror quintic revisited

Let us now show how the results of [43] can be recovered in our framework. Beyond
acting as a check on our computations, this will also serve to illustrate the connection
between vanishing cycles and arithmetic identities in a well-understood example.
The mirror quintic can be obtained as a particular case of the discussion above by
choosing i = i/5 for all i = 1, . . . , 4. Let := e2i/5 . In this case, one has
1
0
0
0
1 0 5/3 2 40 (3)
0 1
0
0
i
7/3 2
2
,
, Tu [] = 0
q(0)
=
0 0
0
1/2
0
0
0
3
0
1/6
which immediately gives
1
0
0
0
2i
1
0
0
,
T [0] =
4 2 2i
1
0
0
0
2i 1
4
5/2i 1 5/4 2
2i
1
0
T [] =
4 2
2i
1
0
0
2i
5/8i 3
0
1
and T [1] = T [0]1 T []. These matrices satisfy (T [0] I )4 = 0, (T [1] I )2 = 0,

T []5 I = 0. To make contact with the results in [43], first note that our convention
for monodromies differs from the one used there by a change of orientation which
implies an inversion of the matrices T [0] and T []. 18 Then notice that the fundamental
period U0 is cyclic for the operator T []1 . Thus one can generate a cyclic set by acting
with T []1 on U0 , which produces 5 periods j related to the Meijer periods through
1
0
1
1

2 = 4

3 6
0
5/2i 1
0
0
5i 1
5/4 2
5i 1
5/2 2
4 5/2i 1
5/4 2
0

3
5/8i U0
U1
15
3
.
8 i
U2
15
3
8 i U
3
5/8i 3
(81)
This corresponds to the procedure used in [43] (as well as in [37,38,40,41]) in order to
generate a basis of periods from the fundamental period. Indeed, the cyclic basis used
in [43] corresponds to the periods 2 , 1 , 0 and 4 denoted by the same letters in that
paper. Up to a rescaling by (2i/5)3 , these form a period vector denoted there by . It
follows that the Meijer period vector is related to the period vector of [43] through:
18 This is due to our use of different coordinates on the moduli space, as we explain below.
2

3
2i
1 = LU
=
0
5
4
with L =
32
125
i 3
8
3
125 i
8
3
125 i
32
125
i 3
8 2
25
4 2
25
211
2
25 i
4 2
25
2
25 i
3
25
1/25
.
0
1/25
In order to make contact with the results of [43], one must also take into account the fact
that the monodromy matrices given there correspond to working on a 5-fold cover of the
moduli space. This is parameterized in [43] by a variable , related to our coordinate
by z = 5 . It follows that our monodromy matrices LT[0]L1 , LT[1]L1 , LT[]L1
1/5 1/5
should be compared, respectively, with the matrices t , t a 1 and a 1 of [43], and
it is easy to see that they agree. As a further check, one can easily show that the || < 1
expansions of j given in Appendix B of [43] agree with the expansions of the cyclic basis
following from (54).
A full sublattice of the integral lattice = H3 (Y, Z) can be identified as explained in
Section 2. This gives a lattice 0 characterized by the fact that the period vector associated
with one of its bases (which we call E) is (up to a global factor):
1
0
0
0
4 5/2i 1
UE = EU =
6
5i 1
5i 1
5/4 2
5/2 2
5/4 2
5/8i 3
U.
15 i 3
(82)
8
15
3
8 i
The integral basis P of used in [43] is characterized by a period vector , which is

related to the Meijer periods U by = U , with:
2
12 2
0
0
125
25 i
8 i 3
0
0
0
= 125
(83)
.
4
4
2
i
0
25
25
0
4 2
25
1/25
Hence the transition matrix from the basis P of H3 (Y, Z) to the basis E of 0 is:
0 5 20 15
3
3

1 4

t
2i
6
4
= 2i
K0 .
K = E 1 =
0 3 11 8
5
5
0 1
3
3
(84)
Since det(K0 ) = 5, we see that 0 = (2i/5)30 is an index 5 sublattice of , i.e., the

finite group /0 has order 5.
The period 19 (2i/5)3(1 0 ) which vanishes at the conifold point can be
1
expressed in terms of the Meijer periods as 25
(4 2 U1 + U3 ). Hence collapse of the
associated 3-cycle at z = 1 is reflected by the relation
U3 (1) = 4 2 U1 (1),
19 This equals z2 in the notations of [43].
(85)
212
which after a few simplifications can be seen to be equivalent to:

1/5, 1/5, 1/5, 1/5

192 5 2 5 + 5 (5 2 5 )(3/5)5 4 F3
(1)
4/5, 2/5, 3/5

2/5, 2/5, 2/5, 2/5

+ 960 5 2 5 + 5 (2 5 5)(4/5)5 4 F3
(1)
4/5, 3/5, 6/5

4/5, 4/5, 4/5, 4/5

(1)
+ 3125(11 5 5 )(3/5)5 (4/5)10 4 F3
8/5, 7/5, 6/5

3/5, 3/5, 3/5, 3/5
+ 3750(4/5)5(3/5)10 4 F3
(1) = 0.
4/5, 7/5, 6/5
(86)
One can of course write down a pair identity by using the expansions (52) around the large
complex structure point, which can be thought of as following from the one above through
analytic continuation. It is a challenging problem to find direct proofs of these identities.
Figs. 6 and 7 compare the classical and quantum volumes of the even-dimensional
cycles on the quintic. In Fig. 6, we plot classical volumes as a function of the classical
Khler parameter s, defined by J = se with e the generator of H 2 (X, Z). In Fig. 7, for
ease of comparison, we plot the quantum volumes as a function of essentially the same
parameter s the so-called algebraic measure on the quantum moduli space (see [34]
1
for a discussion of this concept), which in this case is related to z via s = 2
log(55 /z)
(note that = 55 in this example). In the language of [34], this coordinate is s = Im(talg ),
1
log(z/55 ).
where talg = 2i
For comparison with the classical situation, we consider only the region s 0 in Fig. 7,
even though quantum mechanically we can continue s to negative values as well. The
5
point P , where s = 2
log 5 1.28 and z = 1, corresponds to the conifold. Fig. 7 also
0 )+2 4
displays the values of the special coordinate t = 2(1 5
, which measures the
0
volume of a D2-brane wrapped over the generator of H2 (X, Z). We also plot the absolute
2i
value of the integral period 25
U2 , which can be interpreted as the mass of a wrapped
D4-brane. Note that the scale used for the y-axis is different for the two graphs and
different from the scale used on the x-axis (this is needed in order to fit the interesting
portion of the graphs in the limited space available). Taking this rescaling into account,
one can easily check that the curve m = |t (s)| asymptotes to the line m = s (the diagonal
of the first quadrant) in the the large radius limit s . In this limit, |Uv | asymptotes to
1.923 4.134s + 1.653s 3 .
These figures capture the essential point, one that we will find repeated in various forms
in subsequent examples. Namely, the classical relations between 2, 4, and 6 cycle volumes
are significantly modified in the quantum setting. In particular, we see that a 6-cycle
collapses to zero quantum volume at the conifold point, even though 2 and 4 cycle volumes
stay positive.
213
Fig. 6. Graph of the classical volumes s, 52 s 2 and 56 s 3 of a set of two, four and six cycles vs. the
classical Khler parameter s.
Fig. 7. Graph of |Uv | vs. the classical Khler parameter s. For comparison, we also plot the absolute
value of the special coordinate t , which measures the mass of a wrapped D2-brane, and the absolute
2i U .
value of the (integral) period 25
2
5. Examples II: some degenerate models

Let us now consider the case when the parameters i satisfy conditions (2) and (3),
but they violate condition (1). Namely, consider 0 < i < 1 chosen so that 1 , 2 , 3 are
distinct but 3 = 4 . This situation is encountered for some of the examples studied in
[40] (see Table 3.1 on page 26 of that paper). In particular, it is satisfied for a complete
intersection of a quadric and a quartic in P5 (the third entry of that table), a model which
we will encounter as an interesting sub-locus of a two-parameter example in Section 4.
Another case which satisfies this condition is the complete intersection P6 [2, 2, 3].
214
5.1. The Meijer periods

Clearly nothing qualitatively new happens in such a model when computing the
expansions for |z| < 1, since closing the contour to the right only gives contributions
form the poles in (s) (the A-type poles) in (46), whose nature is independent of the
relative values of i . Hence the expansions around the large complex structure point have
the same form as above. On the other hand, closing the contour to the left gives two types
of contributions from the (B)-type poles:
(B1 ) Contributions from s = n3 (n a non-negative integer), which are double poles.
(B2 ) Contributions from s = n 1 and s = n 2 (n a non-negative integer), which
are simple poles.
Correspondingly, we can write the expansions of Uj (z) for |z| > 1 as
(2)
Uj (z) = u(1)
j (z) + uj (z),
(87)
separating the contributions from these two sub-types. The (B2 )-type contributions are of
exactly the same form as above:
(2)
uj (z) =

2

1
sin(k ) 3j
(k )4
(1 )(2 )(3 )2
k=1

(l k ) (1)j +1 z k
l=1,...,4
4 F3
k , k , k , k
. . . , 1 + k 4
1 + k 1 , , . . . , 1,

(1/z),
(88)
but the B1 -type poles induce logarithmic terms

sin(3 ) 3j
1
u(1)
(1)j +1 z 3
j (z) =
(1 )(2 )(3 )2

(n + 3 )4 (n + 1 3 )(n + 2 3 ) n
z
n!2
n=0

(n + 1 3 ) + (n + 2 3 )
(j + 1)(n + 3 ) (3 j )(1 n 3 ) + 2(1)

n

1
+ log (1)j +1 z .
+2
l
(89)
l=1
5.2. Monodromies of the Meijer basis

The monodromies about z = 0 follow by substituting 4 = 3 in the results of Section 3.
In order to obtain the monodromies about z = , one has to be a bit more careful, as we
now explain.
215
By using the general results of Section 2, it is easy to compute the matrix Rcan []:
0
1
0
0
Rcan [] =
0
0
1 2 32
0
0
21 2 3 1 32 2 32
and its Jordan form
1
0
RJ [] =
0
0
0
2
0
0
0
0
3
0
1
0
1 2 + 21 3 + 22 3 + 32
1
1 2 23
0
0
.
1
3
(90)
These are related by a transition matrix P GL(4, C) satisfying:

Rcan [] = P RJ []P 1 .
(91)
While such a transition matrix is not unique, for what follows it suffices to pick any P with
the property (91), for example, the matrix
2

2
(2 + +3 2 2 )
2 3
( )2 ( )
1
3
1
2
1 2 3 2
(1 3 )2 (1 2 )
P =
1 2 3 2 2
(
2
1
3 ) (1 2 )
1 3 2 3 2
(1 3 )2 (1 2 )
(2 3 )2 (1 2 )
1 3
(2 3 )(1 3 )
1 2 3
1 3
1 2 3 2
(2 3 )2 (1 2 )
1 2 3 2
(2 3 )(1 3 )
1 2 2 3 2
(2 3 )2 (1 2 )
1 2 3 3
(2 3 )(1 3 )
1 2 3 3 2
(2 3 )2 (1 2 )
1 2 3 4
(2 3 )(1 3 )
1 2
2 3
(2 3 )2 (1 3 )2
1 2 3 2 (1 23 +2 )
(2 3 )2 (1 3 )2
1 2
1 2 3 2 (1 2 3 2 )
(2 3 )2 (1 3 )2
1 2 3 3 (1 3 +21 2 2 3 )
(2 3 )2 (1 3 )2
Assuming that we made such a choice, we can define a Jordan basis of periods (which
depends on the choice of P ) through UJ (z) = P t Ucan (z). In terms of fundamental
matrices for the associated PicardFucks system, this relation reads
can (z) = (z)P 1 ,
(92)
which implies that the monodromy of the basis UJ about z = is given by the matrix
RJ []. The associated nilpotent orbits are related through Scan (z) = SJ (z)P 1 , which,
together with the defining property Scan [] = I , allows us to compute
SJ [] = P .
(93)
The Jordan basis is a useful device for extracting the monodromy of the Meijer periods.
To see this, note that Eq. (90) assures us that we can extract the nontrivial behaviour of
periods about z = by writing
U (z) = Z(z)q(z),
UJ (z) = Z(z)qJ (z),
(94)
where q(z), qJ (z) are one-valued and regular at z = while the row vector Z(z) is defined
by 20

Z(z) = z1 z2 z3 z3 log z .
(95)
20 Note that this differs from the vector Z(z) used in Section 3.
216
Since the Jordan monodromy

TJ [] = e2iRJ []
2i1
e
0
=
0
0
0
e2i2
0
0
0
0
e2i3
2ie2i3
0
0
0
(96)
e2i3
is trivially known, it follows that, in order to compute the monodromy of U (z), it suffices
to identify the transition matrix M GL(4, C) such that U (z) = MUJ (z). The latter can
be determined form knowledge of q() and qJ () through the relation:
M = q()t qJ ()t ,
(97)
provided that we can compute the two matrices appearing in this equation. The matrix
qJ () is straightforward to extract by noting that UJ (z)t coincides with the first row of
J (z) = S(z)zRJ and by rewriting Uj (z) = Si (z)(zRJ )ij in the form (94):
0
0
0
S1,1 (z)
0
0
0
S1,2 (z)
,
qJ (z) =
(98)
0
0
S1,3 (z) S1,4 (z)
0
0
0
S1,3 (z)
which together with (93) gives
0
0
P1,1
0
0
P
1,2
qJ () =
0
0
P1,3
0
0
0
0
0
.
P1,4
P1,3
(99)
On the other hand, the matrix q(z) can be determined via its defining property (94) by
using the expansions (88), (89). This amounts to writing
Uj (z) =
4
qij (z)Zi (z),
(100)
i=0
which immediately gives

q() =
1
DC t ,
(1 )(2 )(3 )2
with C and D given by

sin k 3j
Cj k =
j,odd + j,even eik , for k = 0, 1, 3,

sin 3 3j
Cj 2 =
j,odd + j,even ei3

(1 3 ) + (2 3 ) (j + 1)(3 )

(3 j )(1 3 ) + 2(1) + j,even i
(101)
(102)
217
with j = 0, . . . , 3 and
D = diag(d1 , d2 , d3 , d4 ),

(l k ),
dk = (k )4
for k = 1, 2,
l=1,...,4
d3 = d4 = (3 )4 (1 3 )(2 3 ).
These relations allow us to determine
1
M=
CDqJ ()t
(1 )(2 )(3 )2
and hence the Meijer monodromy T [] = MTJ []M 1 . Moreover, it is possible to
show that the matrices D and qJ ()t TJ []qJ ()t commute, so that we can use N :=
CqJ ()t instead of M. We conclude that the Meijer monodromy about the small radius
point is given by
T [] = NTJ []N 1 .
(103)
The reader will easily recognize that the procedure employed above is a natural
generalization of the approach used in Section 3 for extracting the Meijer monodromy
around z = 0. In that case, the canonical matrix Rcan [0] was already in Jordan form, so
that the matrix P was simply the identity (while qJ (0) was denoted there by q (0)). The
procedure presented above can be easily generalized for an arbitrary Jordan form, thus
allowing for a uniform treatment of all possible types of boundary points of the moduli
space.
5.3. The model P5 [2, 4]
The mirror of this model can be described by an orbifold of the complete intersection
Y = {p1 (u) = p2 (u) = 0}, where u = [u0 , . . . , u5 ] P5 and
p1 (u) = u20 + u21 + u22 + u23 2u4 u5 ,
(104)
p2 (u) = u44
(105)
+ u45
4u0 u1 u2 u3 .
The fundamental period of this example was determined in [40] (see also [51] for other
results on this model). By using our techniques, we can go beyond these results and
determine a full basis of periods. We will also find a (weakly) integral period vanishing
at x = 1, which can once again be interpreted as a massless 6-brane state.
The complex variable gives a coordinate on a 6-fold cover of the moduli space,
which is a copy of the Riemann sphere P1 . As before, we prefer to work with the
hypergeometric parameter x = 4/ 6 , which gives a coordinate on the moduli space
itself. 21 The normalization of x is fixed by the requirement that x = 0 corresponds to
the large complex structure point and x = 1 corresponds to the conifold point. Then x =
corresponds to the small radius limit of the model. As mentioned above, logarithmic
21 In this subsection we denote the hypergeometric coordinate by x instead of z, for agreement with Section 7.
218
behaviour of the periods for |x| > 1 prevents us from interpreting x = as Landau
Ginzburg point; rather, it is a more general type of non-geometric point (in the terminology
of last paper in Ref. [33], it is a hybrid point).
The hypergeometric symbol of this model is easily computed by the techniques of [58

2 ,3 ,4
60] with the result that it is given by 1 ,1,1,1
with 1 = 1/4, 2 = 3/4, 3 = 4 =
1/2. Hence the model fits into the scheme of the previous subsection and we can directly
apply the results derived above by simply substituting these particular values of i in
relations (52) and (87), (88), (89). 22
Computation of the Meijer monodromies proceeds as explained above. The monodromy
around x = 0 has the same form as in Section 3, while for the point x = we have
0 1
0
0
0
0
1 0
,
Rcan [] =
0
0
0 1
23
3
7
2
64 16
16
1/4
0
0
0
0
3/4
0
0
.
RJ [] =
(106)
0
0
1/2
1
0
0
0
1/2
A choice for the matrix P
6
2
3/2 3/2
P =
3/8 9
8
3
27
32
32
We can now compute
1/241/24i
3
1/12 2
N =
1/121/12i
1/6
is
3/2 3
3/4
0
.
3/8 3/4
3/16 3/4
1/8+1/8i
3
1/4 2
1/4+1/4i
1/2
(107)
2/3 2i+2i log(2)

3
2/3 i3
2/3 2
,
2i+2i log(2)
i
2/3
2/3
4/3 4/3 log(2)

2/3
thus obtaining the Meijer monodromies
1
0
0
2i
1
0
T [0] =
4 2 2i
1
0
0
2i
4/3 1+log(2)
2
(108)
0
0
,
0
1

22 The reader can easily verify that the fundamental period U = F 1/2,1/2,1/4,3/4 (x) coincides with
0
4 3
1,1,1

(2n)!(4n)!
given in Table 3.1 of [40]. This follows from the identity (2n)!(4n)!
=
the period 0 = n=0
6
8
6
n
(n!) (2 )
(n!)6
(1/2)2n (1/4)n (3/4)n
10n
.
2
(n!)4
219
1 log(x) for
Fig. 8. Graph of |Uv | versus the imaginary part s of the algebraic coordinate talg = 2i
s [0, 2]. The point x = 1 corresponds to s = 5 log 2/ 1.103. We also display the weakly integral
2
periods 4i
U1 and 2 U2 .
5
2i
T [] =
4 2
0
3i 1
1
2i
0
2 2
0
1
2i
i 3
0
,
0
1
and T [1] = T [0]1T []. The monodromy matrices satisfy

(T [0] I )4 = 0,
(T [1] I )2 = 0,
(T []4 I )2 = 0.
Hence the monodromy about x = is neither unipotent, nor of finite order. This confirms
our expectation that the small radius limit of the model has a hybrid character (i.e., is not
a LandauGinzburg orbifold).
A set of periods associated with a basis of a full sublattice of H3 (Y, Z) (up to a common
factor) can be obtained by the procedure discussed above:
1
0
0
0
i
5 3 i
2 12
3
UE (x) = EU (x), with E =

(109)
i
6
i .
8
2 4 3
11
15 13 i
Moreover, it is not hard to check that the period Uv =
it is also easy to see that Uv is weakly integral
Uv (x) = i[3, 2, 2, 1]UE (x).
8
2
1
[U3
3
7 i3
3 2 U1 ] vanishes at x = 1;
(110)
We will see evidence of a different kind for the integrality of this period in Section 6.
Collapse of the associated cycle once again rests on two arithmetic identities, of which we
mention only the one induced by using the expansion of the periods for |x| > 1:
220
(n + 12 )4 (2n 12 )
4n+1 2
n!2
n=0

n 14 + n + 14 n + 12 3(1/2 n) + 2(n + 1)

1
1/4, 1/4, 1/4, 1/4
(1)
+ (1/4)6 4 F3
1/2, 3/4, 3/4
2

3/4, 3/4, 3/4, 3/4
16(3/4)6 4 F3
(1) = 0.
5/4, 5/4, 3/2
1
Fig. 8 plots the values of the vanishing period versus the coordinate s = 2
log(x)
which defines the
algebraic
measure
on
the
(uncomplexified)
Khler
moduli
space.
In

this case = e k=1,...,4 (k )4(1) = 210 . For comparison, we also display the values
of the weakly integral periods 22 U2 and 4i
U1 . Up to a common factor, these periods
correspond to the the integral of over cycles mirror to a 6, 4 and 2-cycle, respectively.
5.4. The model P6 [2, 2, 3]

This mirror of this model is realized as an orbifold of the complete intersection Y =
{p1 = p2 = p3 = 0} of two quartics and a cubic in P6 :
p1 = x12 + x22 + x32 2x6 x7 ,
p2 = x42 + x52 2x1 x2 ,
p3 = x63 + x73 3x3 x4 x5 .
(111)
The hypergeometric coordinate on the moduli space is given by z = 3/(2 7 ), while the

. The large and small radius expansions are
hypergeometric symbol is 1/2,1/2,1/3,2/3
1,1,1
obtained from (52) and (87), (88), (89) by substituting 1 = 1/3, 2 = 2/3 and 3 = 4 =
1/2.
Computation of the Meijer monodromies proceeds as above, so we only list the results
0
1
0
0
0
0
1 0
,
Rcan [] =
0
0
0 1
1/18 17
36
53
36
2/3
0
0
0
0
1/3
0
0
,
RJ [] =
0
0
1/2
1
0
0
0
1/2
9
18
4 8
6
6
2
0
,
P =
4
2
1
2
8/3 2/3 1/2 2
3+3i
3
1/12 2
1/48
N =
1/36 3+3i
1/9
1
2i
T [0] =
4 2
0
6
2i
T [] =
4 2
0
and
T [1] = T [0]1T [].

(T [0] I )4 = 0,
1
96
log(3)
1/4 4i4i log(2)+3i
3
3+3i
3
1/24 2
1
72
0
1
2i
0
log(3)
1/4 44 log(2)+3
2
log(3)
1/4 4i4i log(2)+3i
3+3i
1 log(2) + 3/4 log(3)
1/18
0
0
1
2i
221
1/4 i3
1/4 2
,
1/4 i
1/4
0
0
,
0
1
7/2 i
3 2
3/2 i3
1
2i
0
0
1
2i
0
0
1
The monodromy matrices satisfy

(T [1] I )2 = 0,
(T []6 I )2 = 0.
A set of periods associated with a basis of a full sublattice of H3 (Y, Z) is (up to a common
factor):
1
0
0
0
UE (z) = EU (z),
with E =
17
7/2 i
31
24 i
One can check that the period Uv =

can be expressed as
Uv = 2i[4, 4, 3, 1]UE ,
23/2 i
1
(3U3 7 2 U1 ),
3
3
2
122
24
2
3/2 i3
.
15/2 i3
(112)
18 i3
vanishes at z = 1 (see Fig. 9). This

(113)
which shows that Uv is weakly integral. The behaviour of this model is qualitatively very
similar to that of the previous
example, as we illustrate by drawing the graph of Uv versus
1
1
s = 2
log(z) with = e k=1,...,4 (k )4(1) = 432
= 24 33 .
5.5. A hypergeometric hierarchy
Before leaving the subject of compact one-parameter models, let us make a few
methodological remarks. As illustrated above, starting with the generic family of
Section 3 and taking the limit where some of of the parameters coincide produces new
families of qualitatively different models. In this section, we studied only the first layer of
such degenerate models, namely the case where two of the parameters coincide. The
procedure clearly generalizes, leading to a hierarchy of models on five levels, which are
characterized (up to permutations of the parameters) by the conditions:
(0) all i are distinct,
222
Fig. 9. Graph of |Uv | versus s for s [0, 2]. The point z = 1 corresponds to
1 (4 log 2 + 3 log 3) 0.9658.
s = 2
(1) three of the parameters i are distinct,

(2) 1 = 2 and 3 = 4 but 1 = 3 ,
(3) 1 = 3 = 3 = 4 ,
(4) 1 = 2 = 3 = 4 .
For example, it is easy to see that all of the complete intersection models in projective
spaces investigated in [40] fit into one of these classes. 23 In this section, we considered the
two models of [40] which are of type (1), but it should be clear that a very similar approach
can be followed for the other classes. In fact, one of the major advantages of using Meijer
functions is that it allows for a very systematic treatment of entire families of models, as
we have illustrated in some detail above. In the next section, we study a non-compact oneparameter model, which displays some new features. As we will show, our techniques are
well adapted for this type of model as well.
6. Examples III: an orbifold example

In this section we consider a non-compact one-parameter example, namely the C3 /Z3
orbifold. While non-compact and hence slightly unphysical, global orbifolds can be
realized as local singularities of CalabiYau spaces, or as decompactification limits of
singular CalabiYau varieties. In particular, the orbifold we consider can be realized
as a limit of the 5-parameter model studied in [34]. For an overview of toric and
mirror-symmetry techniques in the study of the moduli space of conformal field theories
23 Class (4) is not realized through compact one-parameter complete intersections in toric varieties, though it
could be realized through different constructions.
223
associated with partial resolutions of orbifolds we refer the reader to [32]. The C3 /Z3
orbifold was also studied in [20] with similar methods.
6.1. Classical geometry of the orbifold
We begin by reviewing the classical geometry of the model. Our orbifold X0 can be
realized as the quotient of C3 by the action
(z1 , z2 , z3 ) ( z1 , z2 , z3 )
where
= e2i/3 .
(114)
This has a straightforward toric description as the quotient
X0 = C4 /C
of C4 by the torus action

(x0 , x1 , x2 , x3 ) 3 x0 , x1 , x2 , x3 ,
(115)
(116)
where C , which corresponds to the fan in R3 consisting of a single simplicial cone

with generators
1 = (3, 1, 1),
(117)
2 = (0, 1, 0),
(118)
3 = (0, 0, 1).
(119)
Note that the interior of this cone contains the integral vector
0 = (1, 0, 0).
(120)
The resolution X of X0 can be achieved by a toric blow-up, which amounts to replacing

X0 by
X = (C4 Z)/C ,
(121)
with the exceptional set Z = {x C4 |x1 , x2 , x3 = 0}. This corresponds to adding 0 to our
list of toric generators and taking the fan to consist of the 3 cones spanned by (0 , i , j )
(1 i j 3), thereby performing a star subdivision of the original cone (see Fig. 10).
Geometrically, this amounts to blowing-up the origin in the space C3 /Z3 , thus producing
an exceptional divisor D0 associated with the generator 0 . The resolved space contains 4
toric divisors D , associated with the four toric generators ( = 0, . . . , 3).
The group Div(X) Pic(X) of divisor classes on X modulo linear equivalence is easy to
compute by the general technology of [31] or by simple manipulations with the associated
linear systems O(D ), with the result that it is a free group generated by either of D1 , D2
or D3 . In particular, we have the relation
D0 = 3D1 = 3D2 = 3D3 ,
(122)
where we denote the linear equivalence class of a divisor D by the same letter. This
corresponds to the fact that O(D0 ) O(D1 )3 O(D2 )3 O(D3 )3 , as can be seen by
considering the charges of x .
224
Fig. 10. Fan for the C3 /Z3 orbifold.
It follows that our model has a 1-dimensional Khler moduli space, which can be
parameterized by writing
k = (B0 + iJ0 )e0 ,
(123)
where k is the complexified Khler class, and e0 is the Poincare dual of D0 . In particular,
the classical volume of D0 is proportional to D0 J 2 = 27J02 , where we used the fact that
D13 = 1. Hence the orbifold point corresponds classically to J0 0, in which limit the
volume of D0 becomes zero.
6.2. The quantum moduli space and massless D-branes
As discussed for example in [32], the moduli space can be obtained without the need of
an explicit construction of a mirror family. For this, it suffices to view the star-subdivision
considered above as a perestroika associated with changing the phase in a mirror secondary
fan, which in this case consist of two opposing semi-axes on the real line. It follows that
the model has two phases, associated with two limits which correspond to blowing up the
singularity until the exceptional divisor acquires very large area (the large radius point),
respectively, to blowing down the divisor to zero area (the deep interior of the orbifold
phase). Hence the compactified moduli space can be identified with a copy of the Riemann
sphere. Analyzing the associated PicardFucks equation (see next subsection), which we
write in terms of a convenient coordinate z gives two regular singular points at z = 0, ,
which with our parameterization correspond to the large radius and deep orbifold points,
respectively, as well as a regular singular point at z = 1, which can be thought of as defining
the boundary between the two regions (in sense which is explained precisely in [33,34]).
The special coordinate t = B0 + iJ0 on the quantum-corrected Khler moduli space is then
225
given by the mirror map as a function t = t (z) which can be determined from the solutions
of the PicardFucks equation.
In the limit z , the conformal field theory is an orbifold theory and hence exactly
solvable. Based on the classical picture, one may expect that the D2-brane wrapping a cycle
in the class e1 , as well as the D4-brane wrapping the exceptional divisor would become
massless at the orbifold point. As we discuss below, this picture is drastically modified by
quantum effects. First, based on the computations of log0 - and log1 -monodromy periods
already performed in [34] (and which will be subsumed by our more detailed results
below), one can immediately see that the value of J0 at the orbifold point is zero, but
the value of B0 equals 1/2 (in units where 2 = 1), so that a D2-brane wrapping a
rational curve in X has nonzero mass at z = . This matches the fact that the conformal
field theory is perfectly well-behaved there; an analogous observation was subsequently
made in [42] in the context of K3 compactifications to explain the physical smoothness
of orbifold points. On the other hand, we will argue that there exists a D4-brane which
becomes massless at the point z = 1, though no D2-brane acquires vanishing mass there.
Following our earlier discussion, we are unable to uniquely fix the lower brane charge on
this D4-brane, so it might well be interpretable as a (D0, D4)-bound state.
6.3. Quantum volumes
6.3.1. The hypergeometric equation and the Meijer periods
The periods of this model satisfy the hypergeometric equation
3

z( + 1/3)( + 2/3) u = 0,
(124)
0,1/3,2/3
which corresponds to the hypergeometric symbol
. The Meijer periods can be
1,1
24
obtained by the procedure discussed in Section 2:
24 The reader may worry that our expression for U (z) is identically zero due to presence of the singular quantity
0
(0) in the prefactor. However,
singularity. One can take U0 (z) to be defined as a limit in which
this is a removable

1,1
(0) in the denominator of 0,1/3,2/3
and (s), (s) in the numerator of the integrand are replaced by (R)
R
R
and (s + 2 ), (s + 2 ) respectively, with a small positive R which is taken to zero after performing the integral.
When closing the contour to right or left, the factor (R) in the denominator will force us to keep only the residue
corresponding to the pole at s = 0. Indeed, we have (R) 1/R and the quantity R Ress0 [(s + R2 )(s + R2 )]
(with s0 a pole of type (A) or (B) of the regularized integrand) has a nonzero limit as R tends to zero only if s0 =
R/2 or s0 = R/2; this limit is 1, which forces the limiting value of the regularized integral to be identically 1
(remembering that the contours about have opposite orientations). One can of course see directly that U0 = 1
is a solution of the hypergeometric equation, but the argument just presented shows how this example fits into
the general theory of Section 2. The solution U1 is obtained from U0 by performing two diagonal operations on
the Meijer symbol. Lifting the parameter 1 along the second diagonal is the usual procedure for trying to make
the (A)-type poles of the integrand become double, which assures that the solution thus produced has log2 (z)
monodromy about z = 0. A further operation is performed along the first diagonal, by lowering the parameter 0.
This replaces the factor (s) in the numerator by a factor (1 s) = s(s) in the denominator, which cancels
a (s) introduced in the numerator by the first operation and leaves a factor of (s)/s;
henceall (A)-type

1,1
1,1
by 1/3,2/3
in
poles remain simple except for the pole at s = 0. We also replace the prefactor 0,1/3,2/3
order to avoid problems from (0). Such a rescaling is always allowed since the choice of normalization for our
solutions is arbitrary. This slightly nonstandard behaviour is due to non-compactness of the model.
226

1, 1
0 0, 1/3, 2/3
I
(z) = 1,
0, 1/3, 2/3
1, 1

1, 1
0, 0 1/3, 2/3
U1 (z) =
I
(z),
1/3, 2/3
1
1

1, 1
0, 0, 0 1/3, 2/3
I
(+z).
U2 (z) =
1/3, 2/3
1
U0 (z) =
(125)
(126)
(127)
For clarity, let us write down the integral representations for the nontrivial Meijer periods:

1
(s + 1/3)(s + 2/3)(s)
U1 (z) =
(128)
ds
(z)s ,
2i(1/3)(2/3)
s(s + 1)
1
U2 (z) =
2i(1/3)(2/3)

ds
(s + 1/3)(s + 2/3)(s)2 s

z .
s
(129)
The reader can verify directly that Uj satisfy Eq. (124).

The expansions of the periods can be obtained as before. Instead of presenting details of
the calculation, let us point out that closing the contour to the right (which is possible for
|z| < 1) leads to j th order poles at s = n (n a strictly positive integer) from the factors of
(s)j in the integrand and to a (j + 1)th order pole at s = 0 from (s)j /s. Computing
the associated residues leads to the expansions for |z| < 1:
U1 (z) = log(z) + [(1/3) + (2/3) 2(1)] +

(1/3)n (2/3)n
n=1
n!2 n
zn ,

1
U2 (z) =
(1/3) + (2/3) + 2 (1)
2

2
+ (1/3) + (2/3) 2(1) + log z

n1

3
1
log(z) + (n + 1/3) + (n + 2/3) 2(1) 2
n
k
n=1
(1/3)n (2/3)n n
z .
n!2 n
The expression for U1 (z) can be simplified by noticing that

(1/3)n (2/3)n n 2
4/3, 5/3, 1, 1
z
z
=
F
(z)
4 3
2, 2, 2
n!2 n
9
k=1
(130)
n=1
(to see this, one must first shift the summation variable according to n n 1 and then
perform some basic manipulations on the Pochhammer symbols). Moreover, applying the
identities (B.3) of Appendix B for n = 3 gives the relations
(1/3) + (2/3) 2(1) = 3 log(3),
5 2
(1/3) + (2/3) + 2 (1) = 10 (1) =
,
3
which allow one to further simplify U1 (z) and U2 (z).
(131)
(132)
227
The expansions of the Meijer periods for |z| > 1 can be obtained similarly. In this case,
one can close the contour to the left, obtaining a Cauchy expansion over the simple poles
s = n 1/3 and s = n 2/3, with n a nonnegative integer. The resulting residues can
be simplified by making use of the completion formula and the identity
(x + 1)n = (x)n
x +n
x
(133)
(x)n
with x1 (x+1)
. In this way, one obtains
n

(1/3)
1/3, 1/3, 1/3
1/3
(z)
F
U1 (z) = 3
(1/z)
3 2
2/3, 4/3
(2/3)2

(2/3)
3
2/3, 2/3, 2/3
(z)2/3 3 F2
(1/z),
+
4/3, 5/3
2 (1/3)(4/3)

(1/3)2 1/3
1/3, 1/3, 1/3
U2 (z) = 3
z
(1/z)
3 F2
2/3, 4/3
(2/3)

3 (2/3)2 2/3
2/3, 2/3, 2/3
z
+
(1/z).
3 F2
4/3, 5/3
2 (4/3)
which allows for the replacement of
1
n+x
(134)
(135)
6.3.2. Monodromies of the Meijer basis

The monodromies of the Meijer basis follow from arguments similar to those presented
in Section 3. In our case, the matrix q(0)
is:
2
1 i 56
q(0)
:= 0 1
0 ,
0
12
where we used (1/3) + (2/3) + 2 (1) = 5 2 /3 and the expansions derived above
for |z| < 1. The monomial-divisor mirror map gives the coordinate w = z/27 (in this
example = (1/3) + (2/3) 2(1) = 33 ), while the matrix q (0) has the form
= q(0)
t q (0)1 . The
q (0) := diag(1, 1, 1/2). This information allows us to compute M
matrix Rcan [0] is again in Jordan form
0 1 0
Rcan [0] = 0 0 1 ,
(136)
0 0 0
leading to the following expression for the Meijer monodromy around z = 0:
1
0
0
1 = 2i
u [0]M
T [0] := MT
1
0
0
2i 1
(137)
(here Tu [0] = e2iRcan [0] ). The monodromy around z = follows easily from the
expansions (134):
1
0
0
T [] = 0
(138)
2
32 i 1 .
0 2i
1
t
228
Finally, the monodromy around z = 1 is given by
1
0
0
T [1] = T [0]1T [] = 2i
2
32 i 1 .
2
4
6i
4
(139)
These monodromy matrices satisfy

(T [0] I )3 = 0,
(T [1] I )2 = 0,
T []3 I = 0.
(140)
6.3.3. Integral structure

Following the general method described in Section 2, we now look for a cyclic period
for this model. In this example we cannot use U0 as a cyclic period, which is yet another
effect of non-compactness. Indeed, U0 is a constant function over the moduli space, which
means that all monodromy operators T [0], T [1] , T [] leave it invariant. In order to be
able to carry out our program, it is thus necessary to identify another cyclic and weakly
integral period. Here we present a possible approach to this problem, which makes use
of the integrality conjecture of Aspinwall and Lutken [48]. It was argued in [48] that the
cohomology class H 3 (Y, C) should align with a vector of the lattice H 3 (Y, Z) in the
large complex structure limit. Taking this conjecture at face value, 25 it is possible to obtain
a candidate period with the desired properties, as we now explain.
Consider the asymptotic form of the period vector in the large complex structure limit
1
log(z/27) is the algebraic coordinate on the moduli space
talg i, where talg = 2i
0

2
+ O talg , eitalg .
U (s) 0 talg
(141)
2 2
According to [48], the leading term in this equation should be proportional to a vector of
the lattice = H3 (Y, Z). It follows that the Meijer period U2 should coincide with the
period of over an integral cycle, up a constant complex factor. Assuming this is the
case, we can use that cycle in order to generate a sublattice of . In fact, it is easy to check
that U2 is cyclic for the action of all monodromies. Hence repeatedly acting on it with
25 In fact, the status of the arguments of [48] is not completely clear to us, for the following reason. In [48],
it is argued that the Hodge subspaces H 3p,p (Y ) (p = 0, . . . , 3) should become rational subspaces of H 3 (Y )
(with respect to H 3 (Y, Z)) in the large complex structure limit. On the other hand, the standard approach to
mirror symmetry computations does not actually use the Hodge subspaces as such but, rather, the subspaces
p
H3p,p (Y ) = FY W 3p where F is the Hodge filtration and W the monodromy weight filtration (see [61]
for a clear explanation of this point). Doing so is desirable because H3p,p (Y ) vary holomorphically over the
moduli space, while the true Hodge subspaces do not. This makes the associated periods amenable for direct
computation. Working with H3p,p (Y ) is permitted at the level of the closed nonlinear sigma model due to
arguments of [24] and [22] which show that working with elements of H3p,p instead of H 3p,p does not
affect the closed conformal field theory correlators. However, the status of this claim needs to be reexamined
when one includes boundary states thereby working at the level of open conformal field theory. While we have
some preliminary calculations that are relevant to this issue, we will not consider it further here. We emphasize,
though, that whereas the reasoning of Section 6.3.3 involves the conjecture of [48], we will provide independent
evidence for integrality of the vanishing period Uv below (see Eq. (147) for an independent argument to that
effect), so that our conclusion that Uv is weakly integral is independent of the considerations of this subsection.
229
T [0], T [1] and T [] produces a basis of periods which are associated (up to a common
factor) with a basis of a full sublattice 0 of . One can take UE (z) = EU (z) as the period
vector associated to such a basis, with the matrix E given by
0
0
1
E := 0
(142)
2i 1 .
2
4i 1
4
The monodromies of the basis E are given by integral matrices, as they should
0 1 0
2
1 1
TE [0] = 0 0 1 ,
1 0,
TE [1] = 0
1 3 3
1 1 0
0
1 0
TE [] = 1 1 0 .
1 5 1
As a further check, let us consider the period U (z) =
with the vector

1
:= 12 2i
0.
1
1
2i U1 (z) 2 , which is associated
(143)
Since this period asymptotes to U s in the large complex structure limit, it is guaranteed
to correspond to an integral cycle up to an unknown complex factor (see, for example, [30]).
It is easy to express this period in the basis UE , with the result that U = intUE , where
int =
1
[3 4
8 2
1],
(144)
which is indeed an element of 0 after rescaling by the factor 8 2 .

6.3.4. A vanishing period and some arithmetic identities
Now consider the period Uv := 81 2 (4 2 U0 + 3U2 ), which corresponds to the vector

:= 12 0 83 2 .
(145)
Then we have Uv = intUE , with
1
[4 2 1].
(146)
8 2
Hence Uv is again a weakly integral period. This can be checked independently by acting
with T [] and T []2 on U , thus producing a weakly integral basis in which Uv has
rational coefficients:
int =
5
1
1
Uv = U TU []U + TU []2U .
6
3
6
We claim that the period Uv vanishes for z = 1:
Uv (1) = 0 U2 (1) =
4 2
.
3
(147)
(148)
230
Fig. 11. Graph of the vanishing period and special coordinate (in absolute value) versus the real
1 log( z ), for z belonging to the real axis. The point z = 1 corresponds
Khler parameter s = 2
27
3
to s = 2 log 3. We also plot the values of |Uv (s)| and |t| for negative s. The region s
corresponds to the small radius limit, which has no classical analogue. In this limit, both |t| and |Uv |
asymptote to the value 1/2.
Using the expansions of U2 (z) for |z| > 1 and |z| < 1, this is equivalent to the following
slightly nontrivial arithmetic identities:

3 3 F2

27
1/3, 1/3, 1/3
2/3, 2/3, 2/3
3 3 F2
(1) +
(1)(2/3)6
2/3, 4/3
5/3, 4/3
16 3
+ (2/3)3 = 0
(149)
and

1
1 2

2
(1/3) + (2/3) + + 9 log(3)
2
3
1
((n + ) + (n + 2 ) 3 2(n))(1/3)n (2/3)n
3
3
n
n!2 n
=
n=1
4 2
(150)
We checked both identities numerically, but we lack an analytic proof. Note that,
given one of the two identities above, analytic continuation automatically produces the
other. Fig. 11 displays the values of |Uv | as a function of the algebraic coordinate s =
1
z
1
2
log( 27
). We also display the values of the special coordinate t = U = 2i
U1 (z) 12 ,
which measures the mass of a D2-brane.
231
6.4. Interpretation
Our results confirm the picture announced above. In particular, we obtained a vanishing
period which has log2 monodromy around z = 0 and thus can be interpreted as a D4brane state in a IIA compactification on X. This object becomes massless at z = 1, and
not at z = 0 as one would have expected based on classical arguments. Moreover, when
it becomes massless, the D2-brane states remain massive, something quite at odds with
classical expectations.
Admittedly, our discussion of integral structure has been less than complete, essentially
because we avoided working with an explicit construction of the mirror, which would necessitate a more careful discussion of the effects of non-compactness on the interpretation
of standard mirror symmetry constructions. In the next section, we consider a compact
model in which the integral structure is fully understood. In order to make the discussion
more realistic, we focus on a two-parameter example, some features of which were discussed in [37].
7. Examples IV: a two-parameter example

In this section, we consider a special locus in the moduli space of a two-parameter
example. We will show that there is a D4-brane state (possibly bound with a D0-brane)
which becomes massless at every point on this locus. Comparison with classical geometry
leads to some surprising conclusions. Two-parameter examples were studied in the papers
[3739]. We will focus on a model discussed in [37], to which we refer the reader for
background.
7.1. The model and its mirror
One of the two models studied in [37] corresponds to the Khler moduli space of a
degree 8 hypersurface in the weighted projective space W P 1,1,2,2,2 . This can be described
as the space of partial resolutions of the Fermat hypersurface
X0 : x18 + x28 + x34 + x44 + x54 = 0,
(151)
which has a curve of singularities of arithmetic genus 3, given by

: x1 = x2 = 0, x34 + x44 + x54 = 0.
(152)
Partial resolutions X are obtained by blowing up X0 along , which produces an

exceptional divisor E (a ruled surface over ). The Picard group of the resolved variety is
generated by two linear systems L (a pencil of K3 surfaces) and H , in terms of which the
(linear equivalence class of the) exceptional divisor can be expressed as
E = H 2L.
(153)
In order to describe the full cohomology ring, consider the two classes l and h in H2 (X, Z)
defined by the fiber of the ruling of E, respectively, by the intersection of H with L,
232
respectively. Then the relations defining the intersection ring, whose derivation can be
found in [37], are
L2 = 0,
H L = 4h,
H 2 = 4l + 8h,
H 3 = 8,
H 2 L = 4.
(154)
In particular, we have
L l = H h = 1,
(155)
H l = L h = 0,
(156)
hence the bases (h, l) of H2 (X, Z) and (H, L) of H4 (X, Z) are dual with respect to the
intersection form. The Khler cone is then easily seen to be given by J = 1 H + 2 L with
, 0, where A denotes the Poincare dual of a homology class A.
The mirror manifold Y can be easily constructed via the methods of [56] or [57]. In the
/G were Y
is the following
second approach Y can be described as the finite quotient Y
1,1,2,2,2
:
hypersurface in W P
: x 8 + x 8 + x44 + x54 8x1 x2 x3 x4 x5 2x14x24 = 0,
Y
1
2
(157)
and the group G is a copy of Z34 acting in a way described in detail in [37]. The moduli
space is the quotient C2 (, )/Z8 , with the generator of Z8 acting by:
(, ) (, ),
(158)
where := e2i/8 . Hence the parameters (, ) used in [37] give an eight-fold cover of
the complex structure moduli space of Y . The discriminant locus has three components at
finite-distance in the variables (, ), which following [37] we denote as follows:
(1) Ccon : ( + 8 4 )2 = 1, where Y acquires a conifold singularity.
(2) C1 : 2 = 1, where Y acquires an isolated singularity not of the conifold type.
(3) C0 : = 0, where the moduli space acquires orbifold singularities.
Beyond these, one must add at least the divisor:
(4) C : ,
in order to form a reasonable compactification of the moduli space. The detailed description
of this divisor depends on the precise compactification under consideration.
The natural compactification for the purpose of mirror symmetry is a toric compactification defined by the secondary fan [28,29]. In the case under consideration, this turns
out to be a certain partial resolution of the weighted projective space W P 1,1,2 , whose toric
diagram we draw in Fig. 12. The model has four phases, the deep interior points of which
are denoted in the figure by A, B, C and D. Point A corresponds to the smooth CalabiYau
phase, point D is a compact singular phase in which the exceptional divisor E has been
blown down, while points B and C correspond to non-geometric phases. The correct singularity structure of the moduli space is obtained upon replacing the generator (0, 1) with
(0, 4), which leads to orbifold singularities at the points B, C and D. Point C is a Landau
Ginzburg point; at that point, the moduli space has a Z8 quotient singularity. In Fig. 12,
233
Fig. 12. The secondary fan.
we indicate the divisors associated with the generators in square brackets. The divisors E,
H and L are as before, while the generator (0, 1) does not correspond to a divisor on X
but must be added in order to build the moduli space of its mirror (this is sometimes called
the non-compact generator). Fig. 12 provides a parameterization of the moduli space by
means of algebraic coordinates [34]. The dotted lines in the figure represent the compactification divisors, which connect the four deep interior points; these are the rational curves
at infinity of [34]. There exists a common point of intersection among C0 , C1 and Ccon ,
denoted by S in the figure. This point corresponds to = 0, = 1. The curve C1 also
intersects the divisor D(0,1) in a point T , which corresponds to = 1, = . We also
indicate the point U of tangency between Ccon and C , which corresponds to ,
8 1 4 1. Note that the Khler cone can be identified with the cone spanned by the
generators (1, 0) and (0, 1); in particular, point A corresponds to the large radius limit of
our two-parameter model. 26
Fig. 13 is a diagram of the curves C1 and Ccon , as well as of the compactification
divisors D(0,1) = C0 , D(1,0) = C and D(0,1) (we do not draw D(2,1) , since it will
not be important below). The curves C1 and Ccon intersect in two points, one of which
is the point S already discussed above. The other is the point R, which corresponds to
8 2 = 2, = 1. These points will play an important role below.
26 Further resolutions some of them non-toric are needed in order to produce a compactification which
is smooth and such that the various boundary divisors intersect with normal crossings [37]. This is useful for an
exhaustive study of monodromies, but will not be important for us.
234
Fig. 13. Components of the discriminant locus.
In order to understand the role of Ccon and C1 in defining the phase structure of the
model, it is useful to determine the image of these components of the discriminant locus
in the (uncomplexified) Khler moduli space, with respect with the algebraic coordinates
defined in [34]. This can be easily achieved by using the asymptotic form of the mirror
map given on pp. 40, 41 of [37] (or by direct consideration of the monomial-divisor mirror
map of [29]). The result is plotted in Fig. 14. We see that the asymptotic branches of Ccon
separate phases A and B, B and C as well as phase C from A and D, but they do not
separate phase A from D. This last phase boundary is induced by the curve C1 , which also
contributes to the separation between phases B and C.
In what follows, we will only be interested in the locus C1 , leaving a discussion of
other interesting loci to future work. As observed in [37], this locus has the interesting
property that for (, ) C1 , the family Y is birationally equivalent to the mirror of a
complete intersection of a quadric and a quartic in P5 , one of the models we studied in
Section 5. Naively, one expects C1 to be mirror to the locus C 1 where the fiber of the
ruling of the exceptional divisor E of X is blown-down to zero size, since, as mentioned
in [37], X becomes birational to a (2, 4) complete intersection on this locus. Therefore, one
naively expects that a D2-brane wrapping a rational curve in the class l becomes massless
everywhere on the locus C 1 . Note that it is possible to keep the area of the P1 fiber l of
the ruling equal to zero while varying the area of the base , which is why we obtain a
one-parameter locus; the classical volume of E is of course identically zero on C 1 . The

entire locus C 1 corresponds to the wall in the Khler moduli space where l J = 0, i.e.,
2 = 0, 1 0.
As we show below by direct computation, this picture receives important modifications
due quantum corrections. Indeed, we will see that the D2-brane wrapping l does not acquire
vanishing mass on the mirror locus to C1 due to the familiar fact that there is a nonzero
B-field flux through
cycles in l at generic points on this locus. Nevertheless, the classical
conclusion that l J = 0 is still valid after quantum corrections are taken into account.
The D-brane which becomes massless everywhere on (the mirror to the locus) C1 , is a
D4-brane, as we will see explicitly. In the last subsection, we also consider the restricted
235
Fig. 14. The image of the discriminant components C1 and Ccon with respect to the algebraic
measure. The image of Ccon is the three-pronged structure at the center of the diagram, extending
towards infinity in directions parallel with the generators (0, 1), (1, 0) and (2, 1). The image of C1
1 log 4 0.2206
is the vertical line parallel with the generator (0, 1), located at a distance given by 2
in the positive direction along this generator. We also display the image of the points R, S, T and
1 log 4, 9 log 2) (0.2206, 0.9928), while the images of
U . The image of R has coordinates ( 2
2
S, T lie at infinity along the image of C1 . The image of U lies at infinity on the horizontal branch of
the image of Ccon . This branch asymptotes to a line parallel with the generator (1, 0) and placed at a
8 log 2 0.882 below it.
distance 2
theory on the locus C1 and show explicitly that it is equivalent to the model P5 [2, 4]. In
particular, this allows us to identify the collapsing 6-brane found in Section 5 with a certain
6-brane state of the ambient two-parameter model.
7.2. Restriction of periods to the locus C1
A basis of periods for our model was obtained in [37]. Such a basis consists of the 6
periods denoted there by 0 , . . . , 5 , whose expansions were computed in [37] in some
regions of the moduli space. The precise form of i depends on whether i is even or odd,
so we think of our collection of periods as being divided into two subsets, {2j }j =0,...,2
and {2j +1 }j =1,...,2 .
We start with the -expansions of i obtained in [37]:

3
1
r
3 r
(1) sin
2j (, ) = 3
2j r r (, ),
4
4
r=1
2j +1 (, ) =

3
1
r
3 r
(2j +1)r r (, ),
(1)
sin
4
4 3
r=1
(159)
236
Fig. 15. Contour of integration for r and r . We also display the poles of the integrands at = n
(n a nonnegative integer) on the positive real axis. The integrands also have poles on the negative
real axis, which are irrelevant for us.
where
1
r (, ) =
2i

d fr (, , ),
1
r (, ) =
2i

d er (, , ),
with the integrands

4 () 12 4
2
u (),
sin ( + r/4) (4)
4 () 12 4
2
er (, , ) =
sin ( + 4r ) (4)
fr (, , ) =
u () sin ( + 4r ) u () sin( r
4 )
.
sin()
The contour is shown in Fig. 15.
The function u () in the expressions above is defined by

1
2 , 1
2
u () := (2) 2 F1
.
1
2
(160)
We refer the reader to [37] for a discussion of the basic properties of the functions u (),
some of which we will use below.
The identity (B.1) of Appendix B, together with the completion formula allows us to
write:
( + 4r )()3
z u (),
fr (, , ) = 8 2
k
k=1,...,3 ( + 4 )
er (, , ) = 8
( + 4r )( + 1)()4
z

k
k=1,...,3 ( + 4 )
237

r
r
u () sin +
+ u () sin
,
4
4

where k=1,...,3 { } := k=1,...,3,k=4r { } and we defined z := (2)4 .
We are interested in the behaviour of the periods near the curve C1 , i.e., in the limit
of (159) for 1. The expansions of these periods in the region |8 4 /( 1)| < 1 are
given in [37], but this is not what we need, since C1 lies in the complement of that region.
In order to extract the information we desire, we must obtain the expansions of (159) for
large , i.e., for |z| 1. This can be achieved once again by a computation of residues.
Let us first discuss the expansion of r in this region. Since |z| 1, we can close
the defining contour to the right, towards +. The resulting Cauchy expansion receives
contributions only from the poles of fr located at = n, with n a nonnegative integer;
these poles are all of order 3. Computation of the associated residues proceeds as before,
giving:
r (, ) = 4
n!
n=0

3
(n + 4r )
k
k=1,...,3 (n + 4 )
(z)n Hr (n, , )un ().
(161)
A similar computation for r (which receives contributions from poles of order 4) gives

(n + 4r )
8
r
(z)n Kr (n, , )un ().
sin
r (, ) =

k
3
3
4
n!
(n
+
)
k=1,...,3
4
n=0
(162)
To arrive at the last formula, we made use of the property
un () = (1)n un (),
(163)
which can be easily deduced from the Euler representation of un (see [37]).
In these expansions, the quantities Hr , Kr are defined by

2
Hr (n, , ) := p1 (n, r) + 1 (n, ) log(z) + p2 (n, r) + 2 (n, ),

3
Kr (n, , ) := q1 (n, r) + 1 (n, r, ) log(z)

+ 3 q1 (n, r) + 1 (n, r, ) log(z) q2 (n, r) + 2 (n, r, )
+ q3 (n, r) + 3 (n, r, ),
where

r
pi (n, r) := (i1) n +
4

n

1
k
i
(i1)
3(i 1)!
(1)
n +
3
(1) ,
4
ki
k=1
k=1,...,3

r
qi (n, r) := (i1) n +
+ (i1) (n + 1)
4

n

k
1
i1
(i1)
(i1)
4
n +
(1) 4(i 1)!

+ (1)
4
ki
k=1,...,3
k=1
238
and

di
log u ()
,
i
d
=n

di
r
r
log u () sin +
i (n, r, ) :=
+ u () sin
i
d
4
4
i (n, ) :=
=n
One can rearrange the result in a form which displays i as expansions in the basis of
functions given by zn (log z)s . Write
Hr (n, , ) :=
2
s (r, n, )(log z)s ,
s=0
Kr (n, , ) :=
3
s (r, n, )(log z)s
s=0
with the coefficients

0 := p2 + 2 + (p1 + 1 )2 ,
1 := 2(p1 + 1 ),
2 := 1
and
0 := (q1 + 1 )3 + 3(q1 + 1 )(q2 + 2 ) + (q3 + 3 ),

1 := 3 (q1 + 1 )2 + (q2 + 2 ) ,
2 := 3(q1 + 1 ),
3 := 1.
We also define 3 := 0 in order to give a concise form to subsequent formulae. Then we

have
i (, ) :=
3
zn (log z)s Mi (s, n, ),
(164)
s=0 n=0
with the expansion coefficients

Mi (s, n, ) =
n!3
3
un ()
k
k=1 (n + 4 )
Li (s, n, ),
(165)
where
L2j (s, n, ) :=
L2j +1 (s, n, ) :=
3

(1)r 2j r sin2
r=1
3
2
3

r
s (n, r, ),
4
(1)r (2j +1)r sin3
r=1

r
s (n, r, ).
4
In order to arrive at these expressions, we made use of the identity

(n + 4r )
1
r
= (1)n
,
sin

k
k
4
k=1,...,3 (n + 4 )
i=1,...,3 (n + 4 )
which follows by applying the completion formula.
(166)
(167)
(168)
239
It is now straightforward to extract the behaviour of the periods in the limit 1. Since
un () is a polynomial for integer n, it has a finite limit at the point = 1, which is easily
computed ([37]):
(2n)!
.
n!2
Moreover, we have
un (1) =
(169)
4 (1/2 + )
,
u (1) =
(1 + )
which allows us to compute the quantities i (n, 1) and i (n, r, 1)
1 (n, 1) = 2 log(2) (n)
2 (n, 1) = (n) +
(170)
1
+ (n + 1/2),
n
1
+ (n + 1/2),
n2
(171)
and
1 (n, 1, 1) = 2 log(2) (n)
1
1
1
+ (n + 1/2) + + i
n
2
2
1
1
= 1 (n, 1) + + i,
2
2
1
+ (n + 1/2) = 1 (n, 1) + i,
n
1
1
1
1 (n, 3, 1) = 2 log(2) (n) + (n + 1/2) + i
n
2
2
1
1
= 1 (n, 1) + i,
2
2
1
1
1

2 (n, 1, 1) = 2 + (n + 1/2) 2 (n) i 2 = 2 (n, 1)) 2 i 2 ,
n
2
2
1
2 (n, 2, 1) = (n) + 2 + (n + 1/2) = 2 (n, 1),
n
1
1
1
2 (n, 3, 1) = (n) + 2 + (n + 1/2) 2 + i 2 = 2 (n, 1) 2 + i 2 ,
n
2
2
3 3
1 3
1

3 (n, 1, 1) = i + (n + 1/2) + (n) 2 3 ,
2
2
n
1

3 (n, 2, 1) = (n + 1/2) (n) 2 3 ,
n
3 3
1
1
3 (n, 3, 1) = i (n) + (n + 1/2) 2 3 3 .
(172)
2
n
2
The relations given above are valid for n = 0. While , are apparently singular at n = 0,
the associated analytic extensions i (, 1), i (, r, 1) have finite limits for 0, which
we list in Appendix B. It follows that each of the periods i has a finite limit for 1, as
long as |z| > 4. 27 These limits are obtained by substituting un (1), i (n, 1) and i (n, r, 1)
in the expansions given above.
1 (n, 2, 1) = 2 log(2) + i (n)
27 For = 1, our expansions converge if |z| > 4.
240
7.3. Integral structure and a vanishing period

A symplectic basis of the lattice H3 (Y, Z) was computed in [37]. If

G0
0
1 G1

2 G2
=
=
0
3 z
z1
4
z2
is the period vector of in this basis, then and

0
1

2
:=
3

4
5
are related by = m, with
1
1
0
0
0
0
1
0
1
1
0
1
3/2
0
0
0 1/2 0
.
m=
1
0
0
0
0
0
1/4 0
1/2
0
1/4
0
1/4 3/4 1/2 1/2 1/4 1/4
(173)
In particular, the periods 4 = z1 , 5 = z2 , which correspond to the integral of along

3-cycles which are mirror to members of the classes h and l, respectively, are given by the
following linear combinations of the the periods :
c1 = [1/4 0
c2 = [1/4 3/4
1/2 0 1/4 0],

1/2
1/2
1/4 1/4].
(174)
(175)
Moreover, the period >3 = z0 coincides with the fundamental period 0 , which
corresponds to the integral of over a 3-cycle mirror to the homology class of a point.
Now let us look for an integral cycle with the property that the associated period of
vanishes identically on the locus C1 . Such a period is of the form:

v = =
(176)
i i ,
i=0,...,5
where the row vector is such that int = m1 has integral entries. By virtue of (164),
the condition that v |C1 is identically zero implies that the following linear combination
of the coefficients L must vanish for all nonnegative integers n and all s = 0, 1, 2, 3:

i Li (s, n, 1) = 0.
(177)
i=0,...,5
241
View Li = Li (s, n, 1) (i = 0, . . . , 5) as sequences defined over the countable set

{0, 1, 2, 3} {0, 1, 2, . . .}.Then we are looking for a nontrivial linear relation among these
five sequences. The existence of such a relation requires that all truncated determinants
associated with the infinite linear system (177) vanish, a condition which is easily seen to
hold numerically. Then a solution of (177) can be found by doing a numerical search for
small integer values of i , with the result that the vector
v := [0 0 1
1 1
1]
(178)
satisfies (177). We conclude that the period

v (, ) = 2 (, ) 3 (, ) + 4 (, ) 5 (, )
(179)
vanishes when restricted to C1 . To check integrality of the associated 3-cycle v , it suffices

to compute the row vector
int = m1 = [0 1
2 2
0 0],
(180)
which is integral indeed.

The nature of the cycle v mirror to v can be determined by finding the position of the
periods v in the monodromy weight filtration of H3 (Y ) associated with the large complex
structure point. By using the monodromy matrices computed in [37], one can easily see that
this filtration can be represented in the basis by: 28
W = [W0 , W1 , W2 , W3 ]

1
1 0 0
1
0 0 3 0 0

0 0 0 2
,0
=
,
0 0 2 0
0

0
0 0 1 0
0
1 0
0
0
1
0
0
0
0
3
0
0
0
1
0
3
0
1
0
0

0
0
0 0
0
,0
0
0
1 0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
0
,
0
0
where the periods associated with the linear combination of 0 , . . . , 5 given by the
columns of each matrix span the corresponding weight subspace. It is easy to see that
v is a linear combination of the columns of W2 , so that the mirror cycle v is a 4-cycle.
Numerical analysis also shows that the periods 1 = G1 , 2 = G2 , 3 = z0 , 5 = z2
have vanishing imaginary part along the locus C1 , though none of them vanishes identically
on this locus. In particular, none of the special coordinates on the moduli space, which are
given by
4 z1
5 z2
(181)
= 0,
t2 =
=
3 z
3 z0
vanishes identically on C1 . The coordinate t2 is real along the locus C1 , while the
coordinate t1 is neither purely real, nor purely imaginary. Since the mirror Khler class
is given by
t1 =
k = B + iJ = t1 H + t2 L,
28 This is the reduced form of the filtration, see Appendix A.
(182)
242
it follows that

k = t2 ,

k = t1 ,
(183)
and since J = Im(k) = Im(t1 )H , this implies

J = 0,
J = Im(t1 ).
l
(184)
As reviewed above, the mirror of C1 is classically a locus in the moduli space of X

where
the exceptional divisor E has been blown down to an curve of genus 3. The result

J
=
0 on this locus confirms that part of this classical geometric intuition is preserved in
l
the quantum
setting: the geometric volume of the fiber of the ruling remains zero. However,
the result | l k| = |t2 | = 0 shows that the B-field modifies this conclusion for the quantum
volume of l, ensuring that the mass of the D2-brane wrapping l does not vanish identically
on C1 . In fact, we have some numerical evidence that no D2-brane becomes identically
massless on this locus. Rather, as shown above, the vanishing period we found:
v = 1 22 + 23 = G1 2G2 + 2z0
(185)
corresponds to a 4-cycle. Hence it is the exceptional divisor E possibly mixed together

with lower dimensional cycles which has zero quantum volume on the locus C1 .
In this example it proves instructive to compare these exact results with semiclassical
calculations. To do so, consider the quantities

2
k = 8t1 (t1 + t2 ),
k 2 = 4(t1 )2 ,
H

k =
and
k = 8t1 t2 ,
(186)

2
J 2 = 8 Im(t1 ) ,

k 2

2
J 2 = 4 Im(t1 ) ,

J 2 = 0.
(187)
This amounts to using the classical relations for the volume, but applied to thequantummechanically correct values of J and k on the locus mirror to C1 . The result E J 2 = 0
agrees with classical reasoning, showing that the mirror of C1 is indeed the locus where
E collapses to zero classical volume. Thus, whereas in all previous examples the location
in the Khler moduli space where we expect to find collapsed cycles based on classical
reasoning is shifted by quantum effects, in this model the classical and quantum loci
agree. (This is somewhat akin to what happens in flop transitions where the flopping
curve has zero quantum volume on the classical flop wall.) However, the D4-brane state
243
which becomes massless at least naively seems to carry some D0-brane charge (though
this statement is, of course, affected by the ambiguities discussed above). Moreover, the
D2-branes wrapping holomorphic subcycles of E remain massive on C1 , a result in stark
contrast with classical geometric reasoning.
7.4. Postscript: quantum geometry of the restricted theory
As mentioned above, it is believed that the restriction of our two-parameter model to
the locus C1 is equivalent to the one-parameter model P5 [2, 4] discussed in Section 4. The
argument given in [37] for this equivalence is based on classical geometry, namely on the
fact that the restrictions of both Y and X to the locus C1 and its naive mirror locus are
birationally equivalent to the one-parameter mirror families Y , respectively, X describing
the model of Section 4.1. 29
Having computed the required analytic continuations of all periods, we are in a position
to perform a detailed check of this statement, which will also serve to complete the
picture given above of what happens quantum-mechanically when one restricts to C1 .
For this purpose, we must understand precisely how the one-parameter model P5 [2, 4] is
embedded in the two-parameter model at the quantum level. We start with the following
natural geometric proposal for the embedding of Y into Y . Consider a period U of the
two-parameter model Y as one approaches the locus C1 . If U has nontrivial monodromy
around C1 , then the process of restricting U to C1 is in general ill-defined. Therefore, it
is natural to postulate that only those periods U which have trivial monodromy around C1
have a counterpart in the restricted model Y . 30
We thus identify H3 (Y ) with the subspace of H3 (Y ) consisting of those classes which
have trivial monodromy around C1 . The latter can be easily determined as the subspace of
period vectors fixed by the associated monodromy operator (denoted by B in [37]), with
the result that it is spanned by the linear combinations of 0 , . . . , 5 defined by the rows
29 As we shall see in a moment, restriction here means more than just restricting the parameter space; we will
indeed show explicitly that it also requires a certain projection on periods. Thus, our one parameter model embeds
as a subsector of the two-parameter model.
30 The alert reader may have noticed that this intuitive argument is not as clear-cut as it may seem. Indeed, it is
perfectly possible to have periods with nontrivial monodromy around some component of the discriminant locus,
which however do have a well-defined limit along that component. A familiar example in lower dimension is
provided by a generic period on the mirror quintic in the vicinity of the conifold point. As shown in [43], such a
period has the form
U = constant (z 1) log(z 1) + g(z),

with g a function regular at z = 1; in that case, U has a well-defined limit at z = 1 even though it has nontrivial
monodromy around the conifold. Hence the argument above is based on something more than the existence of
a well-defined limit on C1 ; essentially, it is an argument about the monodromy weight filtration defined by the
degeneration associated with C1 . In any case, correctness of this ansatz is justified by the consistency of the
results we will obtain.
244
of the following matrix:
0 0 0 0
1 1
0 1 0 3 3 0
.
A :=
1 0 0 3
3 0
0 0 1 1
0 0
Denote the associated periods (restricted to C1 ) by

j (z) :=
Aj i i (z, 1), for all j = 0, . . . , 3.
(188)
(189)
i=0,...,5
In order to understand the precise relation between the two models, we must compute the
transition matrix P from the Meijer basis Uj of P5 [2, 4] to the basis j . The existence
of a constant such matrix on the locus C1 amounts to a proof of the quantum equivalence
between the restriction of Y to C1 and this model.
In order to check this statement and obtain the matrix P , it is convenient to use a matrix
formalism. Considering the column vectors

0
1
:=

2
3
and

U0
U1
U :=
U2 ,
U3
the row vector Z(z) = [1 log z (log z)2 (log z)3 ] as well as the matrix M(n) =
(Msi (n))s=0,...,3,i=0,...,5 with entries Msi (n) := Mi (s, n) (where Mi (s, n) are given
in (164)), we can write
(z)t =
zn Z(z)M (n),
(190)
n=0
with M (n) = M(n)At . Using the results of Section 4, one can similarly write
U (x)t =
x n Z(x)l(n),
(191)
n=0
where x is the coordinate on C1 defined by the property that

1/2, 1/2, 1/4, 3/4
U0 (x) = 4 F3
(x)
1, 1, 1
and l(n) = (lsj (n))s,j =0,...,3 is a matrix with entries
(1)j (1/2)n (1/2)n (1/4)n (3/4)n
vsj (n).
j!
n!4
Here vsj (n) := vj (s, n) with vj (s, n) given in (63) and (64).
lsj (n) =
(192)
245
It is important to notice that the hypergeometric coordinate x on C1 does not coincide

with z. In fact, these two coordinates are expected to differ by the rescaling
4
z= ,
x
(193)
as can be seen by comparing the expression for 0 given on p. 27 of [37] and the expression
of the P5 [2, 4] fundamental period given in Table 3.1 of [40]. In other words, we expect
the relation
(z) = P U (x).
(194)
Instead of testing this relation directly, it proves more convenient to proceed as follows.
1
Define a new coordinate y on C1 by y = w1 = x
, where w = x is the natural variable on
5
the moduli space of P [2, 4] required by the monomial-divisor mirror map; thus > 0 is
given by
log = 2(1/2) + (1/4) + (3/4) 4(1) = 5 log(4).
Hence = 45 and y =
45
x .
(195)
Then we expect z = x4 =
( y ) =
1
4
y with = 4 and we must
1
P U ( y ). It is now easy to see that the
test the existence of a matrix P such that

associated rescalings have the following effect on the matrix function Z:

y
1
Z
= Z(y)C(),
Z
= Z(y)c().
(196)
Here C(), c() are two upper triangular matrices whose nonzero entries are given by

ps p
ps
p p
csp () = (1)
(log ) ,
(log )ps , (197)
Csp () = (1)
s
s
for all 0 s p 3. It follows that there should exist an invertible matrix P GL(4, C)
such that
n C()M(n)At = n c()l(n)P t ,
for all integers n 0.
(198)
To determine a candidate for P , it suffices to consider this equation for n = 0, which gives
P t = l(0)1 c()1 C()M(0)At .
(199)
The reason why we prefer to work with the variable y can now be made clear. Substituting
the formulae for the expansion coefficients of i (z) and Uj (x) gives very complicated
expressions for M(0) and l(0), which are difficult to simplify directly. However, by virtue
of our experience in Section 3, it can be expected that rescaling to the variable y has the
effect of dramatically simplifying the expressions involved. This is because it is much
easier in practice to simplify the linear combinations of the elements of l(0) which form
the matrix c()l(0) than the elements of l(0) themselves; the same turns out to be the
case for C()M(0). Going through this exercise leads to the following remarkably simple
246
expressions
C()M(0)At
7 3 +99i(3)
2/3
3
17i

2
= 2
42
2i 3
61 3 1320i(3)
6 3
13 3 +396i(3)
2 3
3 +22i(3)
29i
3
15i
2
9i
2
11
2
2
9
2 2
3
2
5i
3
3
3i
2 3
i
2 3
1 0 7/6 2 22(3)
11
2
0 1 i
6 ,
c()l(0) =
0 0
1/2
0
0 0
0
1/6
which immediately give
8 2
3 i3
14
6 i
17 i
11 2 10 i3

23
P = 2
17 15 i
9 2
9 i3
(200)
(201)
10 4 i
6 2
3 i3
In order to arrive at these results, we used relations (B.5) of Appendix B. Using this value
of P , one can now check numerically that (198) is indeed satisfied. We did this up to n = 40
and with 60 significant digits, but we were not able to simplify the resulting expressions in
order to give an analytic proof to this claim.
The matrices (188) and (201), give a complete characterization of the quantum relation
between the restricted model P5 [2, 4] and the ambient two-parameter model. It should be
noted that we obtained considerably more detailed information about this relation than
could have been extracted by less direct methods such as those of [34]. For example,
following [34], one can consider the restriction of the PicardFucks equations of the
two-parameter model to the locus C1 and check that the resulting ordinary equation
is equivalent to the hypergeometric equation of the model P5 [2, 4] after performing an
appropriate change of variable. While this procedure suffices to show that the restricted
model is equivalent to P5 [2, 4] at the quantum level, it does not fix the precise isomorphism
between the two. Doing so requires knowledge of the relation between two bases of periods
of these models, which is precisely the information encoded by the matrices P and A
computed above.
Now let us consider the vanishing period Uv = U3 3 2 U1 which we found for the
restricted model in Section 5. Since we know the integral structure of the 2-parameter
model, we can obtain the integral structure of P5 [2, 4] by restriction. In particular, since
= m is the period vector in an integral basis, and since = A = Am1 , while
Uv = U , with := ( 13 )[0 3 2 0 1] and U = P 1 , it follows that the vanishing
period Uv which we found in Section 5 can be expressed as
Uv = P 1 Am1 = int ,
(202)
247
with
int = 86i 2 [86
79 252
190 172
156],
(203)
i.e., we have

Uv = 86i 2 86G 0 79G 1 + 252G 2 190z0 + 172z1 156z2 .
(204)
In particular, we see that Uv is an integral linear combination of integral periods up to

a global factor; hence it is indeed associated with an integral 3-cycle which collapses at
x = 1. The reader can easily check that the large integers appearing in (204) are coprime.
This underscores the rather nontrivial character of this vanishing period, when viewed from
the perspective of the ambient two-parameter model. It would be a difficult task to identify
Uv directly from the data of the two-parameter model, without making use of independent
knowledge of the vanishing period for the model P5 [2, 4].
It is instructive to re-interpret the result from this ambient perspective. The point x = 1
corresponds to z = 4; this is the point R, one of the two points where C1 intersects Ccon
(the other point S corresponds to z = 0 x = , which is the small radius limit of
the model P5 [2, 4]). As shown in [37], there exists an integral cycle vanishing identically
on the locus Ccon , namely the cycle corresponding to the period 1 0 = G 0 ; however,
it is easy to see that this period has nontrivial monodromy around C1 and hence it does
not restrict to a period of the model P5 [2, 4]. Clearly both of our cycles belong to the
highest component W3 of the (reduced) large radius monodromy weight filtration of the
two-parameter model; hence both can be interpreted as 6-brane states in the mirror theory
(the IIA theory compactified on X). It follows that a partial picture of massless D-branes
in the type IIA compactification on X is as follows. First, there is a 6-brane state which
becomes massless everywhere on the conifold locus Ccon ; second, there is another 6-brane
state which becomes massless at one of the intersection points (the point R) between C1
and Ccon ; when restricting to the locus C1 , this state induces the vanishing 6-brane of the
resulting one-parameter model. Finally, there is a 4-brane state which becomes massless
everywhere on the locus C1 .
8. Summary and conclusions

One of the essential lessons of recent years is that much interesting physics can be
learned from studying systems at points in their moduli space which naively appear to
be singular. In a wide range of examples, these singularities signal the appearance of
new massless degrees of freedom that are responsible for yielding non-singular physics.
In the context of compactified string theory, these singularities are often associated with
geometrical degenerations of one form or another, and the new massless degrees of
freedom often arise from D-branes wrapped around collapsing cycles.
In this paper, we have studied in some detail the D-branes that become massless along
various loci in the moduli space of CalabiYau examples compact and noncompact
and used this to extract properties of quantum volume in string theory. The features
248
we find confirm and extend previous results, and highlight significant qualitative and
quantitative departures from classical expectations. Figs. 6, 7, 8, 9 and 11 illustrate some
of the new features of quantum volume and can be thought of as a visual summary
of our results. Although we have not overly stressed this point, we also showed that
the physical arguments which ensure that some D-brane mass vanishes within a phase
boundary can be used to generate interesting and seemingly novel arithmetic identities.
Eqs. (86), (110), (149) and (150) give some representative examples. In fact, one can
reverse the argument and use degenerations of CalabiYau manifolds in order to produce
such identities in a systematic manner. It would be interesting to gain a better understanding
of the mathematics behind this process, as well as of the connection between these results
and the considerations of [21].
From the physical point of view, there are, perhaps, two main issues that remain
unresolved. The first is to determine the precise action of mirror symmetry on the integral
structures of a mirror pair (X, Y ). As we have seen, our inability to resolve this issue
prevents the precise identification of the even-dimensional D-branes on X mirror to
specified odd-dimensional D-branes on Y . Nevertheless, it is only the lower brane charges
that remain ambiguous as we can use the leading logarithmic behaviour of periods on
Y to identify the maximal dimensional component of mirror D-branes on X. Certainly,
though, filling this gap in our understanding is an important subject for the future. The
second issue is that of marginal stability. While we have studied D-branes at geometrical
degenerations, there are other loci in moduli space where a given D-brane state might
be massive, but unstable to decay. Perhaps the techniques of this paper, as well as the
important contributions of [21,46], will allow us to understand the marginal stability loci
in detail for CalabiYau compactifications. We hope to return to this issue in future work.
Appendix A. The monodromy weight filtration

In this appendix, we give a very brief account of role played by monodromies in the
study of moduli spaces of CalabiYau threefolds. This topic is intimately connected with
the theory of variations of Hodge structure. Basic references for this subject (which we
will not attempt to present in detail) are [47] for the general theory of local systems and
the geometric formulation of Fuchsian equations, [23] for the abstract theory of Hodge
structures, [25] as general references and [26] for the monodromy theorem. An excellent
introduction to the mathematical structures underlying mirror symmetry (up to date at the
level of 1997) can be found in [61].
For our purpose, the set-up is that of a family of CalabiYau threefolds (understood
as complex manifolds of vanishing first Chern class), i.e. a (proper) holomorphic map
: X S over a connected base S whose fibers Xs = 1 (s) are all diffeomorphic with
a given CalabiYau manifold X. The complex manifold S plays the role of a parameter
space for the complex structure on Xs . For simplicity, we assume that dim S := r is equal
to h2,1 (X), so that our family captures the full space of deformations of complex structure
of X. In general, one must allow for bases S which are non-contractible; this appears
249
because one is usually interested not only in the family X , but in a compactification :

X
S given by a (proper) map which has singularities above the locus
S S; in
other words, we wish to allow the complex structure of X to degenerate as we vary the
parameters s S. In order to describe the middle cohomology of the various fibers Xs , one
considers the derived sheaf R 3 CX , which is intuitively the sheaf having H 3 (Xs , C) as
its stalk above each point s. The sections of this sheaf are simply the topologically constant
families of cohomology classes. The important subtlety is that, for a general basis S, such
sections will have nontrivial monodromies around the components of
S S; these are
described by actions s : 1 (S, s) Aut(H 3(Xs )) of the fundamental group of S on the
spaces H 3 (Xs ). In other words, if g(s) H 3 (X) is such a section, then upon following a
closed path l : [0, 1] S starting and ending at a given point s0 , the cohomology class g(s)
does not necessarily return to its original value g(s0 ) but rather to the value s0 ([l])g(s0 ),
where [l] is the homotopy class of l in 1 (S, s0 ). An important device for studying this
situation, discussed at length in [47] is to represent the local system R 3 CX as the sheaf
of flat holomorphic sections of a pair (H, ), where H = OS R 3 CX is a holomorphic
vector bundle and is a connection on H which, in our context, is called the GaussManin
connection. In other words, each topologically-constant cohomology class in H 3 (X, C)
can be viewed as a holomorphic section of H which is covariantly-constant with respect
to . The existence of nontrivial monodromies around the components of
S S is then
reflected by the fact that the connection coefficients of in a basis of sections which extend
to smooth sections over a component of
S S will have poles along that locus (this is made
more precise by the nilpotent orbit theorem, as we explain below).
In order to understand the nature of the monodromies around a compactification point
P
S S, we consider the case when P is the intersection of r compactification divisors
having normal crossings at P . This situation can be represented by taking S to be locally a
Cartesian product of r punctured disks
= (D )r = {z = (z1 , . . . , zr ) Cr |0 < |zi | < 1, i = 1, . . . , r},
(A.1)
= {z C | 0 < |z| < 1}. This has a natural compactification =

with D
where
a copy of the unit disc. The compactification divisors intersecting at P = (0, 0, . . . , 0) are
then locally given by the equations zi = 0. The fundamental group 1 () is generated by
the r circles i : {zi = 12 e2it (t [0, 1]), zj = 0 for j = i}. The monodromy theorem [26]
assures us that the automorphisms Ti of H 3 (X) defined by transporting the cohomology
classes around i are quasi-unipotent, i.e., each of them satisfies
ai
r
Ti I i = 0
(A.2)
D
(D)r ,
for some pair of positive integers (ai , ri ).

To avoid lengthening this discussion, let us focus on the case when all Ti are unipotent,
i.e., one can take ai = 1 in (A.2). Then one can define operators Ni = log Ti :=

(1)k1
(Ti I )k (the series defining the logarithm terminates due to the fact
k=1,...,ri 1
k
that Ti is unipotent) as well as an operator-valued function

1
N (z) := e 2 i
(A.3)
(log zi )Ni
i=1,...,r
250
(this series again reduces to a finite sum, since the operators Ni are obviously nilpotent).
We can now state another basic fact of relevance for us, namely the nilpotent orbit
theorem. This states that one can use the monodromy operators Ti in order to generate
over .
More precisely, given
an extension of H to a holomorphic vector bundle H
31
a flat local frame (ej (z))j =1,...,2r+2 of H, the sections ej (z) will generally have
nontrivial monodromies around the curves i (which can be identified with the holonomy
transformations of along these curves). It follows that ei do not directly extend to .

However, the nilpotent orbit theorem states that the sections
j (z) := N (z)ej (z)
(A.4)
are single-valued (i.e., have trivial monodromy) around each of the curves i . It follows
is trivializable
that one can extend the bundle H to trivially, i.e., by declaring that H
on a vicinity of each of i and extending i to a frame over that vicinity by continuity of
their local representations in that trivialization. Then a simple computation in the frame i
32
shows that extends to a connection with regular singular points on H.
We can now describe the associated monodromy weight filtration. This is simply a
filtration
0 W0 W1 W6 = H 3 (X)
(A.5)
of H 3 (X) with the property

Ni Wk Wk2 ,
for all k
and such that for all positive a1 , . . . , ar , the powers of the operator N =
induce isomorphisms
N p : W3+p /W2+p W3p /W2p ,
(A.6)
i=1,...,r ai Ni
(A.7)
for all p = 0, . . . , 3, where W1 is defined to be the null vector space. These properties can
be used in order to determine W .
Finally, let us recall some useful concepts introduced in [30,54,62]. The boundary point
P is called maximally unipotent (or a large complex structure limit point) if:
(1) All monodromy transformations Tj are unipotent;
(2) dim W0 = dim W1 = 1 and dim W2 = 1 + r;
(3) If g0 , . . . , gr is a basis of W2 such that g0 W0 , then the matrix M := (mij )i,j =1,...,r
(defined by Ni gj = mij g0 ) is invertible.
Then (A.7) gives W0 = W1 W2 = W3 W4 = W5 W6 = H 3 (X) so we can work with
the reduced filtration:
0 W0 W1 W2 W3 = H 3 (X),
(A.8)
where Wk := W2k for all k = 0, . . . , 3.

31 Flat with respect to the GaussManin connection.
32 Concretely, the connection matrix in a local frame of H
above the origin has poles along the curves i .
where 1 (log Z) is the space of
1 (log Z) H,
:H
Abstractly, the extended connection gives a map
1-forms in d log zi = dzi /zi and Z := .
251
For a one-parameter model (r = 1), one has a single monodromy operator T around
P and this definition is equivalent to the requirement that T is maximally unipotent, i.e.,
(T I )4 = 0 and (T I )3 = 0. 33 Then (A.8) can be viewed as the natural filtration defined
by the nilpotent operator N = log(T I ). In other words, the fact that T I is nilpotent
of maximal order means that the Jordan form of N is
0 1 0 0
0 0 1 0
NJ =
(A.9)
0 0 0 1,
0 0 0 0
and Wk = e0 , . . . , ek , where e0 , . . . , e3 is any basis of H 4 (X) in which the matrix of N
has this form. In terms of the associated PicardFucks system, such a monodromy matrix
corresponds to a basis of solutions u0 , . . . , u3 with the property that 34

uj (z) =
ai (z) log(z)j ,
(A.10)
0ij
where ai are single-valued around z = 0. In other words, the spaces Wj consist of

the logj -monodromy periods, where the terminology indicates the leading logarithmic
behaviour.
Appendix B. Some useful identities

For the convenience of the reader, we list a few identities which are important for
simplifying some of the expressions encountered in this paper. A first group of identities
can be obtained by starting with the basic equality
n1
(x + k/n) = (2)
n1
2
n1/2nx (nx),
(B.1)
k=0
which immediately leads to

n1
(x + k/n) n(nx) = n log(n),
k=0
n1
(j ) (x + k/n) nj +1 (j ) (nx) = 0.
(B.2)
k=0
33 We assume a compact one-parameter model; the correct generalization for a non-compact case such as the
C3 /Z3 orbifold considered in Section 5 is to require (T I )3 = 0 and (T I )2 = 0.
34 If N has the form (A.9), then T = eN has the form 0
0
0
1
1
0
0
1/2
1
1
0
1/6
1/2
.
1
1
252
These relations can be used for x = 1/n to give

n1
(k/n) (n 1)(1) = n log(n),
k=1
n1
(j ) (x + k/n) = (nj +1 1) (j ) (1).
(B.3)
k=1
We also recall that (1) = , where is Eulers constant and that

(j ) (1) = (1)j 1 j ! (j + 1),
(2k) = (1)k1
(2)2k
B2k
2(2k)!
(B.4)
with Bj the Bernoulli numbers. In particular, we have (2) = 2 /6.

Another set of identities, useful for simplifying some of the expressions encountered in
Section 6, is
(1/4) = 3 log 2
,
2
(3/4) = 56 (3) + 2 3 ,
(3/4) = 3 log 2 +
(1/4) + (3/4) = 2 2 ,
(3/4) = 56(3) + 2 3 ,
(1/4) = 56 (3) 2 3 ,
1
(1/2) = 2 ,
2
(1/2) = 2 log 2,
,
2
(1/4) = 56 (3) 2 3 ,
(1/2) = 14(3).
(B.5)
These follow by applying (B.3) for n = 4 and n = 2, together with the relation
(x) (1 x) = ctg(x)
(B.6)
and the identities following from it by differentiation.

Finally, we list the limiting values of the quantities i (, 1) and i (, r, 1) of Section 7
for 0
1 (0, 1) = 0,
2 (0, 1) = 1/3 2
(B.7)
and
1 (0, 1, 1) = 1/2i + 1/2,
1 (0, 2, 1) = i,
1 (0, 3, 1) = 1/2i 1/2,
2 (0, 1, 1) = 1/2i 2 2/3 2 ,
2 (0, 2, 1) = 1/3 2,
2 (0, 3, 1) = 2/3 2 + 1/2i 2 ,
3 (0, 1, 1) = 3/2i 3 12 (3) + 1/2 3 ,
3 (0, 2, 1) = 12(3),
3 (0, 3, 1) = 1/2 3 12 (3) + 3/2i 3.
(B.8)
253
References
[1] A. Klemm, E. Zaslow, Local mirror symmetry at higher genus, hep-th/9906046.
[2] R.P. Horja, Hypergeometric functions and mirror symmetry in toric varieties,
math.AG/9912109.
[3] M. Gromov, Pseudo-holomorphic curves on almost complex manifolds, Inv. Math. 82 (1985)
307347.
[4] E. Witten, Topological sigma models, Commun. Math. Phys 118 (1988) 411449.
[5] P. Candelas, X. de la Ossa, Moduli space of CalabiYau manifolds, Nucl. Phys. B 355 (1991)
455481.
[6] A. Ashtekar, C. Rovelli, L. Smolin, Weaving a classical geometry with quantum threads, Phys.
Rev. Lett. 69 (1992) 237240, hep-th/9203079.
[7] N. Hitchin, Lectures on special Lagrangian submanifolds, math.DG/9907034.
[8] M. Gross, Topological mirror symmetry, math.AG/9909015;
M. Gross, Special Lagrangian fibrations I: topology, alg-geom/9710006;
M. Gross, Special Lagrangian fibrations II: geometry, math.AG/9809072.
[9] D. Joyce, On counting special Lagrangian homology 3-spheres, hep-th/9907013.
[10] S. Kachru, J. McGreevy, Supersymmetric three-cycles and (super)symmetry breaking, Phys.
Rev. D 61 (2000) 026001, hep-th/9908135.
[11] R. Minasian, G. Moore, K-theory and RamondRamond charge, JHEP 9711 (1997) 002, hepth/9710230;
Y.-K.E. Cheung, Z. Yin, Anomalies, branes, and currents, Nucl. Phys. B 517 (1998) 6991,
hep-th/9710206;
E. Witten, D-branes and K-theory, JHEP 9812 (1998) 019, hep-th/9810188;
G. Moore, E. Witten, Self-duality, RamondRamond fields, and K-theory, hep-th/9912279.
[12] D.R. Morrison, Mirror symmetry and the type II string, Nucl. Phys. Proc. Suppl. 46 (1996)
146155, hep-th/9512016.
[13] A. Strominger, Massless black holes and conifolds in string theory, Nucl. Phys. B 451 (1995)
96108, hep-th/9504090.
[14] J. Polchinski, A. Strominger, New vacua for type II string theory, Phys. Lett. B 388 (1996)
736742, hep-th/9510227.
[15] A. Strominger, S.-T. Yau, E. Zaslow, Mirror symmetry is T-duality, Nucl. Phys. B 479 (1996)
243259, hep-th/9606040.
[16] C. Vafa, Extending mirror conjecture to CalabiYau with bundles, hep-th/9804131.
[17] M. Kontsevich, Homological algebra of mirror symmetry, in: Proc. of the Int. Congress of
Mathematicians, Zurich, Birkhauser, 1994, pp. 120139, alg-geom/9411018.
[18] A. Ceresole, R. DAuria, S. Ferrara, A. Van Proeyen, Duality transformations in supersymmetric YangMills theories coupled to supergravity, Nucl. Phys. B 444 (1995) 92, hep-th/9502072.
[19] J.A. Harvey, G. Moore, On the algebras of BPS states, Commun. Math. Phys. 197 (1998) 489
519, hep-th/9609017.
[20] D.-E. Diaconescu, J. Gomis, Fractional branes and boundary states in orbifold theories,
HEP 0010 (2000) 001.
[21] G. Moore, Attractors and arithmetic, hep-th/9807056;
G. Moore, Arithmetic and attractors, hep-th/9807087.
[22] B.R. Greene, D.R. Morrison, M.R. Plesser, Mirror manifolds in higher dimension, Commun.
Math. Phys. 173 (1995) 559598.
[23] P. Deligne, Theorie de Hodge II, Publ. IHES 40 (1972);
P. Deligne, Theorie de Hodge III, Publ. IHES 44 (1974).
[24] 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.
254
[25] P. Griffiths (Ed.), Topics in transcendental algebraic geometry, Ann. of Math. Stud., Vol. 106,
Princeton Univ. Press, 1984;
W. Schmidt, Variation of Hodge structure: the singularities of the period mapping, Inv. Math. 22
(1973) 211319.
[26] A. Landman, On the PicardLefschetz transformations, Trans. Amer. Math. Soc. 181 (1973)
89126.
[27] D. McDuff, D. Salamon, J-holomorphic Curves and Quantum Cohomology, University Lecture
Ser., Vol. 6, AMS, Providence, 1994;
W. Fulton, R. Pandharipande, Notes on stable maps and quantum cohomology, Algebraic
geometrySanta Cruz, 1995, Proc. Symp. Pure Math., Vol. 62, Part 2, AMS, Providence, 1997,
pp. 4596.
[28] T. Oda, H.S. Park, Linear Gale transforms and GelfandKapranovZelevinski decompositions,
Tohoku Math. J. (2) 43 (1991) 375399.
[29] P.S. Aspinwall, B.R. Greene, D.R. Morrison, The monomial-divisor mirror map, Internat. Math.
Res. Notices (1993) 319337, alg-geom/9309007.
[30] D.R. Morrison, Mirror symmetry and rational curves on quintic threefolds: a guide for
mathematicians, J. Amer. Math. Soc. 6 (1993) 223247, alg-geom/9202004.
[31] W. Fulton, Introduction to toric varieties, Ann. Math. Stud., Vol. 131, Princeton Univ. Press,
1993;
T. Oda, Convex bodies and algebraic geometry: an introduction to the theory of toric varieties,
Ergeb. Math. Grenzgeb. (3) 15 (1988);
D.A. Cox, Recent developments in toric geometry, alg-geom/9606016.
[32] P. Aspinwall, Resolution of orbifold singularities in string theory, in: Mirror Symmetry II,
B. Greene, S.-T. Yau (Eds.), hep-th/9403123.
[33] P.S. Aspinwall, B.R. Greene, D.R. Morrison, CalabiYau moduli space, mirror manifolds and
spacetime topology change in string theory, Nucl. Phys. B 416 (1994) 414480;
P.S. Aspinwall, B.R. Greene, D.R. Morrison, Multiple mirror manifolds and topology change
in string theory, Phys. Lett. B 303 (1993) 249259;
E. Witten, Phases of N = 2 theories in two dimensions, Nucl. Phys. B 403 (1993) 159.
[34] P.S. Aspinwall, B.R. Greene, D.R. Morrison, Measuring small distances in N = 2 sigma
models, Nucl. Phys. B 420 (1994) 184242, hep-th/9311042.
[35] P. Aspinwall, Minimum distances in non-trivial string target spaces, Nucl. Phys. B 431 (1994)
7896, hep-th/9404060.
[36] B.R. Greene, Y. Kanter, Small volumes in compactified string theory, Nucl. Phys. B 497 (1997)
127145, hep-th/9612181.
[37] P. Candelas, X. de la Ossa, A. Font, S. Katz, D.R. Morrison, Mirror symmetry for two parameter
models I, Nucl. Phys. B 416 (1994) 481538, hep-th/9308083.
[38] P. Candelas, X. de la Ossa, A. Font, S. Katz, D.R. Morrison, Mirror symmetry for two parameter
models II, Nucl. Phys. B 429 (1994) 626674, hep-th/9403187.
[39] S. Hosono, A. Klemm, S. Theisen, S.-T. Yau, Mirror symmetry, mirror map and applications to
CalabiYau fypersurfaces, Commun. Math. Phys. 167 (1995) 301350, hep-th/9308122;
S. Hosono, A. Klemm, S. Theisen, S.-T. Yau, Mirror symmetry, mirror map and applications to
complete intersection CalabiYau spaces, Nucl. Phys. B 433 (1995) 501554, hep-th/9406055.
[40] P. Berglund, P. Candelas, X. de la Ossa, A. Font, T. Hubsch, D. Jancic, F. Quevedo, Periods for
CalabiYau and LandauGinzburg vacua, Nucl. Phys. B 419 (1994) 352403.
[41] A. Font, Periods and duality symmetries in CalabiYau compactifications, Nucl. Phys. B 391
(1993) 358, hep-th/9203084.
[42] P.S. Aspinwall, Enhanced gauge symmetries and K3 surfaces, Phys. Lett. B 357 (1995) 329
334, hep-th/9507012.
[43] P. Candelas, X.C. De La Ossa, P.S. Green, P. Parkes, A pair of CalabiYau manifolds as an
exactly soluble superconformal theory, Nucl. Phys. B 359 (1991) 21.
255
[44] H. Ooguri, Y. Oz, Z. Yin, D-branes on CalabiYau spaces and their mirrors, Nucl. Phys. B 477
(1996) 407430, hep-th/9606112.
[45] K. Becker, M. Becker, A. Strominger, Fivebranes, membranes and non-perturbative string
theory, Nucl. Phys. B 456 (1995) 130152.
[46] I. Brunner, M.R. Douglas, A. Lawrence, C. Romelsberger, D-branes on the quintic, hepth/9906200.
[47] P. Deligne, Equations differentielles a points singuliers reguliers, Lecture Notes in Math.,
Vol. 163, Springer, Berlin, 1970.
[48] P.S. Aspinwall, C.A. Lutken, Quantum algebraic geometry of superstring compactifications,
Nucl. Phys. B 355 (1991) 482510.
[49] A. Recknagel, V. Schomerus, Moduli spaces of D-branes in CFT-backgrounds, hep-th/9903139;
A. Recknagel, V. Schomerus, Boundary deformation theory and moduli spaces of D-branes,
Nucl. Phys. B 545 (1999) 233282, hep-th/9811237;
A. Recknagel, V. Schomerus, D-branes in Gepner models, Nucl. Phys. B 531 (1998) 185225,
hep-th/9712186.
[50] N. Ishibashi, The boundary and crosscap states in conformal field theories, Mod. Phys. Lett.
A 4 (1989) 251;
N. Ishibashi, T. Onogi, Conformal field theories on surfaces with boundaries and crosscaps,
Mod. Phys. Lett. A 4 (1989) 161.
[51] A. Libgober, J. Teitelbaum, Lines on CalabiYau complete intersections, mirror symmetry, and
PicardFuchs equations, Internat. Math. Res. Notices 1 (1993) 2939.
[52] O.I. Marichev, Handbook of integral transforms of higher transcendental functions: theory and
algorithmic tables, Ellis Horwood Ser. Math. Appl., Halsted Press, New York, 1983;
A. Erdelyi, W. Magnus, F. Oberhettinger, F.G. Tricomi, Higher transcendental functions;
Y. Luke, The special functions and their approximations, Academic Press, 1969.
[53] E.A. Coddington, N. Levinson, Theory of ordinary differential equations, McGraw-Hill, New
York, 1955.
[54] D.R. Morrison, PicardFuchs equations and mirror maps for hypersurfaces, Essays on mirror
manifolds, Internat. Press, Hong Kong, 1992, pp. 241264, hep-th/9111025.
[55] N.E. Norlund, Hypergeometric functions, Acta Math. 4 (1955) 289349.
[56] V.V. Batyrev, Dual polyhedra and mirror symmetry for CalabiYau hypersurfaces in toric
varieties, J. Alg. Geom. 3 (1994) 493535.
[57] B.R. Greene, M.R. Plesser, Duality in CalabiYau moduli space, Nucl. Phys. B 338 (1990) 15.
[58] V.V. Batyrev, D. van Straten, Generalized hypergeometric functions and rational curves on
CalabiYau complete intersections in toric varieties, Comm. Math. Phys. 168 (3) (1995) 493
533.
[59] I. Gelfand, M. Kapranov, A. Zelevinski, Discriminants, Resultants and Multidimensional
Determinants, Birkhauser, Boston, 1994.
[60] I.M. Gelfand, A.V. Zelevinskii, M.M. Kapranov, Discriminants of polynomials in several
variables and triangulations of Newton polyhedra, Leningrad Math. J. 2 (1991) 449505.
[61] D.R. Morrison, Mathematical aspects of Mirror Symmetry, in: J. Kollar (Ed.), Complex Algebraic Geometry, IAS/Park City Math. Series, Vol. 3, 1997, pp. 265340, alg-geom/9609021.
[62] D.R. Morrison, Compactifications of moduli spaces inspired by mirror symmetry, J. Geom.
Alg. Orsay 218 (1993) 243271, alg-geom/9304007.

Universal hypermultiplet metrics

Sergei V. Ketov 1
Department of Theoretical Physics, Erwin Schrdinger Strasse, University of Kaiserslautern,
67653 Kaiserslautern, Germany
Abstract
Some instanton corrections to the universal hypermultiplet moduli space metric of the type
IIA string theory compactified on a CalabiYau threefold arise due to multiple wrapping of
BPS membranes and five-branes around certain cycles of CalabiYau. The classical universal
hypermultipet metric is locally equivalent to the Bergmann metric of the symmetric quaternionic
space SU(2, 1)/U (2), whereas its generic quaternionic deformations are governed by the integrable
SU() Toda equation. We calculate the exact (non-perturbative) UH metrics in the special cases of
(i) the D-instantons (the wrapped D2-branes) in the absence of five-branes, and (ii) the five-brane
instantons with vanishing charges, in the absence of D-instantons. The solutions of the first type
preserve the U (1) U (1) classical symmetry, while they can be interpreted as the gravitational
dressing of the hyper-Khler D-instanton solutions. The solutions of the second type preserve the
non-abelian SU(2) classical symmetry, while they can be interpreted as the gradient flows in the
universal hypermultiplet moduli space. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
Non-perturbative (instanton) quantum corrections in compactified M-theory/strings are
believed to be crucial for solving the fundamental problems of vacuum degeneracy and
supersymmetry breaking [1]. Some instanton corrections to various physical quantities in
the effective lower-dimensional supergravity theory, originating from the string/M-theory
compactification on a CalabiYau (CY) threefold Y, can be understood in terms of the
Euclidean five-branes wrapped about the entire CY space and the Euclidean membranes
(two-branes) wrapped about special (supersymmetric) three-cycles C of Y [2,3]. 2 Being
E-mail address: ketov@physik.uni-kl.de (S.V. Ketov).
1 Supported in part by the Deutsche Forschungsgemeinschaft.
2 The supersymmetric 3-cycle C is a three-dimensional submanifold of the CY space, which obeys J | =
C
Im()|C = 0, where J is the Khler form and is the holomorphic (3, 0) form in Y. The supersymmetric
cycles minimize volume in their homology class [4], while the corresponding wrapped brane configurations lead
to the BPS states.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 8 4 - 5
S.V. Ketov / Nuclear Physics B 604 (2001) 256280
257
solitonic (BPS) classical solutions to the higher-dimensional (Euclidean) equations of

motion with non-vanishing topological charges, these wrapped branes are localized in the
uncompactified dimensions and thus can be identified with the instantons. The instanton
actions are essentially given by the volumes of the cycles on which the branes are wrapped.
For instance, the compactification of the type IIA superstring theory on Y gives rise to
the four-dimensional (4d) N = 2 superstrings whose low energy effective action (LEEA)
is given by the 4d, N = 2 supergravity coupled to N = 2 vector supermultiplets and
hypermultiplets. Any other N = 2 matter multiplet (of physical spin 1 or 1/2) can
be dualized to an N = 2 vector multiplet or a hypermultiplet, respectively. The numbers
of the N = 2 matter vector multiplets and hypermultiplets are dictated by the topological
data about Y, namely, the Hodge numbers h1,1 and h1,2 + 1, respectively, where hp,q are
the dimensions of the Dolbeaux cohomology groups of Y [1]. The hypermultiplet LEEA
is most naturally described by the non-linear sigma model (NLSM), whose scalar fields
parametrize the quaternionic target space MH [5]. The instanton corrections to the LEEA
due to the wrapped five-branes and membranes can be easily identified and distinguished
from each other in the semi-classical limit, since the five-brane instanton corrections are
2
1/gstring
, whereas the membrane instanton corrections are given
organized by powers of e
1/g
string
, where gstring is the type IIA superstring coupling constant [6].
by powers of e
From the M-theory perspective, the compactified field theory (LEEA) is given by
the five-dimensional supergravity with the same number of unbroken supercharges. The
effective supergravities in four and five dimensions are related via the compacification
of the latter on a circle, which does not affect the NLSM target space MH in the
hypermultiplet sector. The vacuum expectation value of the four-dimensional dilaton field
in the compacitified type IIA superstring is simply related to the CY volume VCY in
M-theory, VCY = e2 , so that the type IIA superstring loop expansion amounts to the
derivative expansion of the M-theory action [7].
The unbroken local supersymmetry with eight supercharges (e.g., N = 2 in 4d) put
severe constraints on the LEEA of matter supermultiplets or, equivalently, on their moduli
space. In fact, it requires the whole moduli space M be the local product, M = MV
MH , where MV is the special Khler manifold associated with the N = 2 vector
multiplets [8], and MH is the quaternionic manifold associated with the hypermultiplets
[5]. 3 Any CY compactification has the so-called universal hypermultiplet (UH) containing
a dilaton, an axion, a complex RR-type pseudo-scalar and a Dirac dilatino [9]. The
constraints imposed by 4d, N = 2 local supersymmetry on the four-dimensional UH
moduli space are even stronger then in higher dimensions: the target space of the universal
hypermultiplet NLSM has to be an Einstein space with the (anti)self-dual Weyl tensor
(these spaces are called the self-dual Einstein spaces or simply the EinsteinWeyl spaces
in the mathematical literature).
The N = 2 vector multiplet moduli space MV is relatively well understood, whereas
much less is known about the hypermultiplet moduli space MH [10]. The conjectured
duality between the type IIA superstring compactification on CY and the heterotic string
3 See, e.g., Ref. [10] for a review.
258
compactification on K3 T 2 , which is supposed to exchange MV and MH in the same

moduli space M, was used in the past to get information about MH on the type II
string side from the knowledge about MV on the heterotic string side, although with only
partial success [11]. Since the UH transforms into the N = 2 supergravity multiplet under
type II mirror symmetry, any quantum corrections to the classical UH metric are essentially
gravitational in nature [7].
The standard instanton calculus gives us another technical device after taking into
account the partial supersymmetry breaking induced by the BPS branes [2]. Being
applied to the bosonic instanton background, broken supersymmetries generate fermionic
(goldstino) zero modes that are to be absorbed by extra terms in the effective field
theory. These instanton-induced interactions are quartic in the fermionic fields (by index
theorems) and thus contribute to the curvature tensor of the supersymmetric NLSM. By
supersymmetry, the instanton corrections to the NLSM curvature imply the corresponding
deformations in the NLSM metric. To calculate these non-perturbative corrections by using
the standard instanton technology, one may integrate over the fermionic zero modes and
then compute the fluctuation determinants, which is apparently the hard problem [12].
In this paper we restrict ourselves to a calculation of the instanton corrections to the
universal hypermultiplet NLSM metric, by analyzing generic quaternionic deformations
of the classical UH metric. We apply the most direct procedure based on the fact that the
(anti-)self-dual Weyl tensor already implies the integrable system of partial differential
equations on the components of the UH moduli space metric. Additional simplifications
arise due to the Einstein condition and the physically motivated isometries. The UH metric
in question is supposed to be regular and complete.
The local data about the underlying CY threefold Y is described in terms of the hp,q
complex moduli. In the compactified theory, the h1,2 moduli are promoted to the nonuniversal matter hypermultiplets. The universal hypermultiplet is universal in the sense
that its own moduli space does not depend upon the CY moduli, or, equivalently, the UH
is merely sensitive to the global information about the CY space (like topology, volume,
charges and symmetries).
Some instanton solutions to the effective Euclidean (UH + supergravity) equations
of motion and the corresponding instanton actions were found in Refs. [3,13,14]. The
instanton solutions carry charges descending from the wrapped brane charges, and they
preserve a part (half) of type IIA supersymmetry. The instanton actions are also dependent
upon the charges of the instanton and the complex structure (or central charge) at the
moduli space infinity. Unfortunately, no exact quaternionic metrics in the UH moduli space
beyond string perturbation theory were constructed, which would amount to a calculation
of the exact LEEA of the universal hypermultiplet. In this paper we report about the results
of our investigation initiated in Ref. [15].
The paper is organized as follows. In Section 2 we review the four-dimensional classical
LEEA (or NLSM) of the universal hypermultiplet in the background of 4d, N = 2
supergravity [16], and discuss its perturbative deformations [7,17] and the origin of
non-perturbative corrections due to the wrapped BPS branes [2,3,13,14]. The symmetry
structure of the classical UH moduli space is given in Section 3. The D-instanton
259
corrections due to the wrapped membranes (D2-branes) in the hyper-Khler limit [18]
are presented in Section 4, whereas their quaternionic generalizations are described in
Section 5. In Section 6 we demonstrate that the zero-charge five-brane instanton corrections
to the UH metric are described by the TodHitchin metric. Its physical interpretation
as the gradient flow in the UH moduli space is also given in Section 6. Generic UH
metrics are briefly discussed in Section 7. In Appendix A we collect basic facts about the
quaternionic symmetric space SU(2, 1)/U (2). Appendix B is devoted to basic definitions
of theta functions and some useful identities between them.
2. M-theory, type IIA string perturbation theory, and the UH classical moduli space
metric
The eleven-dimensional (11d) M-theory is supposed to substitute the ten-dimensional
(10d) type IIA superstring theory at strong string coupling, while the M-theory LEEA
is described by 11d supergravity [20]. The action of 11d supergravity is unique, and its
bosonic part reads [21] 4

1
1
11
1 2

4 F
4 ,
S11 = 2
(2.1)
d x g R 48 F +
A 3 F
2
211
1211
4 = d A 3 , and all quantities with hats refer
where 11 is the 11d gravitational constant, F
to 11d. Though the 11d supergravity action is only the leading (LEEA) part of the full
(unknown) M-theory action, it is already enough to study the BPS states in M-theory. The
latter are given by the solitonic classical solutions (called M-branes) to 11d supergravity,
which preserve some part (say, a half) of 11d supersymmetry. The BPS condition amounts
to the existence of a MajoranaKilling 11d spinor (x) obeying the first-order linear
differential equation
M
D

BC
M + 14 M
BC
1
288
NP QS

N P QS
M
FNP QS = 0,
8M
(2.2)
BC is the 11d supergravity spin connection, and s are

where M = 0, 1, . . . , 10, M
the antisymmetric products (with unit weight) of 11d Dirac matrices. The M-theory
supersymmetry algebra is just the most general super-Poincar algebra in 11d [22],

{Q , Q } = C M PM +
1
2!
C MN
ZMN +
1
5!

C MNP QS YMNP QS ,
(2.3)
where Q , = 1, . . . , 32, are the 11d supersymmetry charges, PM are 11d translations, C
is the 11d charge conjugation matrix, whereas ZMN and YMNP QS stand for the electric
and magnetic charges, respectively,
4 The 11d signature is (, +, +, . . . , +).
260

ZMN =
dx M dx N ,
M2-brane
YMNP QS =
dx M dx N dx P dx Q dx S .
(2.4)
M5-brane
Eq. (2.3) is just the 11d Fierz rearrangement formula for {Q , Q } 528 of Spin(32). The
M-theory solitons are thus given by membranes [23] and five-branes [24], which are in fact
dual to each other under the electricmagnetic duality in 11d.
The physical and topological significance of the abelian (electric and magnetic)
charges in Eq. (2.3) become clear from the 11d equations of motion, which follow from
the action (2.1). In particular, one has

1
4 + 1 F
2

d 11 F
(2.5)
2 4 = d 11 F 4 + 2 A3 F4 = 0,
4 . Hence, the electric charge of an M2-brane [25],
is the 11d dual to F
where 11 F
4

1
1

Qe =
(2.6)
11 F 4 + 2 A3 F4 ,
11 2
S7
is conserved, where S 7 is the asymptotic seven-sphere surrounding the M2-brane (cf. Gauss
law in Maxwell electrodynamics). Similarly, the magnetic charge of the M5-brane [25],

1
4 , d F
4 = 0,
F
Qm =
(2.7)
11 2
S4
is also conserved, where S 4 is the asymptotic four-sphere surrounding the M5-brane (cf.
Dirac monopole charge). The Dirac quantization condition implies
Qe Qm = 2(h c)Z.
(2.8)
Demanding the membrane and five-brane (worldvolume) actions be well defined in

quantum theory gives rise to a quantization of their electric and magnetic charges or,
equivalently, the M-brane tension quantization [26,27],
2T5 = T22
2
and 11
T2 T5 = Z.
A connection to type IIA superstrings

is obtained by compactifying M-theory/11d

supergravity on a circle of radius r = with the KaluzaKlein (KK) ansatz [20]

2
2
2
= e2/3 d s10
+ e4/3 dx 11 Am dx m ,
d s11
(2.9)
2 = g
m
n
where d s10
mn dx dx is the 10d spacetime metric (in the string frame) with m, n =
0, 1, . . . , 9, Am is the KK (type IIA) vector field, and is
the 10d dilaton. The gravitational
2 = 2 /(2 ). Performing the dimensional
constants in 11d and 10d are related by 10
11
reduction of the action (2.1) on the circle with the help of Eq. (2.9) gives rise to the
bosonic part of the type IIA supergravity action in the string frame. The latter is related
to the canonical form of the type IIA supergravity action (in the Einstein frame) via the
261
Weyl rescaling gmn = e/2 gmn . This yields the standard bosonic part of the 10d type IIA
supergravity action,

1
S10 = 2
d 10 x g R 12 (m )2
210

/2
1
1
2
e F4 10 F 4 + e H3 10 H 3 + 2
B2 F4 F4 ,
410
410
(2.10)
4
where H3 = dB2 and F4 = dA3 originate from the three-form A 3 and its field strength F
11
in 11d, e.g., A3 = A3 + B2 dx , etc.

The 10d type IIA supergravity is known to be the LEEA of 10d type IIA strings in the
(string) tree approximation [1]. The double dimensional reduction of the 11d (M-theory)
membrane yields a (type IIA) fundamental string in 10d. The fundamental string in 10d
is dual to a solitonic (NSNS) five-brane. In M-theory compactified on the circle S 1 the
11d five-brane can be wrapped about S 1 , which gives rise to the (RR-charged) type IIA
four-brane called D4-brane [22]. In 10d this D4-brane is dual to the 10d (RR-charged)
membrane called D2-brane [22].
The non-perturbative (instanton) corrections to the four-dimensional UH originate from
the 10d type IIA membranes (D2-branes) wrapped about certain (supersymmetric or
special Lagrangian) 3-cycles C of CY, and the 10d type IIA NSNS five-branes wrapped
about the entire CY space [2,3,13,14]. From the 11d M-theory perspective, the dilaton
expectation value is related to the volume of the CY space (Section 1), whereas the
expectation value of the RR scalar C is related to the CY period C .

The universal sector (UH) of the CY compactification of the 10d type IIA supergravity
in four dimensions is most easily obtained by using the following ansatz for the 10d
metric [3]:
2
2
ds10
= gmn dx m dx n = e/2 dsCY
+ e3/2 g dx dx ,
(2.11)
while keeping only SU(3) singlets in the internal CY indices [19] and ignoring all CY
complex moduli. In Eq. (2.11) (x) stands for the 4d dilaton, g (x) is the spacetime
2 is the (Khler and
metric in four uncompactified dimensions, , = 0, 1, 2, 3, and dsCY
Ricci-flat) metric of the internal CY threefold Y in complex coordinates,
2
= gi j (y, y)
dy i d y j ,
dsCY
(2.12)
where i, j = 1, 2, 3. By definition, the CY threefold Y possesses the (1, 1) Khler form J

and the holomorphic (3, 0) form . The universal hypermultipet (UH) unites the dilaton ,
the axion D coming from dualizing the three-form field strength H3 = dB2 of the NSNS
two-form B2
in 4d, and the complex scalar C representing the RR three-form A3 with
Aij k (x, y)= 2 C(x)ij k (y). When using a flat (or rigid) CY (cf. a six-torus of equal
radii r = ), with
gi j = i j
and ij k = ij k ,
(2.13)
262
the reduction of Eq. (2.10) down to four dimensions with the help of Eqs. (2.12) and (2.13)
yields the so-called FerraraSabharwal action [16]: 5

1
S4 = 2 d 4 x g R 2(m )2 2e2 | C|2
24

4

1
d C 4H3 dD ,
e
H3 4 H 3 2iH3 C
2
(2.14)
44
where the real 4d Lagrange multiplier D has been introduced to enforce the Bianchi
relation dH3 = 0. Removing H3 via its equations of motion from Eq. (2.14) results in
the dual action [3]

1
C 2 .
S4 = 2 d 4 x g 12 R + (m )2 + e2 | C|2 + e4 D + 2i C
4
(2.15)
The equivalent action in the string frame reads [17]

1
2
+ 2( )2 1 H
S4, string = 2 d 4 x g e2 12 R
6
4

1
+ i H (C C
C
C) ,
2 d 4 x g C C
2
4
(2.16)
where H = 61g H is the Hodge dual of the field strength H3 = dB2 . The N =
2 supergravity background in Eqs. (2.14), (2.15) and (2.16) is merely represented by the
4d spacetime metric. The NLSM action of the UH thus has the following target space
metric [16]:

2
d C 2,
= d 2 + e2 |dC|2 + e4 dD + C
42 dsclassical
(2.17)
whose right-hand side is regular and positively definite. This metric is of purely
gravitational origin, with 42 being the only coupling constant.
The perturbative (one-loop) string corrections to the UH metric (2.17) originate from the
(Riemann)4 terms [28] in M-theory compactified on a CY three-fold Y [7]. These quantum
corrections are proportional to the CY Euler number = 2(h1,1 h1,2 ) [29]. The one-loop
corrected action of UH coupled to gravity in the string frame reads [17]

2e4
1
2
+ 2
( )2
S4, one-loop = 2 d 4 x g e2 + 12 R 16 H
e
+
4

1
+ i H (C C
C
C) ,
2 d 4 x g C C
(2.18)
2
4
where is proportional to . 6 The superstring (loop) perturbation theory is controlled by
the powers of e2 , whereas the second line of Eq. (2.18) cannot be multiplied by any
5 The 4d spacetime signature is (, +, +, +). The 4d gravitational constant is related to the 10d gravitational
4
2 /(2 r)6 = 2 /(2 )6 .

constant 10 via the equation 42 = 10
10
6 The relative coefficient varies in the literature, see, e.g., Ref. [7].
263
power of e2 without breaking the perturbative PecceiQuinn type symmetry C C +

const. This non-renormalization argument does not, however, exclude the non-perturbative
(gravitational-type) corrections due to the wrapped branes, which break the PecceiQuinn
type symmetries (Sections 4 and 5).
The string one-loop corrected NLSM metric dictated by Eq. (2.18) is related to the
classical metric of Eq. (2.17) by a local field redefinition, e2 e2 + , so that the
local UH geometry is not affected by perturbative string corrections [7]. String duality,
however, implies some discrete (global) identifications in the UH moduli space, which
are the consequence of the brane charge (and tension) quantization in M-theory [3] (see
Section 3).
3. Classical metric and perturbative deformations

The NLSM of UH with the target space metric (2.17) can be rewritten to the form [2,3]
LNLSM =

1

K,S
+ K,C
S C
C C
S S S + K,S C
S C S + K,C C
2
4
(3.1)
in terms of two complex variables, C and S,

S = e2 + 2iD + C C,
(3.2)
where commas in subscripts denote partial derivatives, and the Khler potential reads

= log 2e2 .
K = log(S +
S 2C C)
(3.3)
Eq. (3.1) makes manifest the Khler nature of the classical NLSM metric,

2
2C
d
.
dsNLSM
= e2K dS d
S 2C dS d C
S dC + 2(S +
S) dC d C
(3.4)
Other useful parametrizations are given in Appendix A. In particular, after the change of
variables (A.9) or its inverse,
S=
1 z1
,
1 + z1
C=
z2
,
1 + z1
(3.5)
and the Khler gauge transformation (A.4) with f (z) = log[ 12 (1 +z1 )], the Khler potential
(3.3) takes the form (A.3) that is associated with the standard (Bergmann) metric (A.1)
of the non-compact symmetric space SU(2, 1)/SU(2) U (1). When using the new
coordinates (A.5), one arrives at the classical UH metric in the form (A.6), with the SU(2)
isometry of the metric being manifest.
The classical quaternionic space Q = G/H SU(2, 1)/SU(2) U (1) has the eightdimensional non-abelian isometry group G = SU(2, 1) whose left action on Q after the
compensating right action of the gauge subgroup H = SU(2) U (1) leaves the metric
(A.1) of Q intact.
264
As far as quantum corrections within the type IIA string (loop) perturbation theory
are concerned, the UH metric is supposed to be invariant under the so-called Peccei
Quinn type symmetries [2,7]. These symmetries are given by constant shifts of the NSNS
axion D,
S S + i,
(3.6)
and constant shifts of the RR field C [30],

C C + i,
S S + 2( + i)C + 2 + 2 ,
(3.7)
where , and are the real parameters. The PecceiQuinn type transformations (3.6)
and (3.7) form the non-abelian Heisenberg group [3]. It is usually assumed [2,3,7] that
the PecceiQuinn type symmetries do not change their form in string perturbation theory.
Since our considerations are purely local, we ignore the global group structure and refer to
the symmetry (3.6) as UD (1).
The instantons originating from the five-branes wrapped over the entire CY space break
both symmetries (3.6) and (3.7) (cf. the breaking of the translational invariance of the parameter in QCD by instantons), whereas the D-instantons (wrapped membranes) break
the symmetry (3.7) but keep the symmetry (3.6) [18].
The U (1) subgroup of the SU(2) H symmetry is given by the duality rotations UC (1)
of the complex RR pseudo-scalar C,
C ei C,
S S,
(3.8)
with the real parameter . The duality rotations (3.8) are believed to be exact in quantum
theory, even when all quantum (or instanton) corrections are taken into account [2,3].
The remaining four classical symmetries of the coset SU(2, 1)/U (2) are given by scale
transformations,
S S + S,
C C + 12 C,
(3.9)
with the real parameter , and [30]

S S + 4i >1 S 2 + 14 (>2 + i>3 )CS,

C C + 4i >1 CS + 12 >2 C 2 12 S + 2i >3 C 2 + 12 S ,
(3.10)
with the real parameters >1 , >2 and >3 . The symmetries (3.9) and (3.10) are always broken
by instanton corrections.
The conserved Noether charges associated with the isometries (3.6) and (3.7) were
calculated in Ref. [3],
i 2K

e (dS d
S + 2C d C),
2
4
2i
+ 2(C + C)J
,
J = 2 eK (dC d C)
4
2
2i(C C)J
.
J = 2 eK (dC + d C)
4
J =
(3.11)
The Noether charge (over a supersymmetric three-cycle C)

Q = 4 J
265
(3.12)
descends from the five-brane charges (2.4), whereas the other two Noether charges,

Q = 4 J and Q = 4 J ,
(3.13)
3
descend from the D2-brane (RR) charges. The existence of two charges is related to
the existence of two homology classes of a three-cycle C [3]. The BPS brane charge
quantization (Section 2) implies that only a discrete subgroup of the continuous Peccei
Quinn type symmetries (3.6) and (3.7) is going to survive in full quantum theory after
taking into account both membrane and five-brane instanton corrections [3]. The discrete
identifications of the UH scalars can be read off from Eqs. (3.6) and (3.7),
S S + in + 2(n + in )C + n2 + n2 ,
C C + n in ,
(3.14)
where now all n,, are integers [3]. The transformations (3.14) define a discrete nonabelian group Z. The global (topological) structure of the UH moduli space M is thus
given by Q/Z [3].
Since the type IIA string loop corrections are invariant under the perturbative Peccei
Quinn type symmetries (3.6) and (3.7), it is natural to rewrite the UH metric in terms of
the new quantities (u, v, ) defined by Eq. (A.14), which are all invariant under (3.6) and
(3.7). When using the notation (A.14), it is not difficult to convince oneself that the unique
quaternionic deformation of the classical UH action within the string perturbation theory
is described by Eq. (2.11) indeed [7]. Hence, the local UH metric (2.17) does not receive
any perturbative quaternionic corrections modulo the UH field redefinitions (Section 2).
4. Hyper-Khler versus quaternionic geometry

The D-instanton corrections to the UH moduli space metric due to the wrapped D2branes were explicitly calculated in the hyper-Khler limit by Ooguri and Vafa [18]. In this
limit the 4d, N = 2 supergravity decouples [5], and the c-map applies [9].
In the absence of five-brane corrections, the unbroken symmetry of the UH metric is
given by UD (1) UC (1) (Section 3). In adapted coordinates with respect to the UD (1)
isometry the UH metric can be written down in the LeBrun form [31],

dsK2 gab d a d b = W 1 (dt + 1 )2 + W eu dx 2 + dy 2 + d2 ,
(4.1)
where two potentials W and u, and a one-form 1 have been introduced. The metric ansatz
(4.1) is valid for any Khler metric gab in four real dimensions, a, b = 1, 2, 3, 4, with a
Killing vector K a that preserves the Khler structure. We use the adapted coordinates a =
(t, x, y, ) where t is the coordinate along the trajectories of the Killing vector associated
266
with UD (1), whereas (x, y, ) are the coordinates in the space of trajectories, W 1 =
gab K a K b = 0. Accordingly, no metric components are dependent upon t in Eq. (4.1).
The Khler condition on the metric (4.1) also implies a linear equation on 1 [31],

d1 = Wx dy d + Wy d dx + W eu dx dy.
(4.2)
In turn, this gives rise to the following integrability condition on W [31]:

Wxx + Wyy + W eu = 0.
(4.3)
The hyper-Khler geometry implies, by definition, the existence of three linearly independent Khler structures (Jk )a b , k = 1, 2, 3, which are covariantly constant, c (Jk )a b = 0,
and obey a quaternionic algebra. Moreover, in four real dimensions, a hyper-Khler metric
necessarily has the anti-self-dual (ASD) Riemann curvature, and vice versa [32]. As regards four-dimensional hyper-Khler metrics, they are just Khler and Ricci-flat, and vice
versa [33].
If the Killing vector K a is triholomorphic (i.e., it is consistent with N = 2 supersymmetry), one may further restrict the metric (4.1) by taking u = 0. The Riemann ASD condition
then amounts to a linear system [34],
W = 0
+

= 0,
and W
(4.4)
2
where is the Laplace operator in three flat dimensions, = 4gstring
z z + 2 , and z =
gstring(x + iy). The complex coordinate z represents the RR-type complex scalar, so that
the unbroken UC (1) symmetry (3.8) implies that a solution to Eq. (4.4) may depend upon
z only via its absolute value |z|.
One may now think of W as the electro-static potential for a collection of electric charges
distributed in three dimensions near the axis z = 0 with unit density in . The unique
regular (outside the positions of charges) solution to this problem in the limit gstring 0,
while keeping |z|/gstring finite, reads [35]
2

2|mz|
cos(2m)
W=
(4.5)
log
K0
,
+
4
zz
gstring
m=1
where K0 is the modified Bessel function. The solution (4.5) can be trusted for large |z|,
where it amounts to the infinite D-instanton/anti-instanton sum [18],
2

2|mz|
1
log
exp
W=
cos(m)
+
4
zz
gstring
m=1
(n + 12 )
gstring
1
4|mz|
n! (n + 2 )
n=0
n+ 1
(4.6)
The exp (1/gstring) type dependence of the solution (4.6) apparently agrees with the
general expectations [6] that it describes the D-instantons indeed.
If merely the vanishing scalar curvature of the Khler metric (4.6) or a nontriholomorphic isometry of the hyper-Khler metric were required, we would end up with
the non-linear equation [31,34]
uxx + uyy + (eu ) = 0.
267
(4.7)
Eq. (4.7) is known as the SU() Toda field equation [36] since it appears in the large-N
limit of the standard (two-dimensional) Toda system for SU(N). See, e.g., Ref. [37] for
more about the Toda equation (4.7), and Ref. [38] for more about the four-dimensional
hyper-Khler NLSM. The EinsteinKhler deformations of the Bergmann metric of
SU(2, 1)/U (2) in Eq. (A.1) were investigated in Refs. [39,40].
5. D-instantons and quaternionic UH metric

b
A quaternionic manifold admits three independent almost complex structures (Jk )a ,

c
c
which are, however, not covariantly constant but satisfy a (Jk )b = (Ta )k n (Jn )b , where
(Ta )k n is the NLSM torsion [32]. This torsion is induced by 4d, N = 2 supergravity
because the quaternionic condition on the hypermultiplet NLSM target space metric is the
direct consequence of local N = 2 supersymmetry in four spacetime dimensions [5]. As
regards four-dimensional quaternionic manifolds (relevant for UH), they all have Einstein
Weyl geometry of negative scalar curvature [5,32], i.e.,
gab ,
(5.1)
2
where Wabcd is the Weyl tensor and Rab is the Ricci tensor for the metric gab . The overall
coupling constant of the 4d NLSM has the same dimension as 42 , while in the N = 2
locally supersymmetric NLSM these coupling constants are proportional to each other with
the dimensionless coefficient < 0 [5]. We take 42 = 1 for simplicity.
Since the quaternionic and hyper-Khler conditions are not compatible, the canonical
form (4.1) should be revised. 7 Nevertheless, the exact quaternionic metric is governed by
the same three-dimensional Toda equation (4.7) [15]. Indeed, when using another (Tod)
ansatz [41]
= 0,
Wabcd
Rab =

1
P u 2
e dx + dy 2 + d2 +
(dt + 1 )2
(5.2)
2
P 2
for a quaternionic metric with an abelian isometry, it is straightforward to prove that the
restrictions (5.1) on the metric (5.2) precisely amount to Eq. (4.7) on the potential u =
u(x, y, ), while P is given by [41]
dsQ2 =
1
(u 2),
2
whereas the one-form 1 obeys the linear equation [41]

2 2 2
d1 = Px dy d Py d dx eu P + P
P dx dy.

P=
(5.3)
(5.4)
The limit 0, where 4d, N = 2 supergravity decouples, should be taken with care.
After rescaling u u in Eq. (4.7) we get
7 A generic EinsteinWeyl manifold does not have a Khler structure.
268
1 u
e = 0.
(5.5)
This equation gives rise to the 3d Laplace (linear!) equation when 0, as expected.
We can, therefore, conclude that the non-linear Toda equation (4.7) substitutes the linear
Laplace equation (4.4) in the presence of 4d, N = 2 supergravity. The cosmological
constant can be considered as the most relevant parameter of the deformation that
converts a given UH hyper-Khler metric into the gravitationally dressed quaternionic
UH metric via Eq. (5.2), based on the same solution to the Toda equation (4.7) [15].
In terms of the complex coordinate = x + iy, the 3d Toda equation (4.7) takes the form

4u + eu = 0.
(5.6)
uxx + uyy +
It is not difficult to check that this equation is invariant under holomorphic transformations
of ,
= f ( ),
(5.7)
with an arbitrary function f ( ), provided that it is accompanied by the shift of the Toda
potential,
u u = u log(f ) log(f ),
(5.8)
where the prime means differentiation with respect to or , respectively. The transformations (5.7) can be interpreted as the residual diffeomorphisms in the NLSM target space of
the universal hypermultiplet, which keep invariant the quaternionic ansatz (5.2) under the
compensating Toda gauge transformations (5.8).
To make contact with particular four-dimensional hyper-Khler geometries (with
isometries) [38], it is natural to search for separable exact solutions to the Toda equation,
having the form
u(, , ) = F (, ) + G().
(5.9)
Eq. (4.7) now reduces to two separate equations,

F +
c2 F
e =0
2
(5.10)
and
2 eG = 2c2 ,
where c2
(5.11)
is a separation constant. After taking into account the positivity of eG , the general
solution to Eq. (5.11) reads

eG = c2 2 + 2b cos + b2 ,
(5.12)
where b and are arbitrary real integration constants.

Eq. (5.10) is the 2d Liouville equation that is well known in 2d quantum gravity [42]. Its
general solution reads
2
4 f
F
e =
(5.13)
(1 + c2 |f |2 )2
269
in terms of arbitrary holomorphic function f ( ). The ambiguity associated with this

function is, however, precisely compensated by the Toda gauge transformation (5.8), so
that we have the right to choose f ( ) = in Eq. (5.13). This yields the following regular
exact solution to the 3d Toda equation:
eu =
4c2 (2 + 2b cos + b2 )
.
(1 + c2 | |2 )2
(5.14)
It is obvious now that the constant c2 is positive indeed. It also follows from Eqs. (5.2),
(5.3) and (5.14) that any separable exact solution to the quaternionic UH metric possesses
the rigid UC (1) duality symmetry with respect to the duality rotations ei of the
complex RR-field .
Though the quaternionic NLSM metric defined by Eqs. (5.2), (5.3), (5.4) and (5.14)
is apparently different from the classical UH metric (2.17), these metrics are nevertheless
equivalent in the classical region of the UH moduli space where all quantum corrections are
suppressed. The classical approximation corresponds to the conformal limit and
| | , while keeping the ratio | |2/ finite. Then one easily finds that P 1 =
const > 0, whereas the metric (5.2) takes the form
ds 2 =

1
1
|dC|2 + d2 + 4 (dD + )2 ,
2
(5.15)
in terms of the new variables C = 1/ and 2 = , after a few rescalings. The metric (5.15)
reduces to that of Eq. (2.17) when using 2 = e2 . Another interesting limit is 0 and
| | , where one gets a conformally flat metric (AdS4 ).
Based on the fact that both hyper-Khler and quaternionic metrics under consideration
are governed by the same Toda equation, a natural mechanism 8 of generating the
quaternionic metrics from known hyper-Khler metrics in the same (four) dimensions
arises: first, one deduces a solution to the Toda equation (4.7) from a given fourdimensional hyper-Khler metric having a non-triholomorphic or rotational isometry, by
rewriting it to the form (4.1), and then one inserts the obtained exact solution into the
quaternionic ansatz (5.2) to deduce the corresponding quaternionic metric with the same
isometry. Being applied to the D-instantons, this mechanism results in their dressing with
respect to 4d, N = 2 supergravity background [15].
The SU() Toda equation is known to be notoriously difficult to solve, while a very
few its exact solutions are known [37]. Nevertheless, the proposed connection to the
hyper-Khler metrics can be used as the powerful vehicle for generating exact solutions
to Eq. (4.7). It is worth mentioning that Eq. (4.3) follows from Eq. (4.7) after a substitution
W = u,
(5.16)
while Eq. (5.2) is solved by

1 = y u(dx) x u(dy).
(5.17)
8 The different mechanism, which associates the quaternionic metric in 4(n + 1) real dimensions to a given
(special) hyper-Khler metric in 4n real dimensions, was proposed in Ref. [43].
270
This is known as the Toda frame for a hyper-Khler metric [34,44,45]. It is not difficult
to verify that the separable solution (5.14) is generated from the EguchiHanson (hyperKhler) metric along these lines [38,44]. Another highly non-trivial solution to the
Toda equation (4.7) follows from the AtiyahHitchin (hyper-Khler) metric [33,45]. The
transform to the Toda frame for the AtiyahHitchin metric is known [46],

/2
1 k 2 sin2
tanh
d
y + ix = K(k) 1 + k 2 sinh2 cos +
,
K(k)
1 k 2 tanh2 sin2
0

2E(k)
1 2
2
2
2
2
2
,
= K (k) k sin + k 1 + sin sin
(5.18)
8
K(k)
where (, , ; k) are the new coordinates (in
four dimensions). The parameter k plays the
role of modulus here, 0 < k < 1, while k = 1 k 2 is called the complementary modulus.
The remaining definitions are

2
+ i, = 2 + arg 1 + k sinh2 ,
log tan
(5.19)
2
in terms of the standard complete elliptic integrals of the first and second kind, K(k) and
E(k), respectively. The solution to the Toda equation (4.7) now reads [38,44]
eu =

1 2
2
K (k) sin2 1 + k sinh2 .
16
(5.20)
The physical significance of the related quaternionic metric solution is apparent in the
perturbative region k 1, where [47]
k eSinst
and Sinst +.
(5.21)
In this limit the AtiyahHitchin metric is exponentially close to the TaubNUT metric
[33], while the exponentially small corrections can be interpreted as the (mixed) Dinstantons and anti-instantons. The D-instanton action Sinst represents here the volume of
the corresponding supersymmetric three-cycle C, on which the D-brane is wrapped [13]. It
is worth mentioning that the same exact solution also describes the hypermultiplet moduli
space metric in the 3d, N = 4 supersymmetric YangMills theory with the SU(2) gauge
group, which was obtained via the c-map in Ref. [48].
6. Five-brane instantons and TodHitchin metric

As was demonstrated in Ref. [13], the BPS condition on the five-brane instanton solution
with the vanishing charge Q of Eq. (3.12) defines a gradient flow in the hypermultiplet
moduli space. The flow implies the SU(2) isometry of the UH metric since the nondegenerate action of this isometry in the four-dimensional UH moduli space gives rise
to the well defined three-dimensional orbits which can be parametrized by the radial
coordinate to be identified with the flow parameter. In the case of the classical UH moduli
271
space SU(2, 1)/SU(2) U (1) the radial coordinate (r) is defined by Eq. (A.5), while its
relation to the complex (S, C) coordinates is given by
r2 =
|1 S|2 + 4|C|2
,
|1 + S|2
(6.1)
where we have used Eqs. (3.5) and (A.5). It is worth mentioning that the non-abelian
SU(2) symmetry includes the abelian duality rotations (3.8). However, it does not imply
the preservation of the rest of the SU(2, 1)/SU(2) U (1) symmetries (other than SU(2))
including the PecceiQuinn type symmetries (3.6) and (3.7). Hence, the SU(2)-invariant
deformations of the classical UH metric, subject to the quaternionic constraints (5.1),
should describe the zero-charge five-brane instanton corrections.
The action of a five-brane instanton with the vanishing charge may merely depend upon
the complex structure modulus at infinity [13]. Therefore, we expect the corresponding
UH moduli space metric be merely dependent upon the complex structure on the boundary
of the coset SU(2, 1)/SU(2) U (1). The conformal boundary metric for the coset
SU(2, 1)/SU(2) U (1) is known to be degenerate, i.e., it possess a zero eigenvalue. This
happens because the conformal structure, associated with the Bergmann metric in the form
(A.6) inside the unit ball in C2 , does not extend across the boundary since the coefficient
at 22 in Eq. (A.6) decays faster than the coefficients at 12 and 32 . However, the conformal
structure survives in the two-dimensional (2d) subspace annihilated by 2 . The 2d complex
structure has a single real parameter the central charge [42] which should appear
in the UH metric. The exact metric solutions, given below in this section, confirm these
expectations.
Lets consider a generic SU(2)-invariant metric in four Euclidean dimensions. In the
Bianchi IX formalism, where the SU(2) symmetry is manifest, the general ansatz for such
metrics reads (see Appendix A for our notation)
ds 2 = w1 w2 w3 dt 2 +
w2 w3 2 w3 w1 2 w1 w2 2
+
+
,
w1 1
w2 2
w3 3
(6.2)
in terms of the (left)-invariant one-forms i defined by Eq. (A.7), and the radial coordinate t. The metric (6.2) is dependent upon three functions wi (t), i = 1, 2, 3. The most general
SU(2)-invariant metric is given by a quadratic form with respect to i . However, it can
always be chosen in the diagonal form, as in Eq. (6.2), without loss of generality.
The quaternionic constraints on the metric amount to the ASDWeyl equation and the
Einstein equation see Eq. (5.1). The exact SU(2)-invariant solutions to these equations
were found by Tod [49] and Hitchin [50]. The main results of Refs. [49,50] are briefly
described below.
Being applied to the metric (6.2), the ASDWeyl condition gives rise to a system of
ordinary differential equations (ODE) [49],
A 1 = A2 A3 + A1 (A2 + A3 ),
A 2 = A3 A1 + A2 (A3 + A1 ),
A 3 = A1 A2 + A3 (A1 + A2 ),
(6.3)
272
where the dots denote differentiation with respect to t, and the functions Ai (t) are defined
by the auxiliary ODE system,
w 1 = w2 w3 + w1 (A2 + A3 ),
w 2 = w3 w1 + w2 (A3 + A1 ),
w 3 = w1 w2 + w3 (A1 + A2 ).
(6.4)
The ODE system (6.3) is known in the mathematical literature as the classical Halphen
system [51]. The Bergmann metric (A.6) is the simplest solution to Eqs. (6.3) and (6.4)
with Ai = 0 for all i = 1, 2, 3. This follows from comparison of Eqs. (6.3), (6.4) and (A.6).
Note that despite of the fact that all Ai = 0, the Bergmann metric (A.6) is, nevertheless,
non-trivial (or non-flat) since Eq. (A.6) is still a non-trivial solution to Eq. (6.4).
Given a metric solution to the ASDWeyl equations, the Einstein equation of Eq. (5.1)
can be easily satisfied after proper Weyl rescaling of the ASDWeyl metric, because any
local Weyl transformation does not affect the vanishing Weyl tensor. Having obtained an
explicit solution to the Halphen system (6.3), it may be substituted into the ODE system
(6.4). To solve Eq. (6.4), it is convenient to change variables as [49]
1 x
,
w1 =
x(1 x)
2 x
w2 =
,
2
x (1 x)
3 x
w3 =
,
x(1 x)2
(6.5)
where the new variables i (x), i = 1, 2, 3, are constrained by an algebraic condition,

22 + 32 12 = 14 .
(6.6)
The algebraic relation (6.6) reduces the number of the newly introduced functions in
Eq. (6.5) from four to three, as it should. In fact, Eq. (6.6) is also dictated by the
quaternionic nature of the metric [49,50].
In terms of the new variables (6.5), the ODE system (6.4) takes the form [49,50]
2 3
,
x(1 x)
3 1
2 =
,
x
1 2
,
3 =
(6.7)
1x
where the primes denote differentiation with respect to x. It is not difficult to verify that
the algebraic constraint (6.6) is preserved under the flow (6.7), so that the highly non-linear
transformation (6.5) is fully consistent. In terms of the new variables (x, i ), the Einstein
condition of Eq. (5.1) on the metric (6.2) in the form

12
(1 x)22 x32
dx 2
2
2u
ds = e
(6.8)
+
+
+ 2
x(1 x) 12
22
3
1 =
273
amounts to the algebraic relation [49]

96 2e2u

8x1222 32 + 21 2 3 x(12 + 22 ) (1 432 )(22 (1 x)12 )
=
,

2
x1 2 + 23 (22 (1 x)12 )
(6.9)
which yields the Weyl factor u(x) in terms of the functions i (x).
The Halphen system (6.3) has a long history [52]. Perhaps, its most natural (manifestly
integrable) derivation is provided via a reduction of the SL(2, C) anti-self-dual YangMills
equations from four Euclidean dimensions to one [53]. A classification of all possible
reductions is known in terms of the so-called Painlev groups that give rise to six different
types of integrable Painlev equations [53]. It remains to identify those of them that lay
behind the ASDWeyl (or quaternionic-Khler) geometry with the SU(2) symmetry. There
are only two natural (or nilpotent, in the terminology of Ref. [53]) types (III and VI)
that give rise to a single non-linear integrable equation. In the geometrical terms, it is the
Painlev III equation that lays behind the four-dimensional Khler spaces with vanishing
scalar curvature [54], whereas the Painlev VI equation is known to be behind the ASD
Weyl geometries having the SU(2) symmetry [49,50,55]. A generic Painlev VI equation
has four real parameters [53], but they are all fixed by the quaternionic property [49,50],
as in Eq. (6.6). This means that the quaternionic metrics with the SU(2) symmetry are all
governed by the particular Painlev VI equation:
y =

1
1
1
1
1 1
1
+
+
+
+
(y )2
y
2 y y1 yx
x x1 yx

x
y(y 1)(y x) 1
x 1
3x(x 1)
,
+
+
+
x 2 (x 1)2
8 8y 2 8(y 1)2 8(y x)2
(6.10)
where y = y(x), and the primes denote differentiation with respect to x.

The equivalence between Eqs. (6.3) and (6.10) via Eqs. (6.4) and (6.5) is known to
mathematicians [49,50,55]. Explicitly, in the Einstein case, it is given by the relations

(y x)2 y(y 1)
=
v
x(1 x)

(y x)y 2 (y 1)
22 =
v
x

(y x)y(y 1)2
32 =
v
(1 x)
12

1
v
,
2y

1
1
v
,
2(y x)
2(y 1)

1
1
v
,
2y
2(y x)
1
2(y 1)
(6.11)
where the auxiliary variable v is defined by the equation

y =

1
1
y(y 1)(y x)
1
+
2v
.
x(x 1)
2y 2(y 1) 2(y x)
(6.12)
274
An exact solution to the Painlev VI equation (6.10), which leads to a regular (and
complete) quaternionic metric (6.2), is unique [50]. The Hitchin solution [50] can be
expressed in terms of the standard theta-functions (z| ), where = 1, 2, 3, 4. 9
In order to write down the Hitchin solution to Eq. (6.10), the theta-function arguments
should be related, z = 12 ( k), where k is an arbitrary (real and positive) parameter. The
variable is related to the variable x of Eq. (6.10) via
x = 34 (0)/44 (0),
(6.13)
where the value of the variable z is explicitly indicated, as usual. One finds [50,57]

34 (0)
1 (0)
1
y(x) =
1
+
+
3 2 44 (0)1 (0) 3
44 (0)

(z)1 (z) 21 (z)1 (z) + 2i 1 (z)1 (z) 12 (z)

.
+ 1
(6.14)
2 2 44 (0)1 (z) 1 (z) + i1 (z)
The parameter k > 0 describes the monodromy of the solution (6.14) around its essential
singularities (branch points) x = 0, 1, . This (non-abelian) monodromy is generated by
the matrices (with the purely imaginary eigenvalues i) [50]

0 i
0
i 1k
0
i k
,
M3 =
. (6.15)
M1 =
,
M2 = 1+k
i
0
i
0
i k
0
Another explicit (equivalent) form of the Hitchin exact solution to the metric coefficients
wi in Eqs. (6.2) and (6.4) was derived in Ref. [58], in terms of the theta functions with
characteristics, by the use of the fundamental Schlesinger system and the isomonodromic
deformation techniques.
The function (6.14) is meromorphic outside its essential singularities at x = 0, 1, ,
while is also has simple poles at x1 , x2 , . . . , where xn (xn , xn+1 ) and xn = x(ik/(2n 1))
for each positive integer n. Accordingly, the metric is well-defined (complete) for x
(xn , xn+1 ], i.e., in the unit ball with the origin at x = xn+1 and the boundary at x = xn [50].
Near the boundary the TodHitchin metric (6.2) has the asymptotical behaviour
ds 2 =
dx 2
4
16
2 +
2
+
(1 x)2 (1 x) cosh2 (k/2) 1 (1 x)2 sinh2 (k/2) cosh2 (k/2) 2
4
+
(6.16)
32 + regular terms.
(1 x) sinh2 (k/2)
As is clear from Eq. (6.16), the coefficient at 22 vanishes faster than the coefficients
at 12 and 32 when approaching the boundary, x 1 , similarly to Eq. (A.6). On the
two-dimensional boundary annihilated by 2 one has the natural conformal structure
sinh2 (k/2)12 + cosh2 (k/2)32 .
(6.17)
The only relevant parameter tanh2 (k/2) in Eq. (6.17) represents the central charge (or the
conformal anomaly) on the boundary. This result is apparently consistent with (i) the fact
9 We use the standard definitions and notation for the theta functions [56] see Appendix B.
275
that the type IIA five-branes can be described in terms of an exact (super)conformal field
theory [59], (ii) the Zamolodchikov c-theorem [60], and (iii) the holographic principle [61].
The constraints (5.1) do not seem to imply any quantization condition on the monodromy
parameter k since the regular metric solutions exist for any k > 0. The central charge (or the
critical exponent k) is quantized in solvable 2d, N = 2 superconformal field theories (the
minimal N = 2 superconformal models) which are associated with compact (simply-laced)
Lie groups [42]. The absence of central charge quantization in our case may be related to
the negative scalar curvature of the metrics. The EinsteinWeyl metrics of the positive
scalar curvature take the similar form given by Eqs. (6.2) or (6.8), while they are known to
be related to the so-called Poncelet n-polygons that give rise to the quantization condition
k = 2/n, where n Z [62]. It is thus the non-compact nature of the Lie group SU(2, 1) that
is responsible for the absence of the central charge quantization in the boundary conformal
field theory.
7. Conclusion
The universal hypermultiplet gives us the unique opportunity to learn more about the
non-perturbative quantum corrections in string/M-theory via better understanding of the
exact quaternionic geometry governing the hypermultiplet LEEA (or NLSM) in the 4d,
N = 2 supergravity background.
The D-instanton corrections to the classical UH moduli space metric are calculable due
to the residual UD (1) UC (1) symmetry (Sections 4 and 5). No such corrections arise
when all the other fields (except UH) are turned off. The five-brane instanton corrections
are calculable at vanishing five-brane charges (Section 6). In a generic case with nonvanishing charges, there may be also contributions from the membranes ending on fivebranes wrapped on a CY space [13]. Then only UC (1) isometry may be left. However, the
quaternionic ansatz (5.2) still applies, this time with respect to the UC (1) isometry. The
absence of any other symmetry means that generic instanton corrections are described
by non-separable exact solutions to the three-dimensional Toda equation (4.7). This
observation is consistent with another observation that the instanton action is not a sum
of membrane and five-brane contributions [13].
More general (than AtiyahHitchin) regular hyper-Khler four-manifolds with a rotational isometry are known [63], though in the rather implicit form (as the algebraic curves).
It is also known that the Toda equation (4.7) can be reduced in a highly non-trivial way to
the Painlev equations [64]. It is not clear to us how to extract the specific information
from this knowledge, which would be relevant for the problem mentioned in the title.
Acknowledgements
I am grateful to Klaus Behrndt, Vicente Cortes, Evgeny Ivanov, Olaf Lechtenfeld,
Werner Nahm, Hermann Nicolai, Paul Tod, Galliano Valent, Stefan Vandoren and Bernard
de Wit for discussions. This paper is based on the invited talks given by the author
276
at the Institute for Theoretical Physics in Hannover and the Max-Planck-Institute for
Mathematics in Bonn, within the national (DFG) programm String Theory.
Appendix A. Coset space SU(2, 1)/SU(2) U (1)
The homogeneous symmetric space SU(2, 1)/SU(2) U (1) is topologically equivalent
to the open ball in C2 with the Bergmann metric [32]
ds 2 =
dz1 d z 1 + dz2 d z 2 (z1 dz1 + z 2 dz2 )(z1 d z 1 + z2 d z 2 )

+
,
1 |z1 |2 |z2 |2
(1 |z1 |2 |z2 |2 )2
(A.1)
where |z1 |2 + |z2 |2 < 1 and |z|2 = zz. The metric (A.1) is Khler,
ds 2 = (a b K) dza d z b = eK dza d z a + e2K (za dza )(zb d z b ),
a, b = 1, 2,
with the Khler potential

K = log 1 |z1 |2 |z2 |2 .
(A.2)
(A.3)
The Khler potential is defined modulo the Khler gauge transformations,

z ) = K(z, z ) + f (z) + f(z),
K(z, z ) K(z,
(A.4)
with an arbitrary holomorphic function f (z1 , z2 ).

Eq. (A.3) clearly shows that the Bergmann metric is dual to the FubiniStudy
metric on the compact complex projective space CP2 = SU(3)/SU(2) U (1) [65]. The
homogeneous space CP2 is symmetric, while it is also an Einstein space of positive
scalar curvature with the (anti-)self-dual Weyl tensor. 10 The non-compact coset space
SU(2, 1)/U (2) is, therefore, also an Einstein space (though of negative scalar curvature),
with the (anti-)self-dual Weyl tensor. In other words, the coset space SU(2, 1)/U (2) is
an EinsteinWeyl (or a self-dual Einstein) space. In four dimensions, the EinsteinWeyl
spaces are called quaternionic by definition [32].
After the coordinate change
z1 = r cos 2 ei(+)/2 ,
z2 = r sin 2 ei()/2 ,
(A.5)
the metric (A.1) can be rewritten to the diagonal form in the Bianchi IX formalism with
manifest SU(2) symmetry,
ds 2 =

r 2 22
dr 2
r2 2
+
+
1 + 32 ,
2
2
2
2
2
(1 r )
(1 r )
(1 r )
(A.6)
where we have introduced the su(2) (left)-invariant one-forms

1 = 12 (sin sin d + cos d ),
2 = 12 (d + cos d),
3 = 12 (sin d cos sin d),
10 The Weyl tensor is the traceless part of the Riemann curvature.
(A.7)
277
in terms of four real coordinates 0 r < 1, 0 < , 0 < 2 and 0 < 4 . The
one-forms (A.6) obey the relations
i j = 12 ij k dk ,
i, j, k = 1, 2, 3.
(A.8)
Another useful parametrization of the coset SU(2, 1)/U (2) arises after the following
change of variables:
z1 =
1S
,
1+S
z2 =
2C
.
1+S
(A.9)
The Khler potential in terms of the new complex variables (S, C) reads

K = log(S +
S 2C C),
(A.10)
with the Khler metric

2C
d
.
ds 2 = e2K dS d
S 2C dS d C
S dC + 2(S +
S) dC d C
(A.11)
This form of the metric was used by Ferrara and Sabharwal [16] in their analysis of the
type II superstring vacua on CalabiYau spaces, in terms of the complex RR-type scalar C
and the complex scalar

S = e2 + 2iD + C C,
(A.12)
where stands for the dilaton field and D stands for the axion field in four spacetime
dimensions. The metric (A.11) can also be rewritten to the form
ds 2 = |u|2 + |v|2 ,
(A.13)
where
u e dC
and v e2
1
2

dC ,
dS C
(A.14)
by using the relations

= 12 ln (S +
S 2C C)/2
(A.15)
eK = 12 e2 ,
(A.16)
and
in accordance with Eqs. (A.10) and (A.12). The notation (A.13) and (A.14) was used
by Strominger [7] in his investigation of the one-loop string corrections to the universal
hypermultiplet. Of course, the metric (A.13) is not flat, since the one-forms u and v of
Eq. (A.14) are not exact. Here are some useful identities [7]
du = 12 u (v + v),
dv = v v + u u,
d = 12 (v + v).
(A.17)
278
Appendix B. Basic facts about theta-functions

The first theta-function 1 (z| ) is defined by the series [56]
+

1 (z) 1 (z| ) = i
2

(1)n exp i n + 12 + (2n + 1)z
n=
=2
+

(1)n q (n+1/2) sin(2n + 1)z,
q = ei ,
(B.1)
n=0
where is regarded as the fundamental complex parameter, whose imaginary part must be
positive, q is called the nome of the theta-function, |q| < 1, and z is the complex variable.
The other theta-functions are defined by [56]
+

2
q (n+1/2) ei(2n+1)z
2 (z| ) = 1 z + 12 | ) =
n=
=2
+

q (n+1/2) cos(2n + 1)z,
(B.2)
n=0
+

2
3 (z| ) = 4 z + 12 | ) =
q n e2inz
n=
=1+2
+

q n cos 2nz,
(B.3)
n=1
and
4 (z| ) =
+

(1)n q n e2inz = 1 + 2
n=
+

2
(1)n q n cos 2nz.
(B.4)
n=1
The identities [56]

34 (0) = 24 (0) + 44 (0),
(B.5)
1 (0) = 2 (0)3 (0)4 (0),
(B.6)
1 (0) 2 (0) 3 (0) 4 (0)

=
+
+
,
1 (0)
2 (0)
3 (0)
4 (0)
(B.7)
and
where the primes denote differentiation with respect to z, may be used to rewrite Eq. (6.4)
to other equivalent forms (cf. [50,57,58]).
References
[1] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, Cambridge Univ. Press, 1987;
J. Polchinski, String Theory, Cambridge Univ. Press, 1998;
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
279
S.V. Ketov, Introduction to Quantum Theory of Strings and Superstrings, Nauka, 1990 (in
Russian).
K. Becker, M. Becker, A. Strominger, Nucl. Phys. B 456 (1995) 130.
K. Becker, M. Becker, Nucl. Phys. B 551 (1999) 102.
S. Kachru, J. McGreevy, Phys. Rev. D 61 (2000) 026001.
J. Bagger, E. Witten, Nucl. Phys. B 222 (1983) 1.
E. Witten, Nucl. Phys. B 474 (1996) 343.
A. Strominger, Phys. Lett. B 421 (1998) 139.
B. de Wit, A. van Proeyen, Nucl. Phys. B 245 (1984) 89;
B. de Wit, P.G. Lauwers, A. van Proeyen, Nucl. Phys. B 255 (1985) 569.
S. Cecotti, S. Ferrara, L. Girardello, Int. J. Mod. Phys. A 4 (1989) 2475.
P.S. Aspinwall, JHEP 9804 (1998) 019.
S. Kachru, C. Vafa, Nucl. Phys. B 450 (1995) 69;
S. Ferrara, J.A. Harvey, A. Strominger, C. Vafa, Phys. Lett. B 361 (1995) 59.
J.A. Harvey, G. Moore, Superpotentials and membrane instantons, hep-th/9907026.
M. Gutperle, M. Spalinski, JHEP 0006 (2000) 037.
M. Gutperle, M. Spalinski, Supergravity instantons for N = 2 hypermultiplets, hepth/70010192.
S.V. Ketov, Gravitational dressing of D-insantons, hep-th/0010255, to appear in Phys. Lett. B.
S. Ferrara, S. Sabharwal, Class. Quant. Grav. 6 (1989) L77.
H. Gnther, C. Herrmann, J. Louis, Fortschr. Phys. 48 (2000) 119.
H. Ooguri, C. Vafa, Phys. Rev. Lett. 77 (1996) 3298.
E. Witten, Phys. Lett. B 155 (1985) 151.
E. Witten, Nucl. Phys. B 500 (1997) 3.
E.J. Cremmer, B. Julia, J. Scherk, Phys. Lett. 76B (1978) 409.
P.K. Townsend, M-theory from its superalgebra, in: Proc. of the NATO Advanced Study
Institute, Strings, Branes and Dualities, Cargese, 1997, p. 141, hep-th/9712004.
E. Bergshoeff, E. Sezgin, P.K. Townsend, Phys. Lett. B 189 (1987) 75.
R. Gven, Phys. Lett. B 276 (1992) 49.
D.N. Page, Phys. Rev. D 28 (1983) 2976.
J.H. Schwarz, Phys. Lett. B 367 (1996) 97.
S.P. de Alwis, Phys. Lett. B 388 (1996) 291.
M.B. Green, M. Gutperle, Nucl. Phys. B 498 (1997) 195;
I. Antoniadis, S. Ferrara, R. Minasian, K.S. Narain, Nucl. Phys. B 507 (1997) 571.
P. Candelas, X.C. de la Ossa, P.S. Green, L. Parkes, Nucl. Phys. B 359 (1991) 21.
B. de Wit, A. van Proeyen, Phys. Lett. B 252 (1990) 221;
B. de Wit, F. Vanderseypen, A. van Proeyen, Nucl. Phys. B 400 (1993) 463.
C.R. LeBrun, J. Diff. Geom. 34 (1991) 223.
A.L. Besse, Einstein Manifolds, Springer-Verlag, 1987.
M.F. Atiyah, N.J. Hitchin, The Geometry and Dynamics of Magnetic Monopoles, Princeton
Univ. Press, 1998.
C.P. Boyer, J.D. Finley, J. Math. Phys. 23 (1982) 1126.
N. Seiberg, S. Shenker, Phys. Lett. B 388 (1996) 521.
M. Saveliev, Commun. Math. Phys. 121 (1989) 283;
R.S. Ward, Class. Quant. Grav. 7 (1990) L95.
A.N. Leznov, Theor. Math. Phys. 117 (1998) 1194.
S.V. Ketov, Quantum Non-Linear Sigma-Models, Springer-Verlag, 2000, Chapter 4.
C.L. Fefferman, Ann. Math. 103 (1976) 395.
R. Britto-Pacumio, A. Strominger, A. Volovich, JHEP 9911 (1999) 013.
K.P. Tod, Twistor Newsletter 39 (1995) 19.
S.V. Ketov, Conformal Field Theory, World Scientific, 1995, Section IX.2.
280
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
B. de Wit, B. Kleijn, S. Vandoren, Nucl. Phys. B 568 (2000) 475.

I. Bakas, K. Sfetsos, Int. J. Mod. Phys. A 12 (1997) 2585.
I. Bakas, Fortschr. Phys. 48 (2000) 9.
D. Olivier, Gen. Rel. Grav. 23 (1991) 1349.
S.V. Ketov, Phys. Lett. B 469 (1999) 136.
N. Seiberg, E. Witten, Gauge dynamics and compactification to three dimensions, hepth/9607163.
K.P. Tod, Phys. Lett. A 190 (1994).
N.J. Hitchin, J. Diff. Geom. 42 (1995) 30.
G.-H. Halphen, Sur un systme dquations diffrentielles, C.R. Acad. Sci. Paris 92 (1981)
1101.
M.J. Ablowitz, P.A. Clarkson, Solitons, Non-Linear Evolution Equations and Inverse Scattering, Cambridge Univ. Press, 1991.
L.J. Mason, N.M.J. Woodhouse, Integrability, Self-Duality, and Twistor Theory, Clarendon
Press, 1996.
H. Pedersen, Y.S. Poon, Class. Quant. Grav. 7 (1990) 1707.
R. Maszczyk, L.J. Mason, N.M.J. Woodhouse, Class. Quant. Grav. 11 (1994) 65.
D.F. Lawden, Elliptic Functions and Applications, Springer-Verlag, 1980.
B.A. Dubrovin, Funct. Anal. Appl. 24 (1990) 280.
M.V. Babich, D.A. Korotkin, Lett. Math. Phys. 46 (1998) 323.
C. Callan, J.A. Harvey, A. Strominger, Nucl. Phys. B 367 (1991) 60.
A.B. Zamolodchikov, JETP Lett. 43 (1986) 731.
S.V. Ketov, Exact renormalization flow and domain walls from holography, hep-th/0009187, to
appear in Nucl. Phys. B.
N.J. Hitchin, A new family of Einstein metrics, in: Proc. of the Pisa Conference in Honour of
E. Calabi, Cambridge Univ. Press, 1995.
A. Dancer, A family of gravitational instantons, Cambridge preprint DAMTP 9213 (1992),
unpublished.
K.P. Tod, Class. Quant. Grav. 12 (1995) 1535.
G.W. Gibbons, C.N. Pope, Commun. Math. Phys. 61 (1978) 239.

Some higher moments of deep inelastic structure

functions at next-to-next-to-leading order
of perturbative QCD
A. Rtey a , J.A.M. Vermaseren b
a Institut fr Theoretische Teilchenphysik, Universitt Karlsruhe, D-76128 Karlsruhe, Germany
b NIKHEF Theory Group, Kruislaan 409, 1098 SJ Amsterdam, The Netherlands
Received 19 March 2001; accepted 21 March 2001
Abstract
We present the analytic next-to-next-to-leading QCD calculation of some higher moments of deep
inelastic structure functions in the leading twist approximation. We give results for the moments
N = 1, 3, 5, 7, 9, 11, 13 of the structure function F3 . Similarly we present the moments N = 10, 12
for the flavour singlet and N = 12, 14 for the non-singlet structure functions F2 and FL . We have
calculated both the three-loop anomalous dimensions of the corresponding operators and the threeloop coefficient functions of the moments of these structure functions. 2001 Published by Elsevier
Science B.V.
PACS: 12.38.-t; 13.85.Hd
1. Introduction
The determination of the next-to-next-to-leading (NNL) order QCD approximation for
the structure functions of deep inelastic scattering has become important for the understanding of perturbative QCD and necessary for an accurate comparison of perturbative
QCD with the increasing precision of experiments. Such calculations however are rather
complicated and hence a complete NNL result does not exist as of yet. The one-loop anomalous dimensions were calculated in [1,2]. In [3] (see also the references therein) the complete one-loop coefficient functions were obtained. Anomalous dimensions at 2-loop order
were obtained in [412]. The 2-loop coefficient functions were calculated in [1321].
Analytical results of the 3-loop anomalous dimensions and coefficient functions of the
moments of F2 and FL are only known for the moments N = 2, 4, 6, 8 in both the singlet
and non-singlet case and additionally for N = 10 in the non-singlet case from [22] and [23].
E-mail address: t68@nikhef.nl (J.A.M. Vermaseren).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 4 9 - 3
282
A. Rtey, J.A.M. Vermaseren / Nuclear Physics B 604 (2001) 281311
In addition the GrossLlevellyn Smith sum rule, which corresponds to the first moment of
p+ p
F3
has been calculated at this order [24].
For a complete reconstruction of the x-dependence of the structure functions via an
inverse Mellin transformation one would need the moments for all N (that is either all
even or all odd integer values). Additionally one needs them both for F2 and F3 in order
to untangle the various quark and gluon contributions. The determination of the NNL
approximation for generic N is work in progress [25], but probably will not be finished
in the near future.
The available moments of F2 have been used by a number of authors to make
a reconstruction of the complete structure functions at NNL by a variety of means [23,
2628]. Additionally they can be used to obtain a better value of S [29]. It should be clear
that it is important to have as large number of moments as possible. First, these results can
be immediately used to increase the precision of phenomenological investigations of deep
inelastic scattering [30]. Second, the moments will be a very important check for the new
methods and programs needed for the determination of the 3-loop results for arbitrary N .
Unfortunately it is not very easy to increase the number of moments, because each new
moment requires roughly five times the computer resources that its predecessor needs.
With the advent of better computers this means that by now it has been possible to obtain
two more moments for the singlet case and one additional moment for the non-singlet
F2 case. This should allow, for instance, a somewhat better determination of S . More
p+ p
to three
important however is the determination of the first seven odd moments of F3
loops. To this end we used the the same programs as in [23], state of the art computers
and a new version of the symbolic manipulation program FORM [31] that supports now
64-bit architectures and to some extend parallel computers (see also [32]). We could push
the limit in these calculations to include two new moments (N = 10, 12) in the calculation
of the flavour singlet structure functions FL and F2 , and two new moments (N = 12, 14)
for the flavour non-singlet structure functions FL and F2 . Additionally we have computed
the moments N = 3, 5, 7, 9, 11, 13 of the structure function F3 . We do not expect more
moments to become available before the complete results for all N will be presented.
2. The formalism
This calculation follows the one presented in [23] (see also [21]) in every detail, so we
only will give a very short review on the methods used.
We need to calculate the hadronic part of the amplitude for unpolarized deep inelastic
scattering which is given by the hadronic tensor

1
2
d4 z eiqz p, nucl|J (z)J (0)|nucl, p
W x, Q =
4

1
= (g q q ) FL x, Q2
2x

4x 2
2x 1
F2 x, Q2
+ g p p 2 (p q + p q ) 2
q
q 2x
+ i

p q
F3 x, Q2 ,
pq
283
(1)
where the J are either electromagnetic or weak hadronic currents and x = Q2 /(2p q)
is the Bjorken scaling variable with 0 < x 1. Q2 = q 2 is the transferred momentum
and |p, nucl is the nucleon state with momentum p. In these equations spin averaging is
assumed. The longitudinal structure function FL is related to the structure function F1 by
FL = F2 2xF1 . For electronnucleon scattering J is the electromagnetic quark current
and F3 vanishes. For neutrinonucleon scattering J is an electroweak quark current which
has an axial-vector contribution and F3 , which describes parity violating effects that arise
from vector and axial-vector interference, will not vanish.
Using the dispersion relation technique one can relate the hadronic tensor to the
following 4-point Green functions:
W (p, q) =
T (p, q) = i
1
Im T (p, q),
2

d4 z eiqz p, nucl|T J (z)J (0) |nucl, p.
Applying a formal operator product expansion in terms of local operators to the timeordered product of two quark currents leads to:

i d4 z eiqz T J1 (z)J2 (0)
=

2

1 N
Q
q 1 q 2
j
g
q
q
C
,
a
s
1 2
1
2 L,N
Q2
q2
2
N,j

g1 1 g2 2 q 2 g1 1 q2 q2
2

j
Q
g2 2 q1 q1 + g1 2 q1 q2 C2,N
, as
2
2

Q
j
q3 qN O j,{1 ,...,N } (0)
+ i1 2 1 3 g 3 4 q4 q2 C3,N
,
a
s
2
+ higher twists, j = , , G.
(2)
Here we have introduced the notation as = s /(4) = g 2 /(4)2 and everything is assumed
to be renormalized. The sum over N runs over the standard set of the spin-N twist-2
irreducible flavour non-singlet quark operators and the singlet quark and gluon operators:

{1 D 2 D N } , = 1, 2, . . . , n2f 1 ,
O ,{1 ,...,N } =
{1 D 2 D N } ,
O ,{1 ,...,N } =
O G,{1 ,...,N } = G{1 D 2 D N1 GN } .
Application of this OPE to Eq. (1) leads to an expansion for the unphysical values x .
From the proper analytical continuation to the physical region 0 < x 1 one finds for the
moments of the structure functions F2 , FL and F3 :
284
1
Mk,N2 =
2

Q
i
i
dx x N2 Fk x, Q2 =
Ck,N
,
a
s Anucl,N ,
2
1
M3,N1 =
k = 2, L, (3)
i=,,G
dx x
N1

i,N Q2
2
F3 x, Q =
C
, as Ainucl,N ,
2
(4)
i=
with the spin averaged matrix elements

p, nucl|O j,{1 ,...,N } |p, nucl = p{1 pN } Anucl,N
j

p2
.
2
(5)
In the derivation of (3) one needs the symmetry properties of T under x x. This
is why one can only find either even or odd moments from these equations, dependent
on the process under consideration. For F3 we will only consider the flavor non-singlet
contributions, due to the properties of the operators O and O G under charge conjugation
there should not be a singlet contribution (see, e.g., [33]).
The scale-dependence of the coefficient functions is then covered by the renormalization
group equations:
2

Q
2
ns
ns
+ (as )
N Ci,N
, as = 0, i = 2, 3, L,
(6)
2
as
2

2

Q
jk
k
2 2 + (as )
j k N Ci,N
,
a
s = 0,
as
2
k=,G
i = 2, L, j = , G.
(7)
The non-singlet coefficient functions and anomalous dimensions do not depend on the
index , and we have adopted the conventional collective denotation ns for them.
3. The even moments of F2 and FL

Eq. (2) is a relation between operators and does not depend on the hadronic states to
which the OPE is applied.
Using the method of projectors [34] one can find both the coefficient functions and the
anomalous dimensions for the even moments of F2 and FL as defined in Eq. (6) from the
following 4-point Green functions:

q q
T = i d4 z eiqz p, quark|T J (z)J (0) |quark, p,
(8)

g g
T
(9)
= i d4 z eiqz p, gluon|T J (z)J (0) |gluon, p.
Applying to Eq. (8) the projectors
{1

N
q q N }

,
PN
N!
p1 pN p=0
(10)
285
and, to project out the different Lorentz projections (these projectors are valid in
D = 4 2 and for the leading twist approximation):
q2
p p ,
PL =
(p q)2

3 2 q 2
1

g
,
P2 =
p
p
+
2 2 (p q)2
2 2
as well as the corresponding flavour projections results in the equations:

Q2
,
a
,

s
2

2

2

1
1
Q
Q
G
,tree
G
= Ck,N
, as , ZN as ,
, as , ZN as ,
+ Ck,N
Aquark,N (),
2

2

2
g g ,s Q
, as ,
Tk,N
2

2

2

Q
Q
1
1
G
G
GG
= Ck,N
+ Ck,N
AG,tree
as ,
, as , ZN as ,
, as , ZN
gluon,N (),

2
2

2
2

Q
1
q q ,ns Q
ns
ns
Ans,tree
, as , = Ck,N
, as , ZN as ,
Tk,N
(11)
quark,N (), k = 2, L.

2
2
q q ,s
Tk,N
ij
From these equations the coefficient functions and from the ZN the anomalous dimensions
can be calculated in the usual way.
It should be mentioned that in the Eq. (11) on the left-hand side after applying the
projectors (10) we are left with only diagrams of the massless propagator type, a problem
solved at 3-loop order long ago [35] and implemented in an efficient way in the FORM
package MINCER [36]. On the right-hand side only the tree level diagrams contributing to
the matrix elements survive.
It turns out that much computing time can be saved when calculating additionally
Green functions with external ghosts to get rid of the unphysical polarization states of
the external gluons instead of using the very complicated projection onto physical states.
G
GG
and ZN only to order s2 . To obtain the
Also, from Eq. (11) one can determine the ZN
s3 -contributions one can calculate additionally Greens functions with external scalar fields
that couple to gluons only at tree level. Altogether, to obtain the coefficient functions and
anomalous dimensions for the even moments with N = 10, 12 of F2 and FL the following
diagrams had to be calculated (q = quark, g = gluon, = photon, h = ghost, = scalar
field):
286
Tree
1-loop
2-loops
3-loops
413
697
366
53
1266
1266
10846
q q
qq
g g
h h
gg
hh
3
1
2
11
11
27
24
20
2
241
241
Total
23
399
Lorentz projections
2
1
2
2
1
1
The q {1 q N } in Eq. (10) are the harmonic (i.e., symmetrical and traceless) part
of the tensor q 1 q N . The number of terms in these harmonic tensors explodes as N
increases and this is the real limitation in these calculations considering the computing time
as well as disk-space usage. In spite of a very efficient implementation of these tensors (see
[37]) for N = 12, singlet and N = 14, non-singlet, individual diagrams had a disk-space
usage up to and over 100 GB. Altogether the calculation of all the above diagrams for N =
10, 12 took approximately 5 weeks on a Compaq Server with 8 Alpha 21264 processors
running at 700 MHz, 4 GB of RAM and 12 17 GB of disk-space. The N = 14 non-singlet
calculation took comparable resources.
4. The odd moments of F3

The coefficient functions and anomalous dimensions of the odd moments of the structure
function F3 can be obtained in the same way as the non-singlet part of F2 and FL but
now considering the time-ordered product of one vector current V and one axial vector
current A . The axial current introduces the appearance of a 5 and some care has to be
taken to treat it correctly within the framework of dimensional regularization. We adopt
the definition used in [24]:
1
5 = i .
6
Projecting out the flavour non-singlet part and the corresponding Lorentz structure with:
P3 = i
p q
1

,
(1 2)(2 2)
pq
one finds products of metric tensors which have to be considered as D-dimensional

objects. Since this definition of 5 in D dimensions violates the axial Ward identity
one needs to renormalize A with a renormalization constant ZA and additionally apply
a finite renormalization with Z5 , both of these constants are given to 3-loop order in [24].
Combining all this finally leads to

2
2

1
1
Q
Q
1
ns
ZA a s ,
,
a
,

=
C
,
a
,

Z
,
a
Z5 (as , )T3,N
Ans
s
3,N
s
s
N,tree ().
N

2
2

287
Due to the 5 insertion at one of the vertices, some of the symmetries that were used to
minimize the number of diagrams could not be applied in this case and to determine the
ns
T3,N
1076 (= 1 + 4 + 55 + 1016) diagrams had to be evaluated, which took about 6 weeks
for the moments N = 1, 3, 5, 7, 9, 11, 13.
5. Results
Using the strategies sketched in the previous section we find the following results for the
coefficient functions and anomalous dimensions. Again following Ref. [23], we present the
combined singlet and non-singlet results for F2 and FL in terms of flavour factors which
are defined in the following table for nf number of flavours:
fl2
Non-singlet
Singlet
1
1
fl11
3
nf
nf
f =1 ef
nf
2
1 ( f =1 ef )
n

f
nf
2
f =1 ef
fl02
fl2
fl11
nf
2
1 ( f =1 ef )
n

f
nf
e2
f =1 f
The numerical values of the anomalous dimensions for the spin-even operators
contributing to F2 and FL that now are known to NNL order are:
10

= 3.555555556 as + as2 48.32921811 3.160493827 nf 1.975308642 fl02 nf

+ as3 859.4478372 133.4381617 nf 1.229080933 n2f

+ fl02 42.21182429 nf 3.445816187 n2f ,

= 6.977777778 as + as2 86.28665021 6.553580247 nf 0.1060246914 fl02 nf

+ as3 1515.562363 244.728592 nf 2.108515775 n2f

+ fl02 5.17013312 nf 0.6789278464 n2f ,

= 9.003174603 as + as2 108.0184697 8.62925674 nf 0.02080408883 fl02 nf

+ as3 1891.827779 307.4236889 nf 2.570638992 n2f

+ fl02 2.526091192 nf 0.2884556035 n2f ,

= 10.45820106 as + as2 123.7764525 10.14583662 nf 0.006586485074 fl02 nf

+ as3 2164.091836 352.3116596 nf 2.882493484 n2f

+ fl02 1.682156519 nf 0.1620816452 n2f ,

= 11.5969216 as + as2 136.2741775 11.34594534 nf 0.00269944007 fl02 nf

+ as3 2379.919952 387.6422968 nf 3.115523145 n2f

+ fl02 1.256562245 nf 0.1054295071 n2f ,
288

12 = 12.53336293 as + as2 146.6771964 12.34063169 nf 0.001302012432 fl02 nf

+ as3 2559.641948 416.9091301 nf 3.300349343 n2f

+ fl02 0.9957297081 nf 0.07498848634 n2f ,

,NS
14
= 13.32896733 as + as2 155.6071962 13.19066078 nf

+ as3 2714.031720 441.9494048 nf 3.452881831 n2f ,
G
= 0.6666666667 as nf 7.543209877 as2 nf

+ as3 37.62337275 nf + 12.11248285 n2f ,
= 0.3666666667 as nf + 1.290703704 as2 nf

+ as3 33.58149273 nf + 6.06027262 n2f ,
= 0.2619047619 as nf + 2.761104812 as2 nf

+ as3 33.41602135 nf + 3.537682102 n2f ,
= 0.2055555556 as nf + 3.243957223 as2 nf

+ as3 28.7612615 nf + 2.225433112 n2f ,
2
4
6
8
10 = 0.1696969697 as nf + 3.407168695 as2 nf

+ as3 23.93704198 nf + 1.449828678 n2f ,
G
12 = 0.1446886447 as nf + 3.438705999 as2 nf

+ as3 19.63230379 nf + 0.9524545446 n2f ,
G
10
12

= 3.555555556 as + as2 48.32921811 + 5.135802469 nf

+ as3 859.4478372 + 175.649986 nf + 4.674897119 n2f ,

= 0.9777777778 as + as2 16.1752428 + 0.6182716049 nf

+ as3 315.276255 + 39.82571027 nf + 1.801843621 n2f ,

= 0.5587301587 as + as2 9.496317796 + 0.08884857647 nf

+ as3 188.9088124 + 19.67944546 nf + 1.087843741 n2f ,

= 0.3915343915 as + as2 6.757603506 0.07061952353 nf

+ as3 134.7055042 + 12.3754454 nf + 0.7536013741 n2f ,

= 0.3016835017 as + as2 5.297576945 0.1348941718 nf

+ as3 104.911278 + 8.796702078 nf + 0.5579674847 n2f ,

= 0.2455322455 as + as2 4.398625917 0.1639529655 nf

+ as3 86.18107998 + 6.735609285 nf + 0.4293379075 n2f ,
2GG = 0.6666666667 as nf + 7.543209877 as2 nf
289

+ as3 37.62337275 nf 12.11248285 n2f ,

4GG = as2 128.178 13.64948148 nf + as 12.6 + 0.6666666667 nf

+ as3 2066.19278 401.3127939 nf 10.43150645 n2f ,

6GG = as2 183.0538144 20.46668466 nf + as 17.78571429 + 0.6666666667 nf

+ as3 2987.042058 566.6373298 nf 10.78060861 n2f ,

8GG = as2 219.6240988 24.69926432 nf + as 21.26666667 + 0.6666666667 nf

+ as3 3609.35419 673.9430658 nf 11.20133837 n2f ,

GG
= as2 247.6655484 27.82178573 nf + as 23.92337662 + 0.6666666667 nf
10

+ as3 4089.236943 755.1340541 nf 11.57068198 n2f ,

GG
= as2 270.6428892 30.31377688 nf + as 26.08168498 + 0.6666666667 nf
12

+ as3 4483.563048 821.1236576 nf 11.88665683 n2f .
The corresponding coefficient functions read:
C2,2 = 1 + 0.4444444444 as

+ as2 17.69376589 5.333333333 nf 2.189300412 fl02 nf

+ as3 442.7409693 165.1971095 nf 24.09201335 fl11 nf + 6.030272415 n2f

+ fl02 79.04486142 nf + 3.325504478 n2f ,
C2,4 = 1 + 6.066666667 as

+ as2 142.3434719 16.98791358 nf + 0.4858308642 fl02 nf

+ as3 4169.267888 901.2351626 nf 18.21884618 fl11 nf + 23.35503924 n2f

+ fl02 16.64834849 nf 2.208630689 n2f ,

C2,6 = 1 + 11.17671958 as + as2 302.398735 28.0130504 nf + 0.4868787285 fl02 nf

+ as3 10069.63085 1816.322929 nf 16.14271761 fl11 nf + 42.66273116 n2f

+ fl02 24.11778813 nf 1.525489143 n2f ,

C2,8 = 1 + 15.52989418 as + as2 470.807419 37.9248228 nf + 0.3859585393 fl02 nf

+ as3 17162.37245 2787.297692 nf 15.09203827 fl11 nf + 61.91177997 n2f

+ fl02 22.33201938 nf 1.036308122 n2f ,
C2,10 = 1 + 19.30061568 as

+ as2 639.210663 46.86131842 nf + 0.3045901308 fl02 nf

+ as3 24953.13497 3770.10212 nf 14.45874451 fl11 nf + 80.52097973 n2f

+ fl02 19.53359559 nf 0.7372434464 n2f ,
290
C2,12 = 1 + 22.62841097 as

+ as2 804.5854321 54.99446579 nf + 0.2451231747 fl02 nf

+ as3 33171.45501 4746.440949 nf 14.03541028 fl11 nf + 98.3483124 n2f

+ fl02 16.98652635 nf 0.5471547625 n2f ,

,NS
C2,14 = 1 + 25.61093284 as + as2 965.8132564 62.46549093 nf

+ as3 41657.11568 5708.215623 nf

13.73240102 fl11 nf + 115.3919490 n2f ,
G
= 0.5 as nf 8.918338961 as2nf
C2,2

g
+ as3 130.7340963 nf + 29.37933515 n2f 0.9007972776 fl11 n2f ,
G
= 0.7388888889 asnf 14.27158692 as2nf
C2,4

g
+ as3 346.4612756 nf + 46.52017564 n2f 1.611816512 fl11 n2f ,
G
= 0.7051587302 asnf 20.06849828 as2nf
C2,6

g
+ as3 715.0372438 nf + 61.28545096 n2f 1.496036938 fl11 n2f ,
G
= 0.6440873016 asnf 23.17873524 as2nf
C2,8

g
+ as3 996.5038709 nf + 68.66467304 n2f 1.286400915 fl11 n2f ,
G
= 0.5861279461 asnf 24.76678064 as2nf
C2,10

g
+ as3 1201.206903 nf + 72.23614791 n2f 1.094394334 fl11 n2f ,
G
= 0.5358430591 asnf 25.51669345 as2nf
C2,12

g
+ as3 1351.047836 nf + 73.7936445 n2f 0.9344248731 fl11 n2f ,

CL,2 = 1.777777778 as + as2 56.75530152 4.543209877 nf 3.950617284 fl02 nf

+ as3 2544.598087 421.6908885 nf 7.736698288 fl11 nf + 11.8957476 n2f

+ fl02 213.9253076 nf + 17.91326528 n2f ,

CL,4 = 1.066666667 as + as2 47.99398931 3.413333333 nf 0.6945185185 fl02 nf

+ as3 2523.73902 383.0520013 nf 5.058869512 fl11 nf + 10.88895473 n2f

+ fl02 55.5530456 nf + 2.348005487 n2f ,

CL,6 = 0.7619047619 as + as2 40.9961976 2.69569161 nf 0.2524824533 fl02 nf

+ as3 2368.193775 340.0691069 nf 3.705612526 fl11 nf + 9.472190428 n2f

+ fl02 24.01322539 nf + 0.7652692585 n2f ,

CL,8 = 0.5925925926 as + as2 35.87664406 2.231471683 nf 0.1217397796 fl02 nf

+ as3 2215.210875 305.4730329 nf 2.913702563 fl11 nf + 8.337149534 n2f

+ fl02 12.97185267 nf + 0.344362391 n2f ,
291
CL,10 = 0.4848484848 as

+ as2 32.01765947 1.908598248 nf 0.06856377261 fl02 nf

+ as3 2081.213222 278.0172177 nf 2.397641695 fl11 nf + 7.452505612 n2f

+ fl02 7.947555677 nf + 0.1841598535 n2f ,

CL,12 = 0.4102564103 as + as2 29.0058065 1.671007435 nf 0.04265241396 fl02 nf

+ as3 1965.791047 255.8431044 nf 2.035689631 fl11 nf + 6.751061503 n2f

+ fl02 5.284573837 nf + 0.1098463658 n2f ,

,NS
CL,14 = 0.3555555556 as + as2 26.5848844 1.488624298 nf

+as3 1866.009187 237.5642566 nf

1.768138102 fl11 nf + 6.182499654 n2f ,
G
= 0.6666666667 as nf + 12.94776709 as2 nf
CL,2

g
+ as3 407.280632 nf 20.23959748 n2f 0.388939664 fl11 n2f ,
G
= 0.2666666667 as nf + 13.81659259 as2 nf
CL,4

g
+ as3 767.7125421 nf 36.78419232 n2f 0.3984298404 fl11 n2f ,
G
= 0.1428571429 as nf + 10.26095364 as2 nf
CL,6

g
+ as3 694.5092121 nf 27.9895081 n2f 0.3055276256 fl11 n2f ,
G
CL,8
= 0.08888888889 as nf + 7.733545104 as2 nf

g
+ as3 592.3307972 nf 21.30333681 n2f 0.2322211886 fl11 n2f ,
G
CL,10
= 0.06060606061 as nf + 6.023053074 as2 nf

g
+ as3 504.5424832 nf 16.70544537 n2f 0.1805482229 fl11 n2f ,
G
CL,12
= 0.04395604396 as nf + 4.830676204 as2 nf

g
+ as3 433.9050534 nf 13.47418408 n2f 0.1439099345 fl11 n2f .
The numerical values of the anomalous dimensions for the odd moments of F3 read:
1ns = 0,

3ns = 5.555555556 as + as2 70.88477366 5.12345679 nf

+ as3 1244.913602 196.4738081 nf 1.762002743 n2f ,

5ns = 8.088888889 as + as2 98.19940741 7.68691358 nf

+ as3 1720.942172 278.1581739 nf 2.366211248 n2f ,

7ns = 9.780952381 as + as2 116.4158903 9.437457798 nf

+ as3 2036.492478 330.8816595 nf 2.739358023 n2f ,
292

9ns = 11.05820106 as + as2 130.3414045 10.77682428 nf

+ as3 2277.19805 370.5905277 nf 3.006446616 n2f ,

ns
= 12.08581049 as + as2 141.6907901 11.86441897 nf
11

+ as3 2473.311857 402.691565 nf 3.212753995 n2f ,

ns
13
= 12.94606135 as + as2 151.2989044 12.78102552 nf

+ as3 2639.409887 429.7314605 nf 3.37996738 n2f ,
and the coefficient functions are:

ns
= 1 4 as + as2 73.33333333 + 5.333333333 nf
C3,1

+ as3 2652.154437 + 513.3100408 nf 11.35802469 n2f ,

ns
C3,3
= 1 + 1.666666667 as + as2 14.25404015 6.742283951 nf

+ as3 839.7638717 45.09953407 nf + 1.747689309 n2f ,

ns
= 1 + 7.748148148 as + as2 173.000629 19.39801646 nf
C3,5

+ as3 4341.081057 961.2756356 nf + 22.24125078 n2f ,

ns
= 1 + 12.72248677 as + as2 345.9910777 30.52332666 nf
C3,7

+ as3 11119.00053 1960.237096 nf + 43.10377964 n2f ,

ns
= 1 + 16.9152381 as + as2 520.0059615 40.35464229 nf
C3,9

+ as3 18771.99642 2975.924131 nf + 63.17127673 n2f ,

ns
= 1 + 20.54831329 as + as2 690.8719666 49.17096968 nf
C3,11

+ as3 26941.47987 3984.411605 nf + 82.24581704 n2f ,

ns
= 1 + 23.76237745 as + as2 857.1778817 57.18099124 nf
C3,13

+ as3 35426.82868 4976.080869 nf + 100.3509187 n2f .
Acknowledgements
A.R. would like to thank Sven Moch, Timo van Ritbergen and Thomas Gehrmann for
many useful discussions and Denny Fliegner for technical support. J.V. would like to thank
the University of Karlsruhe and the DFG for repeated hospitality.
This work was supported by the DFG under Contract Ku 502/8-1 (DFG-Forschergruppe
Quantenfeldtheorie, Computeralgebra und Monte-Carlo-Simulationen) and the Graduiertenkolleg Elementarteilchenphysik an Beschleunigern at the Univeritt Karlsruhe.
Appendix A. Conventions
Here we give the complete expressions for the newly computed moments and coefficient
functions.
293
The notation of the color factors is as usual: The Casimir operators of the fundamental
and adjoint representation are denoted by CF and CA and their values for the color
group SU(3) are 4/3 and 3, respectively. Additionally we are using the symmetric
structure constants of SU(3) for which d abc d abc = 40/3. For the trace normalization of
the fundamental representation we have inserted TF = 1/2. For a generic SU(n) group the
number of generators equals NA = n2 1.
The values of the Riemann function are written as n = (n). The only check for the
newly computed 3-loop results is from a large nf -expansion of [38] where the n2f terms
are calculated. These terms are in agreement with ours. To test the new parallel version of
FORM we have also recalculated the lower moments for F2 and FL and found complete
G there should be the extra term
agreement with [23] except for one missing term. In C2,2
160
3
2
as nf CF ( 3 5 ). The numerical values given in this reference are correct.
It should be noted that the terms in d abc d abc enter for the first time at the three loop level
S PS .
and help in the determination of Pqq
qq
Appendix B. Results for the moments of F2 and FL

18421
2
GG
10
= as CA +
+ as nf +
2310
3

339202487377
9284182
17481908
+ as2CF nf +
+ as2 CA nf
+ as2 CA2 +
12326391000
4492125
1715175

239083526238286750523
+ as3 CA3 +
1578596520362400000

3374081335517123191 17831164
+
3
+ as3 CF CA nf
36243287457300000
190575

1344
3009386129483453
3
+ as3 CF2 nf
3883209370425000 3025

43228502203851731 17746492
3
+ as3 CA2 nf +
2196562876200000
190575

453912946493
2752314359
+ as3 CA n2f
,
+ as3 CF n2f
420260754375
815051160

71203
2
16519839244157
GG
12
= as CA +
+ as nf + + as2 CA2 +
8190
3
549353259000

23220103
74823503
+ as2 CA nf
+ as2 CF nf +
2108106
36540504

6993119873800651614841
+ as3 CA3 +
42112541869725600000

49693541388602890695713 58085396
+
3
+ as3 CF CA nf
498238162032042432000
585585
294
4699124115250144376149 158
3
+ as3 CF2 nf
5480619782352466752000 507

74338654569222233539 57902906
3
+ as3 CA2 nf +
3871314390303360000
585585

50627726543561953
+ as3 CF n2f
45626205314289600

2635361358193
+ as3 CA n2f
,
759677078160
G
10

112
= as CF
495

133349533
88631998
74368
2
2 2
2
+ as CF CA
+ as CF +
+ as CF nf
80858250
121287375
735075

645584
165894431274725803
3
+ as3 CF CA2
48324383276400000
81675

4224421791031474951 645584
3 2
+ as CF CA
+
3
217459724743800000
27225

6457897459084371893 1291168
3 3
3
+ as CF +
326189587115700000
81675

18846629176433 1792
+
3
+ as3 CF CA nf
47069204490000
495

1792
529979902254031
152267426
3 2
3
2
3 + as CF nf +
+ as CF nf +
,
1294403123475000
495
363862125
79
70863259553
9387059226553
G
12 = as CF
+ as2 CF CA
+ as2 CF2 +
429
50645138544
13927413099600

14257247
+ as2 CF nf
115945830

24751969909767240119153 46179137
3
+ as3 CF CA2
14947144860961272960000
6441435

115986599183122809974023 46179137
+
3
+ as3 CF2 CA
5872092623949071520000
2147145

2165927305724992215894703 92358274
3
+ as3 CF3 +
113037783011019626760000
6441435

1264
64190493078139789
+
3
+ as3 CF CA nf
195540879918384000
429

1264
1401404001326440151
3
+ as3 CF2 nf +
13981172914164456000
429

13454024393417
,
+ as3 CF n2f +
41782239298800
28
379479917
373810079
G
10 = as nf
+ as2 CF nf
+ as2 CA nf +
165
125779500
150935400

926990216580622991 643396
3
+ as3 CF CA nf +
24162191638200000
27225

1091980048536213833 17712
+
3
+ as3 CF2 nf
54364931185950000
3025

483988
21025430857658971
+
3
+ as3 CA2 nf
1022738270400000
27225

1584713325754369
1669885489
+ as3 CA n2f +
,
+ as3 CF n2f +
2588806246950000
7906140000

79
G
12 = as nf
546

9256843807
653436358741
+ as2 CA nf +
+ as2 CF nf
3197294100
268572704400

4046032530021008148641959 171207527
3
+ as3 CF CA nf +
104630014026728910720000
8198190

2563
2960118366121154186145047
+
3
+ as3 CF2 nf
143866269286752252240000
507

129763817
10876559659107463644949
+
3
+ as3 CA2 nf
543532540398591744000
8198190

149081947693135635881
+ as3 CF n2f +
249119081016021216000

226617401255197
+ as3 CA n2f +
,
4399220898072000

12055
10 = as CF +
1386

19524247733
9579051036701
+ as2 CF CA +
+ as2 CF2
523908000
1331250228000

2451995507
27284
+ as2 fl02 CF nf
+ as2 CF nf
288149400
13476375

151796299
94091568579766453
+ as3 CF CA2 +
+
3
435681892800000
8004150

151796299
16389982059548833
3
+ as3 CF2 CA
465937579800000
2668050

151796299
2207711300808736405687
+
3
+ as3 CF3
127866318149354400000
4002075

48220
9007773127403
3
+ as3 CF CA nf
389001690000
693
295
296
75522073210471127 48220
+
3
+ as3 CF2 nf
1230075210672000
693

27995901056887
+ as3 CF n2f
11981252052000

1028766412107043 12544
3
+ as3 fl02 CF CA nf
5177612493900000 27225

209966063746798 12544
+
3
+ as3 fl02 CF2 nf +
485401171303125 27225

33230913134
,
+ as3 fl02 CF n2f
420260754375

423424
12 = as CF +
45045

19487270392267
5507868301548461
+ as2 CF2
+ as2 CF CA +
486972486000
731189187729000

90143221429
249775
+ as2 CF nf
+ as2 fl02 CF nf
9739449720
255783528

25648239313
1395004186652448755863
+
3
+ as3 CF CA2 +
6016642459027200000
1352701350

25648239313
98204412073910020058227
3
+ as3 CF2 CA
2634913356900224400000
450900450

25648239313
81630141011772791446330057
+
3
+ as3 CF3
4747586886462824323920000
676350675

3387392
25478252190337435009
3
+ as3 CF CA nf
1052912430329760000
45045

3387392
35346062280941906036867
+
3
+ as3 CF2 nf
526982671380044880000
45045

65155853387858071
+ as3 CF n2f
26322810758244000

12482
69697489543846494691
3
+ as3 fl02 CF CA nf
332158774688028288000 39039

12482
86033255402443256197
+
3
+ as3 fl02 CF2 nf +
219224791294098670080 39039

2566080055386457
+ as3 fl02 CF n2f
,
45626205314289600

180121
ns
= as CF +
14
18018

288858136265399
+ as2 CF CA +
6817614804000
297
481761665447
+ as2 CF nf
48697248600

22819142381313407
2 2
+ as CF
2924756750916000

92531316363319241549 720484
3
3
+ as CF CA nf
3685193506154160000
9009

126653245164236390142889 3663695353
3
2
+
3
+ as CF CA +
515927090861582400000
193243050

68167166257767019
+ as3 CF n2f
26322810758244000

96001333716903621488387 3663695353
3
+ as3 CF2 CA
2459252466440209440000
64414350

37908544797975614512733 720484
3 2
+
3
+ as CF nf
526982671380044880000
9009

40552395746064871242211709 3663695353
3 3
+
3 ,
+ as CF
2373793443231412161960000
96621525

2006299
6124093193824187 104674
3
C2,10 = 1 + as CF +
+ as2 CF CA +
138600
29045459520000
1155

561457267429757
+ as2 CF nf
15975002736000

558708799987324013 88798
2 2
+
3
+ as CF +
14760902528064000
1155

3584203788491
+ as2 fl02 CF nf +
15689734830000

10519793104
24110
21664244926039357214987
+
3
4
+ as3 CF CA nf
23550349033411200000
42567525
693

709221119965457939095237 14713925739913
3
+ as3 CF CA2 +
235503490334112000000
6243237000
151796299
190858
4 +
5
+
16008300
231

57084428047851551911 48220
3
2
+
3
+ as CF nf +
996360920644320000
18711

16350009304926933389608829 1430215936081
+
3
+ as3 CF2 CA +
8369431733412288000000
6163195500
151796299
22658
4
5
5336100
99

24110
1521387460036994061010049 3997754476
3 +
4
+ as3 CF2 nf
2720065313358993600000
42567525
693
298

3247779532370920623770610131 2182208825245282
+
3
+ as3 CF3
92155812816602703168000000
1622461215375
151796299
75212
4
5
+
8004150
99

6272
3566946294536415188593 79527463
3
3
4
+ as fl02 CF CA nf +
861140509985448000000
56600775
27225

148475806971656561
33008
+ as3 fl02 CF n2f
+
3
244642190336775000 735075

422577250875954453771617 12949105012
6272
3
2
3 +
4
+ as fl02 CF nf +
58772839806506826000000
11037151125
27225

3753913187503
81388
d abc d abc
448
+
+
3
5 ,
+ as3 fl11 nf
n
352066176000
606375
33

183473419
C2,12 = 1 + as CF +
10810800

21388499873332252399 1477711
2
3
+ as CF CA +
87742702527480000
15015

9127110915702407798941 1261726
2 2
+ as CF +
+
3
131745667845011220000
15015

57904356630607013
103555928663269
+ as2 fl02 CF nf +
+ as2 CF nf
1403883240439680
563286485361600

83712626229337204073275967
+ as3 CF CA nf
75885504678726462720000
1693696
36842282041
3
4
+
133783650
45045

163181620367687907864404054279
+ as3 CF CA2 +
45531302807235877632000000
641004330821357
25648239313
2642336
3 +
4 +
5
243486243000
2705402700
3003

2003755100099657438423887 3387392
+
3
+ as3 CF n2f +
28457064254522423520000
1216215

4843191986526849157300333804109 4948562279669051
+
3
+ as3 CF2 CA +
1709131279126616756611200000
97491891697200
25648239313
241094
4
5
901800900
1287

848389810670975600831798557
+ as3 CF2 nf
1130004151488672235776000
1326399435709
1693696
3 +
4
12174312150
45045

131770367393773533536363889790633
+ as3 CF3 +
3421680820811486746735622400000
25648239313
88004
42634681331415644
3 +
4
5
+
24926904127125
1352701350
99

86565903158314138806418902631
+ as3 fl02 CF CA nf +
26931765610480021619328000000
6241
3109476222803
3
4
3165321159000
39039

2780
1960867603733060624851
3
2
+
3
+ as fl02 CF nf
4404069467961803640000 95823

33528586559068805780843402423
+ as3 fl02 CF2 nf +
5290168244915718532368000000
6241
751723347541
3 +
4
791330289750
39039

abc
abc
d d
200
809917806143013559 622064791
3
+
3
5 ,
+ as fl11 nf
+
n
67966375276800000
851350500
13
ns
C2,14

90849502
= 1 + as CF +
4729725

1345455725874078602801 315626
2
3
+ as CF CA +
4913591341538880000
3003

1644267296654871017
+ as2 CF nf
35097081010992000

271010
31002322638187643268973
+
3
+ as2 CF2 +
301132955074311360000
3003

28812973254576289068812626927
+ as3 CF CA nf
22575937641921122659200000
31112773830559
360242
3
4
+
103481653275
9009

7823621156350047175125815627621
+ as3 CF CA2 +
1896378761921374303372800000
83168919211026563
3663695353
2738146
3 +
4 +
5
28974862917000
386486100
3003

2361466163828853440218087 720484
+
3
+ as3 CF n2f +
28457064254522423520000
243243

64116556842577537005311631547547
+ as3 CF2 CA +
17091312791266167566112000000
13682992796062061
3663695353
865562
3
4
5
81243243081000
128828700
9009
299
300

358451845381207175240043774137
+ as3 CF2 nf
377340672014967335875200000
12775152582499
360242
3 +
4
103481653275
9009

12924407779985031268428280066600921
+ as3 CF3 +
72710717442244093368131976000000
3663695353
9512108
3556746663996971701
3 +
4
5
+
1695029480644500
193243050
9009

d abc d abc
352
637395762233410021 78376866703
3
+
3
5 ,
+ as fl11 nf
+
n
49587712574400000
65553988500
21

CL,10 = as CF +
11

89670761 48
3
+ as2 CF CA +
8731800
11

163679
+ as2 CF nf
114345

1999510607 96
2 2
+ 3
+ as CF
528273900
11

415796
+ as2 fl02 CF nf
8085825

176183576988227323 55485434
+
3
+ as3 CF CA nf
1699159381920000
1216215

3760
2366034921481985137 95022195887
3
2
3 +
5
+ as CF CA +
6796637527680000
187297110
11

63272639
+ as3 CF n2f +
11320155

323139848004267269 22904191
14240
3 2
+
3
5
+ as CF CA
3354750574560000
17325
11

9048874326307637 1174256
3
+ as3 CF2 nf +
190368782604000
15015

13440
887562386698645967383 357031607224
3 +
5
+ as3 CF3
3166213592269728000
468242775
11

36224
68379915239899511
3
+ as3 fl02 CF CA nf
43491944948760000 147015

21670644503
+ as3 fl02 CF n2f +
156897348300

1373248
319520059852805113
+
3
+ as3 fl02 CF2 nf
282697642166940000 1911195
+ as3 fl11 nf
301

d abc d abc
160
5073093424963 1820773
3 +
5 ,
n
528099264000
363825
11
4
7126442885209 48
+ as2 CF CA +
3
CL,12 = as CF +
13
811620810000
13

2201663
12296141077867 96
+ 3
+ as2 CF nf
+ as2 CF2
1756755
5275535265000
13

16072451
5203557911
+ as2 fl02 CF nf
+ as3 CF n2f +
502431930
1027701675

61335054761825219657 2151978544
+
3
+ as3 CF CA nf
658070268956100000
52026975

5600
14897818968827338964411 5467652405147
3 +
5
+ as3 CF CA2 +
52645621516488000000
10145260125
13

1508964059584735294818791
+ as3 CF2 CA +
26349133569002244000000
3065115509662
21600
3
5
+
2029052025
13

47440688
12143703744427185053
3
+ as3 CF2 nf +
316848648015900000
675675

10180155542576
16091344678479668687707841
3 + 16005
+ as3 CF3
42817342049628646500000
10145260125

561855512
592278098533386728681
3
+ as3 fl02 CF CA nf
576664539388938000000 3381753375

435058406339681
+ as3 fl02 CF n2f +
5280810800265000

1570446544
65983065928499265747263
+
3
+ as3 fl02 CF2 nf
85634684099257293000000 3381753375

abc
abc
d d
160
503821438649257451 302982523
3 +
5 ,
+ as3 fl11 nf
n
62302510670400000
70945875
13

4
ns
= as CF +
CL,14
15

3736751546509 16
+ as2 CF CA +
3
486972486000
5

2263109
+ as2 CF nf
2027025

32
6706232197
+ 3
+ as2 CF2
4969107000
5

16947716309
14479744644122508496943
+
3
+ as3 CF CA nf
170858971648965600000
447972525
302

1704131316277822389591041
+ as3 CF CA2 +
7517794752554486400000
1552
47503400282843
3 +
5
82785322620
3

422957746
+ as3 CF n2f +
91216125

1016513978878471248683819 51799344112951
+
3 20165
+ as3 CF2 CA +
5269826713800448800000
30435780375

438969448
58345189864914275683
3
+ as3 CF2 nf +
1864532428708950000
6891885

640509641943719
5888
8328506007291432016890749
3 +
5
+ as3 CF3
17917410826921525920000
517408266375
3

abc
abc
32
d d
110339419075223314037 17420356529
3 + 5 ,
+ as3 fl11 nf
n
15793686454946400000
4682427750
3
G
C2,10
4352
= as nf
7425

651112454591 54
3
+ as2 CA nf
185952412800 55

72533010722807 108
+
3
+ as2 CF nf
6973215480000
55

18965
30367858943250477461
3
+ as3 CA n2f +
1739677797950400000
137214

114563089982050063118447 427785744377
+
3
+ as3 CA2 nf
669775952210904000000
7924108500
241994
8
4 + 5
+
27225
33

110850413975318446061177 7772983651
+
3
+ as3 CF CA nf
2487739251069072000000
7358100750
321698
244
4 +
5
27225
11

25164738348656825457229 2762978063
3
+ as3 CF n2f +
1399353328726353000000
1226350125

26284376777719892724358177 1782724946402
+
3
+ as3 CF2 nf
235091359226027304000000
77260057875
8856
1048
+
4
5
3025
33

abc
abc
11661390042871 29154339
d
4408
3 g 2 d
3 +
5 ,
+ as fl11 nf
NA
128024064000
107800
11
303
8110049
G
C2,12
= as nf
15135120

39540735563386331 11
2
3
+ as CA nf
10558130155372800 13

6258789011950819 22
2
+ 3
+ as CF nf
598533455520000
13

1574273
1102976289541525512630679
3
2
3
+ as CA nf +
62778008416037346432000
8049132

18295361799349570193991309242431 6726830671058291
+
3
+ as3 CA2 nf
102830377785469173455616000000
132943488678000
129763817
20
4 + 5
+
16396380
21

9699486762665487150445457223019
+ as3 CF CA nf
188522359273360151335296000000
171207527
4390
789241683301607
3
4 +
5
+
132943488678000
16396380
273

25016184592875203289560351501 13373036672
3
2
3
+ as CF nf +
1346588280524001080966400000
7193911725

194382513719914205037949320835327 542538591728921
+
3
+ as3 CF2 nf
1555309464005221248516192000000
33235872169500
2563
2220
4
5
+
1014
91

1656600471440498533 82799792129
d abc d abc
62436
g
3 +
5 ,
+ as3 fl11 n2f
NA
10297935648000000
180589500
91
G
CL,10

2
= as nf +
33

2460678191
2
+ as CA nf +
1056547800

509195549
+ as2 CF nf
704365200

4
140853814103239
3
+ as3 CA n2f
21965628762000
33

5906419
127219094296097749861
80
3 2
+
3 5
+ as CA nf +
1623699278087040000
24012450
11

148061845621707638477 210786157
320
3
3 +
5
+ as CF CA nf
2638511326891440000
31216185
11

491491787586698683 136
3
+ as3 CF n2f +
188465094777960000 429
304
320
9053411269935853949 860883476
+
3
5
+ as3 CF2 nf +
376930189555920000
156080925
11

abc
abc
112883693141257 347273501
d
1280
3 g 2 d
+
+ as fl11 nf
+
3
5 ,
NA
4526565120000
4365900
11

4
4978992299
98593150597
G
+ as2 CA nf +
+ as2 CF nf
CL,12
= as nf +
91
2685727044
179847793125

116919410865341069
8
+ as3 CA n2f
3
22683482755683750
91

16493287
96912603479263207273229
480
3
5
+ as3 CA2 nf +
1467873372990024000000
717341625
91

9006563627
1920
2084160567995424969918773
3 +
5
+ as3 CF CA nf
46709827690503978000000
2152024875
91

1696
49342300647723867029
3
+ as3 CF n2f +
24328035255470821875 6825

1920
281711596115081884853551 2347741862
+
3
5
+ as3 CF2 nf +
15569942563501326000000
717341625
91

abc
abc
2232852976776993919
d d
g
+
+ as3 fl11 n2f
NA
56638646064000000
115891012697
15776
3
5 .
+
993242250
91
Appendix C. Results for the moments of F3

25
535
415
2035
3ns = as CF +
+ as2 CF CA +
+ as2 CF nf
+ as2 CF2
6
27
108
432

62249 100
889433 55
3 + as3 CF CA2 +
+ 3
+ as3 CF CA nf
3888
3
7776
3

2569
311213
553
+ as3 CF n2f
+ as3 CF2 CA
1944
15552

203627 100
244505 110
+
3 + as3 CF3
+
3
+ as3 CF2 nf
7776
3
15552
3

abc
abc
d d
205
+ as3 nf
+
,
n
288

91
73223
7783
2891
+ as2 CF CA +
+ as2 CF nf
+ as2 CF2
5ns = as CF +
15
2700
1350
500

728
1414
38587
305342801
3 + as3 CF CA2 +
+
3
+ as3 CF CA nf
2000
15
1944000
75

1414
215621
63892213
3
+ as3 CF n2f
+ as3 CF2 CA
121500
2430000
25
305
5494973 728
51831073 2828
+
3 + as3 CF3
+
3
+ as3 CF2 nf
135000
15
3037500
75

abc
abc
931
d d
+
,
+ as3 nf
n
4050

1027
67710257
3745727
7ns = as CF +
+ as2 CF CA +
+ as2 CF nf
140
2116800
529200
2054
106801937
2257057261
2 2
3
+ as CF CA nf
3
+ as CF
16464000
106686720
35

219582793861 92741
1369936511
3
2
3
2
+
3 + as CF nf
+ as CF CA +
1185408000
4900
666792000

278223
629922436973
3
+ as3 CF2 CA
20744640000
4900

3150205788689 2054
3 2
+
3
+ as CF nf
62233920000
35

56527729
151689902637457 92741
d abc d abc
+
,
+
3 + as3 nf
+ as3 CF3
8712748800000
2450
n
508032000
9ns
1045
242855129
19247947
2
2
= as CF +
+ as CF CA +
+ as CF nf
126
6804000
2381400
6993510271
63405201707 4180
2 2
3
+ as CF CA nf
3
+ as CF
1000188000
2813028750
63

744184331602493 1253219
20297329837
3
2
3
2
+
3 + as CF nf
+ as CF CA +
3600676800000
66150
9001692000

5022136344247 1253219
3
+ as3 CF2 CA
150028200000
22050

1630263834317 4180
+
3
+ as3 CF2 nf
28005264000
63

13829238556849837 1253219
+
3
+ as3 CF3
793949234400000
33075

abc
abc
d d
13172190779
+ as3 nf
+
,
n
200037600000

31408
1448235599
512808781
ns
= as CF +
11
+ as2 CF CA +
+ as2 CF nf
3465
37422000
57629880

251264
2454220717793
1031510572686647
3
+ as2 CF2
+ as3 CF CA nf
332812557000
43568189280000
3465

151689577
390549244457621303
+
3
+ as3 CF CA2 +
1742727571200000
8004150

28869611542843
+ as3 CF n2f
11981252052000
306
43074459020106191 151689577
3
+ as3 CF2 CA
1198125205200000
2668050

1188145134622636787 251264
3 2
+
3
+ as CF nf
18451128160080000
3465

221430869576690083141 151689577
3 3
+
3
+ as CF
12786631814935440000
4002075

5083969985783
d abc d abc
+
,
+ as3 nf
n
116181838080000

874733
31236494566127
93360116539
ns
13
= as CF +
+ as2 CF CA +
+ as2 CF nf
90090
757512756000
9739449720

22445960639039759
+ as2 CF2
2924756750916000

90849626920977361109 3498932
3
3
+ as CF CA nf
3685193506154160000
45045

41070753377638233401027 3662719609
+
3
+ as3 CF CA2 +
171975696953860800000
193243050

66727681292862571
+ as3 CF n2f
26322810758244000

1400874681707762602284653 3662719609
3
+ as3 CF2 CA
36888786996603141600000
64414350

36688336888519925613757 3498932
3 2
+
3
+ as CF nf
526982671380044880000
45045

40795447722180713788820819 3662719609
3 3
+
3
+ as CF
2373793443231412161960000
96621525

62160363128061559
d abc d abc
+
,
+ as3 nf
n
1984334964852240000
C1ns = 1 + as CF [3]
+ as2 CF CA [23] + as2 CF nf [+4] + as2 CF2
21
+
2
80
3535
10874 440
+ 243 5 + as3 CF CA2
+
5
+ as3 CF CA nf +
27
3
27
3
230
1241 176
133 40
+ as3 CF2 CA +
3 + as3 CF2 nf
3
+ as3 CF n2f
27
9
3
18
3
abc
abc
11
3
d d
+ 83 ,
+ as3 CF3 + as3 nf
2
n
3
5
C3ns = 1 + as CF +
4
5209
4369
34763
333 + as2 CF nf
+ as2 CF2
+ 163
+ as2 CF CA +
144
864
10368

50
24877649 18539
+
3 4
+ as3 CF CA nf
699840
405
3

225337
55
92517547
3 + 4 + 3905
+ as3 CF CA2 +
699840
405
6

100
12125
+
3
+ as3 CF n2f
69984
81

55
222157399 206
+
3 4 2605
+ as3 CF2 CA +
559872
9
2

527
4292227
50
+
3 + 4
+ as3 CF2 nf
34992
9
3

2183
55
1041473
+
3 + 4 405
+ as3 CF3
373248
324
3

abc
abc
d d
19477
+ 53 ,
+ as3 nf
n
5184

523
C5ns = 1 + as CF +
90

30312449 276
2356859
3 + as2 CF nf
+ as2 CF CA +
324000
5
162000

14729261 188
+
3
+ as2 CF2
1620000
5

33317596477 1413442
364
+
3
4
+ as3 CF CA nf
122472000
14175
15

707
142639414763 30424087
3 +
4 + 5485
+ as3 CF CA2 +
163296000
28350
75

784091 728
+
3
+ as3 CF n2f +
54000
405

707
179060877287 512039
3
4 1805
+ as3 CF2 CA +
255150000
7875
25

28068155797 205712
364
+
3 +
4
+ as3 CF2 nf
127575000
4725
15

1414
4045486
29707273013
+ as3 CF3
+
3 +
4 3765
2187000000
10125
75

abc
abc
24801551 86
d d
+ 3 ,
+ as3 nf
n
17496000 45
307
308
48091
C7ns = 1 + as CF +
5040

502042084559 4917
20352710029
2
2
3 + as CF nf
+ as CF CA +
3556224000
70
889056000

972223395949 1836
+
3
+ as2 CF2 +
248935680000
35

1027
3373161364214023 4838581
3
+
3
4
+ as CF CA nf
6721263360000
33075
35

92741
14108
5386502616041647 134407981
3 +
4 +
5
+ as3 CF CA2 +
3360631680000
88200
9800
21

9986153999891 2054
+
3
+ as3 CF n2f +
336063168000
945

2316127855865945089 450138923
278223
4118
3 2
3
4
5
+ as CF CA +
1881953740800000
4630500
9800
21

1027
3455733053370931 483863
+
3 +
4
+ as3 CF2 nf
9801842400000
26460
35

92741
11224
906919428644261623 27358836137
+
3 +
4
5
+ as3 CF3
14637417984000000
37044000
4900
21

abc
abc
d d
1521158314519 529
+
3 ,
+ as3 nf
n
1920360960000 504

17761
C9ns = 1 + as CF +
1400

6247950073621 2858
363260060687
2
2
3 + as CF nf
+ as CF CA +
34292160000
35
12002256000

26950029280781 6698
+
3
+ as2 CF2 +
1008189504000
105

1532454569897512927 121711969
2090
3
+
3
4
+ as CF CA nf
2138802019200000
654885
63

15496996833630287207 7620848351
3
+ as3 CF CA2 +
6805279152000000
3969000
1253219
49634
4 +
5
+
132300
63

3023214594598871 4180
3
2
+
3
+ as CF nf +
68052791520000
1701

1111314791170240118909 4449188411
3
+ as3 CF2 CA +
571643448768000000
41674500
1253219
2678
4
5
44100
9
309
2090
80521228766527611043 22156972
3 +
4
+ as3 CF2 nf
157201948411200000
3274425
63

388781612300063841847 95897878894
+
3
+ as3 CF3
4001504141376000000
93767625
1253219
5092
4
5
+
66150
9

abc
abc
390642677252603
95839
d d
3
+
3 ,
+ as nf
n
756142128000000 141750

12815983
ns
= 1 + as CF +
C11
831600

56653343996617 104947
117825956269669
+ as2 CF CA +
3 + as2 CF nf
259334460000
1155
3195000547200

25436786950269217 84262
+
3
+ as2 CF2 +
461278204002000
1155

125632
5339200983355026311 206111663893
+
3
4
+ as3 CF CA nf
5825689309440000
936485550
3465

8531884151173
5458605391363920639632689
3
+ as3 CF CA2 +
1884027922672896000000
3745942200
151689577
209072
4 +
5
+
16008300
231

251264
58243229827221303967
+
3
+ as3 CF n2f +
996360920644320000
93555

52724308943557
37894668775960020571538737
3
+ as3 CF2 CA +
13600326566794968000000
576875098800
151689577
46958
4
5
5336100
99

2560572011
125632
7504061852485890170936687
3 +
4
+ as3 CF2 nf
10880261253435974400000
85135050
3465

156213323360401
358252596866786733349510147
+
3
+ as3 CF3
3544454339100103968000000
124804708875
151689577
49124
4
5
+
8004150
99

abc
abc
1182009270543683429
6758
d d
+
3 ,
+ as3 nf
n
3220560551577600000 14175

84292133
ns
= 1 + as CF +
C13
4729725

137082775015804598489 1480186
3
+ as2 CF CA +
545954593504320000
15015
310
301032882286150933
+ as2 CF nf
7019416202198400

5233449307353261454457 1210906
+
3
+ as2 CF2 +
60226591014862272000
15015

879328426004243254681619899
+ as3 CF CA nf
796797799126627858560000
1749466
3046809425749
3
4
+
12174312150
45045

37039408217872
645026845788679520269283647601
3
+ as3 CF CA2 +
185919486462879833664000000
14203364175
3662719609
3086186
4 +
5
+
386486100
3003

3498932
2043359170316436833750887
+
3
+ as3 CF n2f +
28457064254522423520000
1216215

2292102879924479
9081431864860519767694223551709
3
+ as3 CF2 CA +
2441616113038023938016000000
48745945848600
3662719609
6487178
4
5
128828700
9009

19506028401623230293412698913
+ as3 CF2 nf
22196510118527490345600000
1749466
313559848919
3 +
4
6087156075
45045

142217314466274683
298680730292949187763497559489677
+
3
+ as3 CF3
4277101026014358433419528000000
99707616508500
3662719609
235868
4
5
+
193243050
693

abc
abc
11802932
d d
65696095155620706347863
+
3 .
+ as3 nf
n
238358315978051068800000 33108075
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
D.J. Gross, F. Wilczek, Phys. Rev. D 8 (1973) 3633.

D.J. Gross, F. Wilczek, Phys. Rev. D 9 (1974) 980.
W.A. Bardeen, A.J. Buras, D.W. Duke, T. Muta, Phys. Rev. D 18 (1978) 3998.
E.G. Floratos, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 129 (1977) 66.
Nucl. Phys. B 139 (1978) 545.
E.G. Floratos, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 152 (1979) 493.
A. Gonzalez-Arroyo, C. Lopez, F.J. Yndurain, Nucl. Phys. B 153 (1979) 161.
A. Gonzalez-Arroyo, C. Lopez, Nucl. Phys. B 166 (1980) 429.
C. Lopez, F.J. Yndurain, Nucl. Phys. B 183 (1981) 157.
W. Furmanski, R. Petronzio, Phys. Lett. B 97 (1980) 437.
G. Curci, W. Furmanski, R. Petronzio, Nucl. Phys. B 175 (1980) 27.
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
311
R. Hamberg, W.L. van Neerven, Nucl. Phys. B 379 (1992) 143.

D.W. Duke, J.D. Kimel, G.A. Sowell, Phys. Rev. D 25 (1982) 71.
A. Devoto, D.W. Duke, J.D. Kimel, G.A. Sowell, Phys. Rev. D 30 (1984) 541.
D.I. Kazakov, A.V. Kotikov, Nucl. Phys. B 307 (1988) 721.
D.I. Kazakov, A.V. Kotikov, Phys. Lett. B 291 (1992) 171.
D.I. Kazakov, A.V. Kotikov, G. Parente, O.A. Sampayo, J.S. Guillen, Santiago de Compostela
Univ., US-FT-90-06 (90, rec. Jul.), p. 6.
W.L. van Neerven, E.B. Zijlstra, Phys. Lett. B 272 (1991) 127.
E.B. Zijlstra, W.L. van Neerven, Phys. Lett. B 273 (1991) 476.
E.B. Zijlstra, W.L. van Neerven, Nucl. Phys. B 383 (1992) 525.
S. Moch, J.A.M. Vermaseren, Nucl. Phys. B 573 (2000) 853, hep-ph/9912355.
S.A. Larin, T. van Ritbergen, J.A.M. Vermaseren, Nucl. Phys. B 427 (1994) 41.
S.A. Larin, P. Nogueira, T. van Ritbergen, J.A.M. Vermaseren, Nucl. Phys. B 492 (1997) 338,
hep-ph/9605317.
S.A. Larin, J.A.M. Vermaseren, Phys. Lett. B 259 (1991) 345.
J.A.M. Vermaseren, S. Moch, hep-ph/0004235.
G. Parente, A.V. Kotikov, V.G. Krivokhizhin, Phys. Lett. B 333 (1994) 190, hep-ph/9405290.
A. Kataev, A.V. Kotikov, G. Parente, A.V. Sidorov, Phys. Lett. B 388 (1996) 179, hepph/9605367.
W.L. van Neerven, A. Vogt, Nucl. Phys. B 568 (2000) 263, hep-ph/9907472.
J. Santiago, F.J. Yndurain, Nucl. Phys. B 563 (1999) 45.
W.L. van Neerven, A. Vogt, hep-ph/0006154.
J.A.M. Vermaseren, Symbolic Manipulation with F ORM Version 2, Tutorial and Reference
Manual, Computer Algebra Nederland, Amsterdam, 1991.
D. Fliegner, A. Retey, J.A.M. Vermaseren, hep-ph/9906426.
A.J. Buras, Rev. Mod. Phys. 52 (1980) 199.
S.G. Gorishnii, S.A. Larin, Nucl. Phys. B 283 (1987) 452.
K.G. Chetyrkin, F.V. Tkachev, Nucl. Phys. B 192 (1981) 159.
S.A. Larin, F.V. Tkachev, J.A.M. Vermaseren, NIKHEF-H-91-18.
S.A. Larin, T. van Ritbergen, J.A.M. Vermaseren, Prepared for 3rd International Workshop on
Software Engineering, Artificial Intelligence and Expert systems for High-energy and Nuclear
Physics, Oberammergau, Germany, October 48, 1993.
J.A. Gracey, Phys. Lett. B 322 (1994) 141, hep-ph/9401214.

Gravitational Lorentz violations and adjustment of

the cosmological constant in asymmetrically
warped spacetimes
Csaba Cski a,1 , Joshua Erlich a , Christophe Grojean b,c
a Theory Division T8, Los Alamos National Laboratory, Los Alamos, NM 87545, USA
b Department of Physics, University of California, Berkeley, CA 94720, USA
c Theoretical Physics Group, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA
Received 15 January 2001; accepted 4 April 2001
Abstract
We investigate spacetimes in which the speed of light along flat 4D sections varies over the extra
dimensions due to different warp factors for the space and the time coordinates (asymmetrically
warped spacetimes). The main property of such spaces is that while the induced metric is flat,
implying Lorentz invariant particle physics on a brane, bulk gravitational effects will cause apparent
violations of Lorentz invariance and of causality from the brane observers point of view. An
important experimentally verifiable consequence of this is that gravitational waves may travel with a
speed different from the speed of light on the brane, and possibly even faster. We find the most general
spacetimes of this sort, which are given by AdSSchwarzschild or AdSReissnerNordstrm black
holes, assuming the simplest possible sources in the bulk. Due to the gravitational Lorentz violations
these models do not have an ordinary Lorentz invariant effective description, and thus provide a
possible way around Weinbergs no-go theorem for the adjustment of the cosmological constant.
Indeed we show that the cosmological constant may relax in such theories by the adjustment of the
mass and the charge of the black hole. The black hole singularity in these solutions can be protected
by a horizon, but the existence of a horizon requires some exotic energy densities on the brane.
We investigate the cosmological expansion of these models and speculate that it may provide an
explanation for the accelerating Universe, provided that the timescale for the adjustment is shorter
than the Hubble time. In this case the accelerating Universe would be a manifestation of gravitational
Lorentz violations in extra dimensions. Published by Elsevier Science B.V.
E-mail address: cmgrojean@lbl.gov (C. Grojean).

1 J. Robert Oppenheimer fellow.
0550-3213/01/$ see front matter Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 7 5 - 4
C. Cski et al. / Nuclear Physics B 604 (2001) 312342
313
1. Introduction
The physics of extra dimensions has recently attracted renewed interest, mainly for the
following three reasons:
The existence of extra dimensions could provide a stable large hierarchy between the
scale of particle physics (the TeV scale) and the scale of gravity (the Planck scale)
[1,2].
One can obtain the 4D Einstein equations from a higher dimensional setup without
compactification or a need for a stabilization mechanism [24], and the expansion of
the brane driven by matter follows the usual 4D Friedmann equations [5].
One may hope that the cosmological constant problem could be partly resolved due
to extra dimensional physics [614].
Randall and Sundrum (RS) studied spacetimes with a single warped extra dimension,
for which the metric is of the form
ds 2 = eA(y) dx dx + dy 2 .
(1.1)
They showed that such geometries can have interesting consequences for particle physics
and gravity. In these scenarios the standard model fields are usually assumed to live at
a particular point in the extra dimension, called a 3-brane. Metrics of the form (1.1) can
reproduce 4D Einstein gravity on the brane without compactification, and can produce a
large hierarchy between the scales of particle physics and gravity due to the appearance of
the warp factor.
In this paper we study the most general backgrounds with one extra dimension, assuming
3D rotational invariance is still maintained 2 (and with the additional assumption that the
only sources in the bulk are a bulk cosmological constant and a U (1) gauge field). The
metric of such backgrounds can be generically written as,
ds 2 = a 2 (r, t) dt 2 + b2 (r, t) d x 2 + c2 (r, t) dr 2 .
(1.2)
However, when the sources are restricted to a bulk cosmological constant and a bulk gauge
field, a five-dimensional version of Birkhoffs theorem holds, and such a metric can always
be transformed into the form,
ds 2 = h(r) dt 2 + r 2 d 2 + h(r)1 dr 2 .
(1.3)
Here d is the unit metric of the 3D sections r, t = const., and h(r) describes a black
hole spacetime. Examining (1.3) one can see that the novel property of such a general 5D
spacetime compared to conventional warped background (1.1) is that the warp factors for
the space and time components of the 4D sections r = const. are generically different. We
will refer to metrics of this form as asymmetrically warped. The induced metric at the 4D
sections may still be flat, implying that (up to quantum gravitational corrections) particle
physics on the brane will see a Lorentz invariant spacetime. However, since every 4D
2 The breaking of 3D rotational invariance in the bulk would transmit its breaking in gravitational interactions
on the brane in contradiction with the isotropy of the cosmic microwave background.
314
section of the metric (1.3) will have a differently defined Lorentz symmetry (one needs
to rescale the time coordinate differently at different points along the extra dimension
to maintain a speed of light c = 1 along each section), the spacetime (1.3) globally
violates 4D Lorentz invariance, leading to apparent violations of Lorentz invariance from
the brane observers point of view due to bulk gravity effects. These Lorentz violations
however produce different effects from explicit Lorentz violations that are sometimes
introduced (see, for example, [15]) in particle physics. In fact, since at the classical level
they only affect gravity, the most striking consequence of this setup would be that the
speed of gravitational waves would be different from the speed of light, which could cause
apparent violations of causality from the 4D brane observers point of view. This possibility
has already been pointed out by Klbermann and Halevi [16], Chung and Freese [17],
Ishihara [18] and Chung, Kolb and Riotto [19]. Note also that the existence of different
speeds of propagation for gravitational and electromagnetic interactions has some common
features with four-dimensional theories of gravity with two light cones as proposed in [20].
In asymmetrically warped spacetimes the reason for the different speeds of propagation for
graviton and photon is that in the background (1.3) the speed of light along the brane is
changing as one is moving along the extra dimension. Thus this setup is analogous to a
medium with a changing index of refraction. If the speed of light away from the brane
is increasing, then by Fermats principle the geodesic between two points on the brane
will bend into the bulk, and the gravitational wave which is not forced to propagate on
the brane will arrive faster than the light signal which is stuck to the brane. In fact, this
difference in the speed of electromagnetic and gravitational waves could be viewed as a
generic prediction of extra dimensions, which can be experimentally verified [19,20] once
gravitational waves are observed by LIGO, VIRGO or LISA.
Asymmetrically warped solutions of the form (1.3) do appear in string theory when
considering near extremal brane backgrounds. In this paper however we will not attempt
to embed the asymmetrically warped spacetimes into string theory by solving the full
set of supergravity equations, but we will rather assume that one can start with a fivedimensional low-energy effective theory which may originate from a particular string
theory compactification (or some other high energy theory).
It has been widely recognized starting with the work of Rubakov and Shaposhnikov [6],
that extra dimensions provide a new approach to the cosmological constant problem: with
the presence of extra dimensions one no longer needs to ensure the vanishing of the 4D
vacuum energy on the brane in order to obtain a static flat brane. The non-vanishing
brane tension can be balanced by the bulk curvature, exactly as it happens, for example, in
the RandallSundrum model. However, in the RS background a tuning between the bulk
cosmological constant and the brane tension is required in order to achieve this balance.
The second important consequence of the metric (1.3) is that it contains parameters in
addition to the ones contained in the solution discussed by RS, namely the mass, charge and
location of the black hole with respect to the brane, which can be thought of as integration
constants for the most general solution in the bulk. Thus one may hope that the fine-tuning
required in the original RS setup in order to ensure the vanishing of the 4D effective
cosmological constant could be eliminated. This would be similar in spirit to the self-
315
tuning models [7,8], where a bulk scalar is coupled to the tension of the visible brane,
and allows the transfer of curvature into the bulk if the brane tension is adjusted, so as
to maintain flatness of the brane. In our case, the adjustment mechanism would imply that
once the brane tension goes through a phase transition, the black hole would adjust its mass,
charge and location in order to balance the new stress tensor on the brane. One important
ingredient in the model of [7,8] was that for a particular coupling of the bulk scalar field
to the brane tension the only maximally symmetric solution to the equations of motion
was the flat brane solution, thus making it plausible that the time dependent solution may
relax to the flat brane. In the black hole backgrounds this feature is automatically present,
without having to tune a coupling: for generic equations of state we will show that the only
maximally symmetric brane sections correspond to the flat branes. For simplicity we will
assume that the spacetime is Z2 symmetric around the brane.
There are two reasons why this approach based on bulk black hole spacetimes may be
preferable compared to the scenario in [7,8]:
Due to the gravitational Lorentz violations explained above, there is no ordinary
Lorentz invariant low-energy effective 4D description of this model, and thus
Weinbergs no-go theorem [21] for 4D adjustment mechanisms of the cosmological
constant cannot be applied here. This reasoning is similar to that used to argue that
models of quasilocalized gravity [22] may be relevant to the cosmological constant
problem.
One may hope that due to the appearance of black holes, the unavoidable [11] naked
singularities appearing in [7,8] would be replaced by black hole singularities shielded
by a horizon.
We will find that it is in fact possible to protect the singularities by a horizon, and the
spacetime can be cut at the location of the horizon without reintroducing fine-tuning
into the theory (that is, without the need to regulate the spacetime with, for example,
an additional brane as in [10]). However, we find that such horizons are consistent with
the presence of a Z2 symmetric brane only in the case of charged black holes, and
require the presence of some exotic (but not fine-tuned) energy density on the brane. Of
course this exotic energy density is not generically required for an asymmetrically warped
background, only for an adjusting solution with a horizon.
In order to gain some insight on how the effective gravity theory would behave from the
brane observers point of view, we investigate some of the basic features of the cosmology
of these models. We find that the Friedmann equation does contain the ordinary expansion
term, but there are additional contributions due to the mass and charge of the black hole,
which was also pointed out in [23,24]. These terms could be useful in explaining the
apparent acceleration of the Universe [25], however in order to obtain a viable cosmology
and solve the cosmological constant problem at the same time the adjustment timescale of
the mass and charge of the black hole must be shorter than the Hubble time.
The paper is organized as follows: in Section 2 we first show what the generic form of the
solutions in the bulk is, and then introduce the brane by solving the junction conditions.
We examine the fine-tuning of the parameters and the existence of a horizon. We close
Section 2 by showing that there are no other maximally symmetric solutions. Section 3
316
is devoted to the physical consequences of the asymmetrically warped spacetimes. We

calculate the geodesics and the speed of gravitational waves, then show how a graviton
zero mode would look like in these theories, and finally consider the cosmology of these
models.
2. Flat branes in black hole backgrounds

Our aim is to find general 4D Lorentz invariant brane (flat brane) solutions to
Einsteins equations in five dimensions, and to investigate the following issues: is it
possible to find flat brane solutions without naked singularities, without fine-tuning the
parameters of the theory, and what would the general properties of such spaces be? In
this section we will first discuss the form of the solution in the bulk, and then impose the
condition for a static brane. We will see that the pure gravity theory will always require a
fine-tuning between the parameters of the theory for a solution to exist. However, if one
also includes a gauge field in the bulk, the situation changes, and one can find flat solutions
for a range of parameters with no tuning. However, for these solutions one of the following
two limitations will apply: either there is no horizon but a naked singularity, or if one does
want to ensure the existence of a horizon one has to assume the presence of some exotic
matter on the brane.
2.1. The solution in the bulk
We model the 5D bosonic sector of the theory by a graviton and a cosmological constant
bk as in the RandallSundrum model, and in addition a U (1) gauge field propagating in
the bulk we will assume that the 4D matter living on the brane is neutral under this
gauge symmetry. So the action describing the dynamics of the system is given by, 3

1
1
5
MN
S = d x |g5 |
R FMN F
bk
4
252

+ d 4 x |g4 | Lmat g , (m) ,
(2.1)
where g5 and g4 are, respectively, the determinants the 5D metric gMN and the 4D metric
induced on the brane, g ; (m) denote some matter fields localized on the brane and
Lmat describes their interactions, and we will assume that this matter can be described as a
perfect fluid of energy density and pressure p; 52 defines the 5D Planck scale. Note that
the brane Lagrangian Lmat includes the brane tension = p.
We will assume that the solution is homogeneous and isotropic in the three spatial
directions of the brane. The general ansatz is,

ds 2 = n2 t, r d t 2 + a 2 t, r dk2 + b2 t, r d r 2 ,
(2.2)
3 Our conventions correspond to a mostly positive Lorentzian signature ( + +) and the definition of the
curvature in terms of the metric is such that a Euclidean sphere has positive curvature. Bulk indices will be
denoted by Latin indices (M, N, . . .) and brane indices by Greek indices (, , . . .).
317
where dk2 = d 2 /(1 kL2 2 ) + 2 d22 is the metric of the spatial 3-sections, with
curvature parameter k, L being a parameter with dimension of length that will be set to
the length scale given by the cosmological constant in the bulk. The brane is located at
r = 0. The 3D sections r , t = const. are either a plane (k = 0), a unit sphere (k = 1) or a
unit hyperboloid (k = 1). The brane will be expanding or contracting as long as the scale
factor a(t, 0) explicitly depends on time. Some of the features of metrics of the above form
(2.2) have also been independently investigated in Refs. [16,17,19].
In the case when there is no gauge field in the bulk, Kraus [26] and later Bowcock et
al. [27] have shown that a generalization of Birkhoffs theorem [28] holds in the bulk,
which implies that there exists a system of coordinates where the 5D metric is static while,
in general, the brane is moving and expanding. In this system of coordinates, the most
general metric is,
ds 2 = h(r) dt 2 + l 2 r 2 dk2 + h(r)1 dr 2 ,
(2.3)
where
r2
1
(2.4)
2 , l 2 = 52 bk .
2
l
r
6
The new coordinates r and t are functions of the brane-based coordinates r and t
and thus the location of the brane is parametrized by r = R(t) and is solution of a
differential equation we will give in the next section. The bulk geometry describes an AdS
Schwarzschild hole (a black wall for k = 0) located at r = 0 and spreading in the three
other spatial directions. The parameter is interpreted as the mass (in units where the fivedimensional Planck scale is equal to one) of the black hole. When = 0, the metric simply
describes 5D anti-de Sitter space and for a non-vanishing positive the r = 0 singularity
is hidden behind a horizon r = rh > 0 where the metric becomes degenerate h(rh ) = 0.
Birkhoffs theorem can also be generalized when the gauge field is turned on. This way
we obtain the AdSReissnerNordstrm metric (see also [23]), which has the same form
as (2.3), except that now the function h also depends on the charge Q of the black hole:
h(r) = k +
r2
Q2
(2.5)
+
.
l2
r2
r4
The non-vanishing component of the bulk field strength tensor F of the gauge field satisfies
the Bianchi identities and is given by,
6Q
Ftr =
(2.6)
.
5 r 3
Later we will mainly be interested in the case k = 0 because in that case the (3 + 1)dimensional sections r = const. are flat. If the charge of a ReissnerNordstrm black hole
is too large, there is no horizon. In our solutions with k = 0, the existence of an inner and
an outer horizon gives the following upper bound on the charge
h(r) = k +
4 32
(2.7)
l .
27
This follows from the fact that the position of a horizon corresponds to the square root of a
positive root of the cubic function f3 (x) = x 3 / l 2 x + Q2 . As long as its discriminant
Q4 <
318
is negative, f3 will have three roots and, as we will argue in Section 2.4, two of them are
positive and thus they are associated to two horizons.
2.2. Matching at the brane
We will assume that a codimension one brane separates two regions of the above
discussed 5D black hole spacetimes. That is, just like in the original RS scenario, we
are gluing two slices of the metric together at the position of the brane. In order for
this to be possible, one has to satisfy the Israel junction conditions (also known as the
jump equations) at the brane. We will restrict ourselves to solutions with a Z2 symmetry
between the two sides of the brane which means that we study the expansion of a brane
located at a fixed point of a Z2 orbifold. A simple way of defining a Z2 symmetric
spacetime by gluing patches together has been explained in Ref. [29]. The idea is that
for a metric of the form
ds 2 = A2 (r) dt 2 + B 2 (r) d2 + C 2 (r) dr 2 ,
(2.8)
one can find another solution using the fact that one still has reparametrization invariance in
r ) = A(f (r )), B(
r ) = B(f (r )), C(
r ) = C(f (r ))f (r ) still
the coordinate r, and thus A(
solves the Einstein equations. Thus one can use this invariance to generate new solutions
by gluing a solution to its image under the reparametrization. A particularly simple way of
doing this is by picking f (r) = r02 /r, and the identification
B(r) = B0 (r),
C(r) = C0 (r); for r r0 ,
A(r) = A0 (r),
2
2
2 2
r0
r0
r r
A(r) = A0
,
B(r) = B0
,
C(r) = C0 0 02 , for r r0 ,
r
r
r r
(2.9)
which will automatically ensure the continuity of the metric functions, and give an r
r02 /r Z2 symmetry of the metric. This is analogous to the y y Z2 symmetry in
the RandallSundrum model, and in fact coincides with it upon doing the coordinate
transformation r = ley/ l . From now on we will always assume the existence of such
a Z2 symmetry in order to simplify the equations and we will apply the previous method
on the metric defined by (2.5) between the black hole and the brane.
In a general case, the brane will not be static, but instead it will be moving in the bulk
spacetime, expressing the fact that the observed brane Universe is expanding (or shrinking).
The junction conditions for this case have been worked out in detail in Refs. [26,27,30,31].
It is assumed that the brane follows a trajectory R( ) in the above bulk spacetime. The
proper time as observed on the brane is defined by the equation

t2 h R( ) R 2 ( )h1 R( ) = 1,
(2.10)
which ensures that the induced metric on the brane will be of the FriedmannRobertson
2 = d 2 + R( )2 d 2 . Keeping the interior (the region next to r = 0)
Walker form dsind
k
319
of the BH spacetime to prevent the volume of the extra dimension to diverge, 4 one then
obtains the junction conditions [27]: 5

6
= 2
h(R) + R 2 ,
(2.11)
5 R
2R + h (R)
2
+p=
,
(2.12)
3
52 h(R) + R 2
where is the energy density of the brane, while p is its pressure. That is, the energy
momentum tensor on the brane is given by TBA = diag(, p, p, p, 0)( grr (r R( ))).
As long as the brane is moving, i.e., R = 0, one combination of these jump equations is
just equivalent to the energy conservation equation on the brane,
R
= 0,
R
and the remaining equation simply reads,
+ 3( + p)
(2.13)
R 2
1
h(R)
(2.14)
= 54 2 2 .
2
R
36
R
Note that in the case we will be mostly interested in of a static brane (with also constant),
the conservation equation is trivially satisfied and in this case the two jump equations are
really independent and the equations simplify. One way of obtaining these equations would
be to simply set R = R = 0 into Eqs. (2.11), (2.12). Alternatively, one can use the simple
jump conditions derived by Bintruy et al. [33] for a general metric of the form
ds 2 = n2 (t, r) dt 2 + a 2 (t, r) dk2 + b2 (t, r) dr 2 ,
(2.15)
where in our case n = h(r), a = r/ l and b = 1/ h(r). The conditions for a static brane
at r = r0 are then simply [33]:
2
[a ]
= 5 ,
a|b|
3
2
[n ]
= 5 (2 + 3p),
n|b|
3
(2.16)
where the above functions should be evaluated at the location of the brane and [f ] stands
for the jump of the derivative of the function f around the brane: [f ] = lim10 (f (r0 +
(r 1)). For the above explained Z symmetric construction [a ] = 2/ l, [n ] =
1) f
0
2
h / h, where we have assumed that for r < r0 we are using the slice of the black hole
metric that includes the singularity. Thus the jump conditions that a static brane has to
satisfy are given by

6 h(r0 ) = 52 r0 and 18h (r0 ) = 54 (2 + 3) 2 r0 ,
(2.17)
4 If one wants to keep the exterior (the region next to r = +) of the BH spacetime, both signs of the jump
equations (2.11), (2.12) have to be flipped. We will discuss this issue in more detail at the end of Section 2.4.
5 Similar differential equations have been obtained in [32] in a different context where the brane just probes the
embedding spacetime and moves along a geodesic. The following equations ensure the consistency of Einstein
equations, and are the Israel junction conditions for this setup.
320
where we have defined = p/. Of course these equations agree with the more general
jump equations (2.11), (2.12). Note that the energy density on the brane, , has to be
positive to cut away the infinity.
2.3. The AdSSchwarzschild black hole solutions
Let us now apply the jump equations for a static brane obtained above to the simple
AdSSchwarzschild black hole case (no vector field in the bulk). Our motivation is
to find a vacuum solution with 4D Lorentz invariance on the brane and thus we will
study static brane solutions 6 with a vanishing induced curvature on the three-dimensional
spatial sections (k = 0). We will postpone consideration of other metrics with maximally
symmetric 4D sections to Section 2.6. In an AdSSchwarzschild black hole spacetime,
the jump equations (2.17) are:
2

2
r0
r0
4 2 2
36 2 + 2 = 54 (2 + 3) 2 r02 .
36 2 2 = 5 r0 ,
(2.18)
l
l
r0
r0
Combining these two equations we obtain an expression for the black hole mass :
1 4
(2.19)
(1 + ) 2 r04 .
24 5
The black hole singularity is shielded by a horizon if its mass, , is positive; otherwise
there is a naked singularity in the metric at r = 0. This condition then translates into <
1. Note that this condition violates the positive energy condition on the brane. However,
it is not immediately obvious, whether such an exotic matter would necessarily lead to an
instability. < 1 could, for example, be obtained if in addition to a positive brane tension
there is a small negative matter energy density (which is not accompanied by pressure,
that is = 0 for this component). Just as a negative tension brane at an orientifold type
fixed point is stable, this violation of the null energy condition at the Z2 symmetric brane
might not be harmful. This issue clearly calls for further investigations. As a corollary of
(2.19), one can also see that with an ordinary brane tension ( = 1), one cannot obtain
a horizon. 7 Moreover, even for the equation of state < 1, the solution is fine-tuned.
The reason is that from the jump equations (2.18) can be expressed in terms of the AdS
length l as
=
54 2
1
,
=
36
r04 l 2
(2.20)
which when substituted back into (2.19) gives the relation 8

6 There is another way to obtain a flat brane solution since a Minkowski metric can be written as an open FRW
space (k = 1) with a scale factor linear in the cosmic time. However, as it will become evident in our general
analysis in Section 2.6, such a solution will not satisfy the jump equations except in some very special cases.
7 A static brane in a Schwarzschild black hole background with a horizon was found in [34]. However, this
brane is not 4D Lorentz invariant but of positive spatial curvature (k = +1).
8 Of course this relation makes sense only if < 1/3. For other equations of state, it is impossible for the
brane to remain static in an AdSSchwarzschild black hole spacetime.
72 1
,
1 + 3 l52
321
(2.21)
which is a fine-tuning for the size of the energy density on the brane, and is the analog
of the RandallSundrum fine-tuning condition for the brane tension of the positive tension
brane. Thus in the presence of the Z2 symmetry the case with a simple AdSSchwarzschild
black hole requires fine-tuning, and is not adjusting its vacuum energy to zero.
2.4. Self-tuning AdSReissnerNordstrm black holes
We have seen above that the simple AdSSchwarzschild black hole solution requires
fine-tuning. We will now show that the situation is different for the AdSReissner
Nordstrm black hole. First we note that there are no additional jump conditions due
to the presence of the bulk gauge field, as long as the SM matter living on the brane
is not charged under this gauge symmetry. The reason is that the Maxwell equations in
the bulk are first order differential equations in terms of the field strength FAB . Thus
there are no second derivatives appearing in the equation, and as long as there are no
delta function sources for this gauge field on the brane (which is ensured with the above
assumption of SM not transforming under the bulk U (1) gauge symmetry), there will be
no additional jump conditions. One may wonder how this fits with the fact that the only
non-vanishing component of the field-strength tensor is Ftr , which from the point of view
of the underlying vector field AM is obtained by taking a derivative with respect to r, and
naively seems to be odd under the Z2 symmetry. The point is that, to have an action Z2
invariant, one has to choose the different components of AM to have different properties
under the Z2 . However, the Z2 parity of Ar is a matter of choice. The Lagrangian (2.1)
will be invariant under the Z2 symmetry with both positive or negative parity assignments
for Ar . If we want to avoid the appearance of an extra jump condition for the gauge fields
one should choose Ar to be even and At,x to be odd under the Z2 parity. Then Ftr is
even, and we recover the above conclusion that there is no additional jump condition for
the gauge field. In fact, such vector fields are indeed present in supergravity theories [35].
However, with such a parity assignment there is an additional ChernSimons term allowed
in the action. The AdSRN solution remains a solution in the presence of this extra term
in the Lagrangian, but it will likely not be the most general solution anymore. It would be
worthwhile to study the most general solutions in the presence of the additional Chern
Simons term that we are setting to zero in this paper.
The remaining part of this section is devoted to the embedding of a flat brane in a charged
black hole background. Not only can such a solution be found without fine-tuning of any
parameter of the action, but in some regions of the plane (, ) describing the brane, the
singularity of the black hole will be protected by two horizons. We have summarized the
results in Fig. 1.
The jump equations for a static brane embedded in a charged black hole spacetime are
now
322
Fig. 1. Phase diagram of a brane sitting in a bulk black hole background. In the upper left zig-zagged
region, the brane will expand. In the other parts of the diagram, a brane can remain static and its
induced metric is 4D Lorentz invariant. In the other filled regions, the black hole singularity is hidden
by two horizons; however for > 0 these horizons are developed in the part of the BH spacetime
cut by the Z2 orbifold. Fig. 2 shows the displacement of the horizons when moving along the dashed
lines varying the energy density or the equation of state on the brane. The region < 1 has been
magnified.

r02
Q2
36 2 2 + 4 = 54 2 r02 ,
l
r0
r0
2

r0
Q2
36 2 + 2 2 4 = 54 (2 + 3) 2 r02 .
l
r0
r0
(2.22)
The existence of the charge as a second constant of integration allows us to evade the
previous fine-tuning because the two jump equations simply fix the mass and the charge of
the black hole (in the case of a flat brane, i.e., k = 0) in terms of brane parameters and r0 :

1
1
= 3 l 2 + 54 2 r04 ,
(2.23)
Q2 = 2 l 2 + 54 (1 + 3) 2 r06 .
36
72
When the parameter is too small, < 1/3, the positivity of Q2 requires

72 1
0 =
.
1 + 3 l52
(2.24)
The black hole singularity at r = 0 will be hidden behind a horizon provided the
inequality (2.7) is fulfilled, which, in terms of the equation of state and the energy density
on the brane, reads 9
9 Note that this inequality automatically implies that the mass of the black hole is positive.
323
Fig. 2. Displacement of the horizons when varying the energy density or the equation of state on
the brane. Both the inner and outer horizons are shown in the plots. (a) is varying in the region
1. When reaches the critical point of a vacuum energy equation of state, the two horizons hit
the black hole singularity; (b) the energy density is varying with < 1 fixed. The two horizons are
between the black hole singularity and the brane. When goes to , the two horizons degenerate
and when goes to 0 , the inner horizon reaches the singularity; (c) fixed, varying in the region
> 0; (d) > 0 fixed, varying. The two horizons belong to the cut region of the BH spacetime.
When goes to 0 , the two horizons degenerate.

2
3
1 4 2
1 4
1 4 2
1 4
2
2
2
2
l 5 + 5 (1 + )
< l 5 + 5 (1 + )
.
36
24
36
36
(2.25)
We immediately see that with an ordinary brane tension equation of state, = 1, this
equation can never be satisfied. However, as long as = 1, there may exist in general
an interval for the energy density where the singularity is protected behind inner and outer
horizons. The constraint (2.25) for the existence of horizons can be written as a quadratic
inequality for the energy density,
4+

1 2 4
1 4 8 3 4
l 1 + 6 32 2
l < 0,
36 5
324 5
(2.26)
324
whose discriminant is 2 = (1 + )3 (1 + 9)58 l 4 /1296. When the discriminant 2 is

positive, the quadratic equation will have two real roots. It can be checked that both
roots are positive when 1, both are negative when 1/9 0 and only one
is positive when 0. Since only positive roots can correspond to a physical value of 2 ,
we conclude that if is positive, the inequality is satisfied for large enough value of the
energy density while in the negative range, has to be less than 1 and the interval for
the energy density where two horizons are developed is included in
< < + ,
with
6
= 2
l5

1
2 (1 + )3 (1 + 9) .
1
+
6
3
83
(2.27)
However this interval has to be reduced further since we need to require that Q2 computed
from (2.23) is positive, i.e., < 0 = 72/(1 + 3)/(l52 ) which only partially overlaps
with the allowed interval. 10 In summary, the singularity at r = 0 will be protected by
horizons iff
( > 0 and > )
or ( < 1 and < < 0 ).
(2.28)
We still have to impose that these horizons sit between the singularity and the brane where
the space is cut. This last condition translates in an upper bound for the parameter
defining the equation of state. This condition is obtained by studying the position of the
brane with respect to the positions of the horizons that correspond to the positive root of
the cubic function f (x) = x 3 / l 2 x + Q2 . It is worth noticing that the sign of f is
changing when crossing a horizon but, from the first jump equation (2.17), we already
know that f is positive at the brane. Moreover, f (0) = < 0 and the sign of f flips
only between the two horizons. Thus it is enough to require that f is positive at the brane,
i.e., 3r04 / l 2 > 0 which from the expression of simply reads < 0.
In conclusion, it is possible without any fine-tuning to embed a static brane in a charged
black hole bulk. Moreover the singularity of this background will be hidden behind
horizons if
< 1

1
2 + (1 + )3 (1 + 9) < <
1
+
6
3
83
72 1
.
1 + 3 l52
(2.29)
Pictorially, the phase diagram of a brane in bulk black hole background is summarized in
Fig. 1. Clearly, the same comments apply here for the < 1 matter as for the AdS
Schwarzschild case. We should note, that it is not clear whether or not the existence
and
6
l52
10 In order to have two horizons, we also need the roots of the cubic equation associated to h to be positive.
When the discriminant is negative, i.e., the inequality (2.25) is satisfied, the cubic equation has three real roots
x1 , x2 and x3 with the properties: x1 + x2 + x3 = 0, x1 x2 + x2 x3 + x3 x1 = l 2 /r04 and x1 x2 x3 = l 2 Q2 /r06 .
Since furthermore the sign of the cubic function at the origin is related to the sign of Q2 , we conclude that in
the interval (2.28), the black hole singularity will be shielded by two horizons. For < 0 and > 0 , only one
horizon would be present but these solutions are non-physical since Q2 < 0.
325
of such exotic matter is a generic requirement for a more complicated asymmetrically

warped background (which could be obtained by introducing more fields into the bulk) to
be self-adjusting. The above arguments seem to be specific enough to hope that a more
complicated theory could avoid the presence of such matter.
So far we have only discussed solutions which include the black hole singularity and
cutting away, by the Z2 orbifold symmetry, the region close to infinity. The motivation was
to obtain an extra dimension of finite volume. The infinity of the black hole spacetime
being asymptotic to an AdS space, just like in a RS model with a single negative brane,
would give a divergent effective 4D Planck scale and therefore was discarded. However,
as suggested in our previous analysis, the possibility to find a brane sitting between the
singularity and the two horizons brings another way to prevent the 4D Planck scale to
diverge: the singularity region is now cut away requiring a negative energy density on
the brane but the space naturally ends at the inner horizon so that the 4D physics on the
brane remains insensitive to the region beyond this inner horizon. Since the signs in the
jump equations (2.17) flip, the previous formulas are also modified. First the mass and the
charge of the BH are now given by

4
1
+ 2 r04 ,
= 3 l 2 54
36
3

1
Q2 = 2 l 2 54 (1 + ) 2 r06 .
(2.30)
24
The horizons will belong to the kept part of the space iff 4/3. And the constraint for
these horizons to exist becomes

6
1
2 27(1 + 3)3 (11 + 9) . (2.31)
< 2
9
37
+
42
+
9
l5 8(4 + 3w)3
Thus the situation is actually quite similar to the previous construction and requires
some exotic vacuum on the brane. Furthermore there is no way to consider these solutions
as perturbations around the RS setup since the allowed region for the parameter is
disconnected from the tension type equation of state. In the remaining part of the paper,
we concentrate on the first type of solutions namely cutting the spacetime near infinity and
keeping the region that includes the singularity.
2.5. Cutting the space at the horizon and the effective cosmological constant
We would like to show in this subsection that if we cut the space at the horizon, the 4D
effective vacuum energy, computed as the integral of the action over the extra dimension,
vanishes. This result may be expected because we were able to construct a solution to the
Einstein equations which is 4D Poincar invariant on the brane. However, unlike in other
self-tuning solutions [7,8] where the spacetime is cut at a naked singularity, the cut at a
horizon is natural since from an observers point of view it will take an infinite time to
reach the horizon and boundary conditions at a horizon are well-defined and do not rely on
the introduction of any ad hoc extra brane where some fine-tunings are reintroduced (see
[10]).
326
So we would like to show that the following quantity is indeed vanishing:

r0
I=
dr

|g5 |

1
1 2
F
R
+
L
g
(r
r
)
.
bk
mat
rr
0
2 5
4
(2.32)
rH
The first step is to evaluate the singular part coming from the brane namely to find the
expression of Lmat in terms of the 4D energy density and pressure that model the brane.
As we have emphasized before, for the case where a horizon is present one needs an
unconventional source with < 1, which cannot be obtained from stable dynamical
fields, but should rather be thought of as a non-dynamical object. However, in order to
calculate the relation between the energy densities and the original Lagrangian we assume
that the results for the region 1 < < 1 also apply for the case < 1. In order to
calculate the relation we are interested in we assume that the matter corresponds to a timedependent scalar field
1
Lmat = V (),
2
from which we obtain that
4 h(r0 )
h (r0 )
Lmat = p = 2
,
5 h(r0 )
52 r0
(2.33)
(2.34)
where we have used the jump equations (2.17) to derive the last equality. We emphasize
again that for the case of matter with < 1 (as we have in the case with a horizon) the
assumption that the matter is dynamical like in (2.33) would imply an instability in the
system, therefore one has to assume that rather than being a dynamical matter it is more
like a topological object, similar to an orientifold. However, we assume that (2.34) holds
even in this case.
In computing the bulk part of the integral, we must include the singular part of the
curvature at the Z2 orbifold fixed point. To evaluate this singular contribution, we can use
the form (2.8) of the bulk metric with A(r) = h(r), B(r) = r and C(r) = 1/ h(r); its
curvature is given by [29] (for k = 0),

B
B 2
B C
2 A
A B A C
+3
+3 2 +3
3
R= 2
A
B
AB
AC
BC
C
B

h(r)
h (r)
6 2 .
= h (r) 6
(2.35)
r
r
But because of the Z2 symmetry at r = r0 , the first derivatives are discontinuous which
translates into delta function type singularities in the second derivatives. So the singular
part of the curvature reads, 11

B
4 A
h(r0 )

+ 3 (r r0 ) = 2h (r0 ) + 12
Rsing = 2
(2.36)
(r r0 ).
C
A
B
r0
11 The sign is fixed by keeping the region of spacetime between the singularity and the brane.
327
And finally, using the expression (2.4) of the bulk cosmological constant and the expression
(2.6) of the gauge field strength, the effective cosmological constant becomes

1
Q2 r0
1
I = 2 l 2 r 4 + 2
+ 2 2r 3 h + 12r 2 h |r
0
r rH 45
5

1
2
+ 2 r 3 h 4r 2 h |r = 52 rH
h(rH ),
(2.37)
0
25
which vanishes precisely because rH is the position of the horizon. We can explicitly see
that, as expected, this cancellation does not require adding anything at the horizon.
2.6. Maximally symmetric solutions on the brane
So far we have only considered flat brane solutions and performed our analysis at a
classical level where the charge of the BH varies continuously. At this point, similar
mechanisms would ensure a vanishing effective cosmological constant without requiring
the existence of extra dimensions. However, as we have discussed in the introduction, it is
important to look for other 4D maximally symmetric metrics, namely de Sitter or anti-de
Sitter 4D spacetime, and this is where extra dimensions become relevant. For instance,
in the 4D adjustment mechanism proposed by Hawking [36], a four form field provides a
contribution to the cosmological constant whose magnitude is not determined by the field
equations but appears as a constant of integration. 12 It was argued that its probability
distribution may be exponentially peaked such that the flat solution will be preferred
among the other maximally symmetric solutions. Unfortunately, Duff has shown [37] that
the vanishing of the effective cosmological constant is the most unlikely possibility and
the anthropic principle has to be invoked [38] to disentangle the different solutions. On
the contrary, in the self-tuning approach to the cosmological constant on a brane, the 4D
Minkowski metric provides the only maximally symmetric solution to Einsteins equations
with an adequate choice of the conformal coupling to the brane [7]. We will argue now that
self-selection of the flat brane solution remains true when the bulk scalar field is replaced
by a vector field, without tuning parameters in the action.
De Sitter and anti-de Sitter solutions can be parametrized as FRW spacetimes,
ds 2 = d 2 + R 2 ( ) dk2 .
(2.38)
The explicit form of the scale factor can be deduced from the fact that (A)dS4 spacetime
are solutions to the usual 4D Friedmann equation with a (negative) positive vacuum energy.
This requires,
k
R 2 1 2
= 4 2 .
R2 3 4
R
(2.39)
Solving this ordinary differential equation leads to the following parameterizations [26,39]:
12 For a modern version of this idea in a supersymmetric/superstring context, see [12,13].
328
and R( ) = R0 eH ,
k = 0
de Sitter:
k = 1 and R( ) = sinh(H )/H,
k=1
and R( ) = cosh(H )/H,
anti-de Sitter: k = 1 and R( ) = cos(H )/H.
(2.40)
(2.41)
If such solutions exist, they must correspond to moving branes in the static bulk metric
and thus must satisfy the jump equations (2.11), (2.12). For a non-static brane, the second
jump equation is equivalent to energy conservation, so R( ) must satisfy,

R
1 4 2
k
1
Q2
R 2
and + 3(1 + ) = 0.
=
+
(2.42)
5
2
2
2
4
6
36
R
R
R
l
R
R
When the parameter appearing in the equation of state is constant, the conservation
equation becomes,

R( ) 3(1+)
= 0
(2.43)
.
R0
Plugging back into the first differential equation in (2.42), we determine in which cases the
(A)dS spacetime are solutions to the jump equations
The 4D de Sitter metrics will be solution only when the equation of state on the brane
corresponds to a vacuum energy, = 1, and when this vacuum energy is bigger
than the contribution from the bulk, in which case the Hubble parameter is given by
1 4 2
5 l 2 . All these solutions are moving in a pure anti-de Sitter bulk,
H 2 = 36
namely = 0 and Q2 = 0.
The 4D anti-de Sitter metrics will be solutions for a vacuum energy type brane ( =
1) with small enough energy density, in which case the Hubble parameter is given
1 4 2
5 while the charge and the mass still vanish. These solutions
by H 2 = l 2 36
are the AdS4 branes in AdS5 bulk studied in [40]. There are other 4D AdS solutions
where the Hubble parameter is given by the 5D vacuum energy, H 2 = l 2 , and the
energy density on the brane is balanced, respectively, by the (negative) mass of the
1 4 4 2
5 l 0 and Q = 0) or by its charge if = 0
5D black hole if = 1/3 ( = 36
1 4 6 2
2
(Q = 36 5 l 0 and = 0).
This analysis shows that the 4D Minkowski metric is the only maximally symmetric
solution on the brane up to, for particular equations of state, a discrete choice of the
constants of integration. Hence, self-tuning in these models is somewhat natural: for a
generic brane equation of state, the only continuous class of solutions is the set of flat
brane solutions. This result offers a selection of a vanishing cosmological constant from
symmetry requirements only. The vanishing of the cosmological constant is ensured by
an adjustment of the charge and the mass of the black hole through gravitational waves
emitted from the brane when a phase transition occurs, for instance. A precise description
of such a phase transition and the response of the bulk would however certainly deserve
further scrutiny.
Our analysis has been performed at the classical level. It is important to comment on
issues at the quantum level. For instance, the original self-tuning model suffers from
quantum instabilities and in particular the flat solution is no longer an isolated vacuum
329
in the quantum moduli space [10,12] due to a conformal anomaly. We do not expect such a
problem in our model since a U (1) gauge symmetry protects the action and only allows to
generate a ChernSimons term which would not modify our analysis. The more important
issue concerns the quantization of the charge. If the BH charge is quantized, it can be tuned
only to the discrete value closest to the continuous value ensuring a vanishing effective 4D
cosmological constant and thus a small non-vanishing effective cosmological constant is
generated. To have a phenomenologically acceptable model, this value should be less than
the observed value of the vacuum energy. In a quantum field theory perspective, this can
always be achieved by adjusting the quantization level of the charge. In a string theory
perspective this quantization level is however related to the string scale [13], and not freely
adjustable. The minimal value of the quantization levels depends on the details of the full
string theory background and is beyond the scope of our effective theory approach.
3. Lorentz violations in black hole backgrounds

We have seen in the previous section that asymmetrically warped spacetimes due to
black holes in the bulk can exist, and perhaps even provide an adjustment mechanism
for the effective cosmological constant. In this section, we investigate some of the physical
consequences of such spacetimes. In this analysis we will not assume that the cosmological
constant problem is resolved by these backgrounds, but we are rather interested in general
consequences of having an asymmetrically warped metric. In particular, we will not need to
have exotic energy densities on the branes (for which the price to pay is the reintroduction
of fine-tuning in the theories). First we calculate the speed of gravitational waves based on
the analysis of lightlike geodesics, then show how the graviton zero mode found by Randall
and Sundrum would be modified in such spacetimes. Finally, we consider the cosmology
of these models in order to gain some insight into the effective gravity observed on the
brane.
3.1. Geodesics in the black hole background and the speed of gravitational waves
One of the most important consequences of an asymmetrically warped spacetime is
that the speed of gravitational waves could differ from the speed of light. The reason for
this is that in asymmetrically warped metrics the local speed of light is a function of the
coordinate along the extra dimension r:

h(r)l 2 1/2
.
c(r) =
(3.1)
r2
Thus one might think of this spacetime as a medium with a continuously changing index
of refraction n(r), in which the propagation of light is the analog of the propagation
of gravitational waves in the asymmetrically warped spacetimes. Propagation of light is
governed by Fermats principle, and if c(r) is increasing away from the brane it may be
advantageous for gravitational waves to bend into the bulk, and arrive earlier than waves
propagating along the brane would. However since electromagnetism is forced to propagate
330
along the brane, it will always keep the local velocity at the brane c(r0 ). Thus we expect
as long as the velocity c(r) increases away from the brane gravitational waves can travel
faster than the local speed of light at the brane. This will lead to an apparent violation
of Lorentz invariance due to the bulk dynamics in the low energy effective theory, and
an apparent violation of causality from the brane observers point of view, in the sense
that gravitational waves emitted from a source will arrive earlier than light signals from
the same source. We should emphasize that this is not a real violation of causality.
Since there are no closed timelike curves in the theory, propagation of massless particles
always proceeds forward in time. Apparent violation of causality simply means that the
regions of spacetime which are in causal contact are different from what one would naively
expect from an ordinary Lorentz invariant 4D theory. This is an experimentally verifiable
prediction for these models, and can be tested by gravitational wave experiments like
LIGO, VIRGO or LISA. Some of these points were made independently in Refs. [1619,
26,41] and it has been argued that this apparent Lorentz symmetry violation may provide
an alternative to inflation to deal with the horizon problem. This is similar to another
alternative to solve the horizon problem proposed in [4244], where the speed of light is
changing in time, thus regions of spacetime may have been in causal contact even if they
appear to be outside each others horizons. Violations of Lorentz invariance similar to the
ones considered in this paper may also appear in non-commutative gauge theories [45,46].
In generic non-commutative theories faster than light propagation will not be suppressed
in the perturbative sector, and can not give a realistic model. In supersymmetric theories
however the Lorentz violation in the perturbative sector vanishes, and will appear only as
faster-than-light propagation in the solitonic sector, which is very weakly coupled to the
perturbative states. This way interesting theories [47] analogous to the brane constructions
presented here can be obtain from non-commutative gauge theories. However, it remains
to be seen whether or not supersymmetry breaking will reintroduce the Lorentz breaking
effects to the perturbative sector.
If c(r) is a decreasing function as one moves away from the brane, for example, in the
case that there are horizons shielding the singularity from the brane, we expect that those
gravitational waves that will be observed are those that remain on the brane, and thus no
discrepancy between the speed of light and the speed of gravity should exist. It would be
interesting to study the graviton wave equation in these cases, which we postpone until
the next subsection. Below we show, that the analysis of lightlike geodesics is in complete
agreement with the physical intuition sketched above.
Here we analyze the geodesics of the theories with asymmetrically warped metrics.
Some of the features of geodesics in asymmetrically warped spacetimes have also been
analyzed in [1618]. The equations for the geodesics are given by
P
Q
d 2xM
M dx dx
= 0,
+
(3.2)
P
Q
d 2
d d
where is an affine parameter along the trajectory (the proper time for the spacelike
geodesic). As usual, these equations can be deduced from the Lagrangian provided by
the proper distance. In the case of the black hole metric:
ds 2 = h(r) dt 2 + l 2 r 2 d x 2 + h(r)1 dr 2 ,
331
(3.3)
the Lagrangian is explicitly given by:

ds 2
2
(3.4)
= h(r)t2 + l 2 r 2 x + h1 (r)r 2 ,
2
d
where the dot represents differentiation with respect to the affine parameter . Furthermore
the EulerLagrange equations are supplemented by the consistency condition defining the
affine parameter:
L=
L = 1,
1 = 1/0/ + 1,
for timelike/lightlike/spacelike geodesics, respectively.
(3.5)
From the t- and x -independence of the black hole metric, we find four Killing vectors
and thus four corresponding conserved quantities:
L
= 2h(r)t = const. 2E,
t
L
= 2l 2 r 2 x i = const. 2pi .
i
x
(3.6)
(3.7)
In terms of these conserved quantities, the consistency equation is

h(r) 2 2
(3.8)
l p h(r)1 = E 2 .
r2
As long as r = 0, this equation is equivalent to the r component of the geodesic
equations (3.2). However, for a straight geodesic parallel to the brane, i.e., r = 0, the first
derivative of (3.8) also has to be satisfied.
From now, we will concentrate on lightlike geodesics only. The motion in the r direction
transverse to the brane is then analogous to Newtonian dynamics for a particle of two units
of mass with energy E 2 moving in the potential
r 2 +
h(r) 2 2
(3.9)
l p .
r2
The behavior of V (r) around the brane in the direction of the singularity, and the position
of the brane with respect to the horizons if they exist, will determine the shape of the
geodesics in the bulk.
From the jump equations (2.23), the expression for the potential becomes

r6
r
4
2 0 ,
V (r) = p 2 1 40 + Q
(3.10)
r
r6
V (r) =
with

1 4 2 2
= l 4 = 3 1 + 5 l ,
36
r0

2
2
l Q
1 4
2
2 2

Q = 6 = 2 1 + 5 (1 + 3)l .
72
r0
2
(3.11)
332
The shape of the potential V (r) will depend on the sign of and whether there exists a
minimum between the singularity and the brane. The condition for to be positive is

36
( > 0) or
< 0 and < =
(3.12)
.
l 2 54
And when > 0, the minimum of the potential will be at

2
3Q
r = r0
.
2
(3.13)
To locate the position of this minimum with respect to the brane, we can study the ratio
2 /,
Q
or equivalently, compute the force, V (r), felt by the analogous Newtonian
particle at the position of the brane:
2

p 2
2 = p (1 + )l 2 54 2 ,
(3.14)
4 + 6Q
r0
6r0
from where we conclude that the minimum will be between the brane and the singularity
iff < 1. In summary, five different shapes for the potential arise and they are drawn
in Fig. 3. This also shows that based on the intuitive picture presented at the beginning of
this section one would expect the speed of gravitational waves can be larger than the speed
of light when 1, since this is the case when the speed of light away from the brane
increases.
Finally, we calculate the average speed of gravitational waves as observed from the brane
observers point of view. This can be calculated for the case when the speed of light away
from the brane increases by first calculating the turning point of the Newtonian particle in
the potential V (r), which is the largest solution 13 between the singularity (r = 0) and the
brane (r = r0 ) of the following equation:

E 2 h(rT ) 2
= 2 l .
(3.15)
|p |
rT
V (r0 ) =
Once the turning point is obtained (numerically), one can eliminate the proper time by
dividing the expressions for x by the expression for r , and integrate the resulting equation
to obtain x(r). With this the distance on the brane xret after which the geodesic returns to
the brane can be expressed as
r0

E
2l 2 dr

xret
(3.16)
,
=
|p |
r 2 (E/|p |)2 h(r)l 2 /r 2
rT
and similarly the time it takes to return can be expressed as

tret
E
|p |
r0
=
rT
2E/|p | dr
h(r) (E/|p |)2 h(r)l 2 /r 2
(3.17)
13 Note that we have cut the space at the brane and kept the region between the singularity and the brane. Since
the potential diverges at the origin, there always exists a point where r = 0 at which the potential is equal to the
energy. This solution corresponds to the turning point where the geodesic starts moving back to the brane.
333
Fig. 3. Shapes of the Newtonian potential. The functions , 0 and defined in (2.27), (2.24)
and (3.12) border the different regions in the plane (, ). The asymptotic value of the potential at
infinity corresponds to the momentum |p |2 along the brane. The space is however cut at the brane
sitting at r = r0 . We identify five different shapes for the potential. (a) < 1 and 0 > > : the
potential vanishes at the two horizons. The geodesics will come back to the brane but, from a brane
observers point of view, it already takes an infinite time to cross the horizon and so the geodesics
will never be seen as coming back. (b) < 1 and > . The 4D local speed of graviton starts
decreasing when the geodesic leaves the brane. (c) 1 < < 0 and > . (d) 1 < < 0 and
> or 0 < and > . (e) 0 < and > . In the three last configurations, the 4D speed of
the graviton is increasing when going into the bulk, and on its return the geodesic may reach a point
on the brane where light emitted with the graviton has not yet arrived.
To obtain the relative speed compared to the speed of light, the average speed cav = xret /tret
has to be compared with the local speed of light at the brane cem = (h(r0 )l 2 /r02 )1/2 .
One can clearly see from (3.17) that the average speed evaluated from the geodesics will
depend on the value of E/|p |, which is equivalent to choosing the oscillation length of the
gravitational wave around the brane (which is also equivalent to how far into the bulk the
gravitational wave is penetrating). The dependence of the average speed on the oscillation
length is given in Fig. 4 for the cases where the speed of light away from the brane is
increasing (that is cases (c), (d) and (e)) of Fig. 3. The full gravitational propagation would
be given by a superposition of these simple oscillating modes.
Depending on the exact structure of the low-energy effective theory, the Lorentz
violation due to bulk effects may be transmitted to particles on the branes by gravitational
loops. At the scale where gravity becomes strongly interacting, i.e., at the fundamental
Planck scale, M , such loops will not be suppressed. In order for particle physics to remain
Lorentz invariant, the Lorentz violations should decrease as the energy scale is increasing,
otherwise one would expect unsuppressed Lorentz violating operators due to gravitational
loops. One can see that in most cases considered here such constraints can be generically
334
Fig. 4. The average speed of gravitons propagating along a geodesic off the brane as function of the
distance on the brane. We clearly see that in the region of the plane (, ) where the Newtonian
potential behaves like the shapes (c), (d) and (e) of Fig. 3, the graviton can propagate faster than the
light and its speed increases with the distance to the source on the brane. We also note in the case
(d) that there are various ways for gravity to propagate between the same points on the brane, i.e.,
different geodesics in the bulk with different values of E/|p | can return to the same point on the
brane. When E/|p | is too large, then the average speed becomes lower than the speed of light. In the
case (b), the graviton will always propagate along the brane with a speed faster than that in the bulk.
satisfied. The reason can be understood from a holographic argument, similar to ones
explained in [48]. As the energy scale is increasing, gravity on the brane is probing less and
less of the bulk region around the brane, indeed gravitational waves simply will have less
time to travel further into the bulk. Thus in order to avoid large Lorentz violating operators
being generated through gravitational loop effects one has to arrange that the region around
the brane at a distance of the order M1 should be very close to ordinary AdS space. Of
course this can always be achieved by moving the black hole far away from the brane. In
the language of the geodesic analysis of this section this constraint would imply that at
small distances the speed of gravitational waves should approach the speed of light on the
brane. One can see that the cases (c) and (e) in Fig. 4 automatically satisfy this requirement,
and thus by adjusting r0 one can satisfy experimental constraints on Lorentz violations in
particle physics. 14
14 We thank John Terning for discussions on this issue.
335
3.2. A perturbative zero mode

We have seen from the analysis of the geodesics that one expects gravity to propagate
with a speed different from ordinary electromagnetism for asymmetrically warped
spacetimes. This can cause apparent violations of causality and Lorentz invariance in the
gravitational sector, without affecting particle physics (except through gravitational loops).
Next we show an analysis different from the geodesic approach to demonstrate the same
effect. We will examine how the graviton zero mode of the RandallSundrum model is
modified in one of the black hole spacetimes considered in this paper. To simplify the
equations, we will consider the black hole metric as a linearized perturbation around the
RS spacetime. We have chosen to analyze the perhaps simplest form of matter on the brane,
the case which simply corresponds to a brane tension ( = 1). As we have seen before,
in this case there is no horizon, but there is a naked singularity away from the brane. In
order to be able to analyze this theory as a perturbation around the RS metric, we have
to cut the spacetime with a second (fine-tuned) brane before the deviation from the RS
spacetime becomes large (and of course before the appearance of the naked singularity).
This way we obtain a metric that is close to the RS metric everywhere, and we will think of
the mass and charge of the black hole as a perturbative expansion around the RS solution.
In order to make this expansion more transparent, we first transform the black hole metric
by an appropriate rescaling of the coordinates t and x and a coordinate transformation
r = r0 exp(ky), where k = 1/ l, to the form
1 dy 2 ,
ds 2 = e2k|y| h(y)
dt 2 + e2k|y| d x 2 + h(y)
(3.18)
with h(y)
= 1 l 2 r04 e4ky + Q2 l 2 r06 e6ky . The location of the brane is now at y = 0. As
stated above, we also assume that l 2 r04 e4ky and Q2 l 2 r06 e6ky remain small everywhere
1 can be expanded in and Q2 . Now we would like to solve
in the bulk, and thus h(y)
the linearized Einstein equation in this background, and in particular find the modified
propagation speed of the graviton zero mode. One can show that the transverse traceless
modes of the graviton h satisfy the following linearized equation:
1
1
1
DM D M h + RM h M + RM h M R h
2
2
2
1
1
+ R h g Rh
2
2
1 2
1
= 5 FM F M h g 52 FMN F MN h 52 bk h ,
(3.19)
2
4
where the covariant derivatives are with respect to the background metric gMN , as are
the Riemann tensor and the Ricci tensor and scalar. We have also checked that these
transverse traceless modes decouple from perturbations of the gauge field and thus only
the background gauge field appears in the right hand side of the equation. In fact, one
can show that after explicitly including the expressions for the covariant derivatives and
the background quantities in (3.19) the equation simply becomes identical to the equation
for a minimally coupled massless scalar in the bulk, just like in the case of the RS-type
backgrounds [49]. Hence, we simply study the propagation of a scalar field in the bulk.
336
We know that for vanishing and Q2 the zero mode solution in the RS model is just
= 0 ei(t q x ) , where 2 = q 2 , and 0 is a normalization constant. Thus we look for a
perturbative solution of the bulk equation of the form,

= 0 1 + (y) ei((+)t q x ) ,
(3.20)
where (y) and are assumed to be proportional to and Q2 . A scalar zero mode
of a similar form has been found for another asymmetrically warped metric in Ref. [19].
Expanding the linearized equation around the background (3.18) one obtains,
4k + l 2 r04 q 2 e6ky Q2 l 2 r06 q 2 e8ky + 2qe2ky = 0.
(3.21)
As explained above, we are assuming for the sake of perturbativity that the spacetime is
made finite by the introduction of a second brane, and does not include the singularity.
Therefore we need to impose a boundary condition on that its derivative at the two
branes vanishes, i.e.,

y=0,y = 0,
(3.22)
R
where yR is the position of the regulator brane. Enforcing these boundary conditions one
obtains the dispersion for the zero mode, which is given by

1
= q 2l 2 r04 + Q2 l 2 r06 1 + e2kyR e2kyR .
(3.23)
4
This dispersion relation remains linear such that the speed of propagation of gravitational
waves is constant and given by

2ky
l 2 Q2 l 2 2
2kyR
Q 1+e
e R,
cgrav = 1 + 4 +
(3.24)
2r0
4r06
which hasto be compared with the propagation speed of electromagnetic waves given
= 0) = 1 1 l 2 r 4 + 1 Q2 l 2 r 6 in these coordinates. For the = 1
by cem = h(y
2
spacetimes and Q are related by the formula = 32 Q2 /r02 , therefore the difference of
the two speeds can be written as

Q2 l 2
0.
cgrav cem = cosh(2kyR ) 1 e2kyR
2r06
(3.25)
Therefore the gravitational speed for the propagation of the zero mode is always larger than
the speed of electromagnetic waves in the scenario considered here. The LIGO experiment
may be able to detect gravitational waves from type II supernovae up to a distance [50] of
about 20 Mpc ( 6 107 ly). For objects of such a distance even a tiny difference in the
speeds of gravitational and electromagnetic waves would cause a huge time difference,
and thus the possible values of and Q could be severely constrained [19]. In fact,
the limitations of such measurements are likely not to lie in the time resolution of the
gravitational and the light signal, but rather the opposite problem: if there is an appreciable
difference in the propagation speeds then due to the huge distance to the expected sources
the arrival time differences could turn out to be way too big to be able to identify the
337
fact that the source for the gravitational wave and the light was the same. For a supernova
20 Mpc away from us, and very conservatively assuming that the arrival time difference
should be less than 5 years, in order to be able to actually detect the different arrival
7
times one needs to have the difference in the speeds to be less than c
c 10 . Otherwise
the gravitational wave experiments will simply not be able to identify the source for the
observed gravitational waves. Type I supernovae could likely be detected by LIGO only
if they happen within our galaxy. These are very rare, however assuming the best possible
scenario one could see a supernova a few hundred thousand light years from us. In this
case (again assuming a very conservative time difference of 5 years) the maximum value
3
of c
c that could be tested is of the order of 10 .
3.3. Cosmological expansion and effective gravity theory
So far we have discussed the interesting physical consequences of asymmetrically
warped spacetimes: the possible difference in the speeds of gravitational and electromagnetic waves, and the possible adjustment of the cosmological constant to give a flat brane.
Next we would like to understand what kind of gravitational theory a 4D observer on the
brane would see. There are at least two distinct possibilities: 15 the effective action for gravity could explicitly break the 4D Lorentz invariance, for instance, by introducing different
coefficients for time and space derivatives in the kinetic terms of the graviton, or the effective action may only violate the weak equivalence principle by the presence of some extra
fields which couple differently to matter and gravitation forcing the graviton to propagate
differently as the other gauge bosons (see [51] for a review on the different tests of Einstein
gravity). Note that the distance of the brane to the singularity or the horizon could be such
a field. Both approaches will manifest themselves experimentally by observing different
speeds of propagation for the graviton and the photon. Instead of trying to understand this
(very important) question of how to incorporate the Lorentz violating effects into the low
energy effective action, we solve the simpler problem (which still gives us some insights
into the long distance behavior of gravity) of finding the cosmological evolution of the
brane in the presence of matter perturbations on top of the vacuum energy. We will express
the sources on the brane as the sum of the (non-dynamical) vacuum energy, plus the matter
perturbations:
tot = 0 + ,
ptot = p0 + p.
(3.26)
The expansion of the brane can then be read off from (2.13), (2.14). Using (2.22) the form
of the expansion equation can be simplified to
2

4 0
54 2
R
1
1
1
1
2
(3.27)
,
= 5 +
Q
+
R
18
R 4 r04
R 6 r06
36
while the conservation of energy equation is given by the slightly unconventional
expression
15 We thank Nemanja Kaloper for discussions on this issue.
338

R
= 0.
+ 3 0 (1 + 0 ) + (1 + )
R
(3.28)
In these equations, R is nothing but the scale factor on the brane and a dot denotes a
derivative with respect to the proper time on the brane, i.e., the usual FRW time; r0 is the
position of the brane before matter was introduced. On comparison with similar terms in
usual 4D Friedmann equation, from (3.27) we can identify the 4D effective Planck scale as
54 0
1
.
=
2
6
MPl
(3.29)
If one assumes that energy densities on the brane are of the order of the TeV scale, then
the required size of the five-dimensional Planck scale would be given by M = 108 GeV,
with 52 = 1/M3 . One can also see that very close to the static solution that is at R = r0
there is no correction to the ordinary FRW equation, which one could perhaps interpret
as the zero mode of this model reproducing ordinary 4D gravity. However, as the brane
moves further away from the static point, the corrections to the Friedmann equation will
start becoming sizeable. This has one positive consequence: a cosmological constant term
/r04 Q2 /r06 , the usual dark radiation term /R 4 and due to the charge of the black
hole a contribution that is similar to the contribution of a 4D massless scalar, Q2 /R 6 is
obtained, which can be used to fit the Friedmann equation to the observed accelerating
Universe [25]. This possibility has also been pointed out by Refs. [23,24]. However, these
terms also pose a problem: if , Q2 and r0 are such that the cosmological constant gets
adjusted to zero for R = r0 , then after just a short expansion period the above mentioned
new terms in the Friedmann equation will start dominating, if and Q2 remain constant.
In order to overcome this problem, one has to assume that the adjustment mechanism
that sets the cosmological constant dynamically to zero also operates during the ordinary
expansion of the Universe, thus making and Q2 time dependent. Let us determine how
small the characteristic time scale for the adjustment mechanism would have to be in order
for the solution to track the vanishing cosmological expansion solution. In such a case
the corrections due to the change in of the form /R
4 should cancel the corrections
5 (up to remaining terms
from the change in the position of the brane of the form R/R
2
2
in the Friedmann equation of order H = (R/R)

). This would lead to /
R/R
H;
thus, the characteristic time scale for the adjustment should be of the order or shorter
than the Hubble scale. If one wants an ordinary FRW Universe after nucleosynthesis
then one should require that this scale is shorter than HBBN . This is not a very strong
restriction for the time scale of adjustment, and could still leave the possibility for early
inflation open. Of course if one is not trying to solve the cosmological constant problem,
then , Q2 and r0 should be viewed as free parameters that are not necessarily related
to the fundamental Planck scale. In this case these parameters can be simply used to fit
the observed accelerating Universe without assuming any time dependence for them. In
that case the acceleration of the expansion of the Universe would be a manifestation of
gravitational Lorentz violations in extra dimensions rather than a consequence of a tiny
bare vacuum energy in four dimensions.
339
4. Conclusions
The work of Randall and Sundrum has revealed that new physics can emerge from a
nontrivial (warped) geometry mixing the four dimensions of our brane-world with an
extra non-compact dimension. While the main attention has been focused on solutions
preserving 4D Lorentz symmetry in the bulk, other solutions exist with different warp
factors for timelike and spatial directions. In fact these asymmetric solutions are the more
generic ones, and they open up new perspectives to low energy gravitational interactions
due to the Lorentz symmetry violation they can mediate.
One of the most striking features is the possibility for gravitational waves to propagate
faster than electromagnetic waves stuck on the brane [1619]. This apparent violation
of causality from a brane observer point of view could be experimentally tested by
gravitational wave detectors. In the same vein, we have shown that the addition of a
vector field in the bulk alleviates the fine-tuning of the cosmological constant problem.
The bulk has a geometry of an AdSReissnerNordstrm black hole, where the relaxation
could be achieved by the adjustment of the mass and charge of the black hole. A horizon
could protect the black hole singularity, however the existence of a horizon requires the
existence of some exotic energy density on the brane. The cosmology in asymmetrically
warped spacetimes can explain the observed acceleration of the Universe which thus would
appear as a manifestation of gravitational Lorentz violations in extra dimensions. If one
wants to solve the cosmological constant problem and explain the accelerating Universe
simultaneously, then one has to assume that the relaxation of the mass and charge is of the
order or faster than the Hubble scale.
Acknowledgements
We thank Nima Arkani-Hamed, Francis Bernardeau, Tanmoy Bhattacharya, Zackaria
Chacko, Daniel Chung, Michael Graesser, Salman Habib, Nemanja Kaloper, Hideo
Kodama, Ann Nelson, Maxim Perelstein, Fernando Quevedo, Riccardo Rattazzi, Yuri
Shirman, Raman Sundrum, Takahiro Tanaka, John Terning and Mark Wise for useful
discussions. We also thank Daniel Chung, Nemanja Kaloper, Maxim Perelstein and John
Terning for helpful comments on the manuscript. We would like to thank the Aspen Center
for Physics where this work was initiated during the workshop New Physics at the Weak
Scale and Beyond. C.C. is a J. Robert Oppenheimer fellow at the Los Alamos National
Laboratory. C.C., J.E. are supported by the US Department of Energy under contract W7405-ENG-36. C.G. is supported in part by the US Department of Energy under contract
DE-AC03-76SF00098 and in part by the National Science Foundation under grant PHY95-14797.
References
[1] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263, hep-ph/9803315;
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257, hepph/9804398;
340
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
See also, 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.
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-ph/9905221;
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690, hep-th/9906064.
J. Garriga, T. Tanaka, Phys. Rev. Lett. 84 (2000) 2778, hep-th/9911055.
S.B. Giddings, E. Katz, L. Randall, JHEP 0003 (2000) 023, hep-th/0002091.
C. Cski, M. Graesser, C. Kolda, J. Terning, Phys. Lett. B 462 (1999) 34, hep-ph/9906513;
J.M. Cline, C. Grojean, G. Servant, Phys. Rev. Lett. 83 (1999) 4245, hep-ph/9906523.
V.A. Rubakov, M.E. Shaposhnikov, Phys. Lett. 125B (1983) 139.
N. Arkani-Hamed, S. Dimopoulos, N. Kaloper, R. Sundrum, Phys. Lett. B 480 (2000) 193,
hep-th/0001197.
S. Kachru, M. Schulz, E. Silverstein, Phys. Rev. D 62 (2000) 045021, hep-th/0001206;
S. Kachru, M. Schulz, E. Silverstein, Phys. Rev. D 62 (2000) 085003, hep-th/0002121.
I. Low, A. Zee, Nucl. Phys. B 585 (2000) 395, hep-th/0004124;
S.P. de Alwis, A.T. Flournoy, N. Irges, hep-th/0004125;
G.T. Horowitz, I. Low, A. Zee, Phys. Rev. D 62 (2000) 086005, hep-th/0004206;
Z. Kakushadze, Nucl. Phys. B 589 (2000) 75, hep-th/0005217;
N. Alonso-Alberca, B. Janssen, P.J. Silva, hep-th/0005116;
C. Zhu, JHEP 0006 (2000) 034, hep-th/0005230;
B. Grinstein, D.R. Nolte, W. Skiba, Phys. Rev. D 62 (2000) 086006, hep-th/0005001;
P. Bintruy, J.M. Cline, C. Grojean, Phys. Lett. B 489 (2000) 403, hep-th/0007029;
K. Uzawa, J. Soda, hep-th/0008197;
H. Collins, B. Holdom, hep-th/0009127;
Z. Kakushadze, Mod. Phys. Lett. A 15 (2000) 1879, hep-th/0009199;
S. Kalyana Rama, hep-th/0010121;
C. Kennedy, E.M. Prodanov, hep-th/0010202;
A. Kehagias, K. Tamvakis, hep-th/0011006;
P.H. Brax, A.C. Davis, Phys. Lett. B 497 (2001) 289, hep-th/0011045;
J.E. Kim, B. Kyae, H.M. Lee, hep-th/0011118.
D. Youm, Nucl. Phys. B 589 (2000) 315, hep-th/0002147;
S. Frste, Z. Lalak, S. Lavignac, H.P. Nilles, Phys. Lett. B 481 (2000) 360, hep-th/0002164;
S. Frste, Z. Lalak, S. Lavignac, H.P. Nilles, JHEP 0009 (2000) 034, hep-th/0006139;
S. Frste, hep-th/0012029.
C. Cski, J. Erlich, C. Grojean, T. Hollowood, Nucl. Phys. B 584 (2000) 359, hep-th/0004133.
S.P. de Alwis, hep-th/0002174.
R. Bousso, J. Polchinski, JHEP 0006 (2000) 006, hep-th/0004134.
P.J. Steinhardt, Phys. Lett. B 462 (1999) 41, hep-th/9907080;
C.P. Burgess, R.C. Myers, F. Quevedo, Phys. Lett. B 495 (2000) 384, hep-th/9911164;
J. Chen, M.A. Luty, E. Ponton, JHEP 0009 (2000) 012, hep-th/0003067;
C. Schmidhuber, Nucl. Phys. B 585 (2000) 385, hep-th/0005248;
Z. Kakushadze, Phys. Lett. B 488 (2000) 402, hep-th/0006059;
Z. Kakushadze, Phys. Lett. B 491 (2000) 317, hep-th/0008041;
S.H. Tye, I. Wasserman, hep-th/0006068;
A. Chodos, E. Poppitz, D. Tsimpis, Class. Quant. Grav. 17 (2000) 3865, hep-th/0006093;
A. Krause, hep-th/0006226;
A. Krause, hep-th/0007233;
J.M. Cline, H. Firouzjahi, hep-ph/0012090.
H.B. Nielsen, I. Picek, Nucl. Phys. B 211 (1983) 269;
S. Chadha, H.B. Nielsen, Nucl. Phys. B 217 (1983) 125;
S. Coleman, S.L. Glashow, Phys. Lett. B 405 (1997) 249, hep-ph/9703240;
S. Coleman, S.L. Glashow, Phys. Rev. D 59 (1999) 116008, hep-ph/9812418;
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
341
D. Colladay, V.A. Kosteleck, Phys. Rev. D 55 (1997) 6760, hep-ph/9703464;

D. Colladay, V.A. Kosteleck, Phys. Rev. D 58 (1998) 116002, hep-ph/9809521;
D. Bear, R.E. Stoner, R.L. Walsworth, V.A. Kosteleck, C.D. Lane, Phys. Rev. Lett. 85 (2000)
5038, physics/0007049;
O. Bertolami, Class. Quant. Grav. 14 (1997) 2785, gr-qc/9706012;
O. Bertolami, C.S. Carvalho, Phys. Rev. D 61 (2000) 103002, gr-qc/9912117.
G. Klbermann, H. Halevi, gr-qc/9810083;
G. Klbermann, Int. J. Mod. Phys. A 15 (2000) 3197, gr-qc/9910063.
D.J. Chung, K. Freese, Phys. Rev. D 62 (2000) 063513, hep-ph/9910235.
H. Ishihara, gr-qc/0007070.
D.J. Chung, E.W. Kolb, A. Riotto, hep-ph/0008126;
D.J. Chung, Talk at Santa Fe 2000 Summer Workshop, Supersymmetry, Branes and Extra
Dimensions.
M.A. Clayton, J.W. Moffat, Phys. Lett. B 460 (1999) 263, astro-ph/9812481;
I.T. Drummond, gr-qc/9908058;
M.A. Clayton, J.W. Moffat, Phys. Lett. B 477 (2000) 269, gr-qc/9910112.
S. Weinberg, Rev. Mod. Phys. 61 (1989) 1.
R. Gregory, V.A. Rubakov, S.M. Sibiryakov, Phys. Rev. Lett. 84 (2000) 5928, hep-th/0002072;
C. Cski, J. Erlich, T.J. Hollowood, Phys. Rev. Lett. 84 (2000) 5932, hep-th/0002161;
G. Dvali, G. Gabadadze, M. Porrati, Phys. Lett. B 484 (2000) 112, hep-th/0002190.
C. Barcel, M. Visser, Phys. Lett. B 482 (2000) 183, hep-th/0004056.
H. Collins, B. Holdom, Phys. Rev. D 62 (2000) 105009, hep-ph/0003173.
A.G. Riess et al., Supernova Search Team Collaboration, Astron. J. 116 (1998) 1009, astroph/9805201;
S. Perlmutter et al., Supernova Cosmology Project Collaboration, Astrophys. J. 517 (1999) 565,
astro-ph/9812133.
P. Kraus, JHEP 9912 (1999) 011, hep-th/9910149.
P. Bowcock, C. Charmousis, R. Gregory, Class. Quant. Grav. 17 (2000) 4745, hep-th/0007177.
G. Birkhoff, Relativity and Modern Physics, Cambridge, MA, 1923.
C. Grojean, Phys. Lett. B 479 (2000) 273, hep-th/0002130.
D. Ida, JHEP 0009 (2000) 014, gr-qc/9912002.
H. Stoica, S.H. Tye, I. Wasserman, Phys. Lett. B 482 (2000) 205, hep-th/0004126.
A. Kehagias, E. Kiritsis, JHEP 9911 (1999) 022, hep-th/9910174.
P. Bintruy, C. Deffayet, D. Langlois, Nucl. Phys. B 565 (2000) 269, hep-th/9905012.
C. Gmez, B. Janssen, P.J. Silva, JHEP 0004 (2000) 027, hep-th/0003002.
M.A. Luty, R. Sundrum, Phys. Rev. D 62 (2000) 035008, hep-th/9910202.
S.W. Hawking, Phys. Lett. B 134 (1984) 403.
M. Duff, Phys. Lett. B 226 (1989) 36.
N. Turok, S.W. Hawking, Phys. Lett. B 432 (1998) 271, hep-th/9803156.
S.W. Hawking, G.F.R. Ellis, The Large Scale Structure of SpaceTime, Cambridge Univ. Press,
1973, pp. 124134.
A. Karch, L. Randall, hep-th/0011156.
D. Youm, Phys. Rev. D 62 (2000) 084002, hep-th/0004144.
A. Albrecht, J. Magueijo, Phys. Rev. D 59 (1999) 043516, astro-ph/9811018;
J. Magueijo, Phys. Rev. D 62 (2000) 103521, gr-qc/0007036.
J.W. Moffat, astro-ph/9811390.
E. Kiritsis, JHEP 9910 (1999) 010, hep-th/9906206;
S.H. Alexander, JHEP 0011 (2000) 017, hep-th/9912037.
K. Landsteiner, E. Lopez, M.H. Tytgat, JHEP 0009 (2000) 027, hep-th/0006210.
D. Bak, K. Lee, J. Park, hep-th/001109.
A. Hashimoto, N. Itzhaki, hep-th/0012093.
342
[48]
[49]
[50]
[51]
C. Cski, J. Erlich, T.J. Hollowood, J. Terning, hep-th/0003076.

C. Cski, J. Erlich, T.J. Hollowood, Y. Shirman, Nucl. Phys. B 581 (2000) 309, hep-th/0001033.
B.C. Barish, gr-qc/9905026.
C.M. Will, gr-qc/9811036.

ChernSimons supergravity plus matter near

the boundary of AdS3
N.S. Deger a , A. Kaya a , E. Sezgin a , P. Sundell b , Y. Tanii c
a Center for Theoretical Physics, Texas A&M University, College Station, TX 77843, USA
b Institute for Theoretical Physics, Nijenborgh 4, 9747 AG Groningen, The Netherlands
c Physics Department, Faculty of Science, Saitama University, Urawa, Saitama 338-8570, Japan
Received 5 March 2001; accepted 15 March 2001
Abstract
We examine the boundary behaviour of the gauged N = (2, 0) supergravity in D = 3 coupled to an
arbitrary number of scalar supermultiplets which parametrize a Khler manifold. In addition to the
gravitational coupling constant, the model depends on two parameters, namely the cosmological
constant and the size of the Khler manifold. It is shown that regular and irregular boundary
conditions can be imposed on the matter fields depending on the size of the sigma model manifold.
It is also shown that the super AdS transformations in the bulk produce the transformations of the
N = (2, 0) conformal supergravity and scalar multiplets on the boundary, containing fields with
nonvanishing Weyl weights determined by the ratio of the sigma model and the gravitational coupling
constants. Various types of (2,0) superconformal multiplets are found on the boundary and in one case
the superconformal symmetry is shown to be realized in an unconventional way. 2001 Elsevier
Science B.V. All rights reserved.
1. Introduction
In probing various aspects of the remarkable connections between anti-de-Sitter and
conformal supergravity theories, the AdS3 /CFT 2 correspondence in particular provides a
relatively more manageable case to study. At the same time, some novel features arise
due to the fact that AdS supergravity in D = 3 is essentially non-dynamical. Nonetheless,
AdS3 supergravity plays a significant role in the description of the matter fields to which
it couples. As a step towards a detailed study of the amplitudes, anomalies and other
significant properties of this type of theories, it is useful to determine precisely the
behaviour of the AdS supersymmetry transformations at the boundary. This problem has
been examined for pure N 2 AdS3 supergravity in [1], pure N = 4 AdS3 supergravity in
E-mail address: sezgin@phys.tamu.edu (E. Sezgin).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 3 9 - 0
344
N.S. Deger et al. / Nuclear Physics B 604 (2001) 343366
[2], 1 the maximal AdS7 supergravity in [5], the maximal F (4) AdS6 supergravity in [6] and
the minimal AdS5 supergravity in [7], where it was shown that the correct transformations
rules of the boundary conformal supergravities indeed follow from a careful study of the
bulk super AdS transformations. However, a similar analysis does not seem to have been
carried out so far for matter coupled AdS supergravities which should shed further light on
the AdS/CFT duality questions in the context of M-theory in backgrounds with less than
maximal supersymmetry. Our aim in this paper is to fill this gap.
AdS supergravities are based on AdS superalgebras. Given the fact that the AdS group
in 2 + 1 dimensions is a product of two factors as SO(2, 2) = SO(2, 1)L SO(2, 1)R , the
super AdS group itself has the factored form GL GR . It turns out that there are many
choices for GL,R , the most typical case being OSp(2, p) OSp(2, q), where p and q are
not necessarily equal. The supergravity theories based on these algebras will be referred
to as the N = (p, q) AdS3 supergravities. They have been constructed as ChernSimons
gauged theories long ago by Achucarro and Townsend [8]. However, very little is known
so far about their matter couplings. In fact, the only cases studied until now seem to be the
N = (2, 0) AdS3 supergravity coupled to an arbitrary number of scalar supermultiplets [9,
10] and N = (16, 0) AdS3 supergravity with an exceptional sigma model sector [11]. The
models constructed in [9] and [10] are significantly different from each other, stemming
from the fact that the scalar fields are neutral under the U (1) R-symmetry group in the
model of [9], but charged in the model of [10]. The IzquierdoTownsend model has only
one free parameter, namely the cosmological constant, in addition to the gravitational
constant, unlike the model of [10], where there is the additional parameter that measures
the size of the sigma model manifold. In fact, the U (1) charge carried by the scalar fields
is related to this size, and as we will show in this paper, the limit in which the U (1) charge
vanishes implies a flat sigma model manifold, and the models of [9] and [10] do indeed
agree in that case.
The model studied here is expected to arise from a compactification of M-theory.
The much studied compactification of Type IIB theory on AdS3 S 3 K, where K is
essentially T 4 or K3, gives rise to N = (4, 4) or N = (4, 0) AdS3 supergravities coupled
to matter. The spectra of these theories are known [12,13] but not their actions so far.
Whether our N = (2, 0) model arises as a consistent truncation of such theories remains to
be seen.
The super AdS transformations in the bulk theory studied here are shown to produce
the transformations of N = (2, 0) conformal supergravity coupled to scalar multiplets with
nonvanishing Weyl weight determined by the ratio of the Khler sigma model manifold
and the gravitational coupling constant. In doing so, the so called regular and irregular
boundary conditions are utilized [14,15]. These choices of boundary conditions result in
the phenomenon in which scalar fields in AdS space of sufficiently negative mass-squared
can be associated with CFT operators of two possible dimensions. An example of this has
been discussed in [15] in the context of AdS5 T 1,1 compactification of Type IIB string
1 Various aspects of the AdS/CFT correspondence for pure supergravities in D = 3, in particular the asymptotic
symmetries, have been studied in [3,4].
345
theory. Here, we provide another example of this phenomenon and show explicitly the
resulting CFT supergravity plus matter symmetry transformations. In doing so, we find
an interesting conformal supermultiplet structure that involves a submultiplet of fields that
transform into each other. In this novel multiplet the superconformal symmetry is realized
in an unconventional fashion.
The (2, 0) model is described in the next section. The relation between the models of
[9] and [10] is described in Section 3. The boundary conditions and the linearized field
equations are given in Section 4, and the bosonic and fermionic symmetries of the boundary
CFT are obtained in Section 5. Concluding remarks are contained in Section 6.
2. The matter-coupled N = (2, 0) AdS3 Supergravity
The N = (2, 0) AdS3 supergravity multiplet consists of a graviton e a , two Majorana
gravitini (with the SO(2) spinor index suppressed) and an SO(2) gauge field
A . The n copies of the N = (2, 0) scalar multiplet, on the other hand, consists
of 2n real scalar fields ( = 1, . . . , 2n) and 2n Majorana fermions r (r =
1, . . . , n and the SO(2) spinor indices are suppressed).
In [10], the sigma model manifold M was taken to be a coset space of the form G/H
SO(2) where G can be compact or non-compact and H SO(2) is the maximal compact
subgroup of G, where SO(2) is the R-symmetry group. In particular, the following cases
are considered [10]
SO(n + 2)
SO(n, 2)
,
M =
.
M+ =
(2.1)
SO(n) SO(2)
SO(n) SO(2)
The results can be readily translated to the case of G/H U (1) with G = SU(n + 1) or
SU(n, 1) and H = SU(n).
Key ingredients in the description of the model are the matrices (LI i , LI r ), where I =
1, . . . , n+2, i = 1, 2, r = 1, . . . , n, which form a representative of the coset M . It follows
that
LI i LIj = ij ,
LI r LI s = rs ,
LI i LI r = 0,
LI i LJ i + LI r LJ r = IJ ,
(2.2)
where correspond to the scalar manifolds M . The SO(n), SO(2) and SO(n + 2) vector
indices are raised and lowered with the Kronecker deltas and the SO(n, 2) vector indices
with the metric I J = diag(+ + + ).
Other important ingredient of the model is the SO(2) gauged pull-back of the Maurer
Cartan form on M which is decomposed into the SO(n) SO(2) connections Qrs
and
ij
Q , and the nonlinear covariant derivative Pir as follows:

ir
1
ij
1
rs
Qij
Qrs
,
Pir = L1 D L ,
= L D L ,
= L D L
where the SO(2) covariant derivative is defined as

1
T
L.
D L = + Aij
ij
2
(2.3)
(2.4)
346
The anti-hermitian SO(2) generator Tij occurring in this definition is realized in terms
of an (n + 2) (n + 2) matrix, which can be chosen as (Tij )I J = ( I i jJ i j ).
Introducing the coordinates ( = 1, . . . , 2n) which parametrize the scalar manifold
G/H , we can also define the coset vielbein Vir and the SO(2) SO(n) connections
ij
A , Ars
on G/H as

ir
1
ij
1
rs
Vir = L1 L ,
(2.5)
Aij
Ars
,
= L L ,
= L L
where / . From the above relations it follows that
Pir = Vir + A S ir ,
(2.6)
Q = A + A C,
(2.7)
Qrs
ij
Ars
+ A C ,
rs
ij
(2.8)
ij
where A = A * ij , A = A * ij , Q = Q * ij and the (C, S ir ) functions are defined as

kl
*ij * kl C = L1 Tij L ,

rs
*ij C rs = L1 Tij L ,

kr
*ij S kr = L1 Tij L .
(2.9)
The matter-coupled N = (2, 0) ChernSimons supergravity Lagrangian which makes
use of these ingredients has been obtained in [10]. Up to quartic fermions the Lagrangian
is as follows [10]: 2
e1
1
e1
1
* D
e1 L = R +
* A A 2 Pir Pir
4
4
2
16ma
4a
1
1
m
r i Pir C 2
+ r D r +
2
2a
2

1
3
ir
2
2ma i r CS m 1 + 4*a r r C 2
2
+ 2ma 2 r 3 s C rs C + 2ma 2 r i j s S ir S j s

+ 2m2 C 2 C 2 2a 2 S ir Sir ,
(2.10)
which has the local N = 2 supersymmetry

= D + m C 2 ,
ea = a ,

A = 4ma 2 3 C 4ma 3 r i S ir ,
Li I LI r = a i r ,

1
r = Pir + 2ma 3 CS ir i .
2a
(2.11)
2 Conventions: = ( + +), * = * i , C and C are symmetric and = 1 * . The SO(2)

ab
0
g
charge conjugation matrix is unity, i is symmetric and { i , j } = 2 ij . A convenient representation is 1 =
1 , 2 = 3 . We define 3 = 1 2 . Note that ( 3 )2 = 1.
The functions C and S ir are defined in (2.9) and 3

1 ab
1
3
D = + ab 2 Q ,
4
2a

1 ab
1
r
3
D = + ab + * + 2 Q r + Q rs s .
4
2a
347
(2.12)
The parameter * = 1 corresponds to the manifolds M defined in (2.1), and the

constant a is the characteristic curvature of M (e.g., 2a is the inverse radius in the case of
M+ = S 2 ). The gravitational coupling constant has been set equal to one, but it can easily
be introduced by dimensional analysis. The constant m is the AdS3 cosmological constant.
Unlike in a typical anti-de-Sitter supergravity coupled to matter, here the constants , a, m
are not related to each other for non-compact scalar manifolds, while a is quantized in
terms of in the compact case as [10].
To conclude this section, and for later purposes, we list the equations of motion which
follow from the Lagrangian (2.10):

R a 2 Pir Pir + 8m2 C 2 C 2 2a 2S ir Sir g = 0,
(2.13)
1
+ 2m[ ] C 2 + 2ma i 3 r CS ir i r Pir = 0,
(2.14)
2a
F 4ma 2 g* Pir S ir = 0,
(2.15)
1
i Pir + 4ma 2 3 s C rs C
D r m(1 + 4*a 2) r C 2 +
2a
+ 4ma 2 i j s S ir S j s 2ma 3 i CS ir = 0,
(2.16)

2

ir
2 2
jr
2
2 ks
D P + 16m a *ij S C 1 + *a C a S Sks
+ 16m2 a 4 C rs Sis C 2 = 0,
(2.17)
where the fermion bilinears in the bosonic field equations have been suppressed and
= D D ,

1
kr
ir
+ Qrs
.
D P ir = gg Pir + *Qik
P
P
g
(2.18)
(2.19)
3. Connection with the IzquierdoTownsend model

The model reviewed above [10] differs from the one constructed by Izquierdo and
Townsend [9], in all the terms containing the C and S-functions. These differences stem
from the fact that the scalar fields in the model above are charged under the R-symmetry
group SO(2) while in the model of [9] they are neutral. Given that this charge is related to
the sigma model radius, taking the zero charge limit in order to compare the two models is
expected to constrain the scalar manifold. Here, we will show the relation between the two
3 The * term in D r was inadvertently omitted in [10].
348
models and show that they indeed agree only in the limit in which the scalar manifold is
flat.
We begin by parametrizing the coset representative L as follows

0
ir
L = exp
(3.1)
,
*( ir )T
0
where ir are 2n real coordinates on M . Next, we perform the rescalings
A a 2 A ,
ir a ir
(3.2)
and consider the limit a 2 0. From the definitions in (2.3) we find

Pir = a ir + ,

1 ir
j
2
Q = a A r *ij + ,
2

2
ri s
ri
s
=
a
Qrs
i i + ,
(3.3)
where denote higher order terms in positive powers of a 2 . Let us define

1
j
d A = ir dr *ij ,
(3.4)
2
so that Q = a 2 (A + A ), where the index represents a pair of indices (ir). We
have
1
1
j
d d F = d ir dr *ij ,
(3.5)
2
2
where F = A A . In the limit a 2 0 the Lagrangian becomes
1
1
1 e1
1
* A A ir ir
e1 L = R + e1 * D
4
2
16 m
4
1
1
ir
+ r D + r i
2
2
1
1
m m r r + 2m2 ,
2
2
and the transformation rules become
ea = a ,
(3.6)
= D + m ,
ir = i r ,
A = 4m ,
1
r = ir i ,
2
where the covariant derivatives are defined by

1
1
D = A 3 A 3 ,
2
2

1
1
D r = + A 3 + A 3 r ,
2
2

1
1
D = A 3 A 3 ,
2
2
3
(3.7)
(3.8)
349
and = + 14 ab ab . Introducing
1
A = A + A = A + ir j r *ij ,
2
the Lagrangian (3.6) becomes
1
1
1 e1
1
* A A ir ir
e1 L = R + e1 * D
4
2
16 m
4
1
1
1
A J + A J + r D r + r i ir
2
2
2
1
1
m m r r + 2m2 ,
2
2
where the current J is defined as
(3.9)
(3.10)
1
(3.11)
* F ,
16me
defined in (3.5). The transformations rules (3.7), on the other hand, become
J =
with F
e a = a ,
A
= D + m ,
= 4m F ,
1
r = ir i ,
ir = i r ,
(3.12)
2
where we have discarded a term in A which can be expressed as a gauge transformation
A = . Of course, the above transformations are up to cubic fermion terms in the
transformation rules of and r , since the Lagrangian (3.10) is up to quartic fermion
terms. Note also that the covariant derivatives have now simplified to

1
1
D r = + A 3 r ,
D = A 3 ,
2
2

1
D = A 3 .
(3.13)
2
3
The formulae (3.10), (3.12) and (3.13) agree with those of the IzquierdoTownsend
model [9] for the flat sigma model. Note that in trying to set the U (1) charge of the scalar
fields equal to zero, we have been forced to flatten the sigma model manifold. This is due to
the fact that the U (1) charge is related to the radius of the scalar manifold. The flat model
discussed here will be used in Section 5.3.
4. Boundary conditions and linearized field equations

In order to examine the properties of the model described above near the boundary,
we shall begin by fixing certain gauges and studying the behaviour of the linearized field
equations near the boundary.
The AdS3 spacetime can be covered by two regions each of which is parametrized by
a set of Poincar coordinates (x 0 , x 1 , x 2 ) in R3 with x 2 > 0. We shall use the notation
350
x = (x 0 , x 1 ) and x 2 r. This patch contains half the boundary of AdS3 in the form of
the Minkowskian plane at r = 0. The other region is behind the horizon at r = . In what
follows we shall work only within one of the regions.
Following [1], we choose the following gauge conditions
1
,
er a = 0,
2mr
Ar = 0,
r = 0,
er 2 =
e 2 = 0,
(4.1)
where a = 0, 1 is the tangent space index in D = 2. Note that the second coordinate is
labeled as r in curved space and as 2 in tangent space. The metric in this gauge takes the
form

1 2
dr + dx dx g ,
ds 2 =
(4.2)
2
(2mr)
where g = e a e b ab . The SO(2, 2) invariant AdS metric corresponds to the case g =
. The components of the spin connection following from the metric (4.2) are
r ab = e[b r e a] ,
ab = ab ,
1
a2 = e a + e(b r e a) eb ,
r a2 = 0.
r
(4.3)
When g = , the only nonvanishing component is a2 = a /r.

We next study the asymptotic behaviour of the solutions of the linearized field equations
near the boundary r = 0. We are going to do this in Euclidean signature. In this signature
the AdS space consists of a single region covered by Poincar coordinates plus a point at
r = . This point is actually a boundary point and the boundary has the topology of the
two sphere, represented in the Poincar coordinates by the Euclidean plane at r = 0 plus
the point at infinity.
We will assume that the dreibein e a behaves as r 1 as in the SO(2, 2) invariant case.
To determine the asymptotic behaviours of the remaining fields, we need to examine their
linearized field equations as expanded around the supersymmetric AdS background in
which the only nonvanishing fields are
g = ,
L = 1.
(4.4)
Next, we use the coset representative L given in (3.1) which leads to the following
expressions at the linearized level
Pir ir ,
C 1,
C rs 0,
S ir ** ij j r .
The field equations for A and linearized around the background (4.4) are
r A = 0,
[ A] = 0,
1
r = r 1 ,
2
4 We use the convention = 1 * 2 .
g
(4.5)
4
(4.6)
(4.7)
351
where the suffixes indicate the eigenvalues of 2 , which in turn indicate the chiralities of
the spinors on the boundary. The equations involving radial derivatives are readily solved
to all orders in r, and in a convenient normalization we have
A = A(0),
= (2mr)1/2 (0)+ + (2mr)1/2(0) .
(4.8)
To give the proper boundary condition for the vector field, we define 5
1
g g * A(0) .
2
The remaining equation for A in (4.6) then amounts to
A(0) = P A(0) =
A+ = + A .
(4.9)
(4.10)
As we shall show in the next section, the anti self dual component A(0) forms an
off-shell d = 2 supermultiplet together with (0)+ and e(0) a . Thus it is natural to treat
A(0) as the independent boundary field, and let A(0)+ be determined from (4.10). The
fact that only one of the Hodge dualities of the vector field is independent can also be
understood by considering the Hamiltonian formulation of the bulk ChernSimons theory,
where A(0) form a pair of canonically conjugate variables. Thus, the proper boundary
conditions for the supergravity multiplet are:
e a (2mr)1 e(0) a ,
+ (2mr)1/2 (0)+
A A(0) .
(4.11)
We now turn to the discussion of the boundary conditions on the matter fields, starting with
the scalar fields.
4.1. Matter scalars
The linearized scalar field equation near the boundary is given by (the r-dependence is
shown explicitly and the SO(2) SO(n) indices of ir and r are suppressed):

r 2 + r 3 r r 1 r m2 = 0,
(4.12)
where
m2 = 4a 2 (a 2 + *) = ( 2),
and equals + or defined by

() = 1 1 + m2 .
Thus, in terms of * and a 2 :

+ () = 2 + 2a 2 ,
* = 1:
() = 2a 2,
(4.13)
(4.14)
(4.15)
5 We use a notation in which the chiralities and Hodge dualities are labeled by lower indices and the regular
and irregular nature of boundary conditions are labeled by upper indices.
352

* = 1:

+ () = 1 + 1 2a 2 ,

() = 1 1 2a 2 .
(4.16)
A free scalar field behaves near the boundary as

+
+
(r, x ) (2mr) () (0)
+ (2mr)2 (2)
+

+ (2mr)2 (2)
+
+ (2mr)+ () (0)
(4.17)
/ Z, and
for 2a 2
+
(r, x ) (2mr) () (0)
+ (2mr)2 (2)
+

+ (2mr)2(2)
+
+ (2mr)+ () ln(2mr) (0)
(4.18)
depend only on x . For 2a 2

/ Z and for
for 2a 2 Z. The expansion coefficients (2n)
* = 1, a 2 = 1/2, that is + = 1, the coefficients (2n)

, n 1, are determined through
the linearized field equations as local expressions in terms of (0)

. For other values of *
+
2
and a , that is for + = 2, 3, . . . , the coefficients (2n) , n + 2 are given in terms of
+
+
(0)
while (2
is undetermined and thus independent. At higher order in r one then
+ 2)
+
+
+
and that (2
, (2
, . . . are given
finds that (0), (2) , . . . are given in terms of (0)
+)
+ +2)
+
in terms of (2+ 2) . The above results follow from the small z expansion of the modified
Bessel functions [16]. Similar results hold for small perturbations around the anti-de-Sitter
background [17].
There are two types of boundary conditions that may be imposed on the scalars:
regular conditions which amount to specifying the leading component at the boundary and
irregular conditions which amount to specifying the independent subleading component
/ Z, a regular
described above [14,15], and which are possible when 0. Thus, for 2a 2
+
boundary condition amounts to specifying (0)
while an irregular boundary condition
amounts to specifying (0)

. For + = 1, that is, for * = 1, a 2 = 1/2, a regular boundary
+
condition amounts to specifying (0)
and an irregular condition amounts to specifying (0)
.
Given that (0) are associated with conformal operators of weight (), the requirement of unitarity imposes the following restriction for irregular boundary conditions:
() 0.
(4.19)
For regular boundary conditions, the unitarity condition is automatically satisfied, while
for irregular conditions (4.19) restricts the possible values of a 2 . Thus it follows that the
following boundary conditions are possible:
* = 1:
+
regular: (2mr) () (0)
irregular:
+
(2mr)2 (2)
* = 1:
regular:
for a 2 0
for a = 0.
2
+
(2mr) () (0)
for a 2 = 12 ,
(2mr) ln(2mr)(0)
for a 2 = 12 .
(4.20)
(4.21)
(4.22)
+ ()
(2mr)
(0)
+
irregular: (2mr)(0)
(2mr)2 +
(2)
for 0 < a 2 < 1, a 2 = 12 ,

for a 2 = 12 ,
for a 2 = 0, 1.
353
(4.23)
4.2. Matter fermions

We now turn to the boundary conditions on the matter fermions. The linearized equations
obeyed by the matter fermions near the boundary is
r + r2 r 2 m = 0,
(4.24)
where the fermion mass is given by

1
m = 1 + 4*a 2 .
2
/ Z + 12 a solution to (4.24) is given by
We find that for mu

= (2mr)1mu (0) + 2mr(1)+ +

+ (2mr)1+mu (0)+ + 2mr(1) + ,

where refers to the 2 eigenvalue. For mu =

= (2mr)1mu (0) + 2mr(1)+ +

+ (2mr)1+mu ln 2mr (0)+ + 2mr(1) + .
1 3 5
2, 2, 2,...,
(4.25)
(4.26)
the solution takes the form
(4.27)
= 12 , 32 , 52 , . . . ,
For
the solution is given by (4.27) with and all the
chiralities flipped. For later use, we also record the following relation
1
(4.28)
(0) , 2 1 = 0.
2m(2m 1)
Note that unlike for the scalars, the coefficients in the logarithmic branch in (4.27) is never
undetermined. Next, following [18], we define the conformal weights of the fermions as

1
() = 1 | | = 1 1 + 4*a 2.
(4.29)
2
Thus, in terms of * and a 2 :

+ () = 32 + 2a 2
* = 1:
(4.30)
() = 12 2a 2 .

+ () = 1 + 12 1 4a 2 ,
* = 1:
(4.31)

() = 1 12 1 4a 2 .
(1) =
Thus, the regular boundary conditions are associated with r () behaviour and the
irregular boundary conditions with r + () behaviour (note that this holds for all values of
a 2 ). It follows from (4.26) and (4.27) that the chirality of the regular and irregular boundary
spinors
(0) is equal to minus the sign of the fermion mass:

2
(4.32)
= 0,
| | (0)
354
where the superscript + refers to regular and to irregular boundary conditions.

Thus, in the case of regular boundary conditions the chirality is negative for positive fermion mass, and positive for negative fermion mass. In the case of irregular
boundary conditions the chirality is positive for positive mass and negative for negative
mass.
Imposing the following unitarity condition
1
() ,
2
it follows that the allowed boundary conditions for the matter fermions are:
(4.33)
* = 1:
for a 2 0,
regular: (2mr) () (0)

irregular: (2mr)
* = 1:
regular:
irregular:
3/2
(1)+
for a = 0.
(2mr) () (0)
for 0 a 2 14 ,
(2mr) () (0)+ for a 2 14 .
(2mr)3/2(1)+
for a 2 = 0,
(2mr)+ () (0)+ for 0 < a 2 < 1 ,

4
(2mr)+ () (0)
(2mr)3/2(1)
for
1
4
< a 2 < 12 ,
(4.34)
(4.35)
(4.36)
(4.37)
for a 2 = 12 .
Note that for * = 1 and a 2 = 1/4 the regular boundary condition can be imposed on
either of the chiralities.
5. The local conformal supersymmetry on the boundary

In this section we shall derive the realization of the d = 2, N = (2, 0) conformal supera ,
symmetry on the boundary supergravity multiplet (e(0)
(0)+ , A(0) ) and boundary
chiral multiplets involving fields that are to be specified case by case in accordance with
the boundary conditions.
The d = 2 symmetries are found by examining the nature of the bulk transformation
rules close to the boundary. To analyze this we first find the D = 3 transformation
parameters which preserve the D = 3 gauge conditions (4.1) near the boundary. We then
evaluate the resulting D = 3 transformations of a solution to the D = 3 field equations
with given set of boundary data (0) . By matching powers of r in the limit when r 0
we thus obtain the resulting d = 2 transformations (0) . In specifying the boundary data
we have to choose between regular and irregular boundary conditions such that (0) is a
local expression in terms of (0) and its derivatives.
For * = 1 and a 2 > 0, the scalar fields diverge at the boundary and the perturbative expansion breaks down, therefore we shall exclude the case * = 1 from now on. For * = 1
355
and a 2 > 0, the matter scalars diverge for a 2 > 1 and the matter fermions diverge for
a 2 > 3/4. Moreover, for 12 < a 2 < 34 , the supersymmetry transformation of the vector field
involve matter contributions which diverge as r 0 (see footnote below (5.9)), which is
not consistent with its r expansion. For 0 < a 2 1/2 the r expansion is well-defined, and
in fact, since all matter fields have positive Weyl weight for this case, the nonlinearities
vanish at the boundary. Finally, for a 2 = 0, the appropriate model to consider is the R2n
sigma model, in which case the scalars have Weyl weight zero. In summary, the perturbative expansion makes sense only for
1
0 a2 ,
2
or * = +1, a 2 = 0.
* = 1,
(5.1)
We remark that in the case of single scalar multiplet coupling, namely, when the sigma
model manifold is S 2 for * = 1 and H 2 for * = 1, the excluded range of the parameter
a 2 coincides with the fact that the scalar potential has the form of a confining well. In the
allowed range, however, the potential is unbounded from below. It would be interesting to
study if the potential exhibits qualitatively similar behaviour for arbitrary number of scalar
multiplets.
Importantly, regularity of the D = 3 solutions determines the irregular boundary
conditions in terms of the regular, or vice versa, which leads to the usual interpretations
of the anti-de- Sitter/conformal field theory duality. This leads to subtleties, however, in
the case of irregular boundary conditions in the matter sector, where the nonlinearities
appear to lead to mixings between the regular and irregular fields in the transformations
of the irregular fields (at least for certain rational values of a 2 ). As a first step towards
understanding this, it is reasonable to begin by examining the nature of the transformations
of the irregular fields among themselves by formally putting the regular fields to zero in
the case of irregular boundary conditions. This is our approach when 0 < a 2 < 1/2, which
we refer to as case 1 below. For the special values a 2 = 1/2 and a 2 = 0, which we treat
separately as case 2 and 3 below, the nonlinearities are however more manageable, and for
these two cases we therefore keep both regular and irregular fields. Thus, in summary, the
boundary conditions in the matter sector are taken as follows:
1
0 < a 2 < , either regular or irregular matter fields;
2
1
Case 2: * = 1, a 2 = , both regular and irregular matter fields;
2
Case 3: a 2 = 0, both regular and irregular matter fields.
Case 1: * = 1,
(5.2)
5.1. Case 1: * = 1 and 0 < a 2 < 1/2

We begin by determining the asymptotic behaviors of the local symmetries of the
bulk which preserve the gauge conditions (4.1) near the boundary. Using the asymptotic
behavior of the supergravity multiplet fields given in (4.11), we find (without linearizing
in the fields) that the residual gauge symmetries are
356
r = rD(0),
= (0) + O(r 2 ),
2
a2
L = O(r ),
= (0) + O(r
= (2mr)
1/2
ab
2
ab
L = L(0) * + O(r ),
4a 2
),

(0) + O(r 2 ) ,
(5.3)
where D(0) , (0) , L(0) , (0) and (0)) denote the parameters of dilatation, reparametrization, Lorentz rotation, SO(2) rotation and supersymmetry, respectively, and the fields
with suffix (0) are arbitrary functions of x . Note that the parameter r is determined fully
and it has only linear r-dependence, while the other parameters have series expansions in
r. The form of r , and a2
L come from the variations of the gauge conditions involving
the dreibein and the form of ab
L can be deduced from the requirement of residual Lorentz
transformations on the boundary. The last two results come from the variation of the gauge
conditions Ar = 0 and r = 0, respectively.
To derive the bosonic transformations in d = 2, we insert (5.3) together with the
expansions (4.11), (4.17)(4.18) and (4.26)(4.27) into the bosonic transformations in D =
3. These do not mix different chiralities and powers of r. It is therefore straightforward
to read off the transformation for the leading components. We find the usual general
coordinate transformations of all the fields with parameter (0) , and (in the rest of
subsection we have dropped the (0) labels for notational simplicity)

e a = D ab L * ab eb ,

1
1
1
+ =
D + L + 2 3 + ,
2
2
2a

1
A = g g* ,
2
ir = ()D ir + * ij j r ,

1
1
r = ()D + L 2 + 1 2 3 r .
(5.4)
2
2a
Note, the superscripts on the matter fields refer to regular/irregular boundary conditions,
and the chiralities of the matter fermions are given by (4.32).
To find the d = 2 supersymmetry transformation rules we substitute the expression for
the supersymmetry parameter given in (5.3) into the D = 3 supersymmetry transformation
rules (2.11) and take the limit r 0. In the supergravity sector we find:
e a = a ,
= D + 2 ,
1
A = a 2 3 + 2a 2
3 ,
2
where we have introduced the notation
= a ea ,
= + ,
1
= m 2 3 A+ + ,
8a
(5.5)

1
1
D = 2 A 3 ,
2
2a
357
(5.6)
and the d = 2 gravitino field strength is defined as in (2.18) but with the covariant
derivative defined above. We note the correction to the special supersymmetry parameter
. In the gravitino transformation rule, this correction arises from the A+ contribution to
the three-dimensional D , while in the vector transformation rule it arises from the D =
3 covariant derivative in the gravitino field strength and from the varying the self-duality
projector according to the following:

A P A = P A A+ ,
(5.7)
where the projection is defined in (4.9). We also note that in obtaining the vector
transformation rule we have eliminated the anti self dual component of using the
boundary limit of the component of the D = 3 gravitino equation (2.14) as follows:
m
1
1
= + 2 3 A+ + ma i r+ +ir ,
2
4
8a
(5.8)
where = D D with the covariant derivative defined as in (5.6). In deriving

(5.8) one notices that the last two terms in the component of the D = 3 gravitino
equation (2.14) add up in the leading order, which follows from the fact that for 0 < a 2 <
1
2 the D = 3 scalar fields obey

rr ir = 2a 2 ir + O(r) .
(5.9)
In the r component of the D = 3 gravitino equation, however, these terms cancel, which
means that there is no matter contribution to the leading order. Thus we have obtained
a local realization of the boundary supersymmetry on the (2, 0) conformal supergravity
multiplet in d = 2 which is off-shell and decoupled from matter. 6
We next study the transformations of the matter fields. We recall from (4.22)(4.23)
and (4.36)(4.37) that both regular and irregular boundary conditions are admissible for
0 < a 2 < 12 . There are two sets of combined regular and irregular boundary conditions
which lead to two types of conformal (2, 0) supermultiplets which will be referred to as
Type 1 and Type 2. The Type 1 multiplet consists of regular scalars, and fermions which
are regular for 0 < a 2 < 14 and irregular for 14 < a 2 < 12 . The Type 2 multiplet, on the
other hand, is a novel multiplet which consists of irregular scalars, and fermions which are
irregular for 0 < a 2 < 14 and regular for 14 < a 2 < 12 .
To find the transformations at the linearized level, we insert (4.11), (4.17) and (4.26) into
the D = 3 transformation rules in (2.11) and match powers of r. In the case of Type 2, we
also use the Dirac equation (4.28) in obtaining the scalar transformation. The results are as
follows:
6 At this point we can see why the range 1 < a 2 < 3 does not yield a local realization of the boundary
2
4
supersymmetry: although the matter fields vanish at the boundary, so that we can trust the r-expansion, (5.9) is
now replaced by rr ir = (2 2a 2 )( ir + O(r)) which results in divergent, matter dependent contributions to
A .
358
Linearized Type 1:
Linearized Type 2:
+ ir
= a + i r ,

r = 1 + ir i + + 4ma + ir i ,
2a
a
ir =
+ i r+ + a i r+ ,
4m(1 2a 2 )
r
+ = 2ma 1(1 2a 2) ir i + .
(5.10)
(5.11)
We now turn to the full transformation rules. In the case of the Type 1 multiplet all
the nonlinearities vanish except the A+ contribution to the special supersymmetry
transformation, and we find the local transformation rules
+ ir = a i r ,
Full Type 1:
(5.12)
1
r = + ir i + 4a + ir i ,
2a
where is given by (5.6). Note that the gauge field A does not appear in the derivative
of the scalars ir in the above formula because the chirality of projects it to its positive
Hodge dual, which is then absorbed into the special supersymmetry parameter . In other
words, the above formulae are U (1) covariant modulo transformations.
In the Type 2 case the first nontrivial order the D = 3 Dirac equation now yields the
following expression for the negative chirality spinor:
r(1) =
1
r+ + 1 3 A+ r+ ,
D
2
4m(1 2a )
8ma 2
(5.13)
is the supercovariant derivative defined by

where D
r = r + 2ma 1 (1 2a 2 ) ir i .
D
(5.14)
has been used and is

The fact that matter fermions have the U (1) charge 1 +
the ordinary Lorentz covariant derivative. Thus the full transformation rules for the Type 2
multiplet reads:
a
r+ + a
ir =
i D
i r+ ,
2)
4m(1
2a
m
Full Type 2:
(5.15)

r
+ = 2ma 1 1 2a 2 ir i .
1
2a 2
In summary, the transformation rules (5.4) and (5.5) are those of N = (2, 0) conformal
supergravity [19] consisting of fields (ea , + , A ) coupled to either one of the
following matter multiplets:
Type 1: scalar multiplets consisting of scalar fields ir with Weyl weight 2a 2 and
negative chirality spinors r with Weyl weight 12 + 2a 2 .
Type 2: scalar multiplets consisting of scalar fields ir with Weyl weight 2 2a 2 and
positive chirality spinors r+ with Weyl weight 32 2a 2 .
The supersymmetry transformation rules (5.5), (5.12) and (5.15) close off-shell as
follows

[1 , 2 ] = + L + ,
(5.16)

2
1
,
[ , ] = D (2) + L (2) + 4a 3 + 2
(5.17)
359
where = 1 *2 . The arguments on the right hand side are the composite parameters
of the relevant transformations. Note the absence of the usual field dependent U (1) gauge
transformation in the commutator of two supersymmetry transformations. In obtaining
(5.16)(5.18) we have used
1
= + 2 ,
(5.18)
2
which directly follows from ab = * ab which is determined from its algebraic equation
of motion as
= e1 * e a e a + .
(5.19)
The result (5.16)(5.17) is up to cubic fermion terms that may arise through some of the
composite and -transformations, since the transformations (2.11) were themselves up
to that order. In the case of Type 1 we have supercovariantized the derivative of +ir . In
the case of Type 2, the supercovariant derivative of r+ is already present, due to the fact
that the three-dimensional fermionic matter field equation (2.16) is already supercovariant.
However, in this case the closure of two supersymmetries on the fermion requires type
terms in that are expected to arise in the complete transformation rules.
5.2. Case 2: * = 1 and a 2 = 1/2
In this case, we recall the boundary behaviors of the matter fields from (4.18) and (4.27)
as follows:
+ir
+ir
ir = 2mr (0)
+ (2mr)2 (2)
+
ir
ir
+ 2mr ln(2mr) (0)
+ (2mr)2 (2)
+ ,

r = (2mr)1/2 r(0)+ + 2mrr(1) +

+ (2mr)3/2 ln(2mr) r(0) + 2mrr(1)+ + .

(5.20)
ir r
As we shall show, the regular boundary fields ((0)
, (0)+ ) form a supermultiplet of
Type 1, using the terminology introduced above, according to which multiplets containing
the regular scalars are called Type 1 and those containing the irregular scalars are called
+ir r
Type 2. As for the irregular fields ((0)
, (1) ), they will be shown to form an extended
multiplet, together with the fields of the Type 1 multiplet.
Substituting (5.20) into the negative chirality component of the field equation we find
that there are indeed no conditions on r(0)+ and r(1) and that r(0) is determined as
follows:
1 r
r(0) =
(5.21)
D (0)+ ,
2m
where
ir
r
r
D
(5.22)
(0)+ = (0)+ + 2 m(0) i (0)+ .
r
As we shall see later, D
(0)+ is supercovariant. Note that the U (1) charge (1 +
2
vanishes here since a = 1/2.
1
)
2a 2
360
Let us now examine the boundary behaviour of the local symmetry transformations,
starting with the bosonic ones. The bosonic transformations of the matter fields are the
same as in the 0 < a 2 < 12 case, except for the dilatation which acts as follows:
ir
ir
= D (0)
,
(0)
1
r = D r ,
(0)+
(0)+
2
Extended Type 1:
(5.23)
+ir
+ir
ir
(0) = D (0) D (0)

,
r
1
r .
D D
(1) = 32 D r(1) 2m
(0)+
We see that due to the logarithmic terms in the expansion and the fact that + =
there is an admixture of the Type 1 fields in the transformations of the irregular fields
+ir r
, (1) ). Thus, the full set of fields are considered to form an extended Type 1
((0)
multiplet. This multiplet structure will also emerge in the conformal supersymmetry
transformation rules.
We next turn to the boundary limits of the supersymmetry transformations. Those of the
supergravity multiplet take the same form as given in (5.5), with the replacement A
A defined as (we suppress the (0) labels in the supergravity sector):
1
A = A r(0)+ 3 r(0)+ .
(5.24)
4
To see this, we begin by noting that there is an additional log term in the expansion of
the RaritaSchwinger field. By substituting (5.20) into the r component of the Rarita
Schwinger field equation we find
+ (r, x ) = (2mr)1/2+ + ,
1
ir
(r, x ) = (2mr)1/2 + (2mr)1/2 ln(2mr) (0)
i r(0)+ + .
2
(5.25)
The component of the RaritaSchwinger field equation gives

1
1
m
= + 3 A+
2
4
4

1
1 ir
+ir
+ m (0) + (0) i r(0)+ ,
2
2
(5.26)
where = D D and D = ( A 3 ). Using this the fermionic

transformation of A becomes
1
1
ir
3 m i 3 r(0)+ (0)
.
A = 3 +
2
2
(5.27)
The matter dependence can be removed by the redefinition(5.24). The field A then
transforms as in (5.5) upon the use of r(0)+ given below. Altogether, we find the conformal
supergravity multiplet transformations (5.5), with a 2 = 1/2.
We next study the supersymmetry transformations of the matter fields near the boundary.
We find after rescaling 1 (and dropping the (0)s for notational simplicity)
2
Extended Type 1:
ir
1
r+ ,
= 2m
i D
r = m ir i ,
+

+ir = i r + m1
i r+ ,

r
= 12 +ir i + 2i +ir + 12 ir
361
(5.28)
where we have used the field equation (5.21) and made the field redefinition
1
A+ 3 r(0)+ .
(5.29)
4m
We observe that ( ir , r+ ) form a Type 1 submultiplet, whose fields are inert under
the conformal supersymmetry transformations, and whose supersymmetry transformations
can be obtained by rescaling
(5.15) as ir (1
the fields of the Type 2 multiplet
2
1
ir
2
and + 2 2+ , and taking the limit a 1/2. Correspondingly, the
2a )

superalgebra closes as in (5.16)(5.17) by setting a 2 = 1/2. The fields ( +ir , r ), on the
other hand, transform into each other under the ordinary supersymmetry transformations,
but transform into ( ir , r+ ) under the conformal supersymmetry transformations. (This
is similar to the situation of the dilatation transformation laws (5.23).) Therefore, we view

the enlarged set of fields ( +ir , ir , r , r+ ) as forming a conformal supermultiplet, with
closure given by (5.16)(5.17) for a 2 = 1/2 (with dilatation transformation (5.23)). This
extended Type 1 multiplet can be truncated consistently to an ordinary Type 1 multiplet by
setting the Type 2 submultiplet equal to zero, but the reverse is not consistent.

r = r(1)
5.3. Case 3: a 2 = 0
In this case the boundary behavior of the matter fields is given by
+
= (0)
+ (2mr)2 (2)
+ + (2mr)2 ln(2mr) (0)
+ (2mr)2 (2)
+ ,

= (2mr)1/2 (0) + 2mr(1)+ + (2mr)2 (2) +

+ (2mr)3/2 ln(2mr) (0)+ + 2mr(1) + .

(5.30)
The supersymmetric variations of the zweibein and gravitino are as in (5.5) while the
transformation of A has a subtlety due to the fact that there is an additional log term
in the expansion of the RaritaSchwinger field. Taking this into account, from the Rarita
Schwinger field equation we find
+ (r, x ) = (2mr)1/2+ + ,
(r, x ) = (2mr)1/2(0)
1
+ir
(2mr)1/2 ln(2mr)i r(0) (0)
+
+ ,
(5.31)
4m
1
1
1
m
+ir
(0) = + + 3 +
A+ i r(0) (0)
,
(5.32)
2
4
8
8
where = D D with the U (1) connection term in D shifted as in (3.8).
Using this the fermionic transformation of A becomes
1
+ir
A = 3 (0)+ + 2
3 (0)+ 3 i r(0) (0)
,
2
(5.33)
362
where we have used (5.7). To cancel the matter contributions, we first need to examine the
transformations of the matter fields. From the matter field equations and the boundary
conditions stated earlier for the case at hand,
we find that the independent fields at

+ir r
+ir r
the boundary are (0)
, (0) , (2)
, (1)+ . Under dilatations, these fields are found to
transform as
1
+ir
(0)
= 0,
r(0) = D r(0) ,
2
+ir
+ir
ir
(2)
= 2D (2)
D (0)
,
3
r(1)+ = D r(1)+ D r(0)+ ,
(5.34)
2
where, again, a mixing of the type observed earlier in the case of a 2 = 1/2 arises here. The
fermionic transformation of the matter fields also exhibit this kind of mixing:
1
+ir
r(0) = i (0)+ (0)
,
2
+ir
(2)
= (0)+ i r(2) + (0) i r(1)+ ,

1
1 ir
+ir
+ir
,
2mi (0)+ (2)
+ (0)
r(1)+ = i (0) (0)
2
2
+ir
(0)
= (0)+ i r(0) ,
(5.35)
ir
+ir
1
r
where (0)
= 8m
2 (0) and (2) is a more complicated function of the independent fields.
Armed with this result, we first redefine A as in (3.9) so that the newly defined field
A transforms precisely as in (5.5), that is without any matter contributions, as expected
from an off-shell conformal supergravity multiplet. As for the interpretation of the matter
multiplet transformations, surprisingly enough, the story is somewhat more complicated.
While the (0) parameter can be redefined into the special supersymmetry parameter
+ir
as in (5.6), a close examination of the transformation rules (2)
and r(1)+ shows that
the dependence of the result on A+ and the special supersymmetry gauge field, which is
an appropriately redefined , cannot be removed unless certain equations of motion are
imposed. However, these gauge fields are determined in terms of the independent fields as
nonlocal expressions. Therefore, in order to realize the conformal supersymmetry on the
boundary in a local fashion, we need to remove the dependence on these dependent gauge
fields. This can be achieved by setting

*
r
+ir = 0,
D
g
(5.36)
D
(0) = 0,
(0)
g
where the supercovariant derivatives are defined in a standard way in accordance with the

r
supersymmetry variations (5.35), and D
(0) contains the shifted field A . These field
equations transform into each other under the supersymmetry variations (5.35), as they
should.
Imposing the on-shell conditions (5.36), and recalling that the derivatives on (5.35) need
to be supercovariantized when considering the higher order fermion terms, we find that the
(0) term in the last equation in (5.35) drops out, and that the boundary evaluation of the
D = 3 matter field equations imply that

ir
(0)
= 0,
r(0)+ = 0,

1 r
1
r
3 r
D (1)+ + A+ (1)+ ,
(2) =
4m
2
363
(5.37)
(5.38)
with the supercovariant derivatives defined in standard way. Substituting these results in
(5.35), the A+ dependent term can be absorbed into redefinition of (0) to yield the
special supersymmetry parameter , and all in all, the supersymmetry transformation rules
(5.35) for the independent field disentangle into those of two separate multiplet of fields
as follows (dropping the (0) and chirality labels on the conformal supergravity fields and
parameters):
+ir
(0) = i r(0) ,
Type 1:
(5.39)
+ir
r(0) = 12 i (0)
,
+ir
1
i
(1)+ + 1
(2)
= 4m
i D
(1)+ ,
m
Type 2:
(5.40)
+ir
r
(1)+ = 2mi (2) .
Again, we have used the terminology of Type 1 and Type 2, according to whether the
multiplet contains regular or irregular scalar fields. The result for Type 1 agrees with that
of [19], where (2,0) conformal supergravity and its coupling to a sigma model in D = 2
is constructed. All the fields occurring in both multiplets above now have definite Weyl
weights since the mixings in the dilatation transformations (5.34) disappear upon the use
of (5.37)(5.38).
The algebra (5.39)(5.40) closes as in (5.17). In interpreting the composite U (1)
transformation, the composite parameter must be rescaled by a 2 , since A has been
rescaled by a 2 , prior to taking the limit a 2 0. The closure of the algebra can be seen
from the fact that the Type 1 and Type 2 multiplets here can be obtained from the Type 1
and Type 2 multiplets found Section 5.1 for 0 < a 2 < 12 , by first rescaling the scalar fields
as a and then taking a 2 0.
6. Conclusions
In this paper we have analyzed the behaviour of (2, 0) gauged supergravity coupled
to matter in D = 3 near the boundary of AdS. We have exhibited the role of the
bulk supergravity and matter field equations in determining the realization of conformal
supersymmetry on the boundary of AdS. We have found that various types of matter
multiplets emerge at the boundary in addition to a universal (2, 0) conformal supergravity
multiplet. These multiplets involve fields whose conformal dimensions depend on the
radius of the Khlerian sigma model coset space and on the gravitational coupling constant
(set equal to 1 in most of the paper). The nature of the boundary conformal multiplets found
depends crucially on the ratio of these constants. Interestingly, the local supersymmetry of
the D = 3 theory does not fix this ratio nor the sign of the sigma model curvature constant,
though the most interesting boundary conditions turn out to be possible for noncompact
364
sigma model coset space whose curvature scalar is restricted to lie in a finite range in
units of the D = 3 Planck length, as discussed in Section 5. In the case of flat sigma
model manifold, we find a connection between the model of [10] and that of Izquierdo and
Townsend [9].
We have seen that there are several subtleties in choosing the boundary conditions for
the matter fields. In particular, we find that both regular and irregular boundary conditions
can be imposed on the matter fields as a consequence of the fact that scalar fields with
sufficiently negative mass-squared can be associated with CFT operators of two possible
dimensions on the boundary. In fact, this phenomenon has already been observed in
[15] in the context of AdS5 T 1,1 compactification of Type IIB string theory. Here, we
provide another example of this phenomenon, and we find the resulting CFT supergravity
plus matter symmetry transformations. Somewhat surprisingly, we also find an interesting
conformal supermultiplet structure on the boundary that involve fields which do not have
definite Weyl weights but rather mix with other fields of the multiplet under dilatations.
In this novel multiplet the superconformal symmetry is also realized in an unconventional
fashion.
In the case of irregular boundary conditions the analysis had to be restricted for certain
values of the sigma model radius, referred to as Case 1 in Section 5, such that the effects
of the nonlinear contributions from the regular fields to the transformations of the irregular
fields were omitted. The inclusion of both regular and irregular fields is necessary for the
interpretations of the AdS/CFT correspondence. The study of these effects is intimately
connected with the identification of the boundary conformal field theory, which lies beyond
the scope of this paper.
We conclude by commenting on some of the interesting open problems. Firstly, it is
clearly desirable to find an M-theoretic origin of the model studied here. The structure of
the conformal supermultiplets that we have found on the boundary provide information
on a class of operators which the boundary CFT must contain but do not provide
the full data required to specify uniquely the the CFT in question. It is conceivable
that an M-theoretic origin of the model exists only for a certain critical value of the
sigma model curvature constant. At any rate, many of the features encountered in the
analysis of the (2, 0), AdS3 supergravity plus matter system studied here are likely to
arise in the (4, 4), AdS3 supergravity plus matter system which arises in the AdS3
S 3 compactification of (2, 0), D = 6 supergravity coupled to tensor multiplets, whose
embedding in M-theory is known. We hope that the results presented here may give a flavor
of what to expect in that case. Indeed, these results may also prove useful in analyzing
higher dimensional AdS supergravity plus matter systems as well.
It would also be interesting to extend the above analysis to a generalized setup in which
the boundary conditions are imposed on a surface which is a finite distance away from
the AdS boundary. This is expected to provide an understanding of how a supergravity
plus matter system can be localized on a brane worldvolume in a RandallSundrum like
scenario. This leads to normalizable bulk modes which correspond to fluctuating boundary
modes. The boundary CFT, which is dual to the matter coupled supergravity in the bulk,
should therefore be supplemented by an off-shell Lagrangian for the (fluctuating) boundary
365
supergravity/matter modes. Thus, the total dynamics is that of the boundary CFT plus
the localized, matter coupled bulk supergravity. In this context the unexpected result of
Section 5.3, in the form of the on-shell constraint given in (5.36), indicates that ordinary
off-shell as well as chiral two-dimensional matter systems may be localized on the brane.
We finally remark that we expect the proper vacuum for this setup to be the black string
(domain wall) solution of [10], rather than the anti-de-Sitter vacuum. It would therefore be
interesting to extend the bulk theory by the inclusion of two-form potentials and to give the
supersymmetric coupling of these black strings to the bulk supergravity [20].
Acknowledgements
We thank Kostas Skenderis for valuable discussions. The work of E.S. has been
supported in part by NSF Grant PHY-0070964, the work of Y.T. in part by the Grantin-Aid for Scientific Research on Priority Area 707 Supersymmetry and Unified Theory
of Elementary Particles, Japan Ministry of Education, and the work of P.S. in full by
Stichting FOM, Netherlands.
References
[1] M. Nishimura, Y. Tanii, Supersymmetry in the AdS/CFT correspondence, Phys. Lett. B 446
(1999) 37, hep-th/9810148.
[2] M. Nishimura, Y. Tanii, Super Weyl anomalies in the AdS/CFT correspondence, Int. J. Mod.
Phys. A 14 (1999) 3731, hep-th/9904010.
[3] M. Banados, K. Bautier, O. Coussaert, M. Henneaux, M. Ortiz, Anti-de-Sitter/CFT correspondence in three-dimensional supergravity, Phys. Rev. D 58 (1998) 085020, hep-th/9805165.
[4] M. Henneaux, L. Maoz, A. Schwimmer, Asymptotic dynamics and asymptotic symmetries of
three-dimensional extended AdS supergravity, Ann. Phys. 282 (2000) 31, hep-th/9910013.
[5] M. Nishimura, Y. Tanii, Local symmetries in the AdS7 /CFT 6 correspondence, Mod. Phys. Lett.
A 14 (1999) 2709, hep-th/9910192.
[6] M. Nishimura, Conformal supergravity from the AdS/CFT correspondence, Nucl. Phys. B 588
(2000) 471, hep-th/0004179.
[7] V. Balasubramanian, E. Gimon, D. Minic, J. Rahmfeld, Four-dimensional conformal supergravity from AdS space, Phys. Rev. D 59 (1999) 046003, hep-th/0007211.
[8] A. Achucarro, P.K. Townsend, A ChernSimons action for three-dimensional AdS supergravity
theories, Phys. Lett B 180 (1986) 89.
[9] J.M. Izquierdo, P.K. Townsend, Supersymmetric spacetimes in 2 + 1 AdS-supergravity models,
Class. Quantum Grav. 12 (1995) 895, gr-qc/9501018.
[10] N.S. Deger, A. Kaya, E. Sezgin, P. Sundell, Matter-coupled AdS3 supergravities and their black
strings, Nucl. Phys. B 573 (2000) 275, hep-th/9908089.
[11] H. Nicolai, H. Samtleben, Maximal gauged supergravity in three dimensions, hep-th/0010076.
[12] N.S. Deger, A. Kaya, E. Sezgin, P. Sundell, Spectrum of D = 6, N = 4b supergravity on
AdS3 S 3 , Nucl. Phys. B 536 (1998) 110, hep-th/9804166.
[13] J. de Boer, Six-dimensional supergravity on S 3 AdS3 and 2d conformal field theory, Nucl.
Phys. B 548 (1999) 139, hep-th/9806104.
[14] V. Balasubramanian, P. Kraus, A. Lawrence, Bulk vs. boundary dynamics in anti-de-Sitter
spacetime, Phys. Rev. D 59 (1999) 046003, hep-th/9805171.
366
[15] I.R. Klebanov, E. Witten, AdS/CFT correspondence and symmetry breaking, Nucl. Phys. B 556
(1999) 89, hep-th/9905104.
[16] P. Minces, V.O. Rivelles, Scalar field theory in the AdS/CFT correspondence revisited, Nucl.
Phys. B 572 (2000) 651, hep-th/9907079.
[17] S. de Haro, K. Skenderis, S.N. Solodukhin, Holographic reconstruction of spacetime and
renormalization in the AdS/CFT correspondence, hep-th/0002230.
[18] M. Henningson, K. Sfetsos, Spinors and the AdS/CFT correspondence, Phys. Lett. B 431 (1998)
63, hep-th/9803251.
[19] E. Bergshoeff, H. Nishino, E. Sezgin, Heterotic -models and conformal supergravity in two
dimensions, Phys. Lett. B 166 (1986) 141.
[20] E. Bergshoeff, R. Kallosh, A. Van Proeyen, Supersymmetry of RS bulk and brane, hepth/0012110.

Identifying the operator content, the homogeneous

sine-Gordon models
O.A. Castro-Alvaredo a , A. Fring b
a Departamento de Fsica de Partculas, Facultad de Fsica, Universidad de Santiago de Compostela,
E-15706 Santiago de Compostela, Spain

b Institut fr Theoretische Physik, Freie Universitt Berlin, Arnimallee 14, D-14195 Berlin, Germany
Received 14 August 2000; accepted 7 February 2001
Abstract
We address the general question of how to reconstruct the field content of a quantum field
theory from a given scattering theory in the context of the form factor program. For the SU(3)2 homogeneous sine-Gordon model we construct systematically all n-particle form factors for a huge
class of operators in terms of general determinant formulae. We investigate how different operators
are interrelated by the momentum space cluster property. Finally, we compute several two-point
correlation functions and carry out the ultraviolet limit in order to identify each operator with its
corresponding partner in the underlying conformal field theory. 2001 Elsevier Science B.V. All
rights reserved.
PACS: 11.10.Kk; 11.55.Ds; 05.70.Jk; 05.30.-d; 64.60.Fr; 11.30.Er
1. Introduction
The central concepts of relativistic quantum field theory, like Einstein causality and
Poincar covariance, are captured in local field equations and commutation relations. As
a matter of fact local quantum physics (algebraic quantum field theory) [1] takes the
collection of all operators localized in a particular region, which generate a von Neumann
algebra, as its very starting point (for recent reviews see, e.g., [2]).
On the other hand, in the formulation of a quantum field theory one may alternatively
start from a particle picture and investigate the corresponding scattering theories. In
particular, for (1 + 1)-dimensional integrable quantum field theories this latter approach
has proved to be impressively successful. As its most powerful tool one exploits here the
bootstrap principle [3], which allows to write down exact scattering matrices. Ignoring
E-mail addresses: castro@fpaxp1.usc.es (O.A. Castro-Alvaredo), fring@physik.fu-berlin.de (A. Fring).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 0 5 5 - 4
368
O.A. Castro-Alvaredo, A. Fring / Nuclear Physics B 604 (2001) 367390
subtleties of non-asymptotic states, it is essentially possible to obtain the latter picture

from the former by means of the LSZ-reduction formalism [4]. However, the question of
how to reconstruct at least part of the field content from the scattering theory is in general
still an outstanding issue. Recently a link between scattering theory and local interacting
fields in terms of polarization-free generators has been developed [5]. Unfortunately, they
involve subtle domain properties and are therefore objects which concretely can only be
handled with great difficulties.
In the context of (1 + 1)-dimensional integrable quantum field theories the identification
of the operators is based on the assumption, dating back to the initial papers [6], that each
solution to the form factor consistency equations [69] corresponds to a particular local
operator. Based on this numerous authors [615] have used various ways to identify and
constrain the specific nature of the operator, e.g., by looking at asymptotic behaviours,
performing perturbation theory, taking symmetries into account, formulating quantum
equations of motion, etc. Our analysis will make especially exploit the conjecture that
each local operator has a counterpart in the ultraviolet conformal field theory.
In the present manuscript we show for a concrete model, the SU(3)2 -homogeneous
sine-Gordon model (HSG), that, by means of the form factor program, it is possible to
reconstruct the field content starting from its scattering matrix. Our analysis is based on
the assumption [12,13] that each solution to the form factor consistency equations [69]
corresponds to a particular local operator. We take furthermore into account that the
SU(3)2 -HSG model, like numerous other (1 + 1)-dimensional integrable models, may be
viewed as a perturbed conformal field theory whose entire field content is well classified.
Assuming now a one-to-one correspondence between operators in the conformal and
in the perturbed theory, we can carry out an identification on this level, that is we
associate each solution of the form factor consistency equations a local operator which
is labeled according to the the ultra-violet conformal field theory. We therefore construct
systematically all possible solutions for the n-particle form factors related to a huge class
of operators in terms of some general building blocks which consist out of determinants of
matrices whose entries are elementary symmetric polynomials depending on the rapidities.
We demonstrate how these general solutions are interrelated by the momentum space
cluster property. In particular we show that the cluster property serves also as a construction
principle, in the sense that from one solution to the consistency equations we may obtain
a huge class, almost all, of new solutions. Finally we compute the corresponding twopoint correlation functions and carry out the ultraviolet limit in order to identify the
corresponding conformal dimensions.
Our manuscript is organized as follows: In Section 2 we recall [16] the solutions for the
minimal form factors and the recursive equation which is central for the determination of
the form factors. We describe the general structure of the n-particle form factors for a huge
class of operators. In Section 3 we provide a rigorous proof for all solutions. In Section 4
we investigate the cluster property. In Section 5 we compute several two-point correlation
functions and carry out the ultraviolet limit on the base of a sum rule and the explicit twopoint correlation function in order to identify the conformal dimensions of each operator.
We state our conclusions in Section 6. The Appendix A contains a collection of useful
369
properties of elementary symmetric polynomials and some explicit formulae for the first
non-vanishing form factors.
2. The SU(3)2-HSG model form factors

The SU(3)2 -HSG model contains only two self-conjugate solitons which we denote,
following the conventions of [16], by + and . The two particle scattering matrix as
a function of the rapidity related to this model was found [17] to be

1
S = 1 and S ( ) = tanh
(1)
i
.
2
2
Here is a real constant and corresponds to a resonance parameter. The system (1)
constitutes probably the simplest example of a massive quantum field theory involving two
particles of distinct type. Nonetheless, despite the simplicity of the scattering matrix we
expect to find a relatively involved operator content, since for finite resonance parameter
the SU(3)2-HSG model describes a WZNW-coset model with central charge c = 6/5
perturbed by an operator with conformal dimension = 3/5. Since this is true as long as
is finite, we shall be content in the following mostly by setting = 0. It is expected from
the classical analysis that for finite value of we always have the same ultraviolet central
charge and therefore the same operator content. The TBA-analysis carried out in [18]
supports this analysis. Thus, a finite variation of at the ultraviolet fixed point and away
from it is not very illuminating and we therefore only distinguish the behaviour
and finite. In the former case one trivially observes that the S-matrices S tend to one
and the theory decouples into two Ising models. The related form factors have to respect
this behaviour and all combinations involving different types of particles vanish, see [16].
One may see also directly that the form factor solutions behave this way by employing the
RiemannLebesgue theorem.
The underlying conformal field theory has recently [19] found an interesting application
in the context of the construction of quantum Hall states which carry a spin and fractional
charges.
Taking the scattering matrix as an input, it is in principle possible to compute
form factors, by solving certain consistency equations [69], and thereafter to evaluate
correlation functions. Form factors are tensor valued functions, representing matrix
elements of some local operator O(
x ) located at the origin between a multiparticle in-state
and the vacuum, which we denote by

FnO|1 ...n (1 , . . . , n ) := 0|O(0)V1 (1 )V2 (2 ) Vn (n ) in .
(2)
Here the V ( ) are some vertex operators representing a particle of species . We

commence now by recalling the basic ansatz for solutions of the form factors for the
SU(3)2 -HSG model from [16]. We used the parameterization
370

l

O | . . . . . .
Fn 1 l l+1 n (1 , . . . , n )

|1 ...n
F i j (ij ),
(x1 , . . . , xn )
= HnO|1 ...n QO
n
i<j
(3)
where
i j
(ij )
F i j (ij ) := Fmin
xi i + xj j
i j
i j
(ij ) denote
The rapidities enter through the variable xi = exp(i ) and the functions Fmin
the so-called minimal form factors. They were found to be
( ) = i sinh ,
Fmin
2
1 i(11) G
Fmin
( ) = 2 4 e 4
(4)

t
dt sin2 (i ) 2
exp
t
sinh t cosh t/2
0
=e
( ).
Fmin
(5)
Here G is the Catalan constant. For the overall constants we obtained

H O|2s+,m = i sm 2s(2sm1+2 )esm/2H O|,m ,
= 0, 1.
(6)
Note that at this point an unknown constant, that is H O|,m , enters into the procedure. This
quantity is not constrained by the form factor consistency equations and has to be obtained
from elsewhere. The polynomials Q have to satisfy the recursive equations
QO|l+2,m (x, x, . . . , xn ) = Dl,m (x, x1 , . . . , xn )QO|l,m (x1 , . . . , xn ),
m

1
Dl,m (x, x1 , . . . , xn ) = (ix)l+1 l+
x k 1 (1)l+k+ k .
2
(7)
(8)
k=0
In particular,

+
D2s+,2t + (x, x1 , . . . , xn ) = (i)2s+ +12s+
t

x 2s2p+ +1 2p+
.
(9)
p=0
Here is related to the factor of local commutativity = (1) = 1. We introduced

also the function which is 0 or 1 for the sum + being odd or even, respectively.
We shall use various notations for elementary symmetric polynomials (see Appendix A for
some essential properties). We employ the symbol k when the polynomials depend on the
variables xi , the symbol k when they depend on the inverse variables xi1 , the symbol k
when they depend on the variables xi e +i/2 and k when we set the first two variables to
x1 = x, x2 = x. The number of variables the polynomials depend upon is defined always
in an unambiguous way through the l.h.s. of our equations, where we assume the first l
variables to be associated with = + and the last m variables with = . In case no
superscript is attached to the symbol the polynomials depend on all m + l variables, in case
of a + they depend on the first l variables and in case of a on the last m variables.
371
Solving recursive equations of the type (7) in complete generality is still an entirely open
problem. Ideally one would like to reach a situation similar to the one in the bootstrap
construction procedure of the scattering matrices, where one can state general building
blocks, e.g., particular combinations of hyperbolic functions whenever backscattering is
absent [20], infinite products of gamma functions when backscattering occurs or elliptic
functions when infinite resonances are present. At least for all operators in the model we
consider here this goal has been achieved. It will turn out that all solutions to the recursive
equations (7) may be constructed from some general building blocks consisting out of
determinants of matrices whose entries are elementary symmetric polynomials in some
particular set of variables. Let us therefore define the (t + s)(t + s)-matrix
+
for 1 i t,
2(j i)+
,

Al,m (s, t) ij :=
(10)
2(j i)+2t + for t < i s + t.

More explicitly, matrix A reads
+
+
+
+2
+4
0
+
+ +2
.
..
..
.
.
.
.
0
0
0
,
Al,m =

+4

+2
0
+2

..
..
..
.
.
.
0
+
+6
+
+4
..
.
0
+6
+4
..
.
..
.
..
.
0
0
..
.
2s+
.
0
..
.
2t+
(11)
The superscripts , may take the values 0 and 1 and the subscripts l, m characterize the
number of different variables related to the particle species +, , respectively. The
different combinations of the integers , , l, m will correspond to different kind of local
operators O.
In addition, the form factors will involve a function depending on two further indices
and
,

gl,m := (l )
lm+
2
(m )
(12)
Here the ,
are integers whose range, unlike the one for , , is in principle not restricted.
However, it will turn out that due to the existence of certain constraining relations, to be
specified in detail below, it is sufficient to characterize a particular operator by the four
integers , , l, m only. Then, as we shall demonstrate, all Q-polynomials acquire the
general form
,

QO|l,m = Ql,m = Q2s+,2t + = i s (1)s( +t +1)g2s+,2t + det A2s+,2t + .

,
(13)
We used here already a parameterization for l, m which will turn out to be most convenient.
The subscripts in g and A are only needed in formal considerations, but in most cases the
number of particles of species + and are unambiguously defined through the l.h.s.
of our equations. This is in the same spirit in which we refer to the number of variables in
372
the elementary symmetric polynomials. We will therefore drop them in these cases, which
leads to simpler, but still precise, notations. To illustrate this with examples, we consider
for instance the solutions to the recursive equations (7) related to the trace of the energy
momentum tensor , the order operator and the disorder operator which were already
stated in [16]
Q|2s+2,2t +2 = i s(2t +3)e(t +1) 1 1 g 0,2 det A1,1 ,
|2s,2t +1
|2s,2t
=i
=i
s(2t +3) 1,1
2s(t +1) 1,1
det A
0,1
det A
0,0
(14)
(15)
(16)
Here A is always taken to be a (t + s) (t + s)-matrix. Notice that in comparison with

(13) the factor of proportionality in (15) and (16) is only a constant, whereas in (14)
also the term 1 1 appears. Terms of this type may always be added since they satisfy
the consistency equations trivially. This is also the reason why in comparison with [16]
we can safely drop in Q|2s,2t +1 the factor (1 )1/2 (1 )1/2 . Additional reasons for this
modification will be provided below. For we were forced [16] to introduce the factor
1 1 in order to recover the solution of the thermally perturbed Ising model for 2s + 2 = 0.
Note that for the value s = 1 formally makes sense.
3. Solution procedure
We shall now recall the principle steps of the general solution procedure for the form
factor consistency equations [69]. For any local operator O one may anticipate the pole
structure of the form factors and extract it explicitly in form of an ansatz of the type (3).
This might turn out to be a relatively involved matter due to the occurrence of higher order
poles in some integrable theories, e.g., [14], but nonetheless it is possible. Thereafter, the
task of finding solutions may be reduced to the evaluation of the minimal form factors and
to solving a (or two if bound states may be formed in the model) recursive equation of the
type (7). The first task can be carried out relatively easily, especially if the related scattering
matrix is given as a particular integral representation [6]. Then an integral representation of
the type (5) can be deduced immediately. The second task is rather more complicated and
the heart of the whole problem. Having a seed for the recursive equation, that is the lowest
non-vanishing form factor, 1 one can in general compute from them several form factors
which involve more particles. However, the equations become relatively involved after
several steps. Aiming at the solution for all n-particle form factors, it is therefore highly
desirable to unravel a more generic structure which enables one to formulate rigorous
proofs. Several examples [8,11,21] have shown that often the general solution may be
cast into the form of determinants whose entries are elementary symmetric polynomials.
Presuming such a structure which, at present, may be obtained by extrapolating from lower
particle solutions to higher ones or by some inspired guess, one can rigorously formulate
1 For the case at hand this is provided for some operators by the well known solutions of the Ising model. In
general this is also a difficult hurdle to take as, for instance, one might need to know vacuum expectation values.
373
proofs as we now demonstrate for the SU(3)2-HSG-model, for which some solutions were
merely stated in [16].
We have two universal structures 2 at our disposal. We could either exploit the integral
representation for the determinant A, as presented in [16], or exploit simple properties of
determinants. Here we shall pursue the latter possibility. For this purpose it is convenient
x (R x ) which acts on the j th column (row) of an (n n)-matrix
to define the operator Ci,j
i,j
A by adding x times the ith column (row) to it
x
Ci,j
A:
Akj Akj + xAki ,
1 i, j, k n,
(17)
x
A:
Ri,j
Aj k Aj k + xAik ,
1 i, j, k n.
(18)
x
x
and Ri,j
on A,
Naturally, the determinant of A is left invariant under the actions of Ci,j
such that we can use them to bring A into a suitable form for our purposes. Furthermore,
it is convenient to define the ordered products, i.e., operators related to the lowest entry act
first,
x
:=
Ca,b
b

p=a
x
Cp,p+1
and Rxa,b :=
x
Rp,p1
.
(19)
p=a
It will be our strategy to use these operators in such a way that we produce as many zeros
as possible in one column or row of a matrix of interest to us. In order to satisfy (7) we have
to set now the first variables in A to x1 = x, x2 = x, which we denote as A thereafter and
,
,
relate the matrices Al+2,m and Al,m . Taking relation (A.4) for the elementary symmetric
,
polynomials into account, we can bring Al+2,m into the form

2
x2
,
Rx
t +2,s+t +1 C1,s+t 1Al+2,m ij
+
,
1 i t,
2(j i)+
t < i s + t,
= 2(j i)+2t + ,
j x 2(j p)
p=1
2(ps1)+ , i = s + t + 1.
(20)
It is now crucial to note that since the number of variables has been reduced by two,
several elementary polynomials may vanish. As a consequence, for 2s + 2 + > l and
2t + 2 + > m, the last column takes on the simple form
x 2

x2
,
Rt +2,s+t +1C1,s+t
1Al+2,m i(s+t +1)
0,
1 i s + t,
= t
(21)
2(t
p)
2p+ , i = s + t + 1.
p=0 x
,
Therefore, developing the determinant of Al+2,m with respect to the last column, we are
,
,
able to relate the determinants of Al+2,m and Al,m as
2 There exist also different types of universal expressions like for instance the integral representations presented
in [9]. However, these type of expressions are sometimes only of a very formal nature since to evaluate them
concretely for higher n-particle form factors requires still a considerable amount of computational effort.
374

,
det Al+2,m
t

x 2(t p) 2p+
det Al,m .
(22)
p=0
We are left with the task to specify the behaviour of the function g with respect to the
reduction of the first two variables
,

x lm++2
l+ gl,m .
gl+2,m = i lm++2
(23)
Assembling the two factors (22) and (23), we obtain, in terms of the parameterization (13)
t

+
,
,,
,
,,
+
2s+
+1
2(sp+1)+
Q
Q
= (i)
.
2s+2+,2t +
2s+
2p+
2s+,2t +
p=0
(24)
We are now in the position to compare our general construction (24) with the recursive
equation for the Q-polynomials of the SU(3)2 -HSG model (9). We read off directly the
following restrictions
=
and = 1.
(25)
A further constraint results from relativistic invariance, which implies that the overall
power in all variables xi of the form factors has to be zero for a spinless operator.
Introducing the short hand notation [FnO ] for the total power, we have to evaluate
,
,

,,

,
Q2s+,2t + = g2s+,2t + + det A2s+,2t + .
(26)
Combining (25) and (26) with the explicit expressions

,
det A2s+,2t + = s(2t + ) + t,
,

+ m( m)/2,
gl,m = l(l m + )/2
,

,,
Q2s+,2t + = l(l 1)/2 m(m 1)/2,
we find the additional constraints
and = ( 1).
= 1 +
(27)
Collecting now everything we conclude that different solutions to the form factor
consistency equations can be characterized by a set of four distinct integers. Assuming
that each solution corresponds to a local operator, there might be degeneracies of course,
,
,
we can label the operators by , , , , i.e., O O, , such that we can also write Qm,l
,
,,
instead of Qm,l
. Then each Q-polynomial takes on the general form

,

O,
|2s+,2t +
QO|2s+,2t + = Q
1, +1
= Q2s+,2t + g2s+,2t +
det A2s+,2t + ,
(28)
and the integers , , , are restricted by

+ = ,
2 + > ,
2 + > .
(29)
375
Table 1
Operator content of the SU(3)2 -HSG model
!
F
0
0
0
0
0
0
1
1
1
1
1
1
0
0
0
1
1
1
0
0
0
1
1
0
0
1
0
0
1
0
1
2
1
2
0
0
0
0
1
1
1
2
1
0
0
2
0
0
0
0
0
1/2
1/2
1/2
0
0
0
1/2
1/2
0
F
0
0
0
0
0
0
1/2
1/2
1/2
1/2
1
1/2
1/10
1/10
1/10
1/10
1/10
1/10
1/10
1/10
1/10
1/10
*
*
We combined here (25) and (27) to get the first relation in (29). The inequalities result from
the requirement in the proof which we needed to have the form (21). We find 12 admissible
solutions to (29), i.e., potentially 12 different local operators, whose quantum numbers are
presented in Table 1.
0,0
Comparing with our previous results, we have according to this notation F O0,0 |2s,2t =
0,1
1,1
F |2s,2t , F O0,1 |2s,2t +1 = F |2s,2t +1 and F O2,2 |2s,2t +1 F |2s+2,2t +2. The last two
solutions are only formal in the sense that they solve the constraining equations (29), but
the corresponding explicit expressions turn out to be zero.
In summary, by taking the determinant of the matrix (11) as the ansatz for the general
building block of the form factors, we constructed systematically generic formulae for the
n-particle form factors possibly related to 12 different operators.
4. Momentum space cluster properties

Cluster properties in space, i.e., the observation that far separated operators do not
interact, are quite familiar in quantum field theories [23] for a long time. In 1 + 1
dimensions a similar property has also been noted in momentum space. For the purely
bosonic case this behaviour can be explained perturbatively by means of Weinbergs power
counting theorem, see, e.g., [6,22]. 3 This property has been analysed explicitly for several
3 There exists also a heuristic argument which provides some form of intuitive picture of this behaviour [15]
by appealing to the ultraviolet conformal field theory. However, the argument is based on various assumptions,
which need further clarification. For instance, it remains to be proven rigorously that the particle creation operator
V ( ) tends to a conformal Zamolodchikov operator for and that the local field factorizes equally into
two chiral fields in that situation. The restriction in there that lim Sij ( ) = 1, for i being a particle which
has been shifted and j one which has not, excludes a huge class of interesting models, in particular the one at
hand.
376
specific models [8,12,21,24]. It states that whenever the first, say , rapidities of an nparticle form factor are shifted to infinity, the n-particle form factor factorizes into a and
an (n )-particle form factor which are possibly related to different types of operators

O
T1,
FnO (1 , . . . , n ) FO (1 , . . . , )Fn
(+1 , . . . , n ).
(30)
For convenience we have introduced here the operator
Ta,b
:= lim
Tp
(31)
p=a
which will allow for concise notations. It is composed of the translation operator Ta which
acts on a function of n variables as
Ta f (1 , . . . , a , . . . , n ) f (1 , . . . , a + , . . . , n ).
(32)
Whilst Watsons equations and the residue equations, see, e.g., [68,16], are operator
independent features of form factors, the cluster property captures part of the operator
nature of the theory. The cluster property (30) does not only constrain the solution, but
eventually also serves as a construction principle in the sense that when given FnO we
may employ (30) and construct form factors related to O and O . Hence, (30) constitutes
a closed mathematical structure, which relates various solutions and whose abstract nature
still needs to be unraveled.
We shall now systematically investigate the cluster property (30) for the SU(3)2 -HSG
model. Choosing w.l.g. the upper signs for the particle types in Eq. (3), we have four
different options to shift the rapidities
FnO|l+,m = T+1<l,n
FnO|l+,m ,
T1,l
(33)
FnO|l+,m
T1,>l
(34)
= T+1l,n
FnO|l+,m ,
which a priori might all lead to different factorizations on the r.h.s. of Eq. (30). The equality
signs in Eqs. (33) and (34) are a simple consequence of the relativistic invariance of form
factors, i.e., we may shift all rapidities by the same amount, for O being a scalar operator.
Considering now the ansatz (3) we may first carry out part of the analysis for the terms
which are operator independent. Noting that

O(1),
( )
++

+
Fmin ( ) = T1,1
Fmin ( ) e 2 ,
T1,1
Fmin ( )
T1,1
(35)
( )
e 2 ,
we obtain for the choice of the upper signs for the particle types in the ansatz (3)

T1,l
F i j (ij )
F ++ (ij )
F i j (ij )
i<j
1i<j
<i<j l+m
l (1l)
(x1 , . . . , x ) 2 e 2
l (x+1 , . . . , xl ) 2
ml+ (lm1)
(x1 , . . . , x ) 2 e
n (x+1 , . . . , xn ) 2
Tn+1<m,n
F i j (ij )
1i<j n
i<j
F i j (ij )
377
F (ij )
n<i<j n
(ml1)
2
n (x1 , . . . , xn ) 2 e
lm+
(xn+1 , . . . , xn ) 2
(1m)
m (xl+1 , . . . , xn ) 2 e 2
m
(xn+1 , . . . , xn ) 2
The remaining cases can be obtained from the equalities (33) and (34). Turning now to the
behaviour of the function g as defined in (12) under these operations, we observe with help
of the asymptotic behaviour of the elementary symmetric polynomials (A.5) and (A.6)

lm+
m
gl,m = e (x1 , . . . , x )l (x+1 , . . . , xl ) 2 (m ) 2 ,

T1,l
,

(36)
lm+
Tn+1<m,n
gl,m = (l )

m
e (xn+1 , . . . , xn )m (xl+1 , . . . , xn ) 2 .
In a similar fashion we compute the behaviour of the determinants

+(1)
,
1,
t (2+ )
(2+ )t (1)t
det Al2,m ,
T1,2+
l det Al,m = e
+
,
,
T1,2+
2t +
det Al2,m ,
l det Al,m =

s
,
,1
s(2+ ) + +(1)
Tn+12
2+ det Al,m2 ,
<m,n det Al,m = e
+

,
,
s +
Tn+12
det Al,m2 .
<m,n det Al,m = (1) 2s+
(37)
(38)
(39)
(40)
(41)
We have to distinguish here between the odd and even case, which is the reason for the
introduction of the integer taking on the values 0 or 1.
Collecting now all the factors, we extract first the leading order behaviour in
F2s+,2t + e(+
T1,l
,
(11) )
2
F2s+,2t + e(+
Tn+1<m,n
,
(11)
2 )
(42)
Notice that, if we require that all possible actions of Ta,b

should lead to finite expressions
on the r.h.s. of (30), we have to impose two further restrictions, namely, and .
These restrictions would also exclude the last two solutions from Table 1. We observe
1,1
tends to zero under all possible shifts. Seeking now solutions for the set
further that F2,2
, , , of (42) which at least under some operations leads to finite results and in all
remaining cases tends to zero, we end up precisely with the first 9 solutions in Table 1.
Concentrating now in more detail on these latter cases which behave like O(1), we find
from the previous equations the following cluster properties
,0
+(12),0
0,0
T1,2+
l F2s+,2t + F2+,0 F2s+ 2,2t + ,
,
,
0,0
T1,2+
l F2s+,2t + F2+,0 F2s+ 2,2t + ,
0,+(12)
0,
0,0
Tn+12
<m,n F2s+,2t + F2s+,2t + 2 F0,2+ ,
(43)
(44)
(45)
378
Fig. 1. Interrelation of various operators via clustering. In this figure we use the abbreviations
, T2 T1,2+1l
, T3 Tn2<m,n
, T4 Tn2<m,n
. We also drop the 2s and
T1 T1,2+1l
2t in the subscripts of the Os. The Ti on the links operate in both directions.
0,0
Tn+12
<m,n F2s+,2t + F2s+,2t + 2 F0,2+ .
,
(46)
We may now use (43)(46) as a means of constructing new solutions, i.e., we can start with
one solution and use (43)(46) in order to obtain new ones. Fig. 1 demonstrates that when
knowing just one of the first nine operators in Table 1 it is possible to (re)-construct all the
others in this fashion.
4.1. The energymomentum tensor
1,1
is rather special. In
As we observed from our previous discussion the solution F2,2
fact this solution is part of the expression which in [16] was identified as the trace of the
energymomentum tensor
1,1
Q|2s+2,2t +2 = i s(2t +3)e(t +1) 1 1 F2s+2,2t
+2 .
(47)
379
The pre-factor 1 1 will, however, alter the cluster property. The leading order behaviour
reads now
T1,2s
F |2s,2t Tn+1<2t,n
F |2s,2t e(1/2).
(48)
We observe that still in most cases the shifted expressions tend to zero, unless = 1 for
which it tends to infinity as a consequence of the introduction of the 1 1 . There is now also
the interesting case = 2, for which the -dependence drops out completely. Considering
this case in more detail we find
1 (x2s+1 , . . . , x2s+2t )
1 (x3 , . . . , x2s+2t )
|2s,2t
|2,0 |2s2,2t
F
F
T1,2 F
(49)
1 (x2s+1 , . . . , x2s+2t )
,
1 (x3 , . . . , x2s+2t )
1 (x1 , . . . , x2s )
(x
1 1 , . . . , x2s+2t 2)
Tn+1,n
(50)
F |2s,2t F |2s,2t 2F |0,2
1 (x1 , . . . , x2s )
.
1 (x1 , . . . , x2s+2t 2)
Note that unless s = 1 in (49) or t = 1 in (50) the form factors do not purely factorize
into known form factors, but in all cases a parity breaking factor emerges. We now turn to
the cases = 2s or = 2t for which we derive
|2s,2t
F
Tn+12t,n
F |2s,2t e(2t s).
T1,2s
(51)
We observe that once again in most cases these expressions tend to zero. However, we
also encounter several situations in which the -dependence drops out altogether. It may
happen whenever t = 2, s = 0 or s = 2, t = 0, which simply expresses the relativistic
invariance of the form factor. The other interesting situation occurs for t = 1, s = 1.
|0,2
= 2m2 , m = m =
Choosing temporarily (in general we assume m = m+ ) H2
2G/
, we derive in this case
m+ e
|2,0
F2
|0,2
F2
(52)
.
2m2
In general when shifting the first 2s or last 2t rapidities we find the following factorization
|2,2
T1,2
F4
|2s,2t
T1,2s
F
Tn+12t,n
F |2s,2t F |2s,0F |0,2t .
(53)
This equation holds true when keeping in mind that the r.h.s. of this equation vanishes once
it involves a form factor with more than two particles. Note that only in these two cases the
form factors factorize purely into two form factors without the additional parity breaking
factors as in (49) and (50).
5. Identifying the operator content

Having solved Watsons and the residue equations one has still little information about
the precise nature of the operator corresponding to a particular solution. There exist,
380
Table 2
Conformal dimensions for O(,) in the SU(3)2 /U(1)2 -coset model
\
1
2
1 2
21
22
21 2
1 22
21 22
1 + 2
21
22
1/10
1/10
1/10
1/10
1/10
1/10
1/10
1/10
1/10
3/5
1/10
1/10
1/10
0
1/2
1/2
0
1/2
1/2
1/2
1/2
0
however, various non-perturbative (in the standard coupling constant sense) arguments
which provide this additional information and which we now wish to exploit for the
model at hand. Basically all these arguments rely on the assumption that the superselection
sectors of the underlying conformal field theory remain separated after a mass scale has
been introduced. We will therefore first have a brief look at the operator content of the
Gk /U(1)r -WZNW coset models and attempt thereafter to match them with the solutions
of the form factor consistency equations. For these theories the different conformal
dimensions in one model can be parameterized by two quantities [26]: a highest dominant
weight of level smaller or equal to k and their corresponding lower weights obtained
in the usual way by subtracting multiples of simple roots i from until the lowest weight
is reached
(, ) =
( ( + 2)) ( )
.
2(k + h)
2k
(54)
Here h is the Coxeter number of G and the Weyl vector, i.e., the sum over all fundamental
weights. Denoting the highest root of G by , the conformal dimension related to the
adjoint representation (, 0) is of special interest since it corresponds to the one of the
perturbing operator which leads to the massive HSG-models. Taking the length of to
be 2 and recalling the well known fact that the height of , that is ht(), is the Coxeter
number minus one, such that ( ) = ht () = h 1, it follows that O (,0) is a unique
operator with conformal dimension (, 0) = h/(k + h). Note that uniqueness demands
in addition that we do not take the multiplicities of the -states into account. For SU(3)2
the expression (54) is easily computed and since we could not find the explicit values in
the literature we report them for reference in Table 2.
Turning now to the massive theory, a crude constraint which gives a first glimpse at
possible solutions to the form factor consistency equations is provided by the bound [14]

O| ...
Fn 1 n (1 , . . . , n ) i O .
(55)
We introduced here limi f (1 , . . . , n ) =: const exp([f (1 , . . . , n )]i i ) as abbreviation and denote the conformal dimension of the operator O in the ultraviolet conformal
381
limit by O . We use the notation [ ] when we take the limit in the variable xi related to
the particle species i = , respectively. For the different solutions we constructed, we
report the asymptotic behaviour in Table 1. When we are in a position in which we already
anticipate the conformal dimensions the bound (55) will severely restrict the possible inclusion of factors like 1 , 1 , 1 , 1+ , which as we mentioned above may always be added
since they trivially satisfy the consistency equations.
More concrete and definite values for O are obtainable when we exploit the knowledge
about the underlying conformal field theory more deeply. Considering an operator which in
the conformal limit corresponds to a primary field we can of course compute the conformal
dimension by appealing to the ultraviolet limit of the two-point correlation function
"

Oi (r)Oj (0) =
(56)
Cij k r 2k 2i 2j Ok (0) + .
k
The three-point couplings Cij k are independent of r. In particular when assuming that 0
is the smallest conformal dimension occurring in the model (which is the case for unitary
models), we have

1/2O
"

CO O 0
4O
lim O(r)O(0) r
(57)
for r
.

r0
CO O O O
Here O is the operator with the second smallest dimension for which the vacuum
expectation value is non-vanishing. Using a Lorentz transformation to shift the O(r) to
the origin and expanding the correlation function in terms of form factors in the usual
fashion

n

"
d1 dn

exp r
mi cosh i
O(r)O (0) =
n!(2)n
...
n=1
i=1

FnO|1 ...n (1 , . . . , n ) FnO |1 ...n (1 , . . . , n ) ,
O
(58)
we can compute the l.h.s. of (57) and extract

thereafter. The disadvantage to proceed
in this way is many-fold. First we need to compute the multidimensional integrals in (58)
for each value of r, which means to produce a proper curve requires a lot of computational
(at present computer) time. Second we need already a relatively good guess for O . Third
for very small r the nth term within the sum is proportional to (log(r))n such that we have
to include more and more terms in that region and fourth we need the precise values of
the lowest non-vanishing form factors, i.e., in general vacuum expectation values or one
particle form factors to compute the r.h.s. of (58). However, the lowest non-vanishing form
factor can be of an arbitrary particle number and one may still extract the value of O .
A short remark is also due concerning solutions related to different sets of s. The sum
over the particle types simplifies considerably when taking into account that form factors
corresponding to two sets, which differ only by a permutation, lead to the same contribution
in the sum. This follows simply by using one of Watsons equations [69], which states
that when two particles are interchanged we will pick up the related two particle scattering
matrix as a factor. Noting that the scattering matrix is a phase, the expression remains
unchanged.
382
5.1. -sum rules

Most of the disadvantages, which emerge when using (57) to compute the conformal
dimensions, can be circumvented by formulating sum rules in which the r-dependence
has been eliminated. Such type of rule has for instance been formulated by Smirnov [21]
already more than a decade ago. However, the rule stated there is slightly cumbersome
in its evaluation and we will therefore resort to one found more recently by Delfino,
Simonetti and Cardy [15]. In close analogy to the spirit and derivation of the c-theorem
[25] these authors derived an expression for the difference between the ultraviolet and
infrared conformal dimension of a primary field O
O
O
uv ir
1
=
2O
"

r (r)O(0) dr.
(59)
Using the expansion of the correlation function in terms of form factors (58) we may carry
out the r-integration in (59) and obtain
O
O
uv ir

1
d1 dn
=
n
2
n
2O
n!(2)
n=1 1 ...n
i=1 mi cosh i

Fn|1 ...n (1 , . . . , n ) FnO|1 ...n (1 , . . . , n ) .
(60)
Notice also that unlike in the evaluation of the c-theorem, which deals with a monotonically
increasing series, due to the fact that it only involves absolute values of form factors, the
series (59) can in principle be alternating. Before the concrete evaluation of the expression
(60) for the various solutions we constructed for the SU(3)2-HSG model, we should
pause for a while and appreciate the advantages of this formula in comparison with (57).
First of all, since the r-dependence has been integrated out we only have to evaluate
the multidimensional integrals once. Second the evaluation of (60) does not involve any
anticipation of the value of O . Third one is very often in the comfortable position that
despite the fact that the vacuum expectation value occurs explicitly in the sum rule, its
explicit form is not needed. Once it is non-vanishing, the next lowest non-vanishing form
factor may be normalized such that O cancels from the whole expression. For a singular
vacuum expectation value the value of the remaining integral guarantees that the sum rule
maintains its form as for instance discussed in [27]. Finally and most important fourth, the
difficulty to identify the suitable region in r which is governed by the (log r)n behaviour
of the nth term in the sum in (58) and the upper bound in (56) has completely disappeared.
There are however little drawbacks for theories with internal symmetries and for the
case when the lowest non-vanishing form factor of the operator we are interested in is
not the vacuum expectation value. The first problem arises due to the fact that the sum
rule is only applicable for primary fields O whose two-point correlation function with the
energymomentum tensor is non-vanishing. Since in our model the n-particle form factors
related to the energymomentum tensor are only non-vanishing for even particle numbers,
0,0
0,1
1,0
1,1
we may only use it for the operators O0,0
, O0,2
, O2,0
and O2,2
, where the latter operator
is plagued be the second problem.
383
0,0
0,1
1,0
We will now compute the sum rule for the operators O0,0
, O0,2
, O2,0
up to the 6-particle
contribution. We commence with the two particle contribution which is always evaluated
effortlessly. Noting that
F2 ( ) = 2im2 sinh(/2)
(61)
and the fact that O

ir is zero in a purely massive model, the two particle contribution
acquires a particular simple form
O (2)
=
i
4O

d

tanh O|++
F2
(2 ) .
cosh
(62)
Using now the explicit expressions for the two-particle form factors (A.8), we immediately
find
# 0,0 $(2) # 0,1 $(2) # 1,0 $(2)
O0,0
(63)
= O0,2
= O2,0
= 1/8.
We recall, see also [16], that in the limit we obtain two copies of the thermally
perturbed Ising model. This means that in the sum over the particle types in (60) there will
be no contributions from terms involving different types of particles. We only obtain two
O |++
O |
0,0
and F2
, such that the operator O0,0
equal contributions, namely 1/16 from F2
plays the role of the disorder operator, as we expect.
0,0
0,1
1,0
, O0,2
, O2,0
from each other we have to proceed to
To distinguish the operators O0,0
higher particle contributions. At present there exist no analytical arguments for this and we
therefore resort to a brute force numerical computation.
Denoting by (O )(n) the contribution up to the nth particle form factor, our numerical
Monte Carlo integration 4 yields
#
# 0,0 $(4)
0,0 $(6)
= 0.0987,
O0,0
= 0.1004,
O0,0
(64)
# 0,1 $(4)
#
0,1 $(6)
O0,2
(65)
= 0.0880,
O0,2
= 0.0895,
#
# 1,0 $(4)
1,0 $(6)
= 0.0880,
O2,0
= 0.0895.
O2,0
(66)
We shall be content with the precision reached at this point, but we will have a look at the
overall sign of the next contribution. From the explicit expressions of the 8-particle form
0,0
the next contribution will reduce the value for . For the other
factors we see that for O0,0
two operators we have several contributions with different signs, such that the overall value
0,0
0,1
1,0
, O0,2
, O2,0
all possess
is not clear a priori. In this light, we conclude that the operators O0,0
conformal dimension 1/10 in the ultraviolet limit. Unfortunately, the values for the latter
two operators do not allow such a clear cut deduction as for the first one. Nonetheless, we
base our statement on the knowledge of the operator content of the conformal field theory
and confirm them also by elaborating directly on (57) and (58).
4 We employed here the widely used numerical recipe routine VEGAS [28]. Typical standard deviations we
achieve correspond to the order of the last digit we quote.
384
5.2. from correlation functions

First of all we do not presume anything about the conformal dimension of the operator
O and multiply its two-point correlation function (58) by r p with p being some arbitrary
power. Once this combination behaves as a constant in the vicinity of r = 0 we take this
value as the first non-vanishing three-point coupling divided by the vacuum expectation
value of O and p/4 as its conformal dimension. This means even without knowing the
vacuum expectation value we have a rational to fix p, but we can not determine the first
0,1
up to the 8-particle
term in (56). Fig. 2(a) exhibits this analysis for the operator O0,2
contribution and we conclude from there that its conformal dimension is 1/10. For the
other operators the figures look qualitatively the same.
The results of the same type of analysis for the energy momentum tensor is depicted
in Fig. 2(b), from which we deduce the conformal dimension 3/5. Recalling that the
energy-momentum tensor is proportional [29] to the dimension of the perturbing field this
is precisely what we expected to find.
Furthermore, we observe that the relevant interval for r differs by two orders of
magnitude, which by taking the upper bound for the validity of (57) into account should
amount to

C 1 1 0 C 3 3 1 C 3 3 0 C 1 1 1 O(102 ).
10 10
5 5 10
5 5
10 10 10
Since to our knowledge these quantities have not been computed from the conformal side,
this inequality can not be double checked at this stage.
In Fig. 3 we also exhibit the individual n-particle contributions. Excluding the two
particle contribution, these data also confirm the proportionality of the nth term to
(log(r))n .
We have carried out similar analysis for the other solutions we have constructed and
report our findings in Table 1. We observe that the combination of the vacuum expectation
value times the three-point coupling for these operators differ, which is the prerequisite for
unraveling the degeneracy.
6. Conclusions
With regard to the main conceptual question addressed in this paper, we draw the overall
conclusion that solutions of the form factor consistency equations can be identified with
operators in the underlying ultraviolet conformal field theory. In this sense one can give
meaning to the operator content of the integrable massive model. The quantity on which
the identification is based is the conformal dimension of the operator. Naturally this implies
that once the conformal field theory is degenerate in this quantity, as it is the case for the
model we investigated, the identification can not be carried out in a one-to-one fashion and
therefore the procedure has to be refined. In principle this would be possible by including
the knowledge of the three-point coupling of the conformal field theory and the vacuum
expectation value into the analysis. The former quantities are in principle accessible by
385
%
&
0,1
0,1
0,1
Fig. 2. Rescaled correlation function G0,2
(R) := O0,2
(R)O0,2
(0) part (a) and (G )(8) (R) :=
(R)(0) part (b) summed up to the eight particle contribution as a function of R = rm.
386
Fig. 3. Rescaled individual n-particle contribution g (n) (R) to the correlation function.
working out explicitly the conformal fusion structure, whereas the computation of the
latter still remains an open challenge. In fact what one would like to achieve ultimately
is the identification of the conformal fusion structure within the massive models.
It would be desirable to put further constraints on the solutions by means of other
arguments, that is exploiting the symmetries of the model, formulating quantum equations
of motion, possibly performing perturbation theory etc.
Technically we have confirmed that the sum rule (59) is clearly superior to the direct
analysis of the correlation function. It would therefore be highly desirable to develop
arguments which also apply for theories with internal symmetries and possibly to resolve
the mentioned degeneracies in the conformal dimensions.
It remains also an open question, whether the general solution procedure presented in
this manuscript can be generalized to the degree that the type of determinants presented
will serve as generic building blocks of form factors.
The specific conclusions for the SU(3)2 -homogeneous sine-Gordon model are as
follows: we have provided a rigorous proof for the solutions of the form factor consistency
equations which were previously stated in [16]. In addition we found a huge number of
new solutions. By means of the sum rule and a direct analysis of the correlation functions
we identified the conformal dimension of these operators in the underlying conformal field
theory. Considering the total number of operators present in the conformal field theory
(see Table 2) one still expects to find additional solutions, in particular the identification
of the fields possessing conformal dimension 1/2 is outstanding. Nonetheless, concerning
the physical picture presented for this model one can surely claim that it rests now on quite
387
firm ground. After the central charge of the conformal field theory had been reproduced
by means of the thermodynamic Bethe ansatz [18] and the c-theorem in the context of the
form factor program [16], we have now also identified the dimension of various operators.
In particular the dimension of the perturbing operator was identified to be 3/5.
Acknowledgements
A.F. is grateful to the Deutsche Forschungsgemeinschaft (Sfb288) for financial support.
O.A.C. thanks CICYT (AEN99-0589), DGICYT (PB96-0960), and the EC Commission
(TMR grant FMRX-CT96-0012) for partial financial support and is also very grateful to
the Institut fr theoretische Physik of the Freie Universitt for hospitality. We would like
to thank J.L. Miramontes and R.A. Vzquez for helpful discussions.
Appendix A
A.1. Elementary symmetric polynomials
In this appendix we assemble several properties of elementary symmetric polynomials
to which we wish to appeal from time to time. Most of them may be found either in [30]
or can be derived effortlessly. The elementary symmetric polynomials are defined as

xl1 xlk .
k (x1 , . . . , xn ) =
(A.1)
l1 <<lk
They are generated by

n

k=1
(x + xk ) =
n

x nk k (x1 , . . . , xn ),
(A.2)
k=0
and as a consequence may also be represented in terms of an integral representation

'
n
1
dz
k (x1 , . . . , xn ) =
(A.3)
(z + xk ),
2i
znk+1
|z|=;
k=1
which is convenient for various applications. Here ; is an arbitrary positive real number.
With the help of (A.3) we easily derive the identity
k (x, x, x1, . . . , xn ) = k (x1 , . . . , xn ) x 2 k2 (x1 , . . . , xn ),
which will be central for us. We will also require the asymptotic behaviours

e (x1 , . . . , x )k (x+1 , . . . , xn ) for < k,
T1, k (x1 , . . . , xn ) k
for k,
e k (x1 , . . . , x )
and
(A.4)
(A.5)
388
(x , . . . , xn )
k +1
for n k,
T1,
k (x1 , . . . , xn ) k+n (x1 , . . . , x )n (x+1 , . . . , xn )
e(k+n)
for > n k
(A.6)
which may be obtained from (A.3) as well.

A.2. Explicit form factor formulae
Having constructed the general solutions in terms of the parameterization (3), it is simply
a matter of collecting all the factors to get explicit formulae. For the concrete computation
of the correlation function, it is convenient to have some of the evaluated expressions at
hand in form of hyperbolic functions.
A.2.1. One particle form factors
0,0
O1,0
|+
F1
0,0
O0,1
|
= F1
0,1
O0,1
|
= F1
1,0
O1,0
|+
= F1
= H 1,0 = H 0,1.
(A.7)
A.2.2. Two particle form factors

O|
F2
= iO tanh ,
2
0,1
O1,1
|+
F2
0,0
0,1
1,0
for O = O0,0
, O0,2
, O2,0
,
+
= H 1,1e21 /2 Fmin
( ),
A.2.3. Three particle form factors

(
i j
(ij )
H 0,1 i<j Fmin
O |
=(
F3
cosh(
ij /2)
1i<j 3
1,0
O1,1
|+
F2
+
= H 1,1 Fmin
( ).
0,0
0,0
0,1
1,0
for O1,0
, O0,1
, O0,1
, O1,0
.
A.2.4. Four particle form factors

m2 2 + i<j cosh(ij ) i j
|++
F4
Fmin (ij ),
=
2 cosh(12/2) cosh(34/2)
i<j
" 0,0
O0,0 cosh(13 /2 + 24 /2) i j
O 0,0 |++
=
Fmin (ij ).
F4 0,0
2 cosh(12 /2) cosh(34 /2)
(A.8)
(A.9)
(A.10)
(A.11)
(A.12)
i<j
A.2.5. Five particle form factors

(
i j
H 0,1 i<j Fmin
(ij )
O |
F5
=(
1i<j 5 cosh(ij /2)
0,0
0,0
1,0
0,1
for O1,0
, O0,1
, O1,0
, O0,1
.
A.2.6. Six particle form factors

m2 3 + i<j cosh(ij ) i j
|++++
F6
= (
Fmin (ij ),
4 1i<j 4 cosh(ij /2)
i<j
(A.13)
(A.14)

O 0,0 |++++
F6 0,0
389
"

0,0 2
O0,0
(2 ) + 4+ + 2+ 2 /(2+ + 2 ) i j
(
=
Fmin (ij ). (A.15)
16 cosh(56 /2) 1i<j 4 cosh(ij /2)
i<j
A.2.7. Seven particle form factors

(
i j
(ij )
H 0,1 i<j Fmin
O |
F7
=(
1i<j 7 cosh(ij /2)
0,0
0,0
1,0
0,1
for O1,0
, O0,1
, O1,0
, O0,1
.
(A.16)
A.2.8. Eight particle form factors

m2 4 + i<j cosh(ij ) cosh(12/2) i j
|++
(
F8
Fmin (ij ), (A.17)
=
i<j
|++++
F8

i j
m2 (4 )1/2 (1 3+ + 1+ 3 ) 4 + i<j cosh(ij )
= 7 + 3/2 (
Fmin (ij ),
(
2 (4 )
1i<j 4 cosh(ij /2) 5i<j 8 cosh(ij /2) i<j
(A.18)
|++++++
F8

m2 4 + i<j cosh(ij ) cosh(78/2) i j
(
Fmin (ij ). (A.19)
=
i<j
References
[1] R. Haag, Local Quantum Physics: Fields, Particles, Algebras, 2nd revised edition, Springer,
Berlin, 1996.
[2] D. Buchholz, R. Haag, J. Math. Phys. 41 (2000) 3674;
R. Haag, Questions in quantum physics: a personal view, in: A. Fokas et al. (Eds.),
Mathematical Physics 2000, Imperial College Press, London, 2000.
[3] B. Schroer, T.T. Truong, P. Weisz, Phys. Lett. B 63 (1976) 422;
M. Karowski, H.J. Thun, T.T. Truong, P. Weisz, Phys. Lett. B 67 (1977) 321;
[4] H. Lehmann, K. Symanzik, W. Zimmermann, Nuovo Cimento 1 (1955) 205.
[5] B. Schroer, Ann. Phys. 275 (1999) 190;
H.-J. Borchers, D. Buchholz, B. Schroer, Polarization-free generators and the S-matrix, hepth/0003243.
[6] P. Weisz, Phys. Lett. B 67 (1977) 179;
M. Karowski, P. Weisz, Nucl. Phys. B 139 (1978) 445.
[7] F.A. Smirnov, Form factors in Completely Integrable Models of Quantum Field Theory,
Advanced Series in Mathematical Physics, Vol. 14, World Scientific, Singapore, 1992.
[8] Al.B. Zamolodchikov, Nucl. Phys. B 348 (1991) 619.
[9] H. Babujian, A. Fring, M. Karowski, A. Zapletal, Nucl. Phys. B 538 [FS] (1999) 535.
[10] J. Cardy, G. Mussardo, Nucl. Phys. B 340 (1990) 387.
[11] A. Fring, G. Mussardo, P. Simonetti, Nucl. Phys. B 393 (1993) 413;
A. Fring, G. Mussardo, P. Simonetti, Phys. Lett. B 307 (1993) 83.
[12] A. Koubek, G. Mussardo, Phys. Lett. B 311 (1993) 193.
390
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
F.A. Smirnov, Nucl. Phys. B 453 (1995) 807.

G. Delfino, G. Mussardo, Nucl. Phys. B 455 (1995) 724.
G. Delfino, P. Simonetti, J.L. Cardy, Phys. Lett. B 387 (1996) 327.
O.A. Castro-Alvaredo, A. Fring, C. Korff, Phys. Lett. B 484 (2000) 167.
J.L. Miramontes, C.R. Fernndez-Pousa, Phys. Lett. B 472 (2000) 392.
O.A. Castro-Alvaredo, A. Fring, C. Korff, J.L. Miramontes, Nucl. Phys. B 573 (2000) 535.
E. Fradkin, C. Nayak, K. Schoutens, Nucl. Phys. B 546 [FS] (1999) 711;
E. Ardonne, K. Schoutens, Phys. Rev. Lett. B 82 (1999) 5096.
P. Mitra, Phys. Lett. B 72 (1977) 62.
F.A. Smirnov, Nucl. Phys. B 337 (1990) 156.
H. Babujian, M. Karowski, Phys. Lett. B 471 (1999) 53.
E.H. Wichmann, J.H. Crichton, Phys. Rev. 132 (1963) 2788.
G. Mussardo, P. Simonetti, Int. J. Mod. Phys. A 9 (1994) 3307.
D. Gepner, Nucl. Phys. B 290 [FS20] (1987) 10;
G. Dunne, I. Halliday, P. Suranyi, Nucl. Phys. B 325 (1989) 526.
D. Fioravanti, G. Mussardo, P. Simon, Phys. Rev. E 63 (2001) 016103.
W.H. Press, B.P. Flannery, S.A. Teukolsky, W.T. Vetterling, Numerical Recipes, The Art of
Scientific Computing, Cambridge Univ. Press, Cambridge, 1992.
J.L. Cardy, Phys. Rev. Lett. 60 (1988) 2709.
I.G. MacDonald, Symmetric Functions and Hall Polynomials, Clarendon Press, Oxford, 1979.

Cancellation of Sudakov logarithms in radiative

decays of quarkonia
F. Hautmann
Department of Physics, Pennsylvania State University, University Park, PA 16802, USA
Abstract
We study infrared QCD effects in radiative quarkonia decays. We examine the endpoint region
z 1 of the photon spectrum. We point out a cancellation mechanism for the corrections in
sn lnm (1 z), m 2n, in the short-distance coefficient for the color-singlet Fock state in the
quarkonium. The cancellation is due to the coherence of the color radiation, and applies even though
logarithmic contributions are present in the jet distributions associated with the decay. We comment
on the implications of our results for the modeling of hadronization in the endpoint region and for
the role of color-octet states in the quarkonium. 2001 Published by Elsevier Science B.V.
1. Introduction
Radiative decays of heavy quarkantiquark bound states (quarkonia) have been
investigated in the framework of perturbative QCD and have been used to measure the
QCD coupling at energies of the order of the heavy quark mass m [14]. The use of
perturbative QCD in this context is based on the fact that, as long as the velocity v of
the heavy quark is small, these decay processes involve two widely separated distance
scales: the scale 1/(mv 2 ) over which the quark and antiquark bind into the quarkonium
and the scale 1/m over which the quarkantiquark pair decays. By expanding about the
nonrelativistic limit v 0, one may treat the process as the product of a long-distance,
nonperturbative factor, containing all of the bound-state dynamics, and a short-distance
factor, describing the annihilation of the heavy quark pair and computable as a power
series expansion in s .
However, near the exclusive boundary of the phase space, where the photons energy
approaches its kinematic limit, both the expansion to fixed order in v and the expansion
to fixed order in s become inadequate to represent correctly the physics of the decay. On
E-mail address: hautmann@phys.psu.edu (F. Hautmann).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 6 2 - 6
392
F. Hautmann / Nuclear Physics B 604 (2001) 391404
one hand, classes of relativistic corrections that by power counting are higher order in v
become enhanced by powers of 1/s [57]. On the other hand, potentially large terms in
ln(1 z) (with 1 z being the parameter that measures the distance from the endpoint)
appear in the coefficients of the expansion in s to all orders [8]. In both cases, reliable
results can only be obtained after resummation of the enhanced contributions.
Infrared QCD effects are responsible for these behaviors. Soft color interactions account
for the heavy quarkantiquark pair turning into a color-octet state, whose contribution,
although subleading in v, becomes important near the endpoint. Sudakov logarithms in
1 z arise as remnants of the imperfect cancellation of the infrared divergences in the
exclusive region, where particle production is suppressed. The need for resummations
noted above is a symptom of the sensitivity of the process to contributions from infrared
scales. A corresponding effect is seen in calculations that incorporate models for the
hadronization of partons [4,9,10], where nonperturbative contributions are found to be
essential to describe the endpoint region of the photon energy spectrum in J / [11] and
[3,12] decays.
In this paper we present a treatment of the infrared QCD radiation near the kinematic
boundary. We describe contributions from soft and collinear gluons with logarithmic
accuracy, and study the color correlations between the jets that accompany the photon.
This treatment takes account of the coherence properties of gluon emission. We determine
the regions in the final-state phase space for the quarkonium decay in which destructive
interference between parton emitters suppresses color radiation.
We use these results to examine potentially large Sudakov logarithms in the photon
spectrum. We find (see Section 4) that Sudakov logarithms cancel order by order in s
in the short distance coefficient for the color-singlet Fock state in the quarkonium. The
cancellation mechanism applies to the leading logarithms as well as to any subleading
logarithms. As a result, no Sudakov suppression factor arises near the endpoint.
The picture underlying this behavior can be understood in simple terms based on the
branching formulas for the decay discussed in Section 2. In the boundary kinematics
the photon recoils against two almost-collinear gluon jets. The cancellation of Sudakov
corrections reflects the fact that color is neutralized already at the level of this two-jet
configuration. This situation may be contrasted with the situation one encounters in the
decay of an electroweak gauge boson into jets. Here Sudakov corrections arise precisely
from the presence of color charges in two-jet configurations.
This picture also indicates that no cancellation should occur for the color-octet Fock state
in the quarkonium. In this case, we expect the usual Sudakov suppression to take place near
the endpoint of the photon spectrum. Then one of the consequences of the results of this
paper concerns the ratio of the color-octet to the color-singlet contributions. This ratio
will be smaller than expected from the power counting in v and s obtained by truncating
perturbation theory to fixed order [5,13,14], owing to the different high order behavior of
the perturbation series in the two cases.
A preliminary discussion of the results presented in this paper appeared in Ref. [15].
The paper is organized as follows. In Section 2 we treat the emission of soft and collinear
gluons associated with the quarkonium decay. This leads us to a general branching formula.
393
In Section 3 we evaluate the phase space in this formula. In Section 4 we compute the
photon spectrum and observe the cancellation of the Sudakov logarithms. In Section 5 we
discuss the implications of our results and give conclusions.
2. Infrared radiation near the phase space boundary

In this section we describe the QCD radiation that accompanies the decay of the
quarkonium near the endpoint of the photon spectrum. The key element in our treatment is
the coherent emission of gluons and the angular ordering of this emission. We arrive at an
expression for the photon spectrum that is valid to leading and next-to-leading accuracy in
the soft and/or collinear logarithms.
We consider the decay of a quarkonium state H of mass M into a photon plus anything:
H + X.
(1)
We concentrate on the case of the leading Fock state for a 3 S1 quarkonium state, that is,
a heavy quarkantiquark pair in a color singlet state.
This decay receives contributions at the leading order of perturbation theory both when
the photon is directly coupled to the heavy quarks (direct term) and when the photon
is produced by collinear emission from light quarks (fragmentation term) [16]. In the
endpoint region the fragmentation contribution is suppressed relative to the direct term.
The discussion of this paper will thus focus on the direct term.
2.1. Notations and tree-level decay
The direct contribution at the leading order is given by the process
H (P ) (k) + g(k1 ) + g(k2 )
(2)
evaluated at the tree level. We introduce dimensionless energy variables for the particles in
the final state, as follows
2P k
2P ki
(3)
,
xi =
, i = 1, 2.
M2
M2
In the quarkonium rest frame z and xi are respectively the photon and gluon energies scaled
by the heavy quark mass m (m M/2 in the lowest nonrelativistic approximation).
The normalized photon spectrum from the process (2) is given by [17]
z=
1
1 d0
= 2
0 dz
9
1
1
dx2 (x1 , x2 , z)(z 2 + x1 + x2 )(x1 + x2 1), (4)
dx1
0
with
(x1 , x2 , z) =
(1 x1 )2 (1 x2 )2 (1 z)2
+
+
.
z2 x22
z2 x12
x12 x22
(5)
394
Over almost the entire range in z the result (4) is well approximated by a spectrum rising
linearly with z. At the endpoint, however, the first derivative of the spectrum (4) is singular.
We will come back to this in Section 4.
2.2. Behavior at z 1
Beyond the tree level, corrections in ln(1z) to the spectrum become possible [8]. These
come from the infrared region: near the endpoint the emission of gluons is suppressed
and terms in ln(1 z) result from the imperfect cancellation of the infrared divergences
between real and virtual graphs. The contribution from the soft and/or collinear gluons
associated with the decay is at most double-logarithmic for each power of s . By power
counting the leading behavior of the photon spectrum as z 1 is of the type [8]

1 d
const. +
ck sk ln2k (1 z),
dz
z 1.
(6)
k=1
In this paper we will see a cancellation mechanism for these corrections that works
order by order in perturbation theory. A qualitative picture of the cancellation that we will
discuss in the next sections can be given at one loop as follows. Consider the emission of
a soft gluon with momentum q from the tree-level graphs for the decay (2). In the leading
infrared approximation, the amplitude A for this process factorizes (see for instance [18])
into a nonabelian, classical current j(q), describing the soft-gluon emission, times the tree
level amplitude A0 :
Ab (P , k1 , k2 ; q) gs (q)jb (q)A0 (P , k1 , k2 ),
(7)
where is the gluon polarization vector, b is the gluon color index, and
j (q) =

i
k
Ti i .
ki q
(8)
Here the sum runs over the final-state colored particles k1 and k2 , and Ti are the color
matrices associated with the coupling of the soft gluon to each of these particles. There
is no contribution from the coupling to quarks: the coupling to the virtual quark lines is
subleading in the infrared limit because these lines are off shell by order M; the coupling
to the quark and antiquark in the quarkonium does not contribute either, although these
quarks are nearly on shell, because in the lowest nonrelativistic approximation they are
almost collinear and their total color charge is zero.
To the approximation at which we are working, the phase space also factorizes [18], so
that one gets a factorized answer for the spectrum, schematically of the form

d0
d
s j2
d
.
(9)
dz
dz
Here d
is the Lorentz-invariant phase space for the soft gluon emission. The standard
power counting in terms of the soft gluon energy gives
d
d d,
j2
k1 k2
1
,
2
f (z; angles)
(10)
395
where d is the angular phase space, and f is a function of z and the angles as 0.
Up to the first order in the correlation of the gluon momenta is

2
k1 k2 M (1 z) + g(z; angles) .
(11)
M
That is, as z 1 the photon recoils against two almost-collinear hard gluons. In this
configuration, the logarithmic integration d/ in Eq. (9) is canceled. The next few
sections are devoted to showing that this is indeed the mechanism that dominates the
radiation accompanying the decay, and that it extends to all orders in s .
The back-to-back kinematics suggests an analogy with the two-jet region in the
annihilation of e+ e into electroweak gauge bosons. Since the two jets are now in
a singlet, though, the pattern of color correlations will be different. Jet event shapes in
e+ e annihilation have been studied extensively by using methods based on branching
graphs. We will see in the remainder of this section that the photon spectrum in decays of
quarkonia can be treated in a similar manner. The subtlest point will be the treatment of
the branching phase space. This will be addressed in Section 3.
2.3. Branching graphs
Methods to calculate multi-parton matrix elements in the leading soft and collinear
orders have been known for a long time. See Ref. [18] for a review. The essential
observation in these methods is that one can replace the calculation of higher-loop
Feynman graphs by the calculation of angular-ordered branching graphs, that is, tree
level graphs in which the angular phase space is subject to certain ordering constraints at
each branching. The method has also been extended to include next-to-leading logarithms:
see for instance Ref. [19]. In this and the next subsection we apply this approach to the
quarkonium decay.
We write the energy spectrum d /dz in the inclusive decay (1) in terms of the exclusive
decay widths dn(excl) for producing + n final state partons q1 , . . . , qn :

2P qi

1
1 d
(excl)
=
(q1 , . . . , qn ) z 2 +
(12)
.
dn
dz
M2
n=2
Here the function expresses the photon energy fraction z in terms of the quarkonium
and parton momenta. The exclusive decay widths are generated by the branching
process [1820].
The first step of the branching corresponds to the lowest order final state, Eq. (2),
in which the photon is accompanied by two gluons with invariant mass (k1 + k2 )2 .
This invariant mass provides the basic mass scale at the hard end of the branching.
The subsequent steps correspond to consecutive parton splittings (Fig. 1a). There are
well-defined probability weights associated with each splitting vertex, and form factors
associated with each line connecting two vertices [18,19]. The basic point about the
branching process is that the phase space for the splittings is restricted to the angular
ordered region in which the branching angles decrease as we go from the parent parton
396
Fig. 1. (a) Branching process; (b) final states in the quarkonium decay.
toward the final state. If we denote, as in Fig. 1a, by yi the energy transfer at the vertex i,
and by ri the transverse momentum flowing between vertices i and i + 1, this region is
defined (for small yi ) by
(k1 + k2 )2
2
r1
y12
2
r2
y12 y22
.
(13)
The reason for the angular ordering is a coherence effect: outside the ordered region
parton emitters interfere destructively [2022]. In this approach, the exclusive decay width
dn(excl) is obtained by taking all angular-ordered branching graphs with n final state
partons, and summing the corresponding probability weights.
Following [19], we integrate the exclusive decay widths over the phase space of the final
state partons keeping k1 and k2 fixed. We rewrite the n-parton phase space as (Fig. 1b)

l
n
l
d 4 qj

d 4 qj
2
4
4
2
+ (qj ) = d k1 k1
qj
+ (qj )
(2)3
(2)3
j =1
j =1
j =1
n

n

d 4 qj
4
4
2
d k2 k2
(14)
qj
+ (qj ) .
(2)3
j =l+1
j =l+1
The integration of the branching matrix element over the final state phase space at fixed
ki2 (i = 1, 2) produces the jet mass distribution Jg (p2 , ki2 ) [19]. This is defined as the
probability to generate a jet with invariant mass ki2 from a parent gluon produced at the
mass scale p2 . We have
397

l
l
d 4 qj

2
2 2
4 4
Jg (k1 + k2 ) , k1 =
+ qj (2) k1
qj
(2)3
j =1
j =1
(branching matrix element),
(15)
Note that the jet mass distributions

and an analogous expression for Jg ((k1 +
2
depend on the mass scale (k1 + k2 ) . As implied by the coherence relation (13), this is the
mass scale at which the initial gluons in the branching are produced.
k2 )2 , k22 ).
2.4. Jet mass distributions and inclusive spectrum

To lowest order, there is only one final state parton in the definition (15) of the jet mass
distribution, and Jg is just given by a delta function:

Jg p2 , k 2 = k 2 + O(s ).
(16)
In general, the analysis of the jet branching process allows one to obtain an evolution
equation [19] for the jet mass distribution. Schematically, this has the form

J p2 , k 2 = k 2 + s p
2 K p
2 , k 2 J p
2 , k 2 ,
(17)
where K is a kernel calculable as a power series expansion in s . Expressions for the
kernel K are known to the leading [18,20,22] and next-to-leading order [19,23]. Solving
the evolution equation to the leading logarithms gives
2
2

p
p
f
,
dk 2 Jg p2 , k 2 K 2 k 2 = exp ln
(18)
ln
0 s
K2
K2
where 0 = (11CA 2Nf )/(12) and f is the function

CA 1
(1 2x) ln(1 2x) 2(1 x) ln(1 x) .
(19)
20 x
It is also useful to introduce the double logarithmic approximation. This is defined by
expanding the exponent in the right hand side of Eq. (18) to the first order in s . In this
approximation, using f (x) = CA x/(20 ) + O(x 2 ), we get
2

p
s
.
dk 2 Jg p2 , k 2 K 2 k 2 exp CA ln2
(20)
2
K2
f (x) =
In order to express the inclusive spectrum in terms of the jet mass distributions, we now
use the definitions (14), (15) in Eq. (12). By taking account of the photon phase space
factor, the inclusive spectrum takes the form

1 d
d 4 k d 4 k1 d 4 k2
2P k
4 4
=
+ k 2
(2) (P k k1 k2 ) z
dz
(2)3 (2)3 (2)3
M2

M0 (P , k1 , k2 )Jg (k1 + k2 )2 , k12 Jg (k1 + k2 )2 , k22 .
(21)
Here M0 is the amplitude for the first step in the branching, with final state gg. Owing
to the constraint set by coherence, we can use the approximation (k1 + k2 )2 k12 , k22 and
evaluate M0 by setting k12 = k22 = 0. The explicit expression is readily computed and reads
398

(P k1 )4 (P k2 )4
8 3 M 2
1
4(k1 k2 )2
.
M0 = 2
+
+
9 (P k1 )2 (P k2 )2 (P k)2 (P k2 )2
(P k1 )2
(22)
We are now going to evaluate the branching formula (21) explicitly. To this end, we need
to analyze the phase space of this formula in detail. This is the object of the next section.
3. Structure of the phase space for small jet masses

We are interested in the angular-ordered, coherent region
k12 , k22 (k1 + k2 )2 M 2 .
(23)
Consider the phase space element in Eq. (21):

2P k
d = d k d k1 d k2 (P k k1 k2 ) z
+ k 2 .
2
M
4
(24)
It is convenient to express this phase space in terms of the energy fractions x1 , x2 defined
in Eq. (3), the jet masses k12 , k22 , and
s12 = (k1 + k2 )2 .
(25)
We can use the momentum conserving in Eq. (24) to integrate over k. From the
positivity of the jet energies and invariant masses we have
0 x1 1,
0 k12
0 x2 1,
x12 M 2 /4,
0 k22
x1 + x2 1,
x22 M 2 /4.
(26)
(27)
The relation (25) between s12 and the jet momenta implies
1
1
k12 + k22 + x1 x2 M 2 2|k1 ||k2 | s12 k12 + k22 + x1 x2 M 2 + 2|k1 ||k2 |
2
2

with |k1 | = x12 M 2 /4 k12 , |k2 | = x22 M 2 /4 k22 .
The phase space element can then be rewritten as
(28)

2M 2
dx1 dx2 dk12 dk22 ds12 (z 2 + x1 + x2 ) M 2 (1 z) s12 ,
(29)
4
with the constraints (26), (27), (28).
Eq. (28) selects in general a subspace of a complicated form. But we need to consider it
in the logarithmic region (23). We may use the functions in Eq. (29) to integrate over s12
and one of the energy fractions, say, x2 . Then Eq. (28) gives
d =
x1 x1 x1+ ,
(30)
with

x1 = (1 z/2) 1 z + k12 /M 2 k22 /M 2 /(1 z)

2

M 4 + (1 z)2 2(1 z) k12 + k22 /M 2 (1 z). (31)
(z/2) k12 k22
399
By approximating the constraints (30) for k12 , k22 M 2 (1 z) M 2 , we get

x1
k12
,
M 2 (1 z)
(32)
and
x1 1
k22
,
M 2 (1 z)
(33)
that is, since 1 x1 x2 for z 1,

x2
k22
.
M 2 (1 z)
(34)
Eqs. (32) and (34) tell us that in the coherent region the phase space available for the
evolution of each of the jets is bounded by the recoiling jet. For z 1 this bound is tighter
than the bound from the fragmentation of the jet itself, Eq. (27). In the next section we
will see that the recoil constraint conspires with the angular-ordered form of the jet mass
distribution to destroy any logarithmic hierarchy in the inclusive photon spectrum.
4. The photon spectrum near the endpoint

We now put together the results of Section 3 for the phase space with the structure of the
spectrum described in Section 2 based on the coherent branching.
By using Eqs. (22), (29), (32) and (34) in Eq. (21), we can rewrite the photon energy
spectrum as follows:
1
1 d
2
dz
9
1
1
dx2 (x1 , x2 , z)(z 2 + x1 + x2 )(x1 + x2 1)
dx1
0

dk12 Jg M 2 (1 z), k12 M 2 x1 (1 z) k12 M 2 x12 /4 k12

dk22 Jg M 2 (1 z), k22 M 2 x2 (1 z) k22 M 2 x22 /4 k22 , (35)
with (x1 , x2 , z) given in Eq. (5). The main structural difference with respect to the case
of jet event shapes in e+ e annihilation is that in Eq. (35) there is a two-dimensional
integration over a distribution in the energy fractions x1 , x2 . This comes from the fact
that in quarkonia decays the parton branching is probed by a non-pointlike source. Eq. (35)
allows us to discuss the logarithmic behaviors in the endpoint region.
Observe first that, by substituting into Eq. (35) the zeroth-order expression (16) for the
jet mass distribution Jg , we recover the lowest order result (4). So at the lowest order
of perturbation theory Eq. (35) gives the correct answer for any z. At higher orders
of perturbation theory, Eq. (35) gives an approximation valid in the region of large z
400
with leading-logarithm and next-to-leading-logarithm accuracy, provided the jet mass

distributions Jg are evaluated with corresponding accuracy. Once expanded to the next to
lowest order in s , Eq. (35) can be matched with the NLO perturbation theory result [24],
and could thus be used to obtain improved predictions, valid over a wider range of z.
Eq. (35) enables us to see that, although Sudakov logarithms are present in the jet
distributions associated with the decay, the corrections in ln(1 z) to the photon energy
spectrum cancel order by order in s . The mechanism for the cancellation can be seen in its
simplest form by using the double-logarithmic approximation (20) for Jg . By expanding
the right hand side of Eq. (20) in powers of s and substituting this into Eq. (35), we note
that the higher order corrections to d /dz involve integrals of the form

1
dx1 ln
1z
M 2 (1 z)
M 2 x1 (1 z)
1
=
dx1 lnk (1/x1).
(36)
1z
That is, logarithmic contributions in x1 arise, which are important at the kinematic limit
x1 0, but these never give rise to logarithms of (1 z) in the photon spectrum. Terms in
ln(1 z) cancel because of coherence, i.e., as a result of the constraint (32) on the jet mass
and of the angular ordering (13) in the branching.
From Eq. (35) we can obtain an explicit expression for d /dz by keeping track of all the
leading logarithms in the jet distributions. This is accomplished through Eq. (18). Using
this formula in Eq. (35) we get
1
1
1 d
1
2
dx1 dx2 (x1 , x2 , z)(z 2 + x1 + x2 )(x1 + x2 1)
dz
9
0
0

exp ln(1/x1)f 0 s ln(1/x1) + ln(1/x2)f 0 s ln(1/x2 ) ,
(37)
where f (x) is given in Eq. (19). It is straightforward to check from Eq. (37), by expanding
the integrand in powers of s , that no logarithms of (1 z) appear in the perturbative
expansion of d /dz.
The above results indicate that the photon spectrum from the decay of the color-singlet
Fock state in the quarkonium is not Sudakov suppressed by higher orders of perturbation
theory in the endpoint region. Higher perturbative corrections give rise to a constant shift
compared to the leading order answer (4).
The first derivative of the spectrum, on the other hand, does get logarithmic corrections
in (1 z). This can be realized just from Eq. (36) by taking the derivative with respect to z.
A singularity in the derivative of the spectrum at z = 1 was indeed noted already in lowest
order (see comment below Eq.(5)). The leading logarithmic behavior at one loop will be of
the type

d 1 d
= a0 ln(1 z) + const. + O(1 z)1
dz dz

+ s a1 ln3 (1 z) + subleading logs + O(1 z)1 + O s2 . (38)
401
It is of much interest to calculate these corrections. Likely, the z = 1 divergence that

appears in the derivative of the spectrum at any fixed order of perturbation theory will
be smoothed out by the all-order summation of the logarithms. Note however that this
calculation will not necessarily involve only the configurations in which the photon and the
gluon jets are at large relative angle, and the jets evolve according to the angular-ordered
branching, but also configurations in which color is emitted at angles comparable to that of
the photon.
To conclude this section, we remark that near the boundary the decay becomes sensitive
to the hadronization process. For the inclusive spectrum, these effects can be factorized in
nonperturbative shape functions [2527]. A very important question concerns the modeling
of these functions. The Monte Carlo calculation [9] provides one such model, based on
parton showering and the assumption of independent fragmentation. Another model is that
of Refs. [4,10], in which nonperturbative corrections are parameterized in terms of an
effective gluon mass mg [28]. We observe that results on the infrared behavior such as those
presented in this paper can be used to investigate models of hadronization. In particular,
if the resummed formulas for the photon spectrum discussed above are combined with
models for the behavior of the running coupling at low energy scales [26,29], they provide
an ansatz for the power corrections in the shape functions [30]. These power corrections are
the analogues of the contributions in mg /M of [10,28], which are found [4] to be necessary
to describe the data in the endpoint region.
5. Conclusions
The photon spectrum in quarkonia decays is a critical issue in QCD phenomenology.
Although, as was realized early on [31], for large enough masses the decay is dominated by
short distances, the observed behavior [3,11,12] of the spectrum at large z is not reproduced
by fixed-order perturbation theory. In this respect the first implication of the results of this
paper is negative: perturbative resummations do not fix the problem. We have seen in the
previous sections that resummation does not produce a large-z suppression of the spectrum,
but a constant shift.
A second, perhaps more general implication is that color coherence effects in the
quarkonium decay are important. As we have seen, they change dramatically the large-z
behavior of the perturbation series compared to what one would conclude from the simple
infrared power counting [8], leading to the cancellation of the Sudakov terms.
Note that the comparison of theory with experiment (see, e.g., [2,3]) has so far involved
the use of models for the parton cascade and hadronization [9] that do not include
coherence. In the case of [9] the emission of color associated with the evolution of the
jets takes place within a cone whose typical size is 1. But the analysis of this paper
indicates that in fact, since the coherence scale in the jet mass distributions is not the
2
2
quarkonium mass
M but rather (k1 + k2 ) , destructive interference occurs outside a cone
with size 1 z. It will be valuable to make Monte Carlo models for quarkonium
decays that take this into account.
402
In the endpoint region relativistic corrections may become important [5,14,32]. Detailed
estimates of color octet contributions have recently appeared [6,7]. In this paper we have
observed the absence of Sudakov suppression for the Fock state of lowest order in the
nonrelativistic expansion, i.e., a color singlet quarkantiquark pair. The mechanism that
produces this result is based on the color correlations of two gluon jets, and therefore is
not at work in the case of the color-octet state. This means that the power counting in the
heavy quark velocity and in s at fixed order is modified by the high order behavior in very
different ways for the color-singlet and color-octet channels. The relative size of the color
octet with respect to the color singlet in the endpoint region will be smaller than indicated
by the power counting at fixed order.
Quarkonia decays are used to measure the QCD coupling and contribute significantly to
the world average value of s [33,34]. In fact Ref. [34] quotes a very precise determination
of s using this method. As we have seen, however, effects from the endpoint region in
radiative decays may be quite dramatic (cancellations at large z, lack of angular ordering
in current Monte Carlo models, need for an all-order power counting for relativistic
corrections), and have yet to be fully understood. The present uncertainty on the extraction
of s from these decays is therefore likely to be larger than previously estimated.
There has recently been progress in next-to-leading-order calculations for the photon
spectrum. NLO corrections have been computed for color octet contributions [6] and
for color singlet, direct contributions [24]. The missing piece are the corrections to the
fragmentation terms [16], to be used with the next-to-leading fragmentation functions
of the photon [35]. A possible choice for future determinations of s will be to use
only data sufficiently away from the endpoint for NLO perturbation theory to be a valid
approximation [7,24]. This raises the question, though, of where the safe region begins and
how much of the data one is left with. Alternatively, note that resummation formulas of
the type presented in this paper can be matched and combined with NLO results. This will
give improved predictions, which could likely be used in a wider range of photon energies
towards the peak region.
Hadronization physics, showing up as power-suppressed corrections to the spectrum,
becomes a dominant effect near the endpoint. Power corrections are indeed found to be
essential to describe the data in this region [4,10]. It becomes important to investigate
models for the nonperturbative shape functions [26,27,36] parameterizing these effects.
The structure of soft color emission discussed in this paper, once combined with an ansatz
for the infrared behavior of the coupling [30], may serve to study these models.
Acknowledgements
I am grateful to S. Catani for collaboration on quarkonium physics and for discussion.
Part of this work was done while I was visiting the University of Oregon. I am grateful to
J. Brau, D. Soper and the University of Oregon Center for High Energy Physics for their
hospitality and support. I thank G. Korchemsky and M. Krmer for useful conversations.
403
This research is funded in part by the US Department of Energy under grant No. DE-FG0290ER-40577.
References
[1] M. Kobel, in: J. Tran Thanh Van (Ed.), QCD and Hadronic Interactions, Proceedings of the 27th
Rencontres de Moriond, Editions Frontires, Gif-sur-Yvette, 1992, p. 145.
[2] J.H. Field, Nucl. Phys. B Proc. Suppl. 54A (1997) 247.
[3] CLEO Collaboration, B. Nemati et al., Phys. Rev. D 55 (1997) 5273.
[4] J.H. Field, hep-ph/0101158.
[5] I.Z. Rothstein, M.B. Wise, Phys. Lett. B 402 (1997) 346.
[6] F. Maltoni, A. Petrelli, Phys. Rev. D 59 (1999) 074006.
[7] S. Wolf, hep-ph/0010217.
[8] D.M. Photiadis, Phys. Lett. B 164 (1985) 160.
[9] R.D. Field, Phys. Lett. B 133 (1983) 248.
[10] M. Consoli, J.H. Field, Phys. Rev. D 49 (1994) 1293;
M. Consoli, J.H. Field, J. Phys. G 23 (1997) 41.
[11] Mark II Collaboration, Phys. Rev. D 23 (1981) 43.
[12] R.D. Schamberger et al., Phys. Lett. B 138 (1984) 225;
CLEO Collaboration, Phys. Rev. Lett. 56 (1986) 1222;
ARGUS Collaboration, Phys. Lett. B 199 (1987) 291;
Crystal Ball Collaboration, Phys. Lett. B 267 (1991) 286.
[13] T. Mannel, S. Wolf, hep-ph/9701324.
[14] M. Gremm, A. Kapustin, Phys. Lett. B 407 (1997) 323.
[15] F. Hautmann, hep-ph/9708496; in: A. Buijs, F.C. Ern (Eds.), Proceedings of Photon 97,
World Scientific, 1998, p. 68.
[16] S. Catani, F. Hautmann, Nucl. Phys. B Proc. Suppl. 39BC (1995) 359.
[17] S.J. Brodsky, T.A. DeGrand, R.R. Horgan, D.G. Coyne, Phys. Lett. B 73 (1978) 203;
K. Koller, T. Walsh, Nucl. Phys. B 140 (1978) 449.
[18] Yu.L. Dokshitzer, V.A. Khoze, A.H. Mueller, S.I. Troyan, Basics of Perturbative QCD, Editions
Frontires, Gif-sur-Yvette, 1991.
[19] S. Catani, L. Trentadue, G. Turnock, B.R. Webber, Nucl. Phys. B 407 (1993) 3.
[20] V.S. Fadin, Sov. J. Nucl. Phys. 37 (1983) 245;
B.I. Ermolaev, V.S. Fadin, JETP Lett. 33 (1981) 269.
[21] A.H. Mueller, Phys. Lett. B 104 (1981) 161.
[22] Yu.L. Dokshitzer, V.S. Fadin, V.A. Khoze, Phys. Lett. B 115 (1982) 242;
Yu.L. Dokshitzer, V.S. Fadin, V.A. Khoze, Z. Phys. C 15 (1982) 325.
[23] J. Kodaira, L. Trentadue, Phys. Lett. B 112 (1982) 66.
[24] M. Krmer, Phys. Rev. D 60 (1999) 111503, hep-ph/9901448.
[25] G.P. Korchemsky, G. Sterman, Phys. Lett. B 340 (1994) 96.
[26] R.D. Dikeman, M. Shifman, N.G. Uraltsev, Int. J. Mod. Phys. A 11 (1996) 571;
I.I. Bigi, M.A. Shifman, N.G. Uraltsev, A.I. Vainshtein, Int. J. Mod. Phys. A 9 (1994) 2467.
[27] A.L. Kagan, M. Neubert, Eur. Phys. J. C 7 (1999) 5;
M. Neubert, Phys. Rev. D 49 (1994) 4623.
[28] G. Parisi, R. Petronzio, Phys. Lett. B 94 (1980) 51.
[29] Yu.L. Dokshitzer, G. Marchesini, B.R. Webber, Nucl. Phys. B 469 (1996) 93.
[30] Yu.L. Dokshitzer, hep-ph/9812252; in: A. Astbury, D. Axen, J. Robinson (Eds.), Proceedings
of the 29th International Conference on High-Energy Physics ICHEP98, Vancouver, Canada,
July 1998, World Scientific, Singapore, 1999, p. 305.
404
[31] T. Appelquist, H.D. Politzer, Phys. Rev. Lett. 34 (1975) 43;

A. De Rjula, S.L. Glashow, Phys. Rev. Lett. 34 (1975) 46.
[32] W.Y. Keung, I.J. Muzinich, Phys. Rev. D 27 (1983) 1518.
[33] S. Bethke, J. Phys. G 26 (2000) 27.
[34] I. Hinchliffe, A.V. Manohar, Ann. Rev. Nucl. Part. Sci. 50 (2000) 643.
[35] L. Bourhis, M. Fontannaz, J.P. Guillet, Eur. Phys. J. C 2 (1998) 529.
[36] T. Mannel, S. Recksiegel, hep-ph/0009268.

Neutrino mass matrix in Anti-GUT with see-saw

mechanism
H.B. Nielsen, Y. Takanishi
The Niels Bohr Institute, Blegdamsvej 17, DK-2100 Copenhagen , Denmark
Abstract
We have investigated the predictions of neutrino oscillations within extended Anti-GUT, based
on a large gauge group the Standard Model and an BL abelian group assumed with separate
(SMGi U (1)B L,i ). We take

gauge fields coupling to each family of quarks and leptons i=1,2,3
into account corrections concerning a crude way of accounting for the number of ways the vacuum
expectation values of the Higgs fields of the model can be ordered in the Feynman diagrams yielding
the mass matrix elements. Performing these corrections in a balanced way between the charged
fermion sector and the neutrino sector (case I) leads to the previous version of our model that was
a fit to the MSW small angle solution region. The fit considered here is marginally worse that the
previous fit inasmuch as it predicts a somewhat lower solar neutrino mass square difference (which
might though be the only way of avoiding the day-night exclusion region). The other two versions
(case II and case III) are characterised by the heaviest of the see-saw neutrino matrix elements that
are composed either of R and R or two R . These cases II and III fit the small mixing angle
MSW scenario very well. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.10.-g; 13.15.+g; 14.60.-z; 14.60.Pq
Keywords: Lepton number violation; Neutrino oscillations; Mass matrices
1. Introduction
The latest results of Super-Kamiokande collaboration [1] suggested that the day-night
effect spectra disfavour the MSW [2] small mixing angle solution (MSW-SMA) at the
level of 95% C.L. However, it might be that the MSW-SMA could not be excluded by the
experiments because the measurement of the day-night effect is rather difficult. Therefore,
we prefer to allow the possibility of the MSW-SMA, since we have recently predicted the
MSW-SMA within extended Anti-GUT [3] framework.
E-mail addresses: hbech@nbi.dk (H.B. Nielsen), yasutaka@nbi.dk (Y. Takanishi).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 7 1 - 7
406
H.B. Nielsen, Y. Takanishi / Nuclear Physics B 604 (2001) 405425
It is also important to study whether the Anti-GUT model could, after all, predict the
MSW large mixing solution (MSW-LMA): in our calculations we have used a technical
correction [4] (factorial factor correction) which takes into account the number of
Feynman diagrams contributing to a given mass matrix element in a crude statistical way.
In fact we used the parameters from a fit to quarks and charged leptons mass matrices
without factorial factors correction, while we partly used them in the neutrino oscillation
calculations. The major aim of the present article is to treat the charged fermions and the
neutrinos on an equal footing with respect to this technical correction. We have assumed
that one of the parameters is fixed to be unity as in earlier works [5]. In this article we
will take into account that this parameter is not a priori unity. However, even these smaller
variations do not bring our model to fit MSW-LMA but still only the MSW-SMA domain.
Although our extended Anti-GUT model is rather restricted with respect to the choice of
charge combinations for the various fermions and even for the Higgs fields introduced to
break the gauge group assumed in order to keep the already achieved good fits, there are a
few possibilities for playing around. One of the most important possibilities is that we can
vary the quantum numbers for the Higgs field, called B L , the vacuum expectation value
(VEV) of which is giving the scale of the see-saw particles (the right-handed neutrinos)
by breaking the (BL) charge (which is assumed to be gauged even in the form of one
(BL)i for each family where i denotes a generation) and thereby the overall scale of the
neutrino oscillations.
In Section 2 we describe briefly the extended Anti-GUT model for the calculation of
neutrino mass squared differences and the mixing angles. Then, in Section 3, we put
forward the mass matrices. In Section 4 we describe the factorial correction. Then in
Section 5 we will present the experimental data and the old fits on the charged lepton
and the quark, while we in Section 6 give the results of our calculations. Section 7 contains
our conclusion and summary.
2. The extended Anti-GUT model

The Anti-GUT model is characterised by a gauge group which for each family has a
set of (family specific) gauge fields. In addition there is an extra U (1) gauge group called
U (1)f . The details of the couplings of the latter is largely specified by the requirements
of the model being free of gauge and mixed anomalies. So each generation gets its own
system of gauge particles, i.e., gauge fields come in generations just like the fermions.
The extension consists in providing each family also with a gauge field coupling to the
(BL)-charge of that family alone. That is to say we postulate for each family 1 the
subgroup of the well-known grand unification group, SO(10), which consists only of those
generators that do not mix the different irreducible representation of the Standard Model,
SMGi U (1)B L,i = SU(3) SU(2) U (1) U (1), where SMG denotes the Standard
1 Really, we should say proto-family because (1) these proto-family get mixed before they are identified with
the true families (2) even the right-handed charm- and top-quarks get permuted.
407
Fig. 1. The energy scale of the vacuum expectation values of the Higgs fields within the extended
Anti-GUT model.
Model gauge group. The U (1)f that seems a strange extra ingredient in the old AntiGUT model although called for in order to be able to fit the quark masses [6]
is in the extended version just a linear combination of (BL)i and yi /2 which happen
to have separate anomaly cancellations for the right-handed neutrinos and for the rest
(true Standard Model). Since the old U (1)f quantum charge is a linear combination
of (B L)2 , (B L)3 , y2 /2 and y3 /2 any anomaly constraint for U (1)f and the extended
model charges must be automatically satisfied. Thus there formally can be no hindrance
for having U (1)f as well as separate U (1)BL,2 and U (1)B L,3 gauge groups. However,
such a possibility would imply that there would be a linear combination of the charges that
would decouple from all the fermions and thus should not really be considered. In fact, it
is a part of the arguments for the Anti-GUT model that one decides to ignore gauge fields
which decouple from all observed fermions on the ground that they would have very little
phenomenological relevance.
The Higgs fields in our model are assigned quantum numbers under the gauge group
(SMGi U (1)B L,i ) which are determined by seeking a

extended Anti-GUT = i=1,2,3
fit to the quark and lepton spectra including its mixing angles. It should, however, be kept
in mind that the possibilities for fitting these charge quantum numbers are few, since they
are only allowed to take quantised values analogous to that these quantum numbers for
the quarks and leptons only take simple quantised values and thus the model remains
very predictive. The model ends up having only 7 Higgs fields falling into four classes
according to the order of magnitude of the expectation values (see Fig. 1): 2
(1) The smallest VEV Higgs field in this model plays the role of the Standard
Model WeinbergSalam
Higgs field, WS , with the VEV at the weak scale being
246 GeV/ 2.
(2) The next smallest VEV Higgs field is also alone in its class and breaks the common
BL gauge group U (1), common to the all the families. This symmetry is supposed
to be broken (Higgsed) at the see-saw scale as needed for fitting the over all neutrino
oscillation scale. This VEV is of the order of 1012 GeV and called B L . This field
was not assumed to exist in the old Anti-GUT model. 3
(3) The next 4 Higgs fields are called , T , W , and and have VEVs of the
order of a factor 10 to 50 under the Planck unit. That means that if intermediate
propagators have scales given by the Planck scale, as we assume, they will give
rise to suppression factors of the order 1/10 each time they are needed to cause a
2 The quantum numbers of the seven boson fields are shown in Table 1.
3 The influence, on the charged lepton mass matrix, of the new Higgs field, , is only on the off-diagonal
elements which are remain suppressed and do not dominate any masses or mixing angles.
408
transition. The field was introduced for the purpose of the study of neutrinos and
was not present in the old model.
(4) The last one, with VEV of the order of the Planck scale, is the Higgs field S,
which gives little or no suppression when it is applied. Therefore, a transition
amplitude in first approximation is not noticeably suppressed by this field S.
Thus it gives rise to ambiguities in the model and its presence are not easily
distinguished in phenomenology. Only if we take the VEV discernibly different
from zero and/or make use of the factorial corrections, there is a possibility to
observe phenomenological consequences of the field S.
Therefore, it is part of the model also that all physics beyond Standard Model is in
unextended Anti-GUT first a couple of orders of magnitude under the Planck scale and
in the extended one begins at the see-saw scale. So, there is pure Standard Model to seesaw scale and consequences of the Standard Model such as the GIM mechanism [7] of no
flavour changing neutral current are valid with the correspondingly very high accuracy, by
far inaccessible to present days experiments.
Since we have non-zero VEVs of scalar fields which are doublets under two different
proto-generation specific SU(2)s such as S, W , T , (while WS has to be doublet
for an odd number of generations), the mass eigenstates of the quarks and leptons do
not belong to definite representations of the proto-generation specific SU(2)s. There is
therefore no hindrance in having non-zero mixing angles, but they are of course suppressed
corresponding to the need for the breaking VEVs being involved.
In our previous article [3] we formulated the quantum numbers of our model with the
use of the quantum number for the subgroup of the full gauge group called U (1)f , but it
is more elegant without this abelian gauge group. The U (1)f quantum number, Qf , used
in earlier articles both in extended Anti-GUT (where it can be avoided), and in old
Anti-GUT (where there are no right-handed neutrinos and thus no anomaly free (BL)s,
so that U (1)f is unavoidable) is related to the quantum numbers of Table 1 by

Qf = (BL)3 (BL)2 + 2

y2 y3
.
2
2
(1)
The calculation of the mass matrices in our model consists in evaluating for each mass
matrix element which of our seven Higgs fields are needed to provide the difference in
quantum numbers between the right-handed Weyl components and the fermion in question
to the left-handed ones. Then, one imagines that the propagator fermions all have Planck
or fundamental mass scale masses in a Feynman diagram which is often a long chain of
interactions with the successive Higgs fields that were needed. The order of magnitude of
the diagram is then of the order of the Planck scale multiplied by a suppression factor
for each Higgs field used. Now we assume that there are of order unity random couplings
all along the chain and we therefore take for the value of the matrix element the product
of the suppression factors times a random factor of order unity. These factors are taken as
random numbers and at the end a logarithmic averaging of the resulting mass eigenvalues
and mixing angles is taken.
409
Table 1
All U (1) quantum charges in extended Anti-GUT model. We presented here the three different
possibilities of the B L quantum charges. The symbols for the fermions shall be considered to
mean proto-particles. Non-abelian representations are given by a rule from the abelian ones, see
below Eq. (3)
SMG2
SMG3
UB L,1
UB L,2
UB L,3
1
6
2
3
13
12
1
3
1
3
1
3
0
0
0
0
0
0
1
1
1
0
0
0
0
0
0
cL , sL
sR
1
3
1
3
1
3
cR
L , L
R
R
0
0
0
1
6
2
3
13
12
0
0
0
0
0
0
1
1
1
0
0
0
tL , bL
tR
bR
1
3
1
3
1
3
L , L
R
R
0
0
0
0
0
0
1
6
2
3
13
12
0
0
0
0
0
0
1
1
1
WS
2
3
16
12
16
16
16
13
23
0
0
0
0
0
0
0
0
0
1
1
1
2
SMG1
uL , dL
uR
dR
eL , eL
eR
eR
1
0
1
6
1
6
1]
B L
2]
B L
3]
B L
0
0
0
0
0
1
0
1
0
1
2
1
3
1
3
2
3
13
13
1
6
0
0
1
0
0
0
1
0
1
0
0
0
0
1
3
3. Mass matrices and Higgs quantum numbers in the extended Anti-GUT model
To write down the mass matrices we need the quantum numbers of the Higgs fields.
However, there is the freedom that we can modify the charge assignments without too
much change in the predictions if we change the fields by adding to their quantum number
assignment the quantum numbers of S. Thus we shall take the point of view that we really
do not know from the already developed phenomenology of the charged sector (quarks and
410
charged leptons) the abelian quantum numbers except modulo those of S.

To take into account the ambiguity of the choice of the quantum charges of Higgs fields,
due to the VEV of S being of order of unity in Planck units, we could parameterise the
quantum number combinations by integer parameters, , , , , and # telling the
number of additional S-quantum number combinations counted from some starting value
of the quantum number combinations of the Higgs fields which are given in Table 1. 4 This
consideration leads to the following Higgs field quantum charges:

2 2
1 1
, , 0, , , 0 ,
S=
6 6
3 3

1 1 1 2
1
W = , , , , 1,
+ S,
6 3 2 3
3

1 2 2
1
T = , 0, , , , 0 + S,
6
6 3 3

1 1
1 1
=
, , 0, , , 0 + S,
6 6
3 3

1
1 1 1 2
, , , , 1,
WS =
+ S,
6 2 6 3
3
= (0, 0, 0, 0, 1, 1) + S,
case I :
B L = (0, 0, 0, 1, 0, 1) + #S,
case II : B L = (0, 0, 0, 0, 1, 1) + #S,

case III : B L = (0, 0, 0, 0, 0, 2) + #S.
(2)
Table 1 and Eq. (2) is simplified by only containing the abelian quantum numbers, but
we do in fact imagine that in our model we have the proto-generation specified by the
following rule: For each proto-generation i the non-abelian representations of SU(2)i
and SU(3)i are the smallest dimension ones obeying the restriction
ti
yi
di
+ + = 0 (mod 1).
2
3
2
(3)
Here ti is the triality being 1 for 3, 1 for 3 and 0 for 1 or 8, respectively. The duality di
is 0 for integer spin SU(2)i representations while 1 for half-integer SU(2)i spin.
Concerning the see-saw scale determining field B L , we have listed three promising
choices of the quantum numbers in addition to the shifting S-quantum numbers choice.
With these quantum number combination we get the parameterised mass matrices as
follows, where we simply write S, W , etc., instead of S , W , etc., for the VEVs:
the uct-quarks:
S 1+2+2 + W T 2 ( )2 S 2+2 + W T 2 S 2 + (W )2 T
WS
S 2+2+ W T 2
S 2+ (W )2 T
S 1+2+3 + W T 2 ( )3
, (4)
MU

2
3
+
3
1+
( )
S
S 1+++ W T
S
4 The starting values were though not the ones of Table 1 as seen from Eq. (12) below.
411
the dsb-quarks:

1+22
2 2
2+2
2
23 3
WS S 1+2 W (T )2 S 2+2 W (T )2 S 23 T3
W (T )
S
W (T )
S
T
S
MD

,
2
S 22+4 W 2 (T )4
S 32+4 W 2 (T )4
S 1 W T
the charged leptons:
1+22 2 2
W (T )
WS S 1+25
W (T )2 5
ME

S
2
S 2+245 (W )2 T 4 5
the Dirac neutrinos:

WS S 1+2+2 + W T 2 ( )2
D
M

S 1+2 + W T 2
2
S 2++ ++ W T
S 2+2+3 W (T )2 ( )3 S 44+3 W T 4 ( )3
S 2+2 W (T )2
S 44 W T 4
S 1+24 (W )2 T 4
S 1 W T
S 2+2+3 + W T 2 ( )3 S 2+2+3 + (W )T 2 ( )3
S 2+2+ W T 2
S 2+2+ (W )T 2
S 1++++ W T
S 1+++ W T
(5)

, (6)

. (7)
Now we have to get the right-handed Majorana neutrino mass matrix to be able
to calculate the effective neutrino mass matrix using the see-saw mechanism [810].
However, there are three different choices of the quantum charge of B L , none of which
give results that can be excluded immediately: the (1, 3)-component of the right-handed
mass matrix is dominant (called case I), another choice is the (2, 3)-component is the
dominant one (called II), and third one is the (3, 3)-component as the dominant one
(called III). In our previous paper, the right-handed neutrino mass matrix is considered
as case I.
The right-handed Majorana neutrinos is given in the case I by:
1 +#

S #
S #
S
S #
S 1+ +#
S 1+ # .
MR
B L
(8)
#
1+
#
1+
#
S
S
S
In the case II:
S 22 +# 2
MR
B L
S 1+ +#
S 1+ #
In the case III:
S 1+ +#
S #
S #
S 1+ #
S # .
#
S
(9)
S 1 +#
.
S #
#
S
(10)
Note that the quantum numbers of our different Higgs fields are not totally independent
in so as far as there is a relation between the quantum numbers,
S 22 +2# ( )2 2
MR
B L S 1 +2# ( )2
S 1 +#
S 1 +2# ( )2
S 2# ( )2
S #
W 9Q
T + (6 3 + 9 + )Q
S,
= 3Q
Q
(11)
and thus the Higgs field combinations needed for a given transition are not unique, so the
choice of the largest contribution for each matrix element must be selected. To compare
with our earlier work (except for [4]) and Table 1, we remark that the quantum number
combination used in that work is obtained by putting
= = 1,
= 0,
= 1,
= 0,
# = 0.
(12)
412
Fig. 2. One of the typical Feynman diagrams getting mass matrix element. The tadpole lines with
crosses symbolise VEV of the different Higgs fields.
4. The number of orderings correction for the Feynman diagrams

The technical detail called factorial correction consists in the following argument
for a Feynman diagram counting correction: the external lines signifying the attachment
of the various Higgs fields used to make the quantum numbers match, can be put in
several different sequences along the chain of fermion propagators in the diagram (see
Fig. 2) providing the transition. For all these different sequences quite different fermion
propagators are used. These differently ordered VEV-attachments are expected to be
statistically independent and should be added with random phases. That means that one has
a random walk in the complex plane (of amplitude values). The average of the numerical
square of the amplitude for the mass matrix element, say, is getting additive contributions
from the various diagrams and so goes up linearly with the number of diagrams. This
means then that the amplitude goes as the square root of the number of diagrams with
different orderings of the attachments of the different Higgs field symbols designating the
action of these Higgs field VEVs. If we, for example, think of a matrix element for the
electron mass, it turns out that in the model, it is necessary to use the vacuum expectation
value for the WeinbergSalam Higgs field WS once, for the twice, for the T twice, for
the W once, and for the S field n times where n depends on the specific assignment of the
quantum numbers to the other fields.
The number of orderings in which we can have the attachments of the VEVs for the
electron mass generating diagram (in first approximation) is then (6 + n)!/(2! 2! n!). This
means that we expect the amplitude statistically to be WS 2 T 2 W S n (6 + n)!/(2! 2! n!).

When these crude estimates of this correction were taken into account in the charged quark
and lepton fits, it turned out that typically a somewhat smaller value of the expectation
value for S was called for, say, around 1/2. Also the other VEVs would be somewhat
changed in the improved fit including this factorial correction (so called because we
have seen that it is square roots of factorials that come in). It is clear that the parameter ,
which roughly plays the role of the Cabibbo angle and also explains the ratio of the mass
of the first- to the second-generation (as being 2 ), will be smaller with the factorial
corrections included in a fit because of the highly suppressed first-generation masses, of
course, tended to have more fields to permute and thus a larger enhancement due to the
factorials than, say, the second-generation. Such a fit, then, must make smaller with
factorials in order to compensate so as to keep the first- to second-generation mass ratio
fixed. Indeed the -VEV without factorials is fit to about 1/10 whereas with factorials
is instead rather 1/30.
413
Here is another little detail: Since the field S has the expectation value near unity (in
Planck units) having one or several S-factors is not distinguishable from the fitting. Hence,
S-factors are not very well determined from phenomenology. Now the number of S-factors
needed to make a given quantum number shift can be changed by modifying the assignment
of the quantum numbers for some of the other Higgs fields. Hence one can compensate for
the changed quantum number by accompanying the other fields with a number of S-fields,
so as to get the same total transition in quantum number.
5. Data and fitting of charged sectors

Before the results are presented, we should review briefly the neutrino experimental
data [1,1118]: the best fit to the Super-Kamiokande atmospheric neutrino data shows
near-maximal mixing angle and (m2atm
3.2 103 eV2 , the 90% C.L. range being
(27) 103 eV2 . As for the solar neutrino data there are four different solutions; the
MSW-SMA require values of the mass square difference and mixing angle that (at 99%
C.L.) lie in the intervals
4.0 106 eV2 (m2 1.0 105 eV2 ,

1.3 103 sin2 2 1.0 102 3.25 104 tan2 2.5 103 ,
whereas the MSW-LMA solution is realised in the intervals
7.0 106 eV2 (m2 2.0 104 eV2 ,

0.50 sin2 2 1.0 0.17 tan2 1.0 .

The LOW solution lies approximately in the region
0.4 107 eV2 (m2 1.5 107 eV2 ,

0.80 sin2 2 1.0 0.38 tan2 1.0 ,

and the Vacuum oscillation (VO) in
5 1011 eV2 (m2 5 1010 eV2 ,

0.67 sin2 2 1 0.27 tan2 4 .

Since the parameter values resulting from the fits to the charged mass matrices for, e.g.,
are of the order of a factor 3 smaller than without the factorial correction, we have
a priori an uncertainty of that order if we are not careful to treat the neutrino part of the
calculation in the same way as the charged particle sector. In our previous paper we made
partial factorial corrections for the neutrinos using the old parameters, of which is
most important, taken from fits to charged masses and mixing angles without factorial
corrections. One can immediately foresee that making factorial corrections for both
charged fermions and neutrinos should have the same effect as correcting the previous
work [4] very roughly as if we simply decrease the -value used for the neutrino
predictions by about a factor 3.
414
Table 2
Fits including O(1) factors to quarks and charged leptons. The VEV of Higgs fields are measured in
the Planck unit. For the notation of , , , , see formulas (2)
i
ii
iii
iv
v
vi
vii
viii
ix
x
xi
xii
xiii
xiv
1
1
1
1
1
0
0
0
0
1
1
1
1
1
1
0
0
1
1
0
0
0
1
1
1
0
0
1
1
1
1
1
1
1
1
0
1
1
1
1
1
1
1
1
1
1
1
0
1
0
0
1
1
0
1
0
W
T
S
0.0741
0.0945
0.0857
0.0894
0.0945
0.0741
0.0857
0.0816
0.0945
0.0857
0.0900
0.0900
0.0816
0.1042
0.0635
0.0522
0.0522
0.0525
0.0474
0.0810
0.0548
0.0735
0.0522
0.0522
0.0497
0.0522
0.0549
0.0522
0.487
0.347
0.686
0.756
0.653
0.286
0.442
0.299
0.721
0.622
0.622
0.537
0.538
0.346
0.0331
0.0331
0.0365
0.0247
0.0365
0.0300
0.0347
0.0331
0.0331
0.0300
0.0383
0.0422
0.0331
0.0444
1.57
1.41
1.59
1.26
1.46
1.27
1.40
1.37
1.62
1.44
1.70
1.79
1.64
1.85
In [4] several fits to the masses and mixing angles for the charged lepton and quark were
found using different quantum number assignments for the Higgs fields 5 , T , W , WS
with respect to changing them by the quantum number combination of S, where we also
vary S rather than fixing S = 1. The best fitting quantum number assignments from
formula within the range || 1, || 1, | | 1 and || 1 are reviewed in Table 2. The
notation 2 in the last column of Table 2 is misleading with respect to its normalisation in
as for as it is defined as what 2 would be if the uncertainty in the logarithms were unity:
2 =
m 2
ln
,
mexp
(13)
where m are the fitted charged lepton and quark masses and mixing angles and mexp are the
corresponding experimental values. The Yukawa matrices are calculated at the fundamental
scale which we take to be the Planck scale. We use the first order renormalization group
equations (RGEs) for the Standard Model to calculate the matrices at lower scales. Running
masses are calculated in terms of the Yukawa couplings at 1 GeV. A typical result of a fit
including averaging over the order of unity (i.e., O(1)) random numbers (see last of the
paragraph of Section 2) is represented in Table 3. 6
5 In the paper [4] the gauge group does not include the see-saw sector; therefore, in fitting the neutrino
masses and mixing angles using right-handed neutrinos, we have to add new gauge fields and Higgs fields which
spontaneously break the U (1)B L gauge group; i.e., we have to introduce the additional parameters, namely, #
and . They can also be taken in the range |#| 1 and || 1. See Section 3 for details.
6 Really Table 3 which is copied from Ref. [4] is said to have O(1) factors in distinction to another table in
that article in which only random phases were used.
415
Table 3
Typical fit including averaging over O(1) factors with
= 1, = 1, = 1 and = 1. All quark masses
are running masses at 1 GeV except the top quark mass
which is the pole mass
Fitted
Experimental
mu
md
me
mc
ms
m
Mt
mb
m
Vus
Vcb
Vub
JCP
3.1 MeV
6.6 MeV
0.76 MeV
1.29 GeV
390 MeV
85 MeV
179 GeV
7.8 GeV
1.29 GeV
0.21
0.023
0.0050
1.04 105
4 MeV
9 MeV
0.5 MeV
1.4 GeV
200 MeV
105 MeV
180 GeV
6.3 GeV
1.78 GeV
0.22
0.041
0.0035
23.5 105
1.46
6. Results and discussion

In Table 4 we have, in addition to presenting the measured numbers for the ratio of the
mass square differences and the mixing angles for neutrinos, given a series of predictions
for these quantities for various quantum number assignments of the Higgs fields in our
model. The first column in this table below the experimental summary part specify
using a capital Roman number I, II or III the quantum number assignment for the BL
charge breaking field B L . The dominant element of the right-handed Majorana mass
matrix, is for the well-fitting scenarios, always one from the third column (the R column).
The capital Roman number denotes then the row number of this dominant element. Which
element dominates depends of course on the B L quantum numbers (see Eqs. (8), (9)
and (10)). Furthermore, there is a small Roman number referring to the quantum number
combination chosen from Table 2 for the Higgs fields , T , W and WS which were
already used in the charged sector fitting. We only use combinations with good fitting
of the charged sector. We also have the freedom to choose the quantum numbers of the
fields only of relevance for the neutrinos and B L by shifting the Higgs fields S as
parameterised by # and . We have marked the best fit case III-ix with the parameter
choice = 0 and # = 1 with the black bullet in Table 4.
The VEV of B L is irrelevant when we only look for the ratio (m2 /(m2atm but not
for the absolute values of these mass square differences. Therefore, we do not have to fit
the VEV of B L . But the -field VEV is fit by hand for each of the quantum
416
Table 4
The numerical results of the three different cases, I, II and III combined with various charged sector
fits denoted by small Roman numbers. The best fit, case III-ix with = 0 and # = 1, is marked
with black bullet. See more detail in the text
Experimental
data
tan2
[sin2 2 ]
(m2
(m2atm
0.231.0
107
[0.61.0]
(0.332.5) 103
3
1.5 +1.5
0.7 10
[(1.310) 103 ]
0.171.0
3
9.4+14
6 10
[0.51.0]
0.381.0
+11
105
3.12.3
[0.81.0]
VO
SMA
LMA
LOW
tan2 atm
[sin2 2atm ]
0.4851.0
tan2 e3
[sin2 2e3 ]
0.026
[ 0.1]
[0.88.1.0]
tan2
(m2
(m2atm
tan2 atm
tan2 e3
I-xiii
I-xiii
= 0, # = 1
= 1, # = 1
7.8 104
7.8 104
6.7 105
7.8 105
1.00
1.00
8.0 104 0.018

7.8 104 0.038
II-ii
II-vi
= 1, # = 0
= 1, # = 0
3.8 103
4.7 103
1.0 102
5.8 103
0.93
0.96
8.8 104 0.016

1.0 103 0.018
II-ix
= 0, # = 1
1.1 103
1.0 102
0.96
2.1 104 0.026
II-v
II-ix
II-xi
= 1, # = 1
= 1, # = 1
= 1, # = 1
1.8 103
1.2 103
2.1 103
8.9 103
9.0 103
8.4 103
1.00
1.00
0.95
2.8 104 0.022

1.9 104 0.022
2.6 104 0.023
II-vi
= 1, # = 1
4.3 103
6.6 103
0.92
1.0 103 0.018
III-v
= 0, # = 0
III-viii = 0, # = 0
III-xi = 0, # = 0
1.1 103
8.8 103
1.4 103
8.7 103
8.0 103
0.93
0.99
0.99
6.6 104 0.032

2.7 104 0.063
7.3 104 0.030
III-v
= 0, # = 1
III-viii = 0, # = 1
III-xi = 0, # = 1
1.2 103
3.0 103
1.5 103
6.8 103
7.8 103
6.3 103
1.00
0.95
0.98
6.4 104 0.031

2.3 104 0.065
6.4 104 0.030
III-ii
III-vi
= 1, # = 1
= 1, # = 1
3.6 103
4.4 103
1.0 102
8.0 103
0.92
0.96
1.7 103 0.018

2.3 103 0.019
III-v
III-ix
= 1, # = 1
= 1, # = 1
1.5 103
1.0 103
6.3 103
6.2 103
0.97
0.95
6.2 104 0.028

4.2 104 0.028
III-ii
= 1, # = 0
3.6 103
1.0 102
0.92
2.4 103 0.019
III-v
III-viii
III-ix
III-xi
= 1,
= 1,
= 1,
= 1,
#=0
#=0
#=0
#=0
1.5 103
3.0 103
1.0 103
1.8 103
6.8 103
7.5 103
6.7 103
6.4 103
0.94
1.00
0.94
0.99
6.2 104
2.5 104
4.2 104
6.3 104
III-v
= 0, # = 1
1.2 103
7.5 103
0.98
6.4 104 0.031
2.9 103
0.028
0.130
0.028
0.028
417
Table 4 continued
tan2
(m2
(m2atm
tan2 atm
tan2 e3
IIIix
III-xi
= 0, # = 1
= 0, # = 1
8.3 104
1.5 103
5.8 103
7.8 103
0.97
0.97
4.3 104
6.4 104
0.035
0.030
III-iii
III-ix
= 1, # = 1
= 1, # = 1
1.5 103
1.3 103
7.2 103
7.5 103
0.97
0.97
4.5 104
2.8 104
0.025
0.039
III-v
III-ix
III-xi
= 1, # = 1
= 1, # = 1
= 1, # = 1
1.5 103
9.8 104
1.8 103
8.2 103
7.9 103
7.9 103
0.99
1.00
1.00
6.5 104
4.5 104
6.4 104
0.027
0.027
0.027
number combinations considered and this VEV relative to the fundamental units (i.e., the
suppression factor really) is presented in the last column.
In spite of the fact that our previous article used only the case where B L had the
quantum numbers of the combination Re R which we here call case I, we do not, upon
using the factorial correction fitted expectation value of , T , W , and WS (especially )
get as good fits to the neutrino mass matrices as before. In fact, the case I calculations,
using various well-fitting assignments in the charged sector, consistently, yield predictions
for the solar to atmospheric mass square difference ratio, (m2 /(m2atm that are too small
by about one order of magnitude compared to the MSW-SMA phenomenological value.
This is due to that we ignoring factorial corrections and order of unity details get
4 for this ratio in the case I quantum number assignments. With the slightly unfair
treatment of our previous article (using , etc., values deduced without use of factorials in
the charged sectors in spite of having factorials in the neutrino part of the calculation)
we managed to get an increase by a factor 6 for this ratio, but that is not sufficient with
the correctly fitted -VEV. We have therefore only presented a few examples with case I in
Table 4.
Both case II and case III have many well-fitting quantum number assignments associated
with them as is seen from Table 4. Generally, the tendency for both case II and case III is
to predict the mass square difference ratio (m2 /(m2atm a bit too high by half an order of
magnitude. Also the solar neutrino angle is predicted a bit too high.
The case III fits a little better than case II because case III predicts the solar mixing
angle a little better, i.e., to be a bit smaller. 7 In all cases our predictions of the solar mixing
angle is in the region of 1.5 103 . It fits very well for the MSW-SMA. It is, however,
not compatible with the other solution regions VO, LMA nor LOW. So it is necessary
unless our model is drastically changed that the MSW-SMA is upheld. Otherwise
our model disagrees significantly. Actually, this decrease in -value insures that the first
generation matrix elements proportional to or 2 have many fields and therefore get large
factorials that must be compensated for. In Table 2 we show some numerical results of
7 Equally good fits to the mass square difference ratio, (m2 /(m2 = 5.8 103 is obtained for solutions IIatm

vi and III-ix, but III-ix has the solar mixing angle closer to the global experimental fit [18], tan2
5.8 104 .
418
Fig. 3. The numerical results in the best case, III-ix with = 0 and # = 1, of the ratio of the solar
neutrino mass square difference to that for the atmospheric neutrino oscillation (thin solid line), and
the squared tangent of the solar neutrino mixing angle (cross-dotted line), the atmospheric neutrino
mixing angle (thick solid line) and the mixing angle e3 (dotted line) as a function of the VEV of
Higgs field. The results are obtained with 100 000 order of unity random number selections.
the most suggestive calculations: the combination ( = 1, = 1, = 1, = 1), i.e.,

case iv, giving the very best fit to the quarks and charged leptons supplemented with the
trivial case (i.e., = 0, # = 0) gave no modification of the already rather simple quantum
number combinations for and B L in Table 1. This fitting of only to neutrinos (case
III-iv with = 0 and # = 0) gives tan2 = 7.5 103 and (m2 /(m2atm = 1.3 102 .
We also considered some variations in and # in order to study other than the very best
fits of the charged sector. The main point is that the different choices do not yield very
different fits to the neutrino oscillations and that there are very many good fits of which a
selection the best mainly has been put in the table.
In the examples with case II and also case III, we tend to get close to the MSW-SMA
solution although the solar neutrino mass square difference (m2 is predicted a bit to the
high side of experiments, i.e., (m2 /(m2atm 6 103 compared to the experimental
(m2 /(m2atm 2 103 . Using the best fitting choice of which S-quantum numbers are
added to the Higgs fields ( = 0, = 1, = 1, = 0, = 0, # = 1) gave results just of
that character (see also Fig. 3):
(m2
3
= 5.8 +30
5 10 ,
(14)
4
tan2 = 8.3 +21
6 10 .
(15)
(m2atm
The deviation from the experimental data of our predictions should be considered within
uncertainly limits. We interpret our being uncertain by order unity as meaning a relative
uncertainty 64% for masses and mixing angles. Note that for the mass square difference
419
ratio (m2 /(m2atm , which involves four mass factors, the uncertainly is larger by a factor
2 2 3 counted in logarithmic uncertainty meaning an uncertainty of the order +511

84 %.
But for the squared mixing angles the uncertainty is only 2 64% meaning +260
72 %.
With these uncertainties the predictions are in good agreement with experiment.
6.1. Crude calculation of the mass square difference
The results of the see-saw mechanism is that the light neutrino masses are quadratic in
the Dirac masses and inversely proportional to the heavy right-handed Majorana masses:

T
Meff MD MR1 MD .
(16)
The mass square difference ratio, (m2 /(m2atm , from Eqs. (7),
(10) and (16) can be
approximately calculated (estimating by rules like a a + b b
a 2 + b2 , where a and
b are order of unity random numbers) in the best case, i.e., III-ix with = 0 and # = 1,
as follows:
WS 2 S 2 W 2 T 2 2 B L 2
case III-ix
det MR
=0, #=1
2
635 ST
60 14
S 3 T 3 60 154
S 3 T 3
635 S 2 T
23 T
2
15 T
2
2 6
6
6 70 S

Meff
33/2
2 6
3 21/2

part of MD
2 6 2
246
12 5 2

part of MR1
6 35 ST 2
3
3
60
14 S3 T 3
60 154 S T

3
21/2
2
12
5 2 2
3 165

2
6 35
S T
2 3T
2 15 T

2
6 70
S
2 6
6

part of (MD )T
WS
2 S 2 W 2 T 2 2 2
det MR
B L
0
0
0 1103 T 2
0 1391 T
0
1391 T ,
2206 2
(17)
where we have used the Higgs fields S, W , T , , and the fields S , W , T , ,

(with opposite quantum charges) equivalently because the non-supersymmetric model is
considered here. Moreover, we have used the numerical value S 2 1/2 and the relation
T to obtain the large atmospheric mixing angle. It is important that the dominant
matrix elements are not uncorrelated. The determinant of the dominant 2-by-2 subgroups
of Meff must be calculated by multiplying the 2 2 determinant of the relevant 2-by-2
sub-matrices of the three matrices, MD , MR1 and (MD )T :
420

det Meff case III-ix
= 0, # = 1
(2, 3) (2, 3)

= det MD
(2, 3) (1, 2) block

= 0, # = 1

T

det MD

= 0, # = 1
(2, 3) (2, 3)
So we get

(m2
(m2atm
(1, 2) (1, 2) block

= 0, # = 1
(1, 2) (2, 3) block

= 0, # = 1
S 8 W 4 T 6 6 4 WS 4 B L 4
(det MR )2
4
4
4
S W T 8 4 WS 4 B L 4
6 106
.
(det MR )2
3 106

2
(Tr Meff ) case III-ix

det MR1
det Meff
S4 1
.
2
8
(Tr Meff )2
(18)
(19)
(20)
Actually, the terms coming from the (1, 1)-component of the MR1 are somewhat dominant
due to the factorial corrections, so that the dominant 2-by-2 subgroup, the (2, 3) (2, 3)
of Meff , is to first approximation degenerate.
7. Conclusion and summary

We have considered a series of extensions so as to include of neutrinos of the earlier
Anti-GUT fit to the charged quarks and lepton masses by introduction of two more Higgs
field, and B L .
To fit the atmospheric neutrino oscillation experimental data it is necessary to fix the
VEV of the Higgs field, , to be in first approximation T in order to arrange
the atmospheric mixing angle to be of order unity. We should therefore not consider the
agreement of the atmospheric mixing angle atm as one of the significant predictions of
this model: we have, however, managed, at least, by such a fitting with , to obtain a large
mixing angle; atm , i.e., the atmospheric mixing angle is essentially an input parameter in
this model.
The success of this model should be rather judged from the following predictions:
(1) The solar mixing angle which comes out proportional to which in turn deviates
only by factorials and order of unity from the Cabibbo angle.
(2) The mass square difference ratio (m2 /(m2atm .
(3) The electron- and tau-neutrino mixing angle being small enough not to be in conflict
with the CHOOZ measurements (Ue3 = sin e3 0.16).
The first of these predictions is the major reason that our model in the present form (i.e.,
with the present quantum number assignments only varied within the limits considered in
this article) is only compatible with the MSW-SMA solar neutrino solution. The smallness
of the Cabibbo angle results in our solar mixing angle being so small that it would stress the
model drastically to seek to fit one of the series of large solar mixing angle fitting regions
such as LMA, LOW or VO.
421
Within the theoretical uncertainly inherent in our predictions being only of order unity
even if taken as specified in Ref. [4] with 64% being one standard deviation our model
fits the data perfectly provided that we allow the MSW-SMA solution as the (possibly)
valid one in spite of the fact that the day-night effect disfavour this solution by two standard
deviation: Extracting from the best fit, III-ix with = 0, # = 1, the result may be stated
as
1
tan c ,
8

2
(m
1
.
2
(matm 13
(21)
(22)
The collective fit of both the neutrinos and the charged sector with 6 fitted VEVs, one of
which S is still close to unity and with the WS determined from the weak interaction
Fermi constant, is so good order of magnitude-wise that it should perhaps be considered
perfect.
Since our mass eigenvalues are not tightly degenerate, it is not likely or rather
it is impossible that the renormalization group running should be very important as
corrections to our predictions. In fact we expect them to have almost no influence to our
order of unity accuracy for mass ratios of particles within the same group such as, say,
the (left) neutrinos. In fact, in the charged sector, the renormalization group mainly just
corrects the masses of the quarks by a factor 3 or so compared to the corresponding charged
leptons.
One could perhaps complain though that we have had too many relatively complicated
quantum number assignments which, although only discrete choices, may be too many
possibilities to make our fit convincing. The Anti-GUT model with its very many gauge
fields can also be complained about as being too complicated, but here it should be pointed
out that indeed it should rather gain its reason for being considered by being the largest
gauge group transforming the Standard Model fermions plus the see-saw neutrinos nontrivially among themselves without unifying irreducible representations of the Standard
Model.
7.1. Quantum number system of fitting results
Presumably the honest motivation for a fit as in the present work is that we have a major
part of the fitting going on by fitting the discrete representation choice for the Higgs fields.
The best fit has its Higgs field (abelian) quantum numbers listed in Table 5. This table, in
this sense, represents quantum numbers that are derived/inspired from experiment. One can
then imagine using this inspiration to look for regularities suggestive of the model beyond
the Standard Model. It is easy to estimate the VEV for B L needed in the fit by using,
say, the example of Section 6.1. Inserting
det MR 72 S 3 4 2 B L 3
into equation (19) yields
(23)
422
Table 5
All U (1) quantum charges of the best fit in extended Anti-GUT model. The VEV are presented in the
Planck unit. Non-abelian representations are given by a rule from the abelian ones, see after Eq. (3)
1
6
1
6
1
6
1
3
16
23
23
2
3
0.721
1
2
1
3
0.0945
2
3
13
1
3
0.0331
1
6
0.0522
0.0345
2
3
23
108
16
1
6
B L
UB L,3
SMG3
UB L,2
SMG2
1
2
16
13
13
16
WS
UB L,1
SMG1
13
VEV
1017
2

W 4 T 4 WS 4
(m2atm
Tr Meff
1.2 102 2
.
S B L 2
Then from Eq. (24) together with the atmospheric neutrino oscillation data we get
B L 6.7 1011 GeV.
(24)
(25)
We present here the order of magnitude of the right-handed Majorana neutrino masses
from the mass matrix, MR , of the best fitting case, III-ix with = 0 and # = 1 including
factorial corrections:
MR1 3 106 GeV,
MR2 4 108 GeV,
MR3 1.4 1011 GeV.
(26)
7.2. Theoretical lesson

It is highly remarkable that we obtain the rather good fit of the mass square difference
ratio, (m2 /(m2atm , in the two well-fitting cases II and III: were it not for the, in
principle, order unity factors coming out of this ratio, it should have simply been unity:
(m2 /(m2atm 1 since the Higgs VEVs drop out except for the Higgs field S. The
remaining in principle order unity ratio comes out of the way the various numbers are
combined in the calculations of inverse matrices and matrix products and the factorial
corrections which are due to diagram counting and the field S.
All of the numbers that are not of order unity in our model, namely the Higgs field VEV
(in Planck units) , , . . . , except S which has anyway S 1, drop out of this ratio.
This is the way that our model in cases II and III solves one of the at first sight almost
hopeless problems with data: how, in fitting the large atmospheric mixing angle, to get
the hierarchically looking large mass square difference ratio between the middle and the
heaviest eigenmass neutrinos when the highest mass has to mix strongly (of order unity)
with other eigenstates.
423
It is, however, indeed possible to interpret this hierarchical mass ratio as really due to a
hierarchical small VEV as case I does, where this ratio goes as 4 . However, case I, does
not fit well precisely because it makes this ratio too small too hierarchical.
7.3. Our qualitative predictions for future neutrino physics
We summarise here our predictions for future neutrino experiments assuming that this
model were/is right:
(1) It should be SMA-MSW that becomes the best fit (the day-night effect should be
tested by the SNO detector [19]).
(2) It is good that the atmospheric mixing angle is of order unity but it should not turn
out to be very close to tan2 atm 1, i.e., tan2 atm = 1 + O(lower order). There are
namely models of a similar spirit (but not our most promising fits) which behave
this way for example when a pair of off-diagonal elements of Meff dominate.
(3) KamLAND should not be able to see any neutrino oscillation because of SMAMSW.
(4) Since this model only contains three light neutrinos, there is no place for the LSND
effect [20].
It is important to confront our model with baryon number production in Big Bang which in
models like ours with pure Standard Model at the weak scale must get the baryon number
as a BL contribution from some other scale. This model is of the see-saw type with the
scale given by the neutrino oscillation scale. After crude estimation of the Baryogenesis
[21,22], we found that this model predicts the right range as Refs. [23,24] studied. Since
we do not have SUSY in our model preferably at least we do not have to worry about
gravitino problems [25]. We can say that they simply do not exist at all or only at such high
energy scales that they are totally irrelevant.
7.4. The efficiency of our model
The number of measured quantities, which are predictable with this model, quark
and charged lepton masses and their mixing angles (three CabibboKobayashiMaskawa
mixing matrix angles) containing Jarlskog triangle area, JCP and also two mass square
differences and the three mixing angles for the neutrino oscillation, Baryogenesis and
neutrinoless double beta decay [26], is 20.
Our model predicted successfully all these quantities using only six parameters 8 which
we fit VEVs of additional Higgs fields: the genuine number of the predicted parameters
is thus 14. However we have taken into the predictions two quantities namely, tan2 e3
and effective Majorana neutrino mass for which only experimental upper bounds exist.
We must though emphasise again that strictly speaking we have almost infinite input
parameters but since we only care for order of magnitude results, we should not count
couplings being order of unity.
8 The VEV of WeinbergSalam Higgs field we have not counted as a parameter because of its relation to the
Fermi constant.
424
The assumptions of quantum numbers are a bit arbitrary or fitted rather and a bit
complicated although still discrete. But otherwise it is a very reasonable and expected
type of assumptions that are used, e.g., the major assumption of all couplings at Planck
scale being of order unity is philosophically one of the easiest to get. It is (almost) just use
of usual dimensionality argumentation. That the only deviation from everything being of
order unity comes from the VEVs may find a bit of support in the well-known fact that in
super-conductors, VEVs tend often to be very small on the atomic physics scale expected
a priori.
Acknowledgements
We wish to thank W. Buchmller, C.D. Froggatt, S. Lola, M. Plmacher and T. Yanagida
for useful discussions. We thank D.L. Bennett for reading the manuscript. We wish to
thank the Theory Division of CERN for the hospitality extended to us during our visits
where most part of this work was done. H.B.N. thanks the EU commission for grants SCI0430-C (TSTS), CHRX-CT-94-0621, INTAS-RFBR-95-0567 and INTAS 93-3316(ext).
Y.T. thanks the ScandinaviaJapan Sasakawa foundation for grants No.00-22.
References
[1] Y. Totsuka, Talk at 8th International Conference on Supersymmetries in Physics, SUSY2K,
26 June 1 July 2000, CERN, Geneva, Switzerland.
[2] L. Wolfenstein, Phys. Rev. D 17 (1978) 2369;
S.P. Mikheyev, A.Yu Smirnov, Sov. J. Nucl. Phys. 42 (1986) 913.
[3] H.B. Nielsen, Y. Takanishi, Nucl. Phys. B 588 (2000) 281.
[4] C.D. Froggatt, H.B. Nielsen, D.J. Smith, in preparation.
[5] See, for example, C.D. Froggatt, H.B. Nielsen, D.J. Smith, Phys. Lett. B 385 (1996) 150;
D.L. Bennett, C.D. Froggatt, H.B. Nielsen, Proceeding paper of the APCTP-ICTP Joint
International Conference (AIJIC 97) on Recent Developments in Nonperturbative Quantum
Field Theory, Seoul, Korea, 2630 May 1997.
[6] C.D. Froggatt, G. Lowe, H.B. Nielsen, Phys. Lett. B 311 (1993) 163;
C.D. Froggatt, G. Lowe, H.B. Nielsen, Nucl. Phys. B 414 (1994) 579;
C.D. Froggatt, G. Lowe, H.B. Nielsen, Nucl. Phys. B 420 (1994) 3.
[7] S.L. Glashow, J. Iliopoulos, L. Maiani, Phys. Rev. D 2 (1970) 1285.
[8] M. Gell-Mann, P. Ramond, R. Slansky, in: P. van Nieuwenhuizen, D. Freedman (Eds.),
Supergravity, Proceedings of the Workshop at Stony Brook, NY, North-Holland, Amsterdam,
1979.
[9] T. Yanagida, in: O. Sawada, A. Sugamoto (Eds.), Proceedings of the Workshop on Unified
Theories and Baryon Number in the Universe, Tsukuba, Japan, 1979, KEK Report No. 79-18.
[10] R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
[11] Y. Fukuda et al., Super-Kamiokande Collaboration, Phys. Lett. B 467 (1999) 185;
H. Sobel, Talk at XIX International Conference on Neutrino Physics and Astrophysics, Sudbury,
Canada, June 2000;
T. Toshito, Talk at the XXX International Conference on High Energy Physics, ICHEP 2000,
July 27 August 2, 2000, Osaka, Japan.
425
[12] Y. Suzuki, Talk at XIX International Conference on Neutrino Physics and Astrophysics,
Sudbury, Canada, June 2000;
T. Takeuchi, Talk at the XXX International Conference on High Energy Physics, ICHEP 2000,
July 27 August 2, 2000, Osaka, Japan.
[13] B.T. Cleveland et al., Astrophys. J. 496 (1998) 505;
R. Davis, Prog. Part. Nucl. Phys. 32 (1994) 13;
K. Lande, Talk at XIX International Conference on Neutrino Physics and Astrophysics,
Sudbury, Canada, June 2000.
[14] J.N. Abdurashitov et al., SAGE Collaboration, Phys. Rev. C 60 (1999) 055801;
V. Gavrin, Talk at XIX International Conference on Neutrino Physics and Astrophysics,
[15] W. Hampel et al., GALLEX Collaboration, Phys. Lett. B 447 (1999) 127.
[16] E. Belloti, Talk at XIX International Conference on Neutrino Physics and Astrophysics,
[17] J.N. Bahcall, P.I. Krastev, A.Y. Smirnov, Phys. Lett. B 477 (2000) 401;
J.N. Bahcall, P.I. Krastev, A.Y. Smirnov, Phys. Rev. D 62 (2000) 093004.
[18] M.C. Gonzalez-Garcia, M. Maltoni, C. Pea-Garay, J.W. Valle, Phys. Rev. D 63 (2001) 033005.
[19] M. Maris, S.T. Petcov, Phys. Rev. D 62 (2000) 093006.
[20] C. Athanassopoulos et al., LSND Collaboration, Phys. Rev. C 54 (1996) 2685;
C. Athanassopoulos et al., LSND Collaboration, Phys. Rev. Lett. 77 (1996) 3082;
C. Athanassopoulos et al., LSND Collaboration, Phys. Rev. Lett. 81 (1998) 1774.
[21] H.B. Nielsen, Y. Takanishi, Contributed to The Scandinavian Neutrino Workshop, Uppsala,
Sweden, 810 February 2001, hep-ph/0101181.
[22] H.B. Nielsen, Y. Takanishi, hep-ph/0101307, Phys. Lett. B, in press.
[23] M. Fukugita, T. Yanagida, Phys. Lett. B 174 (1986) 45.
[24] W. Buchmller, M. Plmacher, Phys. Lett. B 389 (1996) 73;
M. Plmacher, hep-ph/9807557.
[25] T. Asaka, K. Hamaguchi, M. Kawasaki, T. Yanagida, Phys. Rev. D 61 (2000) 083512.
[26] H.B. Nielsen, Y. Takanishi, in preparation.

Duality equivalence between self-dual and

topologically massive non-abelian models
A. Ilha, C. Wotzasek
Instituto de Fsica, Universidade Federal do Rio de Janeiro, 21945, Rio de Janeiro, Brazil
Received 8 November 2000; accepted 7 March 2001
Abstract
The non-abelian version of the self-dual model proposed by Townsend, Pilch and van Nieuwenhuizen presents some well known difficulties not found in the abelian case, such as well defined
duality operation leading to self-duality and dual equivalence with the YangMillsChernSimons
theory, for the full range of the coupling constant. These questions are tackled in this work using
a distinct gauge lifting technique that is alternative to the master action approach first proposed by
Deser and Jackiw. The master action, which has proved useful in exhibiting the dual equivalence between theories in diverse dimensions, runs into trouble when dealing with the non-abelian case apart
from the weak coupling regime. This new dualization technique on the other hand, is insensitive of
the non-abelian character of the theory and generalize straightforwardly from the abelian case. It also
leads, in a simple manner, to the dual equivalence for the case of couplings with dynamical fermionic
matter fields. As an application, we discuss the consequences of this dual equivalence in the context
of 3D non-abelian bosonization. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.10.-z; 11.10.Ef; 11.10.Kk; 11.15.-q
1. Introduction
The bosonization technique that expresses a theory of interacting fermions in terms of
free bosons provides a powerful non-perturbative tool for investigations in different areas of
theoretical physics with practical applications [1]. In two-dimensions these ideas have been
extended in an interpolating representation of bosons and fermions which clearly reveals
the dual equivalence character of these representations [2]. In spite of some difficulties, the
bosonization program has been extended to higher dimensions [3,4].
In particular the (2 + 1)-dimensional massive Thirring model (MTM) has been
bosonized to a free vectorial theory in the leading order of the inverse mass expansion.
Using the well known equivalence between the self-dual [5] and the topologically massive
E-mail address: clovis@if.ufrj.br (C. Wotzasek).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 0 6 - 7
A. Ilha, C. Wotzasek / Nuclear Physics B 604 (2001) 426440
427
models [6] proved by Deser and Jackiw [7] through the master action approach [8], a
correspondence has been established between the partition functions for the MTM and
the MaxwellChernSimons (MCS) theories [9]. The situation for the case of fermions
carrying non-abelian charges, however, is less understood due to a lack of equivalence
between these vectorial models, which has only been established for the weak coupling
regime [10]. As critically observed in [11] and [12], the use of master actions in this
situation is ineffective for establishing dual equivalences. In this paper we intend to fill
up this gap.
We propose a new technique to perform duality mappings for vectorial models in any
dimensions that is alternative to the master action approach. It is based on the traditional
idea of a local lifting of a global symmetry and may be realized by an iterative embedding
of Noether counter terms. This technique was originally explored in the context of the
soldering formalism [13,14] and is exploited here since it seems to be the most appropriate
technique for non-abelian generalization of the dual mapping concept.
Using the gauge embedding idea, we clearly show the dual equivalence between the
non-abelian self-dual and the YangMillsChernSimons models, extending the proof
proposed by Deser and Jackiw in the abelian domain. These results have consequences
for the bosonization identities from the massive Thirring model into the topologically
massive model, which are considered here, and also allows for the extension of the fusion
of the self-dual massive modes [14] for the non-abelian case [15]. We also discuss the
case where charged dynamical fermions are coupled to the vector bosons. For fermions
carrying a global U(1) charge we reproduce the result of [16] but the result for the nonabelian generalization is new.
The technique of local gauge lifting is developed in Section 2 through specific examples.
In Section 3 we show how the gauge embedding idea solves the problem of non-abelian
dual equivalence. For completeness we first discuss the abelian case showing how the well
known results are easily reproduced. The case of dual equivalence between the SD and the
MCS when dynamical fermion fields are coupled to the gauge fields is also considered,
both in the abelian and in the non-abelian cases. The remaining sections are dedicated to
explore this result in the non-abelian bosonization program and to present our conclusions.
2. Noether gauge embedding method

Recently there has been a number of papers examining the existence of gauge invariances in systems with second class constraints [17]. Basically this involves disclosing,
using the language of constraints, hidden gauge symmetries in such systems. This situation may be of usefulness since one can consider the non-invariant model as the gauge
fixed version of a gauge theory. By doing so it has sometimes been possible to obtain
a deeper and more illuminating interpretation of these systems. Such hidden symmetries
may be revealed by a direct construction of a gauge invariant theory out of a non-invariant
one [18]. The former reverts to the latter under certain gauge fixing conditions. The associate gauge theory is therefore to be considered as the embedded one. The advantage in
428
having a gauge theory lies in the fact that the underlying gauge invariant theory allows us
to establish a chain of equivalence among different models by choosing different gauge
fixing conditions.
In this section we shall review a different technique to achieve this goal: the iterative
Noether gauging procedure. For pedagogical reasons, we develop simple illustrations
making use of scalar field theories living in a Minkowski spacetime of dimension two.
This will allow us to discuss some subtle technical details of this method, regarding the
connection between the implementing symmetries and the Noether currents, which are
necessary for its application in the (2 + 1)-dimensional self-dual model.
The important point to stress in this review of the iterative Noether procedure is its
ability to implement specific symmetries leading to distinct models. To avoid unnecessary
complications let us consider the case of a free two dimensional scalar field theory,

1
(0)
S =
(1)
d 2 x ,
2
and choose to gauge either the axial shift,
(i)
+ ,
(2)
or the conformal symmetry,

(ii)
+ ,
(3)
where = 1 (0 1 ) and is a global parameter. In the first case it is simple to identify

2
the Noether current as
J = ,
which comes from the variation of the scalar field action

S (0) = d 2 x J ,
(4)
(5)
which is non-vanishing if is lifted to its local version. To compensate for this nonvanishing result we introduce a counter-term together with an ancillary gauge field B
(also called as compensatory field) as,

(1)
(0)
S = S d 2 x J B ,
(6)
such that its variation reads

(1)
S = d 2 x B J ,
(7)
which is achieved if B transforms as a vector field simultaneously with (2). Introducing

an extra counter-term as,

1
S (2) = S (1) +
(8)
d 2 x B B ,
2
that automatically compensates for (7), will render S (2) gauge invariant, ending the
iterative chain.
429
To implement, on the other hand, the chiral diffeomorphism (3) at a local level we
proceed analogously but starting with the scalar action written in terms of light-cone
variables as,

S (0) = d 2 x + .
(9)
We then obtain,

S (0) = d 2 x T ()+ ,
(10)
after discarding some total derivatives, with

T () = ( )2 ,
(11)
being a component of the energy-momentum tensor. The first iterated action, that
compensates for this non-null result requires a new gauge field as,

(1)
(0)
S = S d 2 x T ().
(12)
After some algebra one finds

S (1) = d 2 x + 2 + ( ) T (),
(13)
which vanishes identically if is chosen to transform as a component of the metric tensor

under the chiral diffeomorphism (3),
= + + .
(14)
We mention that the gauge invariant action (12) has been first proposed by Siegel [19] to
represent the dynamics of two-dimensional self-dual scalar fields.
It is interesting to observe that, in the canonical approach, the Noether currents (4) and
(11) become the generators of the transformations (2) and (3), respectively. In Diracs
nomenclature of constrained systems [20], these currents are first-class gauge generator
constraints derived from the gauge converted model. It is important to mention that the
original models may be reobtained from these gauge invariant ones by choosing the so
called unitary gauge but different gauge choices are now at our disposal.
3. Dual equivalence of SD and MCS theories

In this section we discuss the application of the gauge invariant embedding to show
the dual equivalence between the SD model with the MCS theory both abelian and nonabelian including the coupling with charged dynamical matter fields. As mentioned in the
introduction this has the advantage of possessing a straightforward extension to the nonabelian case for all values of the coupling constant.
Let us recall that the essential properties manifested by the three dimensional self-dual
theory such as parity breaking and anomalous spin, are basically connected to the presence
430
of the topological and gauge invariant ChernSimons term. The abelian self-dual model for
vector fields was first introduced by Townsend, Pilch and van Nieuwenhuizen [5] through
the following action,

1
3

S [f ] = d x
(15)
f f f f ,
2m
2
where the signature of the topological terms is dictated by = 1 and the mass parameter
m is inserted for dimensional reasons. Here the Lorentz indices are represented by greek
letters taking their usual values as , , = 0, 1, 2. The gauge invariant combination of a
ChernSimons term with a Maxwell action

1

(MCS)
3
= d x
f f
f f ,
S
(16)
2m
4m2
is the topologically massive theory, which is known to be equivalent [7] to the self-dual
model (15). f is the usual Maxwell field strength,
f f f .
(17)
The non-abelian version of the vector self-dual model (15), which is our main concern
in this work, is given by

1

2
S = d 3 x tr F F +
(18)

F F F F F ,
2
4m
3
where F = Fa t a , is a vector field taking values in the Lie algebra of a symmetry group
G and t a are the matrices representing the underlying non-abelian gauge group with a =
1, . . . , dim G. The field-strength tensor is given by
F = F F + [F , F ],
(19)
and the covariant derivative is D = + [F , ].

Using the master action approach, the action (18) has been shown to be equivalent to the
gauge invariant YangMillsChernSimons (YMCS) theory

1

2
(YMCS)
3

S
(20)
F F F F F ,
= d x tr
F F +
4m
3
4m2
only in the weak coupling limit g 0 so that the YangMills term effectively vanishes. 1
To study the dual equivalence of (18) and (20) for all coupling regimes and the
consequences over the bosonization program is main contribution of this work.
Next we analyze the dualization procedure in the massive spin one self-dual theories
using the Noether gauging procedure. To begin with, it is useful to clarify the meaning
of the self duality inherent in the action (15). The equation of motion in the absence of
sources is given by,
f = f ,
(21)
m
1 Here we are using the bosonization nomenclature that relates the Thirring model coupling constant g 2 with
the inverse mass of the vector model; see discussion after Eq. (62).
431
from which the following relations may be easily verified,

f = 0,

+ m2 f = 0.
(22)
From (21) and (22) we see that there is only one massive excitation whose value is m.
A field dual to f is defined as,
1
f .
m
Repeating the dual operation, we find,

f =
(23)
1
f = f ,
(24)
m
obtained by exploiting (22), thereby validating the definition of the dual field. Combining
these results with (21), we conclude that,
( f ) =

f = f .
(25)
Hence, depending on the signature of , the theory will correspond to a self-dual or an anti
self-dual model.
To prove the exact equivalence between the self-dual model and the MaxwellChern
Simons theory, we start with the zeroth-iterated action (15) which is non-invariant by gauge
transformations of the basic vector field f . To construct from it an abelian gauge model,
we have to consider the gauging of the following symmetry,
f = ,
(26)
where is an infinitesimal local parameter. Under such transformations, the action (15)
change as,

S = d 3 x J (f ) ,
(27)
where the Noether currents are defined by,
(28)
f ,
2m
and f is the usual Maxwell field strength (17). It is worthwhile to mention that any other
variation of the fields (like f = ) is inappropriate because changes in the two terms of
the Lagrangians cannot be combined to give a single structure like (28). We now introduce
an ancillary vector field B coupled with the Noether currents and transforming in the
usual way,
J (f ) f +
B = .
(29)
Note that in the abelian case, f = B . Then it is easy to see that the modified action,

S (1) = S d 3 x J B ,
(30)
432
transform as
S (1) =
d 3 x J B .
(31)
The final modification consists in adding a term to ensure gauge invariance. This is
achieved by,

1
S (2) = S d 3 x J B + B B ,
(32)
2
which is invariant under the combined gauge transformations (26) and (29). The gauging of
the U(1) symmetry is complete. To return to a description in terms of the original variables,
the auxiliary vector field is eliminated from (32) by using the equation of motion,
B = J .
(33)
Note that taking variations on both sides of this equation and using the gauge invariance
of the ChernSimons form we obtain consistency with the condition f = B . It is now
crucial to note that, by using the explicit structures for the currents, the above action (32)
forms a gauge invariant combination expressed by the action (16) which is the Maxwell
ChernSimons theory. Our goal has been achieved. The iterative Noether dualization
procedure has precisely incorporated the abelian gauge symmetries in the self-dual model
to yield the gauge invariant MaxwellChernSimons theory.
The free case considered above can also be extended to couplings with external, fieldindependent sources. To illustrate this point, we consider a coupling between dynamical
U(1) charged fermions and self-dual vector bosons [16]. In the abelian case, this model is
written as

M) ,
S[f, ] = S [f ] + d 3 x ef J + (i/
(34)
and M is the fermionic mass. S is the self-dual action (15). The
where J =
Noether current associated with this action is
K = f + f eJ .
(35)
m
By introducing an ancillary field B , we can construct a gauge invariant theory which reads

1
SB = S[f, ] d 3 x K B + B B .
(36)
2
After the elimination of B through its equations of motion, we get our final theory,

(MCS)
+ d 3 x e2 J J + ef G + LD ,
S =S
(37)
where S (MCS) is the MaxwellChernSimons action (16) and LD is the free Dirac
Lagrangian. Here G = m1 J is the curl of the fermionic current introduced in [16].
Note the presence of Thirring like interaction e2 J2 and the change of the minimal coupling
into a magnetic coupling, i.e.,
f J f G .
(38)
433
These terms have appeared naturally as a necessary consequences of the dualization

procedure, unlike [16] where these terms were recognized as necessary in order to keep the
fermionic sectors of the two models identical. Extension of this result to bosonic sources
(that are field-dependent) should pose no further requirements under our approach [21] but
were shown to be troublesome under the master action procedure [16].
After this digression over the abelian structure of the theory, we are now in position to
study the non-abelian version of the vector self-dual model whose dynamics is given by
the action (18). To this end we discuss first in which sense this model possess the selfdual property. Following the same reasoning as in the abelian case, we define the duality
operation as,

( + F ) F ,
F
(39)
m
where the operator inside the square brackets in the right-hand side acts on the basic field
F defining F as the dual of F . Repeating this operation, and using the equations of
motion obtained by varying (18) with respect to F
F =

F ,
2m
(40)
we find,
( F ) = F ,

(41)
thus justifying our terminology and showing the self-dual character of this model.
Likewise the abelian case, to proceed with the dualization, we begin with the zerothiterated action (18) whose variation with respect to F is given by

S = d 3 x tr J F ,
(42)
with the Noether currents being defined as,
J = F +
F .
2m
(43)
Our goal is to obtain a non-abelian gauge invariant theory from the above non-invariant
self-dual model. To this end we define the first-iterated action by a coupling between the
currents J and an auxiliary field B ,

S (1) = S d 3 x tr J B ,
(44)
and whose variation is

1
(1)
3
S = d x tr J J J B J B ,
2
(45)
where we have used the following transformation rule for the gauging field,
F = B J .
(46)
434
This prompt us to define the following second iterated action,

1
S (2) = S + d 3 x tr J J F B
2

2
1

= d 3 x tr
F F F F F + B B ,
4m
3
2
(47)
which is gauge invariant after noticing that the transformation rule (46) fixes the B field
as
F .
B =
(48)
2m
Thanks to the structure of the current (43), the action (47) can finally be put in a more
familiar presentation as the gauge invariant theory,
S (2) S YMCS ,
(49)
the YangMillsChernSimons theory defined in (20). This proves that just as in the
abelian case, the non-abelian self-dual action defined in (18) is physically equivalent to (20)
for all regimes of the coupling constant. In the process we have also shown the duality
transformation that correctly defines the inherent self-duality property of action (18).
To complete the analysis we discuss the equivalence for the case where dynamical
fermions carrying non-abelian charge are coupled to the self-dual bosonic vector fields.
Following the same steps as before we found that the effective action obtained from the
second-iterated action, after elimination of the auxiliary field is,

e
e2 2
3

F J + J .
Seff = SMCS + d x tr LD
(50)
2m
2
The result is qualitatively similar to its abelian counterpart. As in (37) we notice the
appearance of a Thirring like currentcurrent interaction in the fermionic sector. However,
the non-minimal magnetic like interaction between the fermionic and bosonic sectors gets
modified in this instance due to the presence of a nonlinear piece in the field strength tensor.
In summary, a general method has been developed that establishes dual equivalence
between self-dual and topologically massive theories based on the idea of gauge
embedding over second-class constrained systems. The equivalence has been established
using an adaptation of the iterative Noether procedure both for abelian and non-abelian
self-dual models, including the cases with coupling to dynamical charged fermions.
4. Application to non-abelian bosonization

In this section we study the consequences of the non-abelian dual equivalence to
bosonization, the mapping of a quantum field theory of interacting fermions onto an
equivalent theory of interacting bosons, in 2 + 1 dimensions. Bosonization was developed
in the context of the two-dimensional scalar field theory and has been one of the main
tools available to investigate the non-perturbative behavior of some interactive field
435
theories [1]. For some time this concept was thought to be an exclusive property of twodimensional spacetimes where spin is absent and one cannot distinguish between bosons
and fermions. It was only recently that this powerful technique were extended to higher
dimensional spacetimes [9,2426]. The bosonization mapping in D = 3, first discussed
by Polyakov [27], shows that this is a relevant issue in the context of transmutation of
spin and statistics in three dimensions. The equivalence of the three-dimensional effective
electromagnetic action of the CP 1 model with a charged massive fermion to lowest order
in inverse (fermion) mass has been proposed by Deser and Redlich [28]. Using their results
bosonization was extended to three-dimensions in the 1/m expansion [9]. These endeavor
has led to promising results in diverse areas such as, for instance, the understanding of the
universal behavior of the Hall conductance in interactive fermion systems [29].
For higher dimensions, due to the absence of an operator mapping a la Mandelstan,
the situation is more complicate and even the bosonization identities extracted from
these procedures relating the fermionic current with the bosonic topological current is a
consequence of a non-trivial current algebra. Moreover, contrary to the two dimensional
case, in dimensions higher than two there are no exact results with the exception of the
current mapping [22,23]. Besides, while the two-dimensional fermionic determinant can
be exactly computed, here it is neither exact nor complete, having a non-local structure.
However, for the large mass limit in the one-loop of perturbative evaluation, a local
expression materializes. This procedure, is in a sense, opposite to what is done in 1 + 1
dimensions where bosonization is a set of operator identities valid at length scales short
compared with the Compton wavelength of the fermions while in D = 3 only the long
distance regime is considered.
In this section we review how the low energy sector of a theory of massive selfinteracting, G-charged fermions, the massive Thirring Model in 2 + 1 dimensions, can
be bosonized into a gauge theory, the YangMillsChernSimons gauge theory thanks to
the results of the preceding section.
4.1. The non-abelian mapping
In the sequence we investigate the problem of identifying a bosonic equivalent of a three
dimensional theory of self-interacting fermions with symmetry group G and show how it
is possible to bosonize the low-energy regime of the theory. We follow the same strategy as
in reference [9] but follow the notation of [10] that is slightly different than [9]. We seek a
bosonic theory which reproduces correctly the low-energy regime of the massive fermionic
theory. To begin with we define the G-current,
j a = i tija j ,
(51)
where i are N two-component Dirac spinors in the fundamental representation of G,

i, j = 1, . . . , N and a = 1, . . . , dim G. Here t a and f abc are the generators and the structure
constants of the symmetry group G, respectively and j a is a G-current.
The (Euclidean) fermionic partition function for the three-dimensional massive Thirring
model is,
436

ZTh =

g 2 a a 3
i
i
j j d x ,
D D exp
(/
+ m)
2N
(52)
with the coupling constant g 2 having dimensions of inverse mass. Next we eliminate the
quartic interaction through a Legendre transformation
2

g
1
exp
(53)
a
a
+
j
a
d 3 x j a ja = Da exp d 3 x tr
2
2g 2
up to a multiplicative normalization constant, where we have introduced a vector field a
taking values in the Lie algebra of G. After integration of the fermionic degrees of freedom,
the partition function reduces to,

1
ZTh = det(i/
(54)
+ m + a/ ) Da exp 2 d 3 x tr a a .
2g
The determinant of the Dirac operator is an unbounded operator and requires regularization. For D = 2 this determinant can be computed exactly, both for abelian and non-abelian
symmetries. Based on general grounds only, one may say that this determinant consists of
a ChernSimons action standing as the leading term plus an infinite series of terms de A , including those terms that are
pending on the dual of the vector field, F
. For the D = 3 the actual computation of this determinon-local and non-quadratic in F
nant will give parity breaking and parity conserving terms that are computed in powers of
the inverse mass,

SCS [a] + IPC [a] + O 2 /m2 .

ln det(i/
+ m + a/ ) =
(55)
16
Here SCS given by

2
SCS [a] = d 3 x i tr f a a a a ,
(56)
3
is the non-abelian ChernSimons action and the parity conserving contributions, in firstorder, is the YangMills action

1
tr d 3 x f f ,
IPC [a] =
(57)
24m
where
f = a a + [a , a ].
(58)
In the low energy regime, only the ChernSimons action survives yielding a closed
expression for the determinant as
lim det(i/
+ m + a/ ) =
SCS [a].
16
Using this result we can write ZTh in the form

ZTh = Da exp SSD [a] ,
(59)
(60)
where SSD is the non-abelian version of the self-dual action introduced in [5],

1
SCS [a].
SSD [a] = 2 d 3 x tr(a a )
2g
16
437
(61)
Therefore, to leading order in 1/m we have established the identification ZTh ZSD .
Now, recalling that the model with dynamics defined by the non-abelian self-dual action
is equivalent to the YangMillsChernSimons theory, we use this connection to establish
the equivalence of the non-abelian massive Thirring model and the YMCS theory as
ZTh ZYMCS .
(62)
It is interesting to observe that the Thirring coupling constant g 2 /N in the fermionic model
is mapped into the inverse mass spin 1 massive excitation, m = /g 2 .
Now comes an important observation. Unlike the master approach, our result is valid for
all values of the coupling constant. The proof, based on the use of an interpolating Action
SI , is seen to run into trouble in the non-abelian case. That the non-abelian extension of
this kind of equivalences is more involved was already recognized in [7] and [10], and
shown that the non-abelian self-dual action is not equivalent to a YangMillsChern
Simons theory (the natural extension of the abelian MCS theory) but to a model where
the YangMills term vanishes in the limit g 2 0 [10].
4.2. Current identities
To infer the bosonization identities for the currents which derive from the equivalence
found in the last section, we add a source for the Thirring current leading to the following
functional generator,

+ m + a/ + b/) + 1 tr a a
ZTh [b] = D DDa exp d 3 x (i/
2g 2

1
1
= exp 2 tr d 3 x b b
Da exp SSD [a] + 2 tr d 3 x b a
2g
g
(63)
after shifting a a b .
In order to connect this with the YangMillsChernSimons system we repeat the steps
of the last section to obtain,
ZTh [b ] ZMCS [b ].
(64)
We have therefore established, to order 1/m, the connection between the Thirring and
self-dual models in the non-abelian context, now in the presence of sources. This is, in its
most general form, the result we were after. It provides a complete low-energy bosonization
prescription, valid for any g 2 , of the matrix elements of the fermionic current. From (64) we
see, from simple differentiation w.r.t. the source, that the bosonization rule for the fermion
current, to leading order in 1/m, reads
a
.
ja i f
(65)
8
438
Eq. (65) gives a natural non-abelian extension of the abelian bosonization rule obtained
in [9], extended to the weak coupling non-abelian case in [10]. In contrast to these results,
the bosonization identity (65), was shown to be an exact recipe in [22,23].
In the abelian case one can interpret the (2 + 1)-dimensional bosonization formula as
(i/ ) while in this case

the analog of the (1 + 1)-dimensional result
it should be considered as the analog of the WessZuminoWitten currents. Notice that
as in the abelian case the bosonized expression for the fermion current is topologically
conserved.
We thus see that the non-abelian bosonization of free, G-charged massive fermions in
2 + 1 dimensions leads to the non-abelian ChernSimons theory, with the fermionic current
being mapped to the dual of the gauge field strength. This result holds only for length
scales large compared with the Compton wavelength of the fermion, since our results were
obtained for large fermion mass. It is important to notice that the limit g 2 0 used in
earlier approaches corresponding to free fermions (but not to an abelian gauge theory) was
not taken here at any stage. This is important since YangMills coupling is proportional
to g 2 , which is why we are left with a YangMillsChernSimons action and not the pure
ChernSimons theory of [10].
5. Conclusions
The rationale of different phenomena in planar physics have greatly benefitted from the
use of (2 + 1)-dimensional field theories including the parity breaking ChernSimons term.
In this scenario it is important to establish connections among different models so that a
unifying picture emerges. In this context we have shown, in earlier work, that the soldering
formalism has established a direct link between self-dual models of opposite helicities
with the Proca model [14]. Other instances includes the recent extension of the functional
bosonization program interpolating from fermions to bosons in a coherent picture [9,10].
In the context of the first it has been argued that the soldering formalism is equivalent
to canonical transformation albeit in the Lagrangian side while for the later the mentioned
mapping between SD and MCS models has been used to establish a formal equivalence
between the partition functions of the abelian version of MTM and a theory of interacting
bosons. The non-abelian extension of this analysis, for the full range of the coupling
constant, has been the main concern of the present work since only partial results were
reported in the literature. Other directions have also been investigated, with new results,
that includes the proof of the self-duality property of (18) and the coupling with G-charged
dynamical matter fields.
Our analysis has shown how the gauge lifting approach sheds light on the question of
dual equivalence between SD and topologically massive theories with new results for the
non-abelian case. This discussion becomes the central issue when deriving bosonization
rules in D = 3, for fermions carrying non-abelian charges since, up to date, only
prescriptions based on the Master action of Ref. [7] were used, apart from [12] with
conclusions consistent with [10]. These derivations of the bosonization mappings suffered
439
from well known difficulties related to the dual equivalence, restricting the results to be
limited to weak coupling constant only. Therefore, regarding the non-abelian bosonization
in D = 3 dimensions, we believe that the method developed here, which is simpler and
better suited to deal with non-abelian symmetries, completes the program initiated in [10]
and confirms the exact identities found in [22,23]. This new approach has also been used
with dynamical fermionic fields leading naturally to the necessity of a Thirring like term
to establish the equivalence of the fermionic sectors in both sides. Such equivalence may
be extended to the scalar case [21].
The bosonization for D 4 poses no difficulties as long as the fermionic determinant
can be evaluated in some approximation and is expected to yield a gauge invariant piece.
This is of importance since the description of charged fermionic fields in terms of gauge
fields has brought new perspectives and a deeper insight on the non perturbative dynamics
of planar physics [30] that might be extended to higher dimensions.
References
[1] S. Coleman, Phys. Rev. D 11 (1975) 2088;
S. Mandelstan, Phys. Rev. D 11 (1975) 3026.
[2] P.H. Damgaard, H.B. Nielsen, R. Sollacher, Nucl. Phys. B 385 (1992) 227;
P.H. Damgaard, H.B. Nielsen, R. Sollacher, Nucl. Phys. B 414 (1994) 541;
P.H. Damgaard, H.B. Nielsen, R. Sollacher, Phys. Lett. B 296 (1992) 132;
P.H. Damgaard, H.B. Nielsen, R. Sollacher, Phys. Lett. B 332 (1994) 131;
P.H. Damgaard, H.B. Nielsen, Nucl. Phys. B 433 (1995) 671.
[3] M. Lscher, Nucl. Phys. B 326 (1989) 557.
[4] E.C. Marino, Phys. Lett. B 263 (1991) 63.
[5] P.K. Townsend, K. Pilch, P. van Nieuwenhuizen, Phys. Lett. B 136 (1984) 38.
[6] R. Jackiw, S. Deser, Templeton, Ann. Phys. 140 (1982) 372.
[7] S. Deser, R. Jackiw, Phys. Lett. B 139 (1984) 2366.
[8] For a recent review in the use of the master action in proving duality in diverse areas
see: S.E. Hjelmeland, U. Lindstrm, UIO-PHYS-97-03, May 1997. e-Print Archive: hepth/9705122.
[9] E. Fradkin, F.A. Schaposnik, Phys. Lett. B 338 (1994) 253.
[10] N. Bralic, E. Fradkin, V. Manias, F.A. Schaposnik, Nucl. Phys. B 446 (1995) 144.
[11] A. Karlhed, U. Lindstrm, M. Rocek, P. van Nieuwenhuizen, Phys. Lett. B 186 (1987) 96.
[12] N. Banerjee, R. Banerjee, S. Ghosh, Nucl. Phys. B 527 (1998) 402.
[13] E.M.C. Abreu, R. Banerjee, C. Wotzasek, Nucl. Phys. B 509 (1998) 519.
[14] R. Banerjee, C. Wotzasek, Nucl. Phys. B 527 (1998) 402.
[15] A. Ilha, C. Wotzasek, Soldering of non-abelian self-dual modes, in preparation.
[16] M. Gomes, L.C. Malacarne, A.J. Silva, Phys. Lett. B 439 (1998) 137.
[17] A.S. Vytheeswaran, Hidden symmetries in second class constrained systems: are new fields
necessary?, Workshop on Current Topics in Quantum Field Theory, Calcutta, India, hepth/0007230.
[18] C. Neves, C. Wotzasek, J. Phys. A 33 (2000) 6447.
[19] W. Siegel, Nucl. Phys. B 238 (1984) 307.
[20] P.A.M. Dirac, Lectures Notes on Quantum Mechanics, Yeshiva University, New York, 1964.
[21] M.A. Anacleto, A. Ilha, J.R.S. Nascimento, R.F. Ribeiro, C. Wotzasek, Phys. Lett. B 504 (2001)
268.
440
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
J.C. Le Guillou, E. Moreno, C. Nnez, F.A. Schaposnik, Nucl. Phys. B 484 (1997) 682.
J.C. Le Guillou, E. Moreno, C. Nunez, F.A. Schaposnik, Phys. Lett. B 409 (1997) 257.
C. Burgess, C.A. Ltken, F. Quevedo, Phys. Lett. B 336 (1994) 18.
J. Frohlich, R. Gtschmann, P.A. Marchetti, J. Phys. A 28 (1995) 1169.
R. Banerjee, Phys. Lett. B 358 (1995) 297.
A.M. Polyakov, Mod. Phys. Lett. A 3 (1988) 325.
S. Deser, A.N. Redlich, Phys. Rev. Lett. 61 (1988) 1541.
D.G. Barci, L.E. Oxman, S.P. Sorella, Nucl. Phys. B 580 (2000) 721.
E. Abdalla, R. Banerjee, Phys. Rev. Lett. 80 (1998) 238.

Coarse-grained field wave function in stochastic

inflation
Mauricio Bellini
Instituto de Fsica y Matemticas, Universidad Michoacana de San Nicols de Hidalgo, AP: 2-82,
(58041) Morelia, Michoacn, Mexico
Received 10 August 2000; accepted 23 April 2001
Abstract
The wave function for the matter field fluctuations in the infrared sector is studied within the
framework of inflationary cosmology. These fluctuations are described by a coarse-grained field
which takes into account only the modes with wavelength much bigger than the size of the Hubble
horizon. The case of a power-law expanding universe is considered and it is found that the relevant
phase-space (cg , Pcg ) remains coherent under certain circumstances. In this case the classical
stochastic treatment for matter field fluctuations is not valid, however, for p > 4.6, the system loses
its coherence and a classical stochastic approximation is allowed. 2001 Published by Elsevier
Science B.V.
PACS: 98.80.Cq; 04.62.+v
1. Introduction
A standard mechanism for galaxy formation is the amplification of primordial fluctuations by the evolutionary dynamics of spacetime. The inflationary cosmology is based on
the dynamics of a quantum field undergoing a phase transition [1]. The exponential expansion of the scale parameter gives a scale-invariant spectrum naturally. This is one of the
many attractive features of the inflationary universe, particularly with regard to the galaxy
formation problem [2] and it arises from the fluctuations of the inflaton, the quantum field
which induces inflation. This field can be semiclassically expanded in terms of its expectation value plus other field, which describes the quantum fluctuations [3]. The quantum to
classical transition of quantum fluctuations has been studied in thoroughly [4].
The infrared matter field fluctuations are classical and can be described by a coarsegrained field which takes into account only wavelengths larger than the Hubble radius. The
E-mail address: mbellini@mdp.edu.ar (M. Bellini).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 9 6 - 1
442
M. Bellini / Nuclear Physics B 604 (2001) 441451
dynamics of this coarse-grained field is described by a second order stochastic equation,

which can be treated using the FokkerPlanck formalism. This issue has been the subject
of intense work during the last two decades [58]. A different approach is the quantum
mechanical treatment of the coarse-grained field [9], where the fluctuations are described
by means of a time dependent quadratic potential with a linear external stochastic force.
Even though an isolated system described by the Schrdinger equation cannot lose its
coherence, the coase-grained field may evolve from a pure to a mixed state. One way to
realize coarse-graining is to let the system interact with an environment. This consists of all
the fields whose evolution we are not interested in. The state of the system is obtained by
tracing over all possible states of the environment. Even if the state describing the system
plus environment is pure, the state of the system alone will in general be mixed. This
is the case of the matter field fluctuations in inflationary cosmology, where the system
is given by the super Hubble modes (infrared sector) while the environment is given
by the short-wavelength modes (ultraviolet sector). For a supercooled expansion of the
universe the environment cannot be considered as a true environment because it is not
thermalized. A true environment appears in warm and fresh inflationary scenarios [10,11]
where the inflaton field interacts with other particles of a thermal bath. In these scenarios
the environment is represented by the thermal bath and the particles in it.
In this work, we aim at studying the phase-space decoherence of the wave function that
describes super Hubble matter field fluctuations during inflation. Decoherence of the phasespace is different to decoherence of the coarse-grained field. While in the latter coarsegrained field decoherence consists in the interchange of degrees of freedom between the
infrared and ultraviolet sectors, in the former there is interference between the fluctuations
of variables that describe the phase-space of the quantum state. This issue has been treated
before using either the Wigner function [12] or the Schrdinger formalism. It is well
known that the evolution of the redefined coase-grained field is described by a secondorder stochastic equation. This is considered in [6] for supercooled inflation and also in
[13] for warm inflation. The effective Hamiltonian related to this stochastic equation can
be expressed in such away that the Schrdinger equation for the system can be written. The
wave function that describes this system is (cg , t), where cg denotes the coordinate and
the variable Pcg cg characterizes the momentum of the phase-space. The interference
2 and P 2 , which arises from the coupling
between the the squared fluctuations cg
cg
between the variables cg and Pcg is the main feature to be analyzed in this paper. It has
been shown that in a de Sitter expansion of the universe the phase-space of the quantum
state remains pure [9]. In this work the decoherence in the phase-space for a power-law
expanding universe in a globally FriedmannRobertsonWalker (FRW) metric is studied.
The paper is organized as follows: in Section 2, a revision of the inflationary formalism
is done. In Section 3, it is introduced the general quantum mechanical formalism for the
coarse-grained field, which describe the redefined matter field fluctuations on super Hubble
scales. In Section 4, the wave function for the particular case of a power-law expansion of
the universe is studied. Finally, in Section 5, some final remarks are given.
443
2. Review of the inflationary formalism

In a previous work [6] we justify the classical behaviour of the order parameter on the
basis of a semiclassical approach. The inflaton field Lagrangian is:

1
L(, , ) = g
g , , + V ()
2

1
1
= a 3 2 2 ()2 V () ,
(1)
2
2a
for a globally flat FRW metric, ds 2 = dt 2 + a(t)2 d r
2 . From here it follows the equation
for the scalar field operator
1 2
+ 3H + V () = 0,
a2
and the Friedmann equation, written in terms of H = a/a,
is

4
1
2
+ 2V () ,
H2 =
2 + 2
3Mp2
a

(2)
(3)
where the overdot represents the time derivative and V () = dV /d. We decompose
the scalar field as its mean value plus the quantum spatially inhomogeneous fluctuations,
(
x , t) = cl (t) + (
x , t) with = 0, up to linear terms in . Hence, the equations of

motion reduce to a set of two classical equations which give the evolution of the field cl
and the Hubble parameter. For simplicity, we consider the Hubble parameter as classical:
H H (c ) = a/a.
To be consistent with the FRW metrics, it is assumed that cl is a

homogeneous field, and thus we have:
cl + 3H cl + V (cl ) = 0,
8 2
H2 =
,
3Mp2
(4)
(5)
2 + V ( ) is the vacuum energy density. The

where V (cl ) = dV (cl )/dcl and = 12 cl
cl
equation for the quantum fluctuations is
1
2 2 + 3H + V (cl ) = 0.
(6)
a
In this last equation H (cl ) and V (cl ) are given by Eqs. (4) and (5), and they both are
functions of t.
The characteristic timescale for the inflaton field can be defined by d = cl / cl . The
Hubble timescale is given by:

d
2 2 cl 1/2
H cl
(7)
= d H =
=
.
H
3 Mp cl
cl
The number of e-folds in a given period of time is given by:
t
0 +t
Nc =
t0
cl

dt H = dcl
.
cl
0
(8)
444
When the scalar field potential is sufficiently flat the field rolls to the minimum of the
potential very slowly, and the following conditions are fulfilled

2 H 2
= 2
(9)
1,
K
H
2 H
1,
= 2
(10)
K H
with K = 8/Mp [14]. When the slow-roll conditions hold, the Eq. (6) can be
approximated by a first-order equation of motion for [15]. However, in this paper we
will consider the exact treatment for the dynamics of . At the end of inflation, when the
end .
scale factor stops accelerating, one obtains (cl ) = 1, which determines cl
2
Hence, we have cl V (cl )/3H and H 2 = 8 2 V (cl ), so that:
3Mp
8 2 cl V (cl )
,
Mp2 V (cl )
(11)
and
8 2
Nc = 2
Mp
cl
0

dcl
)
V (cl
).
V (cl
(12)
The solution to the horizon problem requires Nc 60, which in general implies that d >
H . When that scale crosses the Hubble radius, each length scale c is associated with a
unique value of cl denoted by (cl ). This is given by [14]:
exp[Nc (cl )] ao
,
(13)
H (cl ) ae
with a(cl ) = ae exp[Nc (cl )].
The study of
the quantum component is simplified if we redefine the field with the

3
map = e 2 dt H . The equation of motion for the field operator is:
c (cl ) =
k02
1 2
(14)
= 0,
a2
a2
where k02 = a 2 ( 94 H 2 + 32 H Vc ). Thus can be interpreted as a free scalar field with a

time dependent mass parameter. The field can be expanded in a set of modes k (t)ei k.
r :

1
r
, t =
(15)
d 3 k ak k (t)ei k.
r + h.c. ,
(2)3/2
where the annihilation and creation operators satisfy the usual commutation relations for
bosons:

ak , ak = (3) k
k
,
(16)

[ak , ak ] = ak , ak = 0,
(17)
and the modes are defined for the equation of motion
k + k2 k = 0,
(18)
445
with k2 = a 2 (k 2 k02 ). The function k02 (t) gives the threshold between an unstable
infrared sector (k 2 k02 ), which includes only wavelengths longer than the Hubble radius,
and a stable short wavelength sector (k 2 k02 ). We adopt the normalization condition
k k k k = i for the modes, such that the field operators and satisfy the canonical
commutation relations.
The coarse-grained field that describes the super Hubble spectrum contains only the
modes with wavenumber smaller than the Hubbles wavenumber k0 . This field can be
written as a Fourier expansion in terms of the modes

1
cg r
, t =
(19)
d 3 k (*k0 k) ak k (t)ei k.
r + h.c. ,
3/2
(2)
where denotes the Heaviside function and * 1 is a dimensionless constant.
The Eq. (15) for cg can be written as

cg 2 (t)cg + c x
, t = 0,
d
x , t) = *[ dt
(k0 ) + 2k0 ], with
where c (
=
d 3 k (*k0 k) ak ei k.
r k (t) + h.c. ,
3/2
(2)
=
d 3 k (*k0 k) ak ei k.
r k (t) + h.c. .
3/2
(2)
(20)
(21)
(22)
These noise arises from inflow of short-wavelength modes, produced by the cosmological
evolution of both, the horizon and the scale factor of the universe. Note that in Eq. (20)
we have neglected the term with a12 2 cg because in the infrared sector the following
constraint is fulfilled: k 2 /a 2 k02 /a 2 .
3. Wave function for the coarse-grained field
The effective Hamiltonian related to Eq. (20) is
1 2
2 2
Heff (cg , t) = Pcg
+ c cg ,
2
2 cg
(23)
where Pcg = cg and 2 (t) = k02 /a 2 . Hence, we can write the following Schrdinger
equation

1 2
2 2
(cg , t) + cg + c cg (cg , t),

i (cg , t) =
(24)
2
t
2 cg
2
where (cg , t) is the wave function of the system. The probability density to find the
universe with the configuration (cg , t), is
P (cg , t) = (cg , t) (cg , t),
(25)
where the asterisk denotes the complex conjugate. An elegant way to solve the Eq. (24) is
based on the use of explicitly time dependent invariants of motion for time dependent
446
quadratic potentials. Such quantities appeared in accelerator theory [16] but were first
analyzed by Lewis [17]. An invariant of the form
I (t) = A(t)cg + B(t)Pcg + C(t),
(26)
is proposed, where the time dependent coefficients A, B, C are determined by the condition

I (t) i I (t), Heff = 0.

t
This requires
(27)
A + 2 (t)B = 0,
A + B = 0,
(28)
(29)
C Bc = 0.
The Eqs. (28) and (29) can be combined to give (for
B 2 B = 0,
(30)
2
= k02 /a 2 )
(31)
and, since
t
C(t) =

dt B t c x
, t ,
(32)
the only equation that we need to solve is (31). From Eq. (29), the invariant now can be
written as
cg + C(t).
I (t) = B(t)Pcg B
(33)
In order for I (t) to be an Hermitian operator, only real solutions of Eq. (31) are admissible.
The eigenvalue equation for this operator is
I (t) (t) = (t),
(34)
where the eigenfunction [18]

i 1 2
1
(t) =
e B [ 2 Bcg +(C)cg ] ,
2B
satisfies the orthogonality condition

(t) (t) = .
(35)
(36)
Here, the eigenvalues are independent of the time. Since the action of I (t) does not
involve time derivatives, the function (t) is always arbitrary up to a time dependent
phase factor. Hence, we can write the eigenfuctions of I (t)
(t) = ei (t ) (t).
(37)
Taking the time derivative in Eq. (37), is obtained

I
(t) = ( I ) (t).
t
t
(38)
This can be written in terms of (t) as

(t) Heff i
(t) = (t) | ,
t
t
447
(39)
where (t) comes from solving
t
(t) =
dt
[ C(t )]2
.
2B 2 (t )
(40)
The solutions of the time dependent Schrdinger equation are

i 1 2
1
e B [ 2 Bcg +(C)cg + ] ,
(t) =
2B
(41)
where
(t) = B(t) (t).
(42)
The general solution (cg , t) can be represented as
(cg , t) = d (cg , t),
(43)
where the coefficients are

= (t0 ) (t0 ) ,
(44)
and t0 is the initial time. In our case t0 is the time when inflation starts, and corresponds
0 ) = 0, and C(t0 ) = 0. With
to 1 in Planckian units. We will take (t0 ) = 0, B(t0 ) = 1, B(t
these choice is obtained
1
=
(45)
dcg eicg (cg , t0 ),
2
which is the Fourier transform of the initial state. In the case of an squeezed harmonic
oscillator, (cg , t0 ) can be written as

1/4

(0) 2
cg cg
1
(0)
s (cg , t0 ) =
(46)
exp
+ iPcg cg .
2
2 2
(0)
(0)
Here, is the width of the Gaussian and Pcg
= cg (
x , t0 ). By setting z = Pcg
, the
wave function becomes

1/2
(0)
i 1 2
e B [ 2 Bcg +(Pcg C)cg ]

(cg , t) =
2 B
(0)
2 2
dz e z +i(cg cg /B)z+iz /B ,
(47)
where z is given by
z = B(t)[z2 R(t) + zU(t) S(t)],
(48)
448
being
t
R(t) =
t
U(t) =
t
S(t) =
dt
1
,
2B 2 (t )
(49)
dt
[Pcg C(t )]
,
B 2 (t )
(50)
dt
[Pcg C(t )]2

.
2B 2 (t )
(51)
(0)
(0)
In our case, cl = cg and Pcl = cg imply that

cg (t) = B(t) (0) + U(t) ,

1
(0)
Pcl =
C(t) .
Bcl + Pcg
B
(52)
(53)
Furthermore, we can define the parameter

2 (t) =

B 2 (t) 4
+ R2 (t) ,
2
(54)
such that the wave function can be written as

(cg , t) =
1
(2)1/4 1/2
e
cg
2
1
[cg cl ]2
42
[2 (Pcl B
B cl )
R(t)cl
]
2 2
2
cg
2 R(t)
[2 B
B + 2 ]
2
ei (t ),
(55)
where (t) is an arbitrary phase. Furthermore, it can be verified that the expectation value
of the effective energy in the infrared sector is
2

R
1
2 2 1 B
Eeff = Ecl +
(56)
+
,
82
2
2 B
2 2
where
1
2 2
Ecl = Pcl2
+ c cl .
2
2 cl
(57)
The contribution of the quantum fluctuations to Eeff is represented by the second, third,
and fourth terms of (56). The fourth term describes the decoherence of the system, in such
away that it is zero for coherent states and positive for decoherentized states. The parameter
= k0 /a depends on the cosmological model. On the other hand, the squared fluctuations
2 and P 2 are
cg
cg

2
cg
= cl2 (t) + 2 (t),
2

2
R
1
B
2
(t) +
.
+
Pcg = Pcl (t) +
B
42 (t)
2 2
(58)
(59)
449
4. Wave function in a power-law expansion of the universe

If mass parameter is time independent but the external classical force is nonzero, the
solutions are coherent states [19]. This case corresponds to a de Sitter expansion of the
universe, which was studied in a previous paper by Habib and Mijic [9]. In this work we
are interested in the study of the particular case of a power-law expansion of the universe in
which the scale factor is a t p . In this case the squared parameter of mass is given by [6]
2 (t) = M 2 t 2 ,
(60)
where M 2 = 94 p2 15
given by H (t) = p/t [6]. The
2 p + 2 and the Hubble parameter being
condition to get an unstable sector is M 2 > 0, or p > (5 + 17)/3 3.04. Furthermore,
since H (c ) = H0 e/Mp , the slow-roll parameters and become both of the order of
101 , so that the slow-roll regime is guaranteed. Hence, the equations which characterize
the system are
B M 2 t 2 B = 0,
t

C(t) = dt B t c x
, t .
The general solution of Eq. (61) is
1
1
2
2
B(t) = c1 t 2 [1+ 1+4M ] + c2 t 2 [1 1+4M ] ,
(61)
(62)
(63)
0 ) = 0 imply
where the initial conditions B(t0 ) = 1 and B(t
1 + 4M 2 1
1 + 4M 2 + 1
,
c2 =
.
c1 =
(64)
2 1 + 4M 2
2 1 + 4M 2
Since it is difficult to know exactly the functions R(t), U(t) and
S(t), we can make the
1
[1+
1+4M 2 ]
calculation for late times. For t 1, one obtains B(t)|t 1 t 2
and

1 + 4M 2
2 1 + 4M 2
R(t)|t 1
(65)
ln
,
( 1 + 4M 2 1)(1 + 1 + 4M 2 )
1 + 4M 2 1
2
2 (t)|t 1 t 1+ 1+4M .
(66)
R 2
This means that the decoherence function D(t) = 12 ( B
B + 2 2 ) in the expectation
effective energy (56), will be
1
2
D(t)|t 1 t 2 ( 1+4M 1) ,
(67)
which always increases because M 2 > 0 during inflation. Replacing (65), (66) and (67) in
2 increases
Eqs. (58) and (59) one observes that, since 2 (t)|t 1 increases with time, cg
but the second term in (59) decreases as t increases. The second terms in (58) and (59)
describe the evolution of super Hubble fluctuations due to the exchange of degrees of
freedom between the infrared (k 2 k02 ) and the ultraviolet (k 2 k02 ) sectors. The third
term in Eq. (59) is the most interesting one. This term describes the decoherence of the
phase-space (cg , Pcg ) during inflation. Note (see Eq. (67)), that it increases with t, so
450
the phase-space loses its coherence at the end of inflation. This term comes from the
coupling between cg and Pcg . This is the physical origin of decoherence in the phasespace (cg , Pcg ). However, the relevant phase-space during inflation is (cg , Pcg ), where
cg = a 3/2 cg and Pcg = cg . This implies that the function that describes decoherence
between cg and Pcg can be written, to a good approximation, as
1
2
D(cg , Pcg , t)|t 1 a 3 D(t)|t 1 t 2 [ 1+4M (6p+1)],
(68)
where D(t)|t 1 is given by (67). A numerical calculation shows that D(cg , Pcg , t) grows
for p > 4.6, but decreases in the range 3.04 < p < 4.6. It is well known that power-law
inflation takes place if p > 3.04. This means that for 3.04 < p < 4.6 the phase-space
(cg , Pcg ) remains coherent during inflation.
5. Final remarks
In this paper we have considered the wave function for the coarse-grained field. It
describes the redefined matter field fluctuations in the infrared sector for a globally flat
FRW background metric. The Hamiltonian of the Schrdinger equation is given by an
effective Hamiltonian. It has a quadratic contribution with a time dependent mass plus
an effective stochastic force, which describes the interaction between both, ultraviolet
and infrared sectors. However, this is not a true interaction in the sense of a thermalyzed
environment. Here, continuously we have new modes are crossing the ultraviolet sector
increasing the number of degrees of freedom of the infrared sector. This effect appears in
the stochastic equations as an effective noise c , which is responsible for the uncertainty of
the quantum state.
On the other hand the lost of coherence in the phase-space is a consequence of a nonzero
correlation between the variables of such a space (in this case cg and Pcg ). As was
shown in this work, the decoherence function D(t) increases with time in such away that
the phase-space (cg , Pcg ) is decohered at the end of power-law inflation. However, the
relevant phase-space to describe decoherence during inflation is (cg , Pcg ). We found that
in the range 3.04 < p < 4.6 the system does not decohere. This implies that, under these
conditions, the classical treatment of super Hubble matter field fluctuations developed in
stochastic inflation should be revised. On the other hand for very large p (more exactly for
p > 4.6) the phase-space (cg , Pcg ) loses its coherence at the end of inflation, and then
stochastic inflation provides a good treatment for large-scale matter field fluctuations when
the scale factor grows very rapidly. This calculation was performed in the framework of
supercooled inflation, which does not take into account dissipative effects produced by the
interaction between the inflaton field and other particles of a thermal bath. In the warm
inflation [20,21], the phase-space could not either remain coherent, but as a consequence
of additional thermal and dissipative effects. A more detailed treatment has to deal with
complicated nonlinear effects of super Hubble fluctuations. This goes beyond the scope of
this paper. I hope to consider this topic elsewhere.
451
Acknowledgements
I would like to acknowledge F. Astorga for his careful reading of the manuscript. Also, I
would like to acknowledge CONACYT (Mxico) and CIC of Universidad Michoacana for
financial support in the form of a research grant.
References
[1] A.H. Guth, Phys. Rev. D 23 (1981) 347.
[2] For reviews on inflation see, A.D. Linde, Particle Physics and Inflationary Cosmology, Harwood
Academic Publishers, New York, 1990, and references therein.
[3] A.A. Starobinsky, in: Fundamental Interactions, MGPI Press, Moscow, 1983;
A.A. Starobinsky, in: H.J. de Vega, N. Snchez (Eds.), Current Topics in Field Theory Quantum
Gravity, and Strings, Springer, New York, 1986.
[4] D. Polarski, A.A. Starobinsky, Class. Quant. Grav. 13 (1996) 377;
C. Kiefer, J. Lesgourdes, D. Polarski, A.A. Starobinsky, Class. Quant. Grav. 15 (1998) L67.
[5] S. Habib, Phys. Rev. D 46 (1992) 2408.
[6] M. Bellini, H. Casini, R. Montemayor, P. Sisterna, Phys. Rev. D 54 (1996) 7172.
[7] M. Mijic, Phys. Rev. D 49 (1994) 6434.
[8] H. Casini, R. Montemayor, P. Sisterna, Phys. Rev. D 60 (1999) 063512.
[9] S. Habib, M. Mijic, Stochastic dynamics of coarse-grained quantum fields in the inflationary
universe, unpublished.
[10] A. Berera, Phys. Rev. D 54 (1996) 2519.
[11] M. Bellini, Fresh inflation: a warm inflationay model from a zero temperature initial state, grqc/0101062, Phys. Rev. D, submitted for publication.
[12] D. Polarski, A.A. Starobinsky, Class. Quant. Grav. 13 (1996) 377;
C. Kiefer, J. Lesgourgues, D. Polarski, A.A. Starobinsky, Class. Quant. Grav. 15 (1997) L67.
[13] M. Bellini, Class. Quant. Grav. 16 (1999) 2393;
M. Bellini, Nucl. Phys. B 563 (1999) 245.
[14] E.J. Copeland, E.W. Kolb, A.R. Liddle, J.E. Lidsey, Phys. Rev. D 48 (1993) 2529.
[15] See, for example:
A. Mezhlumian, A.A. Starobinsky, astro-ph/9406045;
S.-J. Rey, Nucl. Phys. B 284 (1987) 706.
[16] E.D. Courant, H.S. Snyder, Ann. Phys. 3 (1958) 1.
[17] H.R. Lewis Jr., Phys. Rev. Lett. 18 (1967) 510;
H.R. Lewis Jr., J. Math. Phys. 9 (1968) 1976.
[18] R. Vaidyanathan, J. Math. Phys. 23 (1982) 1346.
[19] P. Carruthers, M.M. Nieto, Amer. J. Phys. 33 (1965) 537.
[20] A. Berera, Phys. Rev. Lett. 75 (1995) 3218.
[21] M. Bellini, Phys. Lett. B 428 (1998) 31.
Nuclear Physics B 604 [FS] (2001) 455478

Thermodynamic Bethe ansatz for generalized

extensive statistics
Andrei G. Bytsko
Steklov Mathematics Institute, Fontanka 27, St. Petersburg 191011, Russia
Received 7 November 2000; accepted 15 March 2001
Abstract
We investigate properties of the entropy density related to a generalized extensive statistics and
derive the thermodynamic Bethe ansatz equation for a system of relativistic particles obeying such a
statistics. We investigate the conformal limit of such a system. We also derive a generalized Y-system.
The Gentile intermediate statistics and the statistics of -ons are considered in detail. In particular,
we observe that certain thermodynamic quantities for the Gentile statistics majorize those for the
HaldaneWu statistics. Specifically, for the effective central charges related to affine Toda models
we obtain nontrivial inequalities in terms of dilogarithms. 2001 Published by Elsevier Science
B.V.
PACS: 05.30.-d; 05.70.Jk; 11.10.Kk
1. Introduction
Although all experimentally observed particles are either bosons or fermions, the general
principles of quantum mechanics do not prohibit existence of particles obeying other types
of statistics [1]. The motivation to consider an exotic statistics is that it may provide an
effective description of the dynamics of particles if their interaction is so strong that the
available occupancy for a given state depends on the number of particles already present
in the state [2]. An exotic statistics can also arise if particles possess a hidden internal
degree of freedom that is invisible in the Hamiltonian but that can become dynamically
relevant. Finally, an exotic statistics can emerge in description of physical models in one or
two spatial dimensions. The reason is that in this case the multi-particle wave function is
not necessarily even or odd and particles (anyons) may obey a fractional statistics [3].
Moreover, in low-dimensional theories the exchange statistics of the fields present in
Research supported in part by INTAS grant 99-01459 and by RFFI grant 99-01-00101.
E-mail address: bytsko@pdmi.ras.ru (A.G. Bytsko).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 2 4 - 9
456
A.G. Bytsko / Nuclear Physics B 604 [FS] (2001) 455478
a Hamiltonian is not directly related to the exclusion statistics of the corresponding

particles. For instance, although particles in the one-dimensional real coupling affine Toda
field theories (ATFT) are bosons, one has to impose the Pauli principle on their momenta in
the thermodynamic Bethe ansatz analysis in order to obtain correct values of corresponding
central charges [4]. Another example is the conformal field theory, where quasi-particle
spectra can be constructed with use of an exotic exclusion statistics [5].
The first attempt to introduce an exotic statistics is attributed to G. Gentile [6], who
proposed an intermediate statistics, in which at most G indistinguishable particles can
occupy a given state. This intermediate statistics interpolates between fermions and bosons,
that are recovered for G = 1 and G = . A decade after Gentiles work H.S. Green
introduced the parafermi statistics [7] that also fixes maximal occupancy for a given state.
Since then, various aspects of systems governed by the Gentile or parafermi statistics were
discussed in the literature, among them possible applications to the elementary particles
theory [8] and to the quark confinement problem in the QCD [9].
Many authors discussed thermodynamic properties of an ideal gas obeying the Gentile or
parafermi statistics [6,10,11]. In the present paper we will study the thermodynamic limit
of one-dimensional relativistic integrable field models governed by a Gentile-like statistics.
More precisely, the main purpose of this manuscript is to implement systematically
a generalized extensive exclusion statistics into the thermodynamic Bethe ansatz (TBA)
analysis of systems where the interaction of particles is relativistic, short-range and
characterized by a factorizable scattering matrix.
The TBA was originally developed in the seminal papers by Yang and Yang [12] for
treating a one-dimensional ideal gas. In later works [13,14] this technique was adopted
to one-dimensional relativistic models. The boundary condition for a many-particle wave
function leads to what is commonly referred to as the Bethe ansatz equation, that provides
the quantization condition for possible momenta of the system. The TBA constitutes an
interface between massive integrable models and conformal field theories. From the TBA
one can extract information about the ultraviolet limit of the underlying massive model, in
particular, find the corresponding (effective) central charge.
In the derivation of the TBA equation the underlying exclusion statistics is usually taken
to be either of bosonic or fermionic type [4,12,13]. In the present paper the TBA equation
will be derived and studied for significantly more general situation. Namely, let W (N, n)
denote the total dimension of the Hilbert space for a system of n indistinguishable particles
that can occupy N different states in the Fock-space. Assume that there exists a function
such that its N th power is a generating function for the dimensions W (N, n), i.e.,

N
W (N, k)t k .
f (t) =
(1)
k0
In the statistical mechanics, if the variable t is understood as fugacity, the sum on the righthand side is the grand partition function of the system. The property of a grand partition
function to be an N th power of a function independent of N is called extensivity. As we
are going to demonstrate below, this property allows us to develop the TBA analysis for
a rather general choice of f (t). Let us stress that for the purpose of deriving the TBA
457
equation even asymptotical extensivity suffices, i.e., Eq. (1) should hold in the large N
limit.
Of course, if the explicit form of f (t) is known, we can use it to simplify the general
formulae. For instance, an instructive example is the Gentile statistics of order G which is
defined by the following choice of f (t):
fG (t) = 1 + t + t 2 + + t G .
(2)
For G = 1 and G = this statistics describes ordinary fermions and bosons, respectively.
Although fG (t) is the simplest non-trivial choice of f (t), it has all features of a generalized
extensive statistics. Therefore, the Gentile statistics will be our main working example
below.
The paper is organized as follows. In Section 2 we describe possible types of
a generalized extensive statistics and study properties of the corresponding entropy
densities in the thermodynamic limit. The Gentile statistics and the -ons statistics are
discussed in this context. In Section 3 we compare properties of the Gentile statistics
and the HaldaneWu statistics. In Section 4 we derive the thermodynamic Bethe ansatz
equation and the finite-size scaling function for a relativistic multi-particle system in
which statistical interaction is governed by a generalized extensive statistics. In Section 5
we derive the Y-system related to such a generalized statistics. In Section 6 we study
the ultraviolet limit of the generalized TBA equation and obtain an expression for
the corresponding effective central charge. In Section 7 we compute finite-size scaling
functions and central charges related to some affine Toda models for the Gentile statistics
and the -ons statistics and compare them with their counterparts for the HaldaneWu
statistics. Our conclusions are stated in Section 8.
2. Types of extensive statistics and entropy density

Consider a system of n indistinguishable particles that can occupy N different states in
the Fock-space. Assume that the particles obey the Gentile statistics of order G, i.e., that
each state can be occupied by at most G particles (with G being a positive integer number).
There are several combinatorial ways to compute the total dimension of the Hilbert space
for such a system. For instance, we first choose m1 states which are occupied by at least
one particle, then among these m1 states we choose m2 m1 states which are occupied
by at least two particles, etc. This way of counting yields the following expression for the
total dimension of the Hilbert space

mG1 nm1 mG1
m1 m2
CN
Cm1 CmG2
CmG1
,
WG (N, n) =
(3)
Nm1 mG1 0
where Ckm = k!/(m! (k m)!) are binomial coefficients (and Ckm = 0 if m > k or if
m < 0). Now, let us compute N th power of the polynomial fG (t) defined in (2). Thanks
to the obvious recursive relation fG (t) = 1 + tfG1 (t), we can do this by G consecutive
458
applications of the binomial formula. Then it is easy to see that N th power of fG (t) is a
generating function for the dimensions WG (N, n), i.e.,
fG (t)
N
GN
WG (N, n)t n .
(4)
n=0
Thus, the Gentile statistics is a particular case of an extensive statistics.

Let us stress that the order G of the Gentile statistics must be a positive integer. Although
we could try to extend it to other values by formal replacement of (2) with
1 t G+1
,
(5)
1t
the resulting function fG (t) will not satisfy the important positivity property (see below).
Consider now a more general system of n indistinguishable particles that possesses, at
least for large N , the extensivity property (1) with
fG (t) =
f (t) =
d
Pk t k ,
(6)
k=0
where d can be a finite positive number or infinity. The Taylor coefficients Pk are fractional
dimensions of levels in a single-state Hilbert subspace and can be regarded as probabilities
of occupation of the given state by k particles (see related discussions in [1517]). The
total dimension of the single-state subspace is f (1). It is natural to require positivity of the
quantities Pk :
P0 = 1 and Pk 0 for k 1.
(7)
The first condition here implies that the vacuum is realized with probability one
independently of the size N of the system. Furthermore, we assume that the series (6)
converges in the complex plane for 0 |z| < R, where R . This implies that f (t)
belongs to one of the following types:
I.
d < ,
R = ,
II.
d = ,
R = ,
IIIa.
d = ,
1 < R < ,
IIIb. d = ,
0 < R 1.
(8)
The types I, II and III consist of finite degree polynomials, analytic functions with infinite
Taylor series, and meromorphic functions, respectively. For example, a finite order Gentile
statistics belongs to the type I, the Boltzmann statistics (f (t) = exp t) belongs to the type II,
and the bosonic statistics (f (t) = (1 t)1 ) is of the type IIIb. The reason we divided the
type III into two subtypes will become clear below.
Let us remark that, in addition to (7), one may require that P1 = 1 with the motivation
that the statistics should not be deformed if only one particle is present in the system.
Although all the concrete cases which we consider below do fulfill this requirement, it
does not seem to be crucial for the general considerations.
459
The entropy of the system under consideration is S = k ln W (N, n), where k is the
Boltzmann constant. We need to compute the entropy in the thermodynamic limit, i.e.,
when N and n/N is fixed. The extensivity property (1) allows us to express W (N, n)
as the following contour integral

N
1
W (N, n) =
dz f (z) zn1
2i
|z|=
1
=
2i
|z|=

dz
exp N ln f (z) ln z ,
z
(9)
where < R, and we denoted = n/N . Now we can find an asymptotic expression for
the entropy in the thermodynamic limit by application of the saddle point method to the
integral (9). The saddle point of the exponent in (9), denote it x, is found as the positive
root of the equation (prime stands for a derivative)
xf (t)|t =x = f (x).
(10)
Choosing = x in (9), so that the saddle point belongs to the integration contour, we apply
the standard result (see, e.g., [18]) for the Laplace integral:

exp{N(ln f (x) ln x)} 1

+ O(1/N)
as N ,
W (N, N) =
(11)
x
2Nh (x)
where h (t) = f (t)/(f (t)) + (1 )/t 2 . For any f (t) satisfying (7) and x obeying
(10) we have h (x) > 0 due to the Hadamard theorem [18,19]. From (11) we obtain the
entropy density s() for a given value of
1
ln W (N, N) s() = ln f (x) ln x,
N N
lim
(12)
where x is positive and satisfies (10).

Two remarks are in order here. Strictly speaking, if f (t) is degenerated, i.e., if there
exists integer p > 1 such that f (t) = k Ppk t pk , then the above derivation modifies since
f (t) has p saddle points. However, the corresponding physical interpretation is simply
that particles can be added only in p-tuples. Therefore, it is natural to introduce cluster
variables t = pt, n = pn, etc. In terms of these new variables formulae (9)(11) remain
valid.
More important remark is that for deriving (12) the property Pk 0 is crucial. In general,
without the positivity property (9) does not have a real and uniform in limit. Moreover,
if we allow f (t) to have negative Taylor coefficients, then some W (N, n) in the expansion
of (f (t))N may also be negative. This leads to an immediate problem with the definition
of the entropy as a logarithm of W (N, n). However, if the first negative coefficient in
the expansion of (f (t))N appears only at sufficiently large power n0 (of order O(N)),
then f (t) violating the positivity condition can still have well defined entropy density
for certain range of . For instance, such is the case of the HaldaneWu statistics with
statistical interaction g (see Section 3), where n0 N/g. However, in the present paper
460
we will restrict our consideration, for simplicity, only to partition function strictly obeying
the positivity condition.
The entropy density is a key quantity for developing the TBA analysis. From the
definition of it is clear that 0 d for f (t) of the type I and 0 for the
other types. Formally this follows from Eq. (10) which shows that varies from (x = 0)
to (x = R). Moreover, (x) is a strictly increasing function because x = xh (x) > 0
for x > 0.
Since f (t) is regular at zero, it is easy to see that ln x tends to zero for small .
Together with the condition P0 = 1 this yields s(0) = 0. In order to analyze further
behaviour of s(), we employ (10) and derive from (12) that
s() = ln x.
(13)
2 s()
< 0, that is s() is a convex up function.

Since x > 0, we conclude that
Furthermore, (13) implies that max s() = ln f (min{1, R}). Note, however, that s() is
not guaranteed to be positive for the whole range of . More precisely, s() may have one
root at the interval (1) < (R). Indeed, for f (t) of the type I we derive from (10)
and (12) that
s(d) = ln Pd .
(14)
That is, s() is non-negative everywhere at 0 d only if the last Taylor coefficient
Pd is greater or equal to one. For the other types of f (t) we have the following properties
s() =
s() =
for type IIIb,

for types II and IIIa.
(15)
(16)
Indeed, in the first case x R 1, hence, according

to (13), s() is strictly increasing.

In the second case we have s() = ln f (1) (1) d ln x. The integral here apparently

diverges since it exceeds (R0 ) d ln R0 , where we can choose any R0 such that
1 < R0 < R.
Investigation of thermodynamic properties of models in which the entropy density
becomes negative is rather problematic. Therefore, Eqs. (14)(16) suggest that we should
restrict the choice of f (t) to the type I with Pd 1 (the Gentile-like statistics) and to
the type IIIb (the bose-like statistics). However, let us notice that in the TBA framework
the variable x depends on the rapidity . For some relativistic models with factorizable
scattering the typical picture [4] is that x( ) falls off as | | grows (for instance, x( ) =
exp{mr cosh } for the trivial S-matrix). In this case x( ) < x(0) and we can consider
f (t) of the types II and IIIa if s() remains non-negative for (x(0)).
In the context of this discussion it is instructive to consider the so-called -ons [11,20].
These are particles with statistics interpolating between fermions ( = 1) and bosons ( =
1) in the following simple way
x
.
(x) =
(17)
1+x
Solving equation (10) for this choice of , it is easy to find the corresponding single-state
partition function:
461
Fig. 1. Entropy densities for G = 2 Gentile statistics (the upper curve), for g = 1/2 HaldaneWu
statistics (the middle curve) and for the fermionic statistics (the lower curve).
f (t) = (1 + t)1/ = 1 + t +
[k 1]
k2
tk
,
k!
(18)
where [k] = m=1 (1 m ). For > 0 the positivity condition (7) is fulfilled only if
= 1/d, with d positive integer. In this case f (t) is of the type I, the corresponding
maximal occupancy is = d. Since ln Pd < 0 for d > 1, the entropy density becomes
negative for certain range of . The case of = 0 describes the Boltzmann statistics. Here
f (t) = exp t belongs to the type II and s() is negative for > e. Finally, for < 0 the
positivity condition (7) is always fulfilled and f (t) is of the type IIIa or IIIb depending on
the value of .
Returning to the Gentile statistics, we can summarize the properties of the corresponding
entropy density sG () as follows (see Fig. 1 for illustration). sG () is a convex up function
defined for 0 G and vanishing at the end points of this interval. It attains the
maximum at = G/2:
sG (G/2) = ln(G + 1).
(19)
Furthermore, sG () possesses the following symmetry

sG (G ) = sG ().
(20)
This is a consequence of the equality WG (N, GN n) = WG (N, n) which, in turn, follows

from the relation (4).
3. Comparison with HaldaneWu statistics

It is interesting to compare thermodynamic properties of a system of particles obeying
the Gentile statistics with those of a system obeying the so-called HaldaneWu statistics.
For the latter, the total dimension of the Hilbert space is postulated to be [21]
g (N, n) = (N + (1 g)n + g 1)! .
W
n! (N gn + g 1)!
(21)
462
This expression interpolates between the bosonic (g = 0) and the fermionic (g = 1)

statistics. Introduction of such an interpolating statistics was motivated by Haldanes
generalization [2] of the Pauli exclusion principle to the form
$D/$n = g.
(22)
Here D is the number of available states (holes) before the nth particle has been added
to the system. The parameter g is called the statistical interaction. Properties of an ideal
gas obeying the HaldaneWu statistics were actively discussed in the literature [15,21,22].
The corresponding TBA equation for relativistic integrable models was obtained in [23]
and investigated in [23,24].
g (N, n) is problematic if the
A direct comparison of the quantities WG (N, n) and W
system contains a finite number of particles. Indeed, formula (3) has a bulky form for
generic G (i.e., for 2 G < ). Moreover, it requires additional conventions to make
exact sense of expression (21) for generic g (i.e., for 0 < g < 1). Actually, there exists no
prescription for counting of states on the microscopical level that would lead to (21) (see
related discussions in [1517]). To overcome these difficulties we can compare the two
statistics in the thermodynamic limit. More precisely, let us compare the entropy density
g (N, N) that is readily derived
sG () with its counterpart sg () limN N1 ln W
from (21) with the help of the Stirling formula

sg () = 1 + (1 g) ln 1 + (1 g) ln (1 g) ln(1 g).
(23)
Let us notice that the HaldaneWu statistics is extensive but the corresponding function
fg (t) does not satisfy the positivity property (7). Indeed, we can reconstruct fg (t) first as
a function of by the following formula

f () = exp s() s() ,
(24)
that follows from (12) and (13). For sg () given by (23) this yields
fg () =
1 + (1 g)
,
1 g
(25)
which is a well-defined function. But reexpressing it in terms of t with the help of (10)
we will always obtain a function that breaks the positivity property (7). For
instance, for
g = 1/2 we find (which coincides with the result of [15]) fg (t) = 1 + t 2/2 + t 1 + t 2 /4 =
1 + t + t 2 /2 + t 3 /8 t 5 /128 + . However, the first negative term in expansion of the
N th power of this series appears for n = 2N + 3. This implies that in the large N limit
thermodynamic quantities are well-defined for 2.
For our purposes we can regard (23) as a definition of the HaldaneWu statistics. Then
we have a finite maximal occupancy N/g for a given number of particles N simply because
sg () is defined for 0 1/g. Furthermore, sg () is a convex up function vanishing
at the end points of this interval. Thus, we see that properties of the entropy density sg ()
are similar to those of a type I extensive statistics and, specifically, to those of the Gentile
statistics. Therefore, it is natural to ask how much the behaviour of sG () differs from
sg () if g = 1/G (when the corresponding maximal occupancies coincide).
463
First, notice that the function s1/G () cannot coincide with sG () identically for
generic G since sg () does not possess a symmetry like (20) if g = 0, 1. Next, we can
G
compare the mid-point value s1/G (G/2) = G+1
2 ln(G + 1) 2 ln G with the corresponding
value of sG () given by (19). Taking into account the inequalities
t ln t > (t 1) ln(t + 1) for t > 1,
t ln t < (t 1) ln(t + 1) for 0 < t < 1,
(26)
we establish that sG (G/2) > s1/G (G/2) for G > 1. Actually, numerical computations for
various values of G > 1 show that sG () majorizes s1/G () everywhere at 0 < < G,
sG () > s1/G ()
for G > 1.
(27)
For illustration, the case of G = 2 is presented in Fig. 1.

Thus, we see that for generic G the difference between s1/G () and sG () is not small
if is not close to = 0 or = G. It is instructive to compare these functions also near
the end points (in particular, this will provide an additional support to the assertion (27)).
In a vicinity of = 0 Eq. (10) is solved as x = + (2G,1 1)2 + O(3 ), and we find
that sG () = (1 ln ) + (1/2 G,1)2 + O(3 ), where m,n stands for the Kronecker
symbol. Comparing this expansion with sg () = (1 ln )+(1/2 g)2 +O(3 ) which
follows from (23), we infer that for small functions sg () and sG () take close values
(up to the order 2 ) for any choice of G and g. Actually, employing (10), it is easy to
derive the following expansion for the entropy density of an arbitrary extensive statistics
with f (t) satisfying (7):

s() = (P1 ln ) + O 2 .
(28)
Thus, for small in the thermodynamic limit, the HaldaneWu and Gentile statistics are
close not only to each other but to any extensive statistics for which two first Taylor
coefficients of f (t) are P0 = P1 = 1. For instance, for small the HaldaneWu and
Gentile statistics are close also to the statistics of -ons (18).
Consider now = G ' for small positive '. Then, employing (28) and the
symmetry (20), we derive: sG (G ') = (1 ln ')' + O(' 2 ). On the other hand, setting
= 1/g ' in (23), we obtain sg (1/g ') = (1 2 ln g ln ')g' + O(' 2 ). We see that in
a vicinity of = G the difference between sG () and s1/G () is not even of first order in
' but of order ' ln ' (so that the corresponding plots are visibly different here, see Fig. 1).
Summarizing, it appears plausible that the Gentile statistics of order G > 1 majorizes the
HaldaneWu statistics with parameter g = 1/G and these statistics are not close except for
small values of = n/N .
4. Thermodynamic Bethe ansatz equation

Now we will consider a relativistic multi-particle system of l different species of
particles confined to a finite region of the size L. We denote by na the number of particles
and by Na the dimension of the Fock-space related to the species a. In the thermodynamic
464
limit both the dimensions of the Fock-space and the size L of the region were the system
is confined approach infinity but the ratios na /L remain finite [12]. For the fraction of
particles with rapidities between $/2 and + $/2 it is convenient to introduce
densities
$Na = a ( )$ L,
$na = ar ( )$ L.
(29)
As usual, the rapidity parameterizes the two-momentum P = m(cosh , sinh ) and the
total energy of the system is given by

=L
l
d ar ( )ma cosh .
(30)
a=1
According to (29) the variable in (12) is now related to the particle densities
a ( ) = ar ( )/a ( ).
(31)
Thus, we can express the entropy as the following functional

S ,
= kL
l
d a ( )s(a )
a=1
= kL
l

d a ( ) ln fa xa ( ) ar ( ) ln xa ( ) ,
(32)
a=1
where each xa ( ) satisfies (10) and hence it is a function of ar ( )/a ( ). The subscript of
fa means that different species may have different single-state partition functions.
In the formulation of the TBA equation for ordinary statistics it is common to introduce
the so-called pseudo-energies 'a ( ) such that bosonic (upper sign) and fermionic (lower
sign) distributions are ar ( )/a ( ) = (exp('a ( )) 1)1 . In view of (10), it is natural to
define the pseudo-energies for an arbitrary extensive statistics as follows
'a ( ) = ln xa ( ),
(33)
so that ar ( )/a ( ) = 'a () ln fa (e'a () ). For instance, the distribution for the Gentile
statistics acquires the following form in terms of the pseudo-energies (with a slightly
different meaning it was introduced by Gentile [6] for an ideal gas)
Ga
k exp{(Ga k)'a ( )}
ar ( )
= k=1
G
(34)
.
a
a ( )
exp{k'a ( )}
k=0
According to the fundamental principles of thermodynamics the equilibrium state of a

system is found by minimizing the free energy F . Hence, keeping the temperature constant
we obtain the equilibrium condition by minimizing F [, r ] = E[ r ] T S[, r ] with
respect to ar . The equilibrium condition reads
l

S b
F
E
S
=
T
= 0.
r
r
r
a
a
a
b ar
b=1
(35)
465
The admissible momenta in the system are restricted by the boundary conditions. To
derive the corresponding quantization equations one takes a particle in the multi-particle
wave function on a trip through the whole system [12,13]. The particle will scatter with
all other particles, which yields the following set of equations determining admissible
rapidities
exp(iLmi sinh i )
N

Sij (i j ) = i ,
i = 1, . . . , N,
(36)
j =i
where N = a Na , and Sij ( ) is a two-particle scattering matrix. The constant phases

i on the r.h.s. of (36) may depend on the particles species; for our purposes their exact
values are irrelevant. Taking the logarithmic derivative of (36) and employing densities as
in (29), we obtain
ma cosh + 2
l

ab br ( ) = 2a ( ).
(37)
b=1

Here (u v)( ) 1/(2) d u( )v( ) stands for the convolution and we introduced
as usual the notation ab ( ) = i (ln Sab ( )).
Employing equations (10), (12) and (31), we compute variations of the entropy (32):

S/ar = ln xa ( ),
(38)
S/a = ln fa xa ( ) .
Substitution of these relations into the equilibrium condition (35) together with (37)
yields the desired thermodynamic Bethe ansatz equations for a system in which statistical
interaction between particles of ath species is described by a single-state partition
function fa ,

1
ma cosh = ln xa ( ) +
ab ln fb (xb ) ( ).
kT
l
(39)
b=1
In general, this set of integral equations cannot be solved analytically. However, one can
try to solve these TBA equations numerically with the help of the iteration method (as
we will do for some examples in Section 7). For the ordinary statistics and the Haldane
Wu case it is known that such an iterative procedure converges provided that ab ( ) falls
off sufficiently fast. The most complete proof of this assertion is presented in [24,25] and
is based on the fixed point theorem. Presumably this proof extends to the case of (39)
(although some further restrictions on fa (t) may be required), but we will not discuss this
here. Let us notice only that if (39) can be solved iteratively and ab ( ) is symmetric in
(which is always the case for a unitary S-matrix), then xa ( ) is also symmetric in .
We now substitute the expressions for the total energy (30) and the entropy (32) together
with Eqs. (37) and (39) back into the expression for the free energy, and obtain

l

LkT
ma
d cosh ln fa xa ( ) .
F (T ) =
2
a=1
(40)
466
The relation between the free energy and the finite-size scaling function is well-known to
be c(T ) = 6F (T )/(LT 2 ) [26]. As usual we introduce a variable r = 1/T and set now
the Boltzmann constant to be one. Then the finite-size scaling function acquires the form
c(r) =

l

6ma r
a=1

d cosh ln fa xa ( ) .
(41)
Here we assumed that xa ( ) = xa ( ). Once more for f (t) = 1 + t and f (t) = 1/(1 t)
we recover the well-known expressions for the fermionic and bosonic scaling functions.
It is instructive to compare the above TBA equation (39) and the finite-size scaling
function (41) with those for the HaldaneWu statistics [23]:
l

1
ma cosh = ln 1 + ya ( ) +
ab ln 1 + yb1 ( ),
kT
(42)
b=1
cg (r) =

l

6ma r
a=1

d cosh ln 1 + ya1 ( ) .
(43)
Here ab ( ) = ab ( ) 2gab ( ), and g is a matrix which appears on the r.h.s. of (22) if

we consider a system of several species. As was discussed in [23], the TBA equation (42)
leads to an equivalence principle: two systems having the same mass spectra and identical
quantities ab ( ) are thermodynamically equivalent (as seen from (43), their finite-size
scaling functions coincide). Our TBA equation (39) does not possess such a feature. More
precisely, for (39) there exists no way to compensate a difference in statistics by a change
of ab ( ) independent of the explicit form of the S-matrix.
5. Y-systems
For some class of models it is possible to carry out certain manipulations on the TBA
equations such that the original integral TBA equations acquire the form of a set of
functional equations in new variables Ya [14]. These functional equations (commonly
referred to as Y-systems) have the further virtue that unlike the original TBA equations they
do not involve the mass spectrum. In the Y -variables certain periodicities in the rapidities
are exhibited more clearly. These periodicities may then be utilized in order to express
the quantity Ya ( ) as a Fourier series, which in turn is useful to find solution of the TBA
equations and expand the scaling function as a power series in the scaling parameter r. We
will now demonstrate that similar equations may be derived for a multi-particle system in
which the dynamical scattering is governed by the scattering matrix related to the ADEaffine Toda field theories and the statistical interaction is of a general type considered
above.
Consider the minimal part (i.e., independent of the coupling constant) of the scattering
matrix of the ADE-affine Toda field theories. As was shown in [27], upon appropriate
choice of CDD-ambiguities, these S-matrices satisfy the identity

l
i
i
+ ab
=
ab +
Iac cb ( ) 2Iab ( ),
h
h
467
(44)
c=1
where h denotes the Coxeter number, l the rank and I the incidence matrix of the
corresponding Lie algebra. It is then straightforward to derive the Y-system

l

I
i
i
Ya +
(45)
zb (Yb ( )) ab ,
Ya
=
h
h
b=1
where Ya ( ) = xa1 ( ) and

za (t) = tfa t 1 .
(46)
i
Eq. (45) follow upon first adding (39) at + i
h and h and subtracting I times (39) at
from the sum. Thereafter we employ the fact [28] that the masses of an ADE affine Toda
field theory are proportional to the PerronFrobenius vector of the corresponding Cartan
matrix, i.e., b Cab mb = 4 sin2 (/(2h))ma . Then, with the help of (44), Eq. (45) follow.
In comparison with (39) Eq. (45) have already the virtue that they are simple functional
equations and do not involve the mass spectrum. For the Gentile statistics we have z(t) =
t + 1 + t 1 + + t 1G . In particular, for G = 1 we have z(Y ) = 1 + Y and recover the
known fermionic Y-system. The bosonic case (G = ) corresponds to z(Y ) = Y 2 /(Y 1).
And the case of G = 2 seems to be particularly interesting since here z(Y ) = 1 + Y + Y 1
possesses an additional symmetry z(Y ) = z(Y 1 ) or, equivalently, z('a ( )) = z('a ( )).
6. Ultraviolet limit
In the small r limit the scaling function c(r) becomes the effective central charge of a
conformal field theory [26] which describes the ultraviolet limit of a given massive model.
That is, limr0 c(r) = ceff c 24h , where c is the conformal anomaly and h is the
lowest conformal weight in the corresponding conformal field theory.
In order to evaluate c(r) in the r 0 limit we substitute the derivative of (39) into (41),
replacing in both equations cosh by 12 e . Then we integrate the resulting equation by
parts, assuming that ab ( ) is symmetric and falls off sufficiently fast (typically, ab ( ) =
O(e ) as ). Next we substitute back equation (39). Finally we change variables
from to x( ), taking into account that x() = 0 (as follows from (39)), and again
integrate by parts. All this yields
ceff =
l
6
ca ,
2
(47)
a=1
1
ca = ln xa ln fa (xa ) +
2
xa
dt
ln fa (t).
t
(48)
Here we denoted xa xa ( = 0). These quantities can be determined from the constant
TBA equations that follow from (39) when r 0 (this derivation requires that xa ( )
468
become constant in a region of of order ln r when r is small; see more detailed

explanations in [4,25])
ln xa =
l
Nab ln fb (xb ),
a = 1, . . . , l,
(49)
b=1

where 2Nab = d ab ( ).
As a simple example we consider the statistics of -ons (18), assuming for simplicity
that is the same for all species (generalization to different a is obvious). Here the
integral in (48) can be computed explicitly in terms of the Euler dilogarithm which is
defined for t 1 as Li2 (t) = k1 t k /k 2 . We have

x
1
dt
ln f (t) =
t
x
1
1 ( x)k
dt
= Li2 ( x)
ln(1 + t) =
t
k2
k=1
and hence
ca =
1
1
Li2 ( xa )
ln xa ln(1 + xa ),
(50)
where xa satisfies (49). Since Li2 (t) = 12 Li2 (t 2 ) Li2 (t), we can rewrite (50) for
positive employing the Rogers dilogarithm which is defined as L(t) = Li2 (t) +
1
t
2
2 ln t ln(1 t) for 0 t 1. Then, using the Abel identity L(t ) = 2L(t) 2L( 1+t ),
we obtain

xa
1
1
ca = L
(51)
ln ln(1 + xa ).
+
1 + xa
2
Notice that this expression does not diverge for small ; lim 0 ca = xa (1 12 ln xa ) is
the value corresponding to the Boltzmann statistics.
Consider now the Gentile statistics, assuming that the order G is the same for all species.
In this case we also can compute the integral in (48) explicitly:
x
dt
ln fG (t) =
t
x

1 t G+1
dt
ln
t
1t

k=1

1
x k(G+1)
Li2 (x (G+1))
k
=
Li
,
x
(x)
2
G+1
G+1
k2
and we see that for the Gentile statistics expression (48) acquires the following nice form
involving only the Rogers dilogarithms:

1
L xaG+1 .
(52)
G+1
Introducing, in agreement with (33), the constants 'a = ln xa , we obtain from (52) for
G = and G = 1

1

and ca = L e'a L e2'a .
ca = L e'a
(53)
2
ca = L(xa )
469
These are the well-known values for the bosonic and fermionic statistics [4]. Notice that
1
in the latter case one usually derives ca = L 1+e
'a . This is equivalent to our formula (53)
due to the Abel identity. In fact, the Abel identity can be used in a similar way for any G
of the form G = 2m 1. In this case we can rewrite (52) as follows

k1
m

(xa )2
1
ca =
(54)
L
.
k1
2k1
1 + (xa )2
k=1
In this form ca is a manifestly positive and monotonic function of xa (the dilogarithm L(t)
grows monotonically for 0 t 1). Actually, this property holds not only for the Gentile
statistics but in general case as well. To prove this statement we take a derivative of (48)
and use (10) and the formula (12) for the entropy density. This yields
xa ca =
s(a (xa ))
.
2xa
(55)
As we discussed in Section 2, we should consider only such models where s((x)) > 0.
In this case the r.h.s. of (55) is positive. Hence, ca is always a monotonically increasing
function of xa (and therefore a monotonically decreasing function of 'a ). Moreover, taking
into account (7), we infer from (48) that ca (0) = 0, thus we also proved positivity of ca (xa ).
Since these properties are crucial for a physical interpretation of ca , let us underline that
they hold only if the first condition in (7) is fulfilled. Let us notice also that the properties
of ca together with (49) imply that for any model such that Nab 0 (which is the case,
for instance, for all the ADE-affine Toda models [4]) the value of ca does not exceed that
of the corresponding free model (where all Nab = 0 and consequently all xa = 1). Hence,
(48) evaluated for xa = 1 yields the upper bound on ca ,
1
ca
dt
ln fa (t).
t
(56)
Notice, however, that for a statistics of the type IIIb with R < 1 we have xa < 1, so that
in this case the upper bound on ca will be lower than (56). From a physical point of view
this implies that such a statistics cannot emerge in models with weak interaction and, in
particular, in a free model.
It was argued in [4] that the quantity 62 ca (for the ordinary statistics) can be interpreted
as a massless degree of freedom associated to the ath species in the massive model.
The properties of ca which we proved above show that the interpretation of 62 ca can
be extended to the case of a generalized extensive statistics as well. In this context, it is
important to remark that for the ordinary statistics (actually, for the Gentile statistics of any
order) we have 62 ca 1 as follows from (52) (recall that L(1) = Li2 (1) = 2 /6) if all Nab
are non-negative. However, for a general statistics the upper bound (56) on 62 ca depends
on the choice of fa and may not equal to one. For instance, taking fa (t) = (1 t)m with
m > 1, we obtain 62 ca = m for the corresponding free theory. In such a case, the massless
degree of freedom of the corresponding free particle exceeds that of a free boson. This can
2
possibly imply that we have to restrict the choice of statistics to such fa that ca 6 .
470
7. Examples related to affine Toda models

7.1. Ising and KleinGordon models
The most simple examples which illustrate the features outlined above more concretely
are the Ising model (A1 -minimal affine Toda field theory) with S( ) = 1 and the Klein
Gordon model with S( ) = 1. In both cases equation (39) is solved trivially, x( ) =
exp{rm cosh }. Then for the Gentile statistics (2) we can compute the entire scaling
function (41):
6rm
cG (r) = 2

d cosh ln 1 e(G+1)rm cosh ln 1 erm cosh
6rm 1
k=1

K1 (krm) K1 (G + 1)krm ,
(57)
where K1 (t) is the modified Bessel function. We depict cG (r) in Fig. 2, referring to them by
a slight abuse of notation as A1 . One observes that the difference in statistics affects most
severely the ultraviolet region. The scaling functions for different G converge relatively
fast towards each other in the infrared regime. These appear to be common features of a
general deformed statistics.
As seen from Fig. 2, the behaviour of cG (r) in the ultraviolet region (there is no first
order term in the small r expansion of cG (r)) is similar for any finite G to that of the
fermionic scaling function and is different from the bosonic case (where the first order
term is present). To explain this we observe that, as follows from (57),
cG (r) = cb (r)

1
cb (G + 1)r ,
G+1
(58)
where cb (r) is the bosonic scaling function. Now it is obvious that the first order terms on
the r.h.s. of (58) cancel each other for any finite G.
Fig. 2. Scaling functions of the A1 -minimal affine Toda model for the Gentile statistics of order
G = 1, 2, 3, 4 (solid curves, from down to up), for g = 1/2 HaldaneWu statistics (the dashed curve)
and for the bosonic statistics (the dotted curve).
471
The well-known property of the asymptotic behaviour of the modified Bessel function,
limt 0 tK1 (t) = 1, allows us to derive from (57)

6 1
1
G
eff
,
cG
(59)
= lim cG (r) = 2
1
=
2
r0
k
G+1
G+1
k=1
eff
which is in agreement with (52) since xa = 1 now. It is worth noticing that cG
can be identified for every positive integer G as the effective central charge of a
minimal conformal model M(s, t) (such that st = 6(G + 1)). In particular, G = 1, 2,
3, 4, correspond to the M(3, 4), M(2, 9), M(3, 8), M(5, 6), minimal models, respectively.
For the -ons (18) the scaling function related to the A1 -models can also be computed:
6rm
c (r) =
2
6rm

1
( )k1 K1 (krm). (60)
d cosh ln 1 + erm cosh = 2
k=1
Using again the asymptotic behaviour of the modified Bessel function, we obtain ceff =
limr0 c (r) = 6 2 Li2 ( ) in agreement with (50). Here, unlike in the Gentile case,
it is not easy to see whether the effective central charge ceff has rational value for a given
. As we discussed in Section 2, positive is to be of the form = 1/d with d positive
integer. It appears then that the only d leading to rational value of ceff is d = 1. On the
other hand, varies continuously in the range 1 0, so that any rational value
of ceff not exceeding 62 occurs here. For the Boltzmann statistics c0eff = 62 is irrational.
Finally, < 1 is the case of a type IIIb statistics with R < 1, which, as we discussed
above, cannot emerge in a free model.
7.2. Scaling Potts and YangLee models
Next we consider the scaling Potts model, which was studied previously in the TBA
framework for the fermionic statistics in [13] and for the HaldaneWu statistics in [23].
The two particles in the model are conjugate to each other and consequently their masses
are the same, m1 = m2 = m. The conjugate particle occurs as a bound state when two
particles of the same species scatter. The S-matrix of the scaling Potts model equals the
minimal S-matrix of the A2 -affine Toda field theory. The corresponding quantities ab ( )
that enter the TBA equation are given by
3
3
and 12 ( ) =
.
11 ( ) = 22 ( ) =
(61)
2 cosh + 1
1 2 cosh
Consider the scaling Potts model with both species of particles obeying the Gentile
statistics of the same order G. In this case the Z2 -symmetry of the model is preserved so
that x1 ( ) = x2 ( ) x( ) and (39) reduces to a single integral equation. It appears to be
impossible to find an analytic solution x( ) to this equation, but it is straightforward to
solve it numerically. Taking x [0] ( ) = exp{rm cosh } as the first approximation, we can
iterate (39) as follows
472

ln x [n+1] ( )

2 3
cosh( )
= rm cosh
ln fG x [n] ( ) .
d

1 + 2 cosh 2( )
(62)
Convergence is achieved relatively quickly (it depends on the value of mr). The results for
G = 1 (which is in complete agreement with the calculation in [13]) and for G = 2 are
shown in Fig. 3. To make contact with the literature, we introduced the quantity L( ) =
ln fG (x( )). One observes the typical plateau L( ) = const for some region of when
mr 0, which is required to derive (49). We see also that for the same value of mr the
functions L( ) corresponding to different G have similar profiles but different heights of
the plateau.
Having solved the generalized TBA equation, we can substitute x( ) into (41) and
compute the entire scaling function cG (mr). The result of the numerical computation is
shown in Fig. 4. The behaviour of the scaling functions is similar to what we have already
observed in the A1 case.
Fig. 3. Solution L() of the TBA equation for the A2 -minimal affine Toda model for the scaling
parameter mr = 0.1 (dotted curves), mr = 0.01 (dashed curves), and mr = 0.001 (solid curves). For
the same value of mr the lower curve corresponds to the fermionic statistics and the upper curve to
G = 2 Gentile statistics.
Fig. 4. Scaling functions of the A2 -minimal affine Toda model for the Gentile statistics of order
G = 1, 2, 3, 4 (solid curves, from down to up) and for the HaldaneWu statistics with gab = 12 ab
(the dashed curve).
473
Due to the fact that the scattering matrices of the scaling Potts model and the scaling
A2
A2
YangLee model are related as S YL ( ) = S11
( )S12
( ), the generalized TBA equations
for the two models coincide. The only difference is that the scaling YangLee model has
only one particle. Therefore, its scaling function equals a half of that for the scaling Potts
model.
7.3. Comparison with HaldaneWu statistics
In Section 3 we have seen that the entropy density corresponding to the Gentile statistics
majorizes that for the HaldaneWu statistics. It appears (at least for certain models) that an
analogous relation holds for the corresponding finite-size scaling functions. Specifically, if
cG (r) is the scaling function (41) of A1 or A2 minimal affine Toda model where all species
obey order G Gentile statistics and cg (r) is the scaling function (43) of the same model
1
ab , then for all r 0
where the HaldaneWu statistical interaction is of the form gab = G
we have
1
cG (r) > cg (r), where gab = ab and G > 1.
(63)
G
For the A1 case we can rigorously prove relation (63) since the solutions of the generalized
TBA equations (39) and (42) are found explicitly:

g ln y( ) + (1 g) ln 1 + y( ) = mr cosh = ln x( ).
(64)
This equation allows us to express x( ) in terms of y( ):

g1
x( ) = y g ( ) 1 + y( )
.
(65)
Moreover, since mr cosh > 0, we infer from (64) that

y( ) > y0 ,
where y0 = (1 + y0 )g1 .
(66)
As seen from (41) and (43), in order to establish (63) it suffices to show that
fG=1/g (x( )) > 1 + y 1 ( ). The latter relation is equivalent to the inequality

1/g

g
1 1 + y( )
(67)
> y g ( ) 1 + y( ) ,
as can be verified by simple algebraic manipulations using formula (65). Now it is
elementary to check that (67) is valid provided that (66) holds. Thus, we proved the
assertion (63) in the A1 case. For illustration, the scaling function cg (r) for g = 1/2
HaldaneWu statistics is shown in Fig. 2. It lies everywhere below cG (r) for G = 2 Gentile
statistics with the maximal difference reached in the ultraviolet region.
For a higher rank case, already for A2 , the generalized TBA equations are integral and
it is not clear whether we can compare xa ( ) and ya ( ) analytically. However, numerical
computations for various G show that (63) holds in the A2 case as well. For illustration, we
present in Fig. 4 the scaling function cg (r) for the HaldaneWu statistics with gab = 12 ab .
A particular consequence of (63) is the following inequality for the corresponding
central charges:
eff
> cgeff ,
cG
for gab =
1
ab
G
and G > 1.
(68)
474
eff is found from (47), (49) and (52) whereas c eff is the effective central charge
Here cG
g
1
related to the HaldaneWu statistics with gab = G
ab and is given by [23]
cgeff =

l
6
1
L
,
2
1 + ya
a=1
where
ln(1 + ya ) =
l

(Nab + gab ) ln 1 + ya1 .
b=1
(69)
eff
Again, in the A1 case we can prove relation (68) directly. Here cG
has a simple form (59)
and, therefore, (68) reduces to

6
1
1
>
(70)
,
L
1 + g 2
1 + y0
where y0 is defined as in (66). Notice that g < y0 < 1 for g = 0, 1 (since y0 g would
1
1
< 1+g
. Using now a property of the dilogarithm,
contradict (26)). Therefore, 12 < 1+y
0
1
2
1
2
t for
< t < 1,
L(t) >
t for 0 < t < ,
(71)
6
2
6
2
we establish the desired inequality (70).
As we discussed above, the order G of the Gentile statistics is to be a positive integer.
eff
, then (70)
However, if we regard (47), (49) and (52) as a definition of the quantity cG
extends to any G > 1 (since the proof does not require that G be an integer) as shown in
Fig. 5. Furthermore, considering G in the range 0 < G < 1, we can prove with the help of
1
1
) > 1+g
due to (71). Thus, for this range of G
(26) that here y0 < g and hence 62 L( 1+y
0
we have a reverse inequality, i.e.,
L(t) <
1
ab and 0 < G < 1.
(72)
G
As we will see below, this inequality holds for higher rank cases as well. But let us stress
eff in (72) is a formally defined quantity which is not related directly to the
again that cG
Gentile statistics.
The inequality for the central charges (68) is a weaker statement than the inequality
for the scaling functions (63). Furthermore, (68) involves dilogarithms, which makes its
eff
< cgeff ,
cG
for gab =
Fig. 5. Effective central charge of the A1 - and A2 -minimal affine Toda models for the Gentile
statistics of order G (lower and upper solid curves) and for the HaldaneWu statistics with
1 (lower and upper dashed curves).
gab = G
ab
475
proof more complicated (already in the A1 case we used a rather subtle property (71)).
But the advantage of (68) for a numerical verification is that it relates numbers and not
functions as (63). Moreover, since c(r) is a continuous function, validity of (68) implies
that (63) holds at least in some ultraviolet region. Therefore, we will discuss below validity
of (68) for several affine Toda models. As a mathematical by-product, this provides us with
interesting dilogarithm inequalities.
In the A2 case, Eq. (68) contains dilogarithms in a more involved way. As seen from (49),
(52) and (69), it is equivalent to the following inequality

1+ 1
1
g
L x0 g > L
, for 0 < g < 1.
L(x0 )
(73)
1+g
1 + y0
Here x0 and y0 are determined from the following equations that follow from (49) and (69)
upon noticing that N11 + N12 = 1
x0 f1/g (x0 ) = 1
1+g
and y0
= (1 + y0 )g .
(74)
It is worth remarking that, as was noticed in [23], 62 L( 1+1y ) coincides with the effective
0
central charge of the CalogeroSutherland model with the coupling constant = g.
Numerical computations show that (73) indeed holds for any 0 < g < 1, that is for any
G > 1 (here again we need not to restrict G to be an integer). Furthermore, like in the A1
case, inequality (73) reverses for 0 < G < 1, i.e., for g > 1. For illustration, we depict the
eff
and cgeff in Fig. 5. This result provides an additional support to our claim
corresponding cG
that in the A2 case the entire scaling function cG (r) majorizes cg (r) for G > 1.
It is interesting to remark that, as follows from (52) and (69), the bosonic version
(G = or, equivalently, g = 0) of the A2 -minimal affine Toda model has ca = 12 . That
is, then each species in the scaling Potts model (or the only species in the scaling Yang
Lee model) appears to be a free fermion, whereas the entire central charge is that of a free
boson (see also related comments in [23]). That is why all curves on Fig. 5 converge to the
same value c = 1.
7.4. Higher rank cases
Now we discuss briefly the simplest minimal affine Toda models which have particle
eff and c eff again
species with different masses. We will compare numerical values of cG
g
setting gab = G1 ab . For this purpose we use (49) and (69), taking into account [4] that
the N -matrix in these equations is given by Ng = Ig (2 Ig )1 , where Ig stands for the
incidence matrix of the corresponding Lie algebra g.
In the A3 case the masses of the three species are m1 = m3 = m2 / 2. Then, computing
eff by formulae (49) and (52), we find: c eff 0.77, c eff = 1, c eff 1.12, c eff 1.16 for
cG
G = 12 , 1, 2, , respectively. The values of cgeff corresponding to the HaldaneWu statistics
1
with gab = G
ab were found in [23]: ceff 0.89, ceff = 1, ceff 1.07, ceff 1.16 for
1
g = 2, 1, 2 , 0, respectively. Carring out analogous computations
for the A4 case, where
the masses of the four species are m1 = m4 , m2 = m3 = m1 ( 5 + 1)/2, we find (Gentile
statistics): ceff 0.92, ceff = 8/7, ceff 1.25, ceff 1.28 for G = 12 , 1, 2, , respectively;
476
and (HaldaneWu statistics): ceff 1.05, ceff = 8/7, ceff 1.20, ceff 1.28 for g = 2, 1,
1
2 , 0, respectively.
(2)
One can also consider a folded algebra case with two different masses, namely A4 ,
(2)
where m2 = m1 ( 5 + 1)/2. In general, one assigns to the A2n minimal affine Toda model
a tad-pole graph [4,27] such that the corresponding incidence matrix differs from that of
An only by 1 in the lower-right entry. Then one can check that xa in (49) and ya in (69)
coincide with those for the A2n case (a well-known fact for the fermionic statistics [4,13]).
1 eff
Consequently, for any statistics we have ceff (A(2)
2n ) = 2 c (A2n ). Therefore, the data for
(2)
A4 follow from the above results for A4 .
Finally,
in the D4 case, where the masses of the four species are m1 = m3 = m4 =
m2 / 3, we obtain for the Gentile statistics: ceff 0.82, ceff = 1, ceff 1.08, ceff 1.09
for G = 12 , 1, 2, , respectively; and for the HaldaneWu statistics: ceff 0.93, ceff = 1,
ceff 1.04, ceff 1.09 for g = 2, 1, 12 , 0, respectively.
We see that all the above results are in agreement with (68) and (72). This, together
with the more detailed results we obtained in the A1 and A2 cases, allows us to conjecture
that these inequalities and the inequality (63) for the scaling functions hold actually for all
simply-laced minimal affine Toda models.
We remark also that in all the computations of ceff in this subsection we always have
the rule that the value of ca is smaller for the species which has heavier particle (this was
known for the fermionic statistics [4]). This observation provides an additional support
to the above discussed interpretation of 62 ca as a massless degree of freedom of the
corresponding particle.
8. Conclusion
Summarizing, we considered possible types of a generalized extensive statistics and
established properties of the corresponding entropy densities. In particular, we established
numerically (and gave some supporting analytical arguments) that the entropy density
for the Gentile statistics of order G > 1 majorizes that for the HaldaneWu statistics if
they correspond to the same maximal occupancy. Further, we derived the thermodynamic
Bethe ansatz equation and the finite-size scaling function c(r) for a relativistic multiparticle system obeying a generalized extensive exclusion statistics. We put particular
emphasis on the ultraviolet limit of such a system. We derived an expression for an effective
central charge in the general case and showed that for the Gentile statistics it acquires
an elegant form involving dilogarithms. We discussed a physical interpretation of the
partial central charges ca and argued that it possibly leads to restrictions on the choice of
statistics. Finally, we observed (and partially proved) that the majorizing properties of the
Gentile statistics with respect to the HaldaneWu statistics extend also to finite-size scaling
functions and central charges related to the minimal simply-laced affine Toda models.
In the presented analysis of thermodynamic properties of a relativistic multi-particle
system the single-state partition function f (t) played a key role. For a given model, if
we know f (t) (at least in the thermodynamic limit) and the quantities ab ( ), then we
477
can compute, at least in principle, the entire scaling function which provides complete
thermodynamic description of a system. In principle, f (t) should be determinable from the
corresponding Hamiltonian. However, this might be a non-trivial task since the exclusion
statistics of particles can differ from the exchange statistics of the fields present in the
Hamiltonian. Therefore, it may be more practical for a given system to fix an exotic
statistics a priory and investigate whether it leads to realistic physical properties. In
particular, one can obtain information about the ultraviolet limit and see if it corresponds
to an appropriate conformal field theory. As we have seen above, the different types (8)
of f (t) lead to significantly different thermodynamic properties. The question, which of
these statistics can emerge in physical models, remains open.
Another open problem is to prove rigorously the assertion (27) that the entropy density
for the Gentile statistics majorizes that for the HaldaneWu statistics. Such a proof will
probably provide a deeper insight into general properties of exotic exclusion statistics.
Also it would be interesting to verify the majorizing inequalities for the scaling functions
(63) and central charges (68) related to the affine Toda models and to understand whether
such a majorization is a model independent feature of the two involved statistics.
Acknowledgements
I am grateful to A. Fring for proposing the problem and participation in the initial stage
of the research. This work was supported in part by INTAS grant 99-01459 and by Russian
Fund for Fundamental Investigations grant 99-01-00101.
References
[1] O.W. Greenberg, A.M.L. Messiah, Phys. Rev. B 136 (1964) 248.
[2] F.D.M. Haldane, Phys. Rev. Lett. 67 (1991) 937.
[3] F. Wilczek, Phys. Rev. Lett. 48 (1982) 1114;
F. Wilczek, Phys. Rev. Lett. 49 (1982) 957;
Y.S. Wu, Phys. Rev. Lett. 52 (1984) 2103.
[4] T.R. Klassen, E. Melzer, Nucl. Phys. B 338 (1990) 485.
[5] K. Schoutens, Phys. Rev. Lett. 79 (1997) 2608;
A.G. Bytsko, A. Fring, Nucl. Phys. B 521 (1998) 573;
A.G. Bytsko, A. Fring, Phys. Lett. B 454 (1999) 59;
P. Bouwknegt, K. Schoutens, Nucl. Phys. B 547 (1999) 501.
[6] G. Gentile, Nuovo Cimento 17 (1940) 493;
G. Gentile, Nuovo Cimento 19 (1942) 109.
[7] H.S. Green, Phys. Rev. 90 (1953) 270.
[8] H.S. Green, Prog. Theor. Phys. 47 (1972) 1400;
Y. Ohnuki, S. Kamefuchi, Prog. Theor. Phys. 52 (1974) 1369;
R.Y. Levine, Y. Tomozawa, Phys. Lett. B 128 (1983) 189.
[9] A.J. Bracken, H.S. Green, J. Math. Phys. 14 (1973) 1784;
P.G.O. Freund, Phys. Rev. D 13 (1976) 2322;
M. Cattani, N.C. Fernandes, Nuovo Cimento B 87 (1985) 70;
M. Cattani, N.C. Fernandes, Phys. Lett. A 124 (1987) 229.
478
[10] J.D. Anand, S.N. Biswas, M. Hasan, Phys. Rev. D 18 (1978) 4529;
M.C. de Sousa Vieira, C. Tsallis, preprint CBPF-NF-038/86, 1986;
P. Suranyi, Phys. Rev. Lett. 65 (1990) 2329;
A.K. Rajagopal, Phys. Lett. A 214 (1996) 127.
[11] K. Byczuk, J. Spalek, G.S. Joyce, S. Sarkar, Acta Phys. Polon. B 26 (1995) 2167.
[12] C.N. Yang, C.P. Yang, Phys. Rev. 147 (1966) 303;
C.N. Yang, C.P. Yang, J. Math. Phys. 10 (1969) 1115.
[13] Al.B. Zamolodchikov, Nucl. Phys. B 342 (1990) 695;
Al.B. Zamolodchikov, Nucl. Phys. B 358 (1991) 524.
[14] Al.B. Zamolodchikov, Phys. Lett. B 253 (1991) 391.
[15] C. Nayak, F. Wilczek, Phys. Rev. Lett. 73 (1994) 2740.
[16] K.N. Ilinski, J.M.F. Gunn, A.V. Ilinskaia, Phys. Rev. B 53 (1996) 2615.
[17] A.P. Polychronakos, Phys. Lett. B 365 (1996) 202.
[18] M.V. Fedoruk, Asymptotics: Integrals and Series, Nauka, Russian, 1987.
[19] E.C. Titchmarsh, The Theory of Functions, Oxford University Press, Oxford, 1939.
[20] R. Acharya, P. Narayana Swamy, J. Phys. A 27 (1994) 7247;
A.V. Ilinskaia, K.N. Ilinski, J.M.F. Gunn, Nucl. Phys. B 458 (1996) 562.
[21] Y.S. Wu, Phys. Rev. Lett. 73 (1994) 922.
[22] D. Bernard, Y.S. Wu, cond-mat/9404025 (1994);
M.V.N. Murthy, R. Shankar, Phys. Rev. Lett. 73 (1994) 3331;
K. Hikami, Phys. Lett. A 205 (1995) 364;
S.B. Isakov, D.P. Arovas, J. Myrheim, A.P. Polychronakos, Phys. Lett. A 212 (1996) 299.
[23] A.G. Bytsko, A. Fring, Nucl. Phys. B 532 (1998) 588.
[24] A. Fring, C. Korff, B.J. Schulz, Nucl. Phys. B 549 (1999) 579.
[25] C. Korff, hep-th/0008200 (2000).
[26] H.W.J. Blte, J.L. Cardy, M.P. Nightingale, Phys. Rev. Lett. 56 (1986) 742;
I. Affleck, Phys. Rev. Lett. 56 (1986) 746.
[27] F. Ravanini, A. Valleriani, R. Tateo, Int. J. Mod. Phys. A 8 (1993) 1707;
M.J. Martins, Nucl. Phys. B 394 (1993) 339.
[28] M.D. Freeman, Phys. Lett. B 261 (1991) 57;
A. Fring, H.C. Liao, D.I. Olive, Phys. Lett. B 266 (1991) 82.

Probability distributions of line lattices in

random media from the 1D Bose gas
Thorsten Emig a , Mehran Kardar a,b
a Physics Department, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
b Institute for Theoretical Physics, University of California, Santa Barbara, CA 93106, USA
Abstract
The statistical properties of a two-dimensional lattice of elastic lines in a random medium are
studied using the Bethe ansatz. We present a novel mapping of the dilute random line lattice onto
the weak coupling limit of a pure Bose gas with delta-function interactions. Using this mapping, we
calculate the cumulants of the free energy in the dilute limit exactly. The relation between density
and displacement correlation functions in the two models is examined and compared with existing
results from renormalization group and variational anstze. 2001 Published by Elsevier Science
B.V.
PACS: 05.30.-d; 05.70.-a; 64.60.Cn; 74.60.Ge
1. Introduction
Disorder, in the form of quenched impurities in a sample, is likely to modify various
measurements. However, from a theoretical point of view, characterizing the behavior
of disordered systems has proven difficult, making this one of the most challenging and
controversial fields of statistical mechanics. The directed polymer in a random medium
(DPRM) is one of the rare cases of a disorder-dominated system whose statistical properties
can be determined in great detail both analytically and numerically [1], and thus serves as
a prototype for other such systems. Furthermore, its behavior is related to a plethora of
other problems in statistical mechanics, such as the kinetic roughening of surfaces [2], the
hydrodynamics of the Burgers equations [3], and surprisingly the biological sequencealignment problem for genes and proteins [4,5].
The DPRM is directly relevant to vortex lines in superconductors with point impurities.
Most research on the vortex phases in these systems has focused on the structural
E-mail addresses: emig@mit.edu (T. Emig), kardar@mit.edu (M. Kardar).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 0 2 - X
480
T. Emig, M. Kardar / Nuclear Physics B 604 [FS] (2001) 479510
properties of the flux lattice in a sample containing many vortices. Recently, the
statistics of single vortex lines was studied experimentally on a two-dimensional flux
line lattice oriented parallel to a thin micrometer-sized film of 2HNbSe2 [6]. Magnetic
response measurements show interesting sample dependent fine structure in the constitutive
relation B(H ), providing a fingerprint of the underlying random pinning landscape.
More generally, the random potential is expected to modify the measurement of various
thermodynamic and structural quantities, making it desirable to characterize the probability
distribution functions (PDFs) for the outcome. At one extreme, microscopic quantities
such as two-point correlation functions are quite sensitive to randomness, inducing in
some cases complicated multiscaling behavior of their PDFs [7]. On the other hand, the
free energy and other macroscopic quantities are expected to have simpler self-averaging
behavior, while non-trivial finite-size effects are still present in their corresponding PDFs.
To date, the PDF of the free energy is known exactly only for a single line, the DPRM
[811]. However, as in superconductors, in most cases elastic lines appear at a finite
density, leading to novel effects resulting from a competition between randomness and line
interactions. Similar line lattice structures occur in domain walls in the incommensurate
phase of charge-density-waves, in monolayers absorbed on crystal surfaces, and in the
form of steps on the vicinal surface of a crystal [12]. As pointed out by de Gennes
in the context of polymers [13], and later on by Pokrovsky and Talapov for domain
walls in incommensurate phases [14], the statistical mechanics of a pure two-dimensional
line lattice maps onto one-dimensional free fermions if the repulsive line interactions
are replaced by non-crossing constraints. Applying the same constraint, Kardar used
the replica method to show that the line lattice in a random potential maps onto onedimensional U (n) fermions with an attractive interaction [8]. Calculating the ground state
energy of this quantum system for integer n by Bethe ansatz, and analytically continuing
to n 0, the quench-averaged free energy was calculated in this reference. Although
the PDF was not calculated explicitly, a scaling ansatz for the replica free energy was
proposed based on dimensional arguments. Following up on this mapping, there have been
attempts [15,16] to calculate the density correlations of the line lattice from the excitation
spectrum of the U (n) fermions by Bethe ansatz, leading to different results.
In this paper we examine the analogy between the replicated line lattice and U (n)
fermions further. A considerable difficulty in the replica approach is the need for analytical
continuation to small n of the results obtained for integer n. Analyticity is usually assumed
in replica theories, since it cannot be proved in most cases. Here we develop a new method
for analytical continuation in the line lattice problem which is based on Eulers gamma
function, the natural continuation of the factorial to real valued arguments. This technique
provides an analytic expression for the replica free energy of the line lattice at small n in
terms of an integral equation, which in general has to be solved numerically. Interestingly,
in the dilute limit of the line lattice, the analytically continued Bethe ansatz equations are
identical to that of the 1D Bose gas with delta-function repulsions. This novel mapping
allows us to obtain exact expressions for the free energy cumulants of the line lattice, and
to reveal new relations between density correlations in the 1D Bose gas and the replica
theory of the line lattice, respectively. The scaling predictions for the replica free energy
481
proposed in [8], and used in [17] in a wider context of disordered systems, are strikingly
confirmed.
The outline of the paper is as follows. Using the replica method and the analogy between
elastic lines and world-line of quantum particles, we first show that the random line system
is described by a quantum problem of colored fermions, which interact upon contact.
This analogy is explained in detail in Section 2. Exact solutions for the ground state energy
of this colored fermion model in the thermodynamic limit are then obtained via the series
of nested Bethe anstze introduced by Yang [18]. For an integer number n of colors, the
resulting sets of Bethe ansatz equations can be reduced to a single set of equations by
considering n-clusters of fermions, as showed by Takahashi [19] and Kardar [8]. These
integer n equations are derived in Section 3.1. In the following Section 3.2, we develop a
new technique to continue the discrete Bethe ansatz equations directly to continuous n.
We find that for small n, and in the limit of low line density, the complex n-color
fermion problem can be mapped onto the well-studied one-dimensional Bose gas with
delta-function repulsions [20] (see Section 4). It turns out that the Bose gas interaction
strength is proportional to the number n of replicas, thus mapping the interesting regime
n 1 into the weak coupling limit of the Bose gas. Therefore, the problem of calculating
disorder averaged cumulants of the free energy of the line system becomes equivalent to a
perturbative expansion of the ground state energy of the one-dimensional Bose gas in the
weak coupling limit. The exact results for the cumulants of the free energy are presented
in Section 5. Finally, in Section 6 we apply our novel mapping to the question of what can
be learned about the densitydensity correlation function of the line lattice from the known
results for corresponding correlation functions of the 1D Bose gas.
2. Mapping fluctuating lines to fermions

Configurations of directed elastic lines can be regarded as the world lines of quantum
particles, where the coordinate along the line plays the role of time. To develop a concrete
picture of this analogy, we consider the canonical partition sum ZN of a (1 + 1)dimensional lattice of N interacting lines in a random environment of lateral length L.
To simplify the sum over all configurations of the lines, we assume periodic boundary
conditions in the direction parallel to the lines. This is justified since the statistical
mechanics of the line lattice is insensitive to the particular choice of boundary conditions
in the limit where L is much larger then the spacing between line collisions [21]. Now the
partition sum can be written as the path integral
N

ZN
ZN V (x, y) = 0
drj
N!
P
j =1
xj (L)=rPj

H [V (x, y)]
Dxj (y) exp
,
T
(1)
xj (0)=rj
where we have summed over all permutations P and have used Feynmans definition of
the path-integral measure Dxj (y) [22]. Here Z0 denotes the partition function of a single
482
line in a pure system with only one end of the line fixed. 1 The Hamiltonian in Eq. (1) is
given by

L
N
N

g dxj 2
H V (x, y) = dy
(2)
+
U (xi xj ) +
V (xj , y) ,
2 dy
0
j =1
i<j
j =1
which is a functional of the random potential V (x, y). The first term measures the elastic
energy of individual lines with line tension g. The lines interact via a short ranged repulsive
pair potential U (x), which is local in y. Effectively, a line interacts only with the lines in the
same constant y cross section. This assumption is valid if the line coordinates xj (y) vary
slowly with y. The last term couples the potential from randomly distributed impurities
to the local line positions, and makes the Hamiltonian time dependent. Without this
last term, the expression for ZN is formally identical to the Feynman path integral for
the canonical partition sum of an interacting 1D Bose gas [21]. But there is an additional
restriction for the configurations of elastic lines compared to the boson world lines. Usually
the overlap of lines is associated with a high energy cost and, therefore, we treat the lines
as non-crossing, corresponding to an infinite potential U (x) at x = 0. As a consequence,
only the identity permutation contributes to the sum in Eq. (1).
The free energy F [V (x, y)] = T ln ZN is itself a random variable. To obtain the
probability distribution function (PDF) of the free energy, we use the replica method.
Moments of the partition function are obtained by introducing n replicas of the original
system, described by line coordinates xj, (y), = 1, . . . , n, and then averaging over the
random potential. Assuming that V (x, y) is Gaussian distributed with zero mean and
spatially uncorrelated with variance

V (x, y)V x , y = x x y y ,
(3)
where [. . .] denotes impurity averaging, the average over the random potential yields

n
ZN
n
N
Z0nN
=
drj,
(N!)n
=1 j =1

Hn
,
Dxj, (y) exp
T
xj, (L)=r
j,
(4)
xj, (0)=rj,
with the replicated Hamiltonian

L
n
n
N

g dxj, 2
Hn = dy
+
U xi, (y) xj, (y)
2
dy
=1 j =1
=1 i<j
0

n
N

xi, (y) xj, (y) .
2T
(5)
,=1 i,j =1
The sum over permutations has been eliminated from Eq. (4) due to the non-crossing
condition. The impurity averaged moments of ZN are obtained from the time-independent
1 Z depends on the lattice constants of the discretization of the path integral. The partition function of the
0
interacting random line system will be expressed relative to Z0 [21].
483
Hamiltonian Hn at the cost of introducing a new attractive interaction proportional to

the variance of the random potential. This attraction can be understood as follows.
In the original random system the lines prefer to wander through regions of favorable
values of the random potential V (x, y). Due to the impurity average the system becomes
translationally invariant and there are no longer energetically favored regions of space
time. Instead, in calculating nth moments of ZN , the system of n copies of the lines
gains due to [exp(V (x, y))] = exp(/2) an energy /2T per crossing of two lines (of
different replicas), thus leading to an attraction.
The time-independence of the replicated Hamiltonian, and the choice of periodic
boundary conditions, allows us to write the partition function in terms of the symmetric
, . . . , x ; L) as an integral over all
statistical density matrix sym (x1,1 , . . . , xN,n ; x1,1
N,n
boundary positions of the lines at y = 0,

n
= Z0nN
ZN

N
n
dxj, sym (x1,1, . . . , xN,n ; x1,1, . . . , xN,n ; L).
(6)
j =1 =1
The density matrix itself can be written as an imaginary time Feynman path integral for
the world lines of Nn particles with fixed boundary positions x1,1 , . . . , xN,n at y = 0
, . . . , x
and x1,1
N,n at y = L [22]. The density matrix can alternatively be expressed in
coordinate representation as

, . . . , xN,n
;L
sym x1,1 , . . . , xN,n ; x1,1
=
sym

,
eqm Ei i (x1,1 , . . . , xN,n )i x1,1
, . . . , xN,n
(7)
with the mapping qm cl L = L/T , h cl1 = T between quantum mechanical and

classical parameters. 2 The sum runs over all symmetric states with energy eigenvalues Ei
and eigenfunctions i of the 1-dimensional quantum-Hamiltonian
2
= T
H
2g
n
n
n
N
N

2

+
U
(x
x
)
(xi, xj, ).
i
j
2
2T
xj,
=1 j =1
=1 i<j
,=1 i,j =1
(8)

From a classical statistical mechanics point of view, the operator exp(H /T ) is the
n ].
transfer matrix corresponding to the partition function [ZN
Using the representation of the density matrix in Eq. (7), we can perform the integrations
in Eq. (6) easily. Taking the thermodynamic limit qm L , the free energy Fn of the
replicated system is given by the ground state energy E0 of the quantum system in Eq. (8),
n
= T nN ln(Z0 ) + E0 L.
Fn = T ln ZN
(9)
Therefore, the statistical properties of the line lattice in the thermodynamic limit L
are dominated by the ground state wave function 0 , which must be of bosonic
2 Here and in the following, we set k = 1.
B
484
symmetry [22]. To obtain the disorder averaged free energy [F ] of the original line system
n ] 1)/n is used.
within the replica method usually the identity [ln ZN ] = limn0 ([ZN
n
However, calculating the average [ZN ] for all values n > 0 by analytic continuation of
E0 (n) yields much additional information about the PDF of the free energy beyond [F ].
n
] is the characteristic function of the random variable [ln Z], expansion with
Since [ZN
respect to n shows that
nN ln Z0
n
(n)j [F j ]c
E0 (n)L
= ln ZN
=
,
T
j!
Tj
(10)
j =1
where [F j ]c is the j th cumulant of the free energy. This identification relies on the
possibility of an analytic expansion around n 0. That this is indeed the case for the
line system considered here will be further discussed below.
Determining the PDF of the free energy is thus reduced to calculating the ground state
energy of the quantum-mechanical system in Eq. (8). Before we can proceed with the
calculation of E0 (n), we have to specify the pair potential U (x). We will consider the
limit where the characteristic length scale of the short ranged pair potential goes to zero,
leaving just a non-crossing condition for the lines. This replacement of the interaction
potential by a pure geometric constraint for the line configurations does not change the
physical mechanisms involved in the non-trivial characteristics of the line system. The
crucial competition between the elasticity of single lines, the tendency of lines to follow
energy valleys of the random potential, and the restriction of configurations by adjacent
lines are preserved. Within the quantum-mechanical description of the lines, there is a
convenient way to implement the non-crossing condition: as noted first by Pokrovsky and
Talapov [14] for pure systems, the Pauli exclusion principle can be used to represent the
elastic lines as world lines of fermions, which automatically avoids crossings. We thus
look for a ground state wave function 0 with a suitable antisymmetry under particle
permutations, which describes Nn particles with just an attractive interaction on contact
between any two particles. It is interesting to note that we change the symmetry of the
originally bosonic ground state function by hand by applying the exclusion principle.
However, the ground state energy we are interested in is the same both in the fermion
and the impenetrable boson description. Introducing the new subscript i = (i 1)n +
instead of (i, ) to number consecutively all nN particles, the Hamiltonian (8) can now be
written as
nN
2
= h
H
2m
i=1

2
2c
(xi xj ),
0
xi2
i<j
(11)
with mass m = g and interaction amplitude c0 = /2T . The explicit distinction between
lines in the same or different replicas has disappeared from the Hamiltonian, but is
hidden in the required transformation properties of the wave function under particle
permutations. The appropriate symmetry of the ground state wave function is restricted
by two conditions: (i) 0 must be antisymmetric if two of the N particles in the same
replica (color) are interchanged (Fermi statistics). (ii) 0 must be symmetric if two of
485
the n different replica (colored) particles corresponding to the same line are exchanged
since replicas are equivalent. Due to its antisymmetry, 0 tends to zero if two particles
of the same color approach each other and the attractive delta-function potential becomes
inactive. This guarantees automatically that the impurity averaging results effectively in
an attractive interaction on contact between particles of different color only. Therefore, the
sum over pairs of lines interacting via a delta-function in Eq. (8) yields no contribution if
= , resulting in the sum over all pairs in Eq. (11). To conclude, the original quantum
Hamiltonian (8) with U (x) replaced by the non-crossing condition for lines belonging to
the same replica is equivalent to the Hamiltonian (11), if one solves for a ground state wave
function of appropriate symmetry.
3. The Bethe ansatz

In one dimension, a number of interacting quantum systems can be solved exactly by
the Bethe ansatz [23,24]. Lieb and Liniger were the first to apply the Bethe ansatz method
to a continuum quantum field theory by solving for the ground state energy and the excitation spectrum of the one-dimensional interacting Bose gas with repulsive delta-function
pair potentials [20]. While their Hamiltonian is exactly given by Eq. (11), the ground state
wave function that we are looking for has a more complicated structure since it has to describe fermions of n different colors. The case n = 2 corresponds to spin- 21 fermions and
was solved some time ago by Gaudin [25] for attractive interactions, and by Yang [18] for
repulsive interactions. Choosing n = 3 colors, we obtain a model for nuclear matter with an
attractive quarkquark interaction, which has been discussed in connection with quantum
chromodynamics and also studied by Bethe ansatz methods [26]. Since we are interested
in an analytical continuation of the ground state energy for small n 1, first we find a
solution for integer n and then continue it analytically. According to the symmetry requirements discussed above, the allowed spatial wave functions of the U (n) fermions have to
transform like the irreducible representation of the symmetric group SnN corresponding
to the Young diagram [nN ], which consists of N rows of equal length n. Sutherland has
constructed a general solution of the eigenvalue problem for Hamiltonian (11) with c0 < 0
when the wave function transforms like an arbitrary irreducible representation of the symmetric group [27]. The method involves a nested series of Bethe anstze and leads to coupled integral equations for momentum densities. The same method has been applied later
to the attractive case by Takahashi [19] and Kardar [8], leading to considerable simplifications in the solution of the Bethe ansatz equations. Before we turn to the interesting regime
n 1, we explain briefly the idea of the method of nested Bethe anstze, in a simplified
version for the relevant irreducible representation [nN ] with integer values n 1.
3.1. Bethe ansatz equations for integer n
The eigenstates of the Hamiltonian (11) are given by Bethes hypothesis [23]. For
0 < xQ1 < < xQnN < W , with W the transversal system size, we can write the wave
function as
486

nN

=
[Q, P ] exp i
Pj xQj .
P
(12)
j =1
Here Q and P are permutations in SnN , and the numbers [Q, P ] form a (nN)! (nN)!
matrix. In the following we denote the columns of this matrix by the vector P = [ , P ].
Next, we define the permutation operators Plm , which act on the P . Their effect is to
interchange the Ql component with the Qm component of the vector P , i.e., Plm [Q, P ]
[Q , P ] where the new permutation Q is defined by Q l = Qm , Q m = Ql , and Q j = Qj
for j = l, m. These operators form a (nN)!-dimensional representation of the symmetric
group SnN . But this representation is reducible, i.e., for a wave function of a given
symmetry, many coefficients [Q, P ] are identical. Therefore, in the following we consider
implicitly only the subspace of the original P -space which corresponds to the irreducible
representation [nN ] of SnN . On this subspace the operators Plm act as matrices of the
dimension corresponding to the irreducible representation.
The requirement of a continuous wave function, which has cusps wherever two of its
coordinates are equal in order to compensate the delta-function interaction in Eq. (11),
imposes relations between the coefficients [Q, P ]. In terms of two vectors P , P
corresponding to permutations P and P , respectively, which differ by a transposition of
i and j , i.e., Pl = i = Pm , Pm = j = Pl , and Pj = Pj for j = l, m, the relations can be
written as
P = Yijlm P ,
(13)
with the operator

Yijlm =
i j
ic
+
Plm ,
i j ic i j ic
(14)
and the rescaled interaction strength c = 2mc0 /h 2 = g/T 3 . If these relations are
fulfilled, the wave function (12) is an eigenstate of the Hamiltonian (11) for any set of
momenta i . The allowed momenta are then restricted by imposing periodic boundary
conditions on the wave function. Expressed in terms of I corresponding to the identity
element I of SnN , the conditions force I to be simultaneously an eigenvector of the nN
matrix equations
eij W I = Xj +1 j Xj +2 j XN j X1 j Xj 1 j I ,
ij
(15)
with j = 1, . . . , nN and Xij ({j }) Pij Yij . The j which solve these equations depend,
of course, on the chosen irreducible representation.
So far, we have considered only the spatial part (12) of the full fermionic wave function.
The full wave function is given by the product of (12) and the color wave function .
The product has to be antisymmetric under simultaneous interchange of both position
and color of two particles. To guaranty this property, the color wave function has to
transform like the conjugate irreducible representation [N n ]. One can think of the basis
vectors of this representation as all the allowed color sequences of length nN which form
by linear combination the function . Now, the crucial point of Yangs and Sutherlands
487
method for determining the momenta i is to solve the matrix equations in color space
by regarding the color wave function as a spatial wave function describing (n 1)N
distinguishable particles on a discrete cyclic chain with N identical vacancies. By making
a generalized Bethe ansatz introduced by Yang, this new problem can be cast into matrix
equations identical in form to the original one, but corresponding to the lower dimensional
representation [N n1 ] for the color wave function . This procedure is then continued
until the problem has been reduced to one for particles of a single color only. To be more
concrete, the periodic boundary conditions for read

j = Xj +1 j Xj +2 j XN
j X1 j Xj 1 j ,
(16)
where
j = eij W ,
(17)
and Xij ({j }) is obtained from Xij ({j }) by the replacement Pij Pij . This change
of sign for the permutation operator comes from the fact that in general two representation
matrices of a permutation P corresponding to the original and conjugated representation,
respectively, differ by the parity of P only. To solve Eq. (16) we use the generalized Bethe
hypothesis of Yang for on the discrete cyclic chain. It consists in the ansatz

(2)

[Q, P ]G (2)
=
(18)
P1 , yQ1 G P(n1)N , yQ(n1)N ,
P
where the integers 1 yQ1 < < yQ(n1)N nN , denote the coordinates of the
(2)
distinguishable particles, and (2)
1 , . . . , (n1)N , is a set of unequal complex quasimomenta. P and Q are now elements of S(n1)N . The coefficients [Q, P ] form now
a (n 1)N (n 1)N matrix, the columns of which we denote by P . Again, we
lm that act on so that they interchange Ql and Qm .
define permutation operators P
P
By considering a suitable subspace of the P -space, these operators are chosen to form
the irreducible representation [N n1 ] of S(n1)N , in order to assure that the color wave
function has the appropriate symmetry. Physically, the number associated with a
particular set {y1 , . . . , y(n1)N } and permutation Q can be interpreted as the amplitude
for finding a particular arrangement of colors for the nN particles. Indeed, the positions
of all the particles of one color, say red, determine the yi of the remaining particles, and
the color arrangement of these remaining (n 1)N particles can be identified with the
permutation Q. To complete the ansatz, we have to specify the function G(, y),
G(, y) =
y1

j =1
i(j ) + c/2
.
i(j +1 ) c/2
(19)
The requirement that this ansatz for is a simultaneous eigenvector of the nN matrix
equations (16) imposes relations between the P , as the original eigenvalue problem for
the Hamiltonian (11) does for the P . These relations can again be written as
P = Y lm P ,
ij
(20)
488
with the operator

Y lm =
(2)
ic
ij
(2)
(2)
i j + ic
(2)
i j
(2)
(2)
i j + ic
lm ,
P
(21)
and P , P defined as above Eq. (13). Finally, we have to impose periodic boundary
conditions on . Using the relations (20), we obtain for the vector I the (n 1)N matrix
equations
nN
(2)

i(l j ) c/2
(2)
j =1 i(l j ) + c/2

I = Xl+1
l X(n1)N l X1 l Xl1 l I ,
(22)
for l = 1, . . . , (n 1)N , and with

Xij
(2) (2)
(2)
j ic (2)
ij Y ij = i
Xij l
l
P
.
ij
(2)
(2)
i j + ic
(23)
Substituting Eq. (23) in Eq. (22), we obtain new matrix equations, which parallel in form
precisely the original equations (16) for the color wave function, but with operators Xij , in
(2)
which the quasi-momenta l

read
have replaced the original momenta l . The new equations

l I = Xl+1
l X(n1)N l X1 l Xl1 l I ,
(24)
with the new set of l defined via

l
(n1)N
i(l m ) c
m=1
(2)
i((2)
l m ) + c
(2)
(2)
nN
(2)

i(l j ) c/2
j =1
i((2)
l j ) + c/2
(25)
Thus Eq. (16) is reduced to the lower dimensional Eq. (24), and the eigenvalue j in
Eq. (17), which is obtained from the ansatz (18), is
j =
(n1)N
i(j m ) + c/2
m=1
i(j (2)
m ) c/2
(2)
(26)
To reduce Eq. (24) further and to determine the eigenvalue l , we apply the procedure
described above again in the representation [N n2 ], introducing the quasi-momenta
(3)
j . This process terminates after n 1 iterations and yields Sutherlands Bethe ansatz
equations for the Nn(n + 1)/2 complex momenta ()
i , with = 1, . . . , n, and i =
(1)
1, . . . , (n + 1 )N . Here we have denoted the original momenta j by j .
Comparing Eq. (17) with Eq. (26), we obtain
e
(1)
ii W
(1)
(2)
(n1)N
i(i j ) + c/2
j =1
(2)
i((1)
i j ) c/2
with (i = 1, . . . , nN).
(27)
489
Continuing the iteration, we obtain from Eq. (25) with l given by Eq. (26), with j and
(2)
m replaced by corresponding higher order quasi-momenta, the set of equations
mN i((nm+2) (nm+1) ) c/2

i
j
j =1
(nm+2)
i(i
(nm+1)
) + c/2
(m1)N
i((nm+2)
(nm+2)
)c
i
l
l=1
i((nm+2)
(nm+2)
)+c
i
l
(m2)N
i((nm+2)
(nm+3)
) + c/2
p
i
p=1
i((nm+2)
(nm+3)
) c/2
p
i
with
i = 1, . . . , (m 1)N
m = n, . . . , 3
(28)
Finally, the iteration ends at the representation [N], which corresponds to Xij = 1 in
Eq. (24). Therefore, this equation is trivially fulfilled with l = 1, corresponding together
with Eq. (25) to
2N i((n) (n1) ) c/2

i
j
(n)
(n1)
) + c/2
j =1 i(i j
N

i((n) (n) ) c
i
(n)
(n)
l=1 i(i l ) + c
with (i = 1, . . . , N).
(29)
()
In the following, we will make a simple ansatz [8,19] for the complex momenta i ,
which solve Eqs. (27)(29) in the limit W , leaving only one set of N independent
equations for the N different real parts of the original momenta i (1)
i . To motivate
this ansatz, we consider for the time being the case of a single line (N = 1) coming in
n colors. This corresponds to the representation [n] for n bosons without spin. Under the
attractive interaction they form a bound cluster, which is described by complex momenta
= ic(n + 1 2)/2 ( = 1, . . . , n) forming a so-called n string [24]. Switching back
to the general case of N lines, but now for vanishing disorder strength so that c = 0, the
problem becomes that of N free fermions without spin (n = 1) represented by the Young
diagram [1N ]. In this case, the momenta are equally spaced along the real axis between
kF and kF = , where = N/W is the density of particles. With this two limiting
cases in mind, it is reasonable to make the general ansatz for large W
(n)
(n)
j = kj + iBj,1
(j = 1, . . . , N),
(30)

m
(nm)
+ iBj,
= kj + ic 1 +
2
(j = 1, . . . , N; = 1, . . . , m + 1; m = 1, . . . , n 2),
(31)

n1
(1)
(1)
i=(j 1)n+ = kj + ic 1 +
+ iBj, (j = 1, . . . , N; = 1, . . . , n),
2
(32)
(nm)
i=(j 1)(m+1)+
(nm)
where the kj are real valued momenta and the correction terms Bj,
will be proven
to vanish in the limit W below. In constructing this ansatz we start with the quasimomenta (n)
i , which were introduced in the last iteration step of the nested Bethe ansatz.
490
In this last step the original problem of nN interacting particles was completely reduced
to that of N colorless cluster fermions. Therefore, in analogy to the usual fermionic case
discussed above, we make the ansatz that the quasi-momenta (n)
i correspond to the real
valued cluster momenta kj . But due to the complex structure of the effective clustercluster
interaction, the momenta kj are, of course, no longer homogeneously distributed. To get a
physical picture of the internal structure of a single cluster of fermions, we think of it as
an n string composed of n bosons which are hold together by the attractive interaction.
From the above discussion of the N = 1 case we know that the internal boson momenta
of the cluster are equally spaced along the imaginary axis. Therefore, the single cluster
wave function is localized in space within a range of size 1/c = T 3 /g. The anstze (30)
(32) show that also for a finite cluster density in the limit W , the overlap of clusters
can be neglected, particularly with regard to the internal structure of the clusters. Thus each
clusters momenta are simply given by the momenta of an unperturbed n string plus the
(1)
collective real valued cluster momentum kj . This leads to the nN original momenta i of
Eq. (32). The intermediate quasi-momenta of Eq. (31) can be understood as that of smaller
sub-clusters which appear during the iterative scheme of the nested Bethe anstze.
We now prove that the anstze (30)(32) indeed solve Eqs. (27)(29) in the limit
W , leaving a new set of only N equations to determine the kj s. In substituting
(nm)
only in factors which
the ansatz in Eqs. (27)(29) we keep the correction terms Bj,
would become zero otherwise. The resulting equations for the correction terms have two
important properties: (i) From the first set of nN equations (Eq. (27)) one observes that the
(nm)
Bj,
have to vanish exponentially fast if W . (ii) The solutions depend formally
on the real momenta kj , which are still free parameters. In the following we will show that
these kj have to fulfill necessarily N independent equations in order to make the equations
(nm)
for the Bj,
consistent and thus solvable.
Multiplying the n equations of Eq. (27) with the ansatz substituted and i = (i 1)n +
with i fixed and = 1, . . . , n, we obtain
eiki nW = (1)n1
n1

=1
(1)
(2)
N n1

i(ki kj ) + c
Bi,
Bi,
.
(1)
(2)
i(ki kj ) c
B
B
i,+1
(33)
i, j =1 =1
j =i
(1)
(2)
In the next step we eliminate all correction terms Bi, , Bi, from this equation without
calculating them explicitly. Thus we substitute the ansatz into the set of Eq. (28). The
cluster momenta kj drop out and we get for m = 3, . . . , n; i = 1, . . . , N , and =
2, . . . , m 2, the new set of equations
(nm+2)
Bi,
(nm+2)
Bi,+1
Bi,
Bi,
(nm+1)
(nm+1)
(nm+2)
(nm+2)
(nm+2)
(nm+3)
Bi,1
Bi,
Bi,
Bi,
= (nm+2)
(nm+2)
.
(nm+2)
(nm+3)
Bi,
Bi,+1
Bi,
Bi,1
(34)
In the special case = 1 we have
(nm+2)
(nm+1)
Bi,1
Bi,1
(nm+2)
(nm+1)
Bi,1
Bi,2
(nm+2)
(nm+3)
Bi,1
Bi,1
(nm+2)
(nm+2)
Bi,1
Bi,2
(35)
491
and similarly for = m 1,
(nm+2)
(nm+1)
Bi,m1
Bi,m1
(nm+2)
Bi,m1
(nm+1)
Bi,m
(nm+2)
(nm+2)
Bi,m1
Bi,m2
(nm+2)
Bi,m1
(nm+3)
Bi,m2
(36)
Now we take for fixed m {3, . . . , n} and i {1, . . . , N} the product of the m 1
Eqs. (34)(36). Introducing the following ratio of correction terms,
Bi (m) =
m1
(nm+2)
(nm+1)
Bi,
Bi,
=1
(nm+2)
(nm+1)
Bi,
Bi,+1
(37)
we obtain the simple recursion relation

Bi (m) = Bi (m 1).
(38)
From the definition of Bi (m) it is easy to realize that the coefficient of the right-hand side
of Eq. (33) is given by (1)n1 Bi (n) = Bi (2). Finally, to obtain Bi (2) we substitute the
anstze of Eqs. (30)(32) into Eq. (29) keeping again the correction terms only in factors
which would become zero otherwise. For i = 1, . . . , N , we thus obtain
Bi (2) =
(n)
(n1)
(n)
(n1)
Bi,1 Bi,1
Bi,1 Bi,2
= 1.
(39)
This result shows that the correction terms dependent coefficient of Eq. (33) has to be one
(nm)
. This in turn determines the
in order to have a consistent set of equations for the Bj,
allowed real parts kj of the original momenta i (1)
i . They have to fulfill Eq. (33), which
now reads
einkj W =
N n1

i(kj kl )/c +
i(kj kl )/c
(j = 1, . . . , N).
(40)
l=1 =1
l=j
These are the final Bethe ansatz equations 3 which have to be solved to obtain the ground
in Eq. (11) in the thermodynamic limit, as
state energy of the quantum Hamiltonian H
E0 =
nN
N

nh 2 2
h 2 2 h 2 c2
n 1 n2 N +
i =
kj .
2m
24m
2m
i=1
(41)
j =1
In the last equation, we have used Eq. (32). The Bethe ansatz Eq. (40) has been analyzed
in Ref. [8] to obtain the free energy of the line lattice to leading order in the density =
N/W . The central result of this reference is an integral equation for integer n, which
was obtained from Eq. (40) in the limit W, N with fixed, by using the fact that
[W (kj +1 kj )]1 becomes a continuous function in that limit. By taking the limit n 0,
and subsequently the limit 0, the kernel of the integral equation becomes of Hilbert
3 The same equations have been obtained for the case n = 3 in the context of quantum chromodynamics [26]
by a more direct approach which does not use Sutherlands nested series of Bethe anstze. Instead it was a priori
assumed that the overlap of clusters can be neglected in calculating the momenta in the thermodynamic limit.
492
type, thus enabling an analytic solution [28]. To obtain the probability distribution function
of the free energy, we have to determine the ground state energy as a function of n for
small but finite arguments. Since extracting this information by an exact method from the
integer n integral equation of Ref. [8] is not obvious, here we start with a direct analytical
continuation of the Bethe ansatz Eq. (40) to non-integer n.
3.2. Analytic continuation in n
In what follows, we show how to evaluate the right-hand side (rhs) of Eq. (40) for noninteger values of n. The product over the replica index represents simply the extension
of a factorial to complex numbers. Therefore, we can make use of the recursion relation
for the complex gamma function, 4(z + 1) = z4(z), to obtain for real valued k the relation
n1

=1
ik/c +
4(1 ik/c)4(n + ik/c)
= (1)n1
ik/c
4(1 + ik/c)4(n ik/c)
= (1)n
4(ik/c)4(n + ik/c)
,
4(ik/c)4(n ik/c)
(42)
which is well defined for all n. Using this representation of the finite product, and taking
the logarithm of both sides of Eq. (40), the proper extension of the Bethe ansatz equations
to real valued n reads

N

4(n + i(kj kl )/c)4(i(kj kl )/c)
nkj W = i
(43)
ln
,
4(n i(kj kl )/c)4(i(kj kl )/c)
l=1
l=j
where some care has to be taken to conserve the property that the ground state has a total

momentum of zero, i.e., N
j =1 kj = 0. To express the argument of the logarithm in terms
of more elementary functions, we rewrite it as the infinite product
4(n + ik/c)4(ik/c) m2 + nm + (k/c)2 + ink/c

=
.
(44)
4(n ik/c)4(ik/c)
m2 + nm + (k/c)2 ink/c
m=0

z/m of
Here we have used the product representation 1/ 4(z) = ze z
m=1 (1 + z/m)e
the gamma function with the Euler constant [29]. Next we write the last expression as a
product of exponential factors to compensate the logarithm in Eq. (43). Using arctan(z) =
1
2i ln((1 + iz)/(1 iz)) we obtain

nk/c
4(n + ik/c)4(ik/c)
=
exp 2i arctan
.
(45)
4(n ik/c)4(ik/c)
m2 + nm + (k/c)2
m=0
This result shows that the rhs of Eq. (43) is indeed a real valued expression, and the final
Bethe ansatz equation for arbitrary real n becomes

N

kj kl
gn
,
nkj W =
(46)
c
l=1
l=j
493
Fig. 1. Both the function gn (x) (solid curve) and its limiting form 2 arctan(n/x) = sgn(x)
2 arctan(x/n) (dashed curve) for x 1, n 1 show a jump of 2 at x = 0.
with the function

gn (x) = 2

m=0

arctan
nx
.
m2 + nm + x 2
(47)
This is the central result of this section. The function gn (x) has no poles for real valued x, is
bounded between and converges to n for x , see Fig. 1. Notice that the jump
of 2 at x = 0 is produced solely by the m = 0 term in Eq. (47), whereas the remaining
terms are continuous.
This Bethe ansatz equation is valid for all n and for all line densities . To compare
it with results already existing in literature we consider two limiting cases. For n = 1
we have gn (x) = sgn(x), where sgn(x) is the sign function. The solution of Eq. (46)
then simply consists of momenta kj which are equally spaced between and for
W , N . This is the free fermion situation corresponding to lines which wander due
to thermal fluctuations only. Notice that with only one color of lines (n = 1) present,
the disorder induced inter-color interaction is, of course, inactive. The same situation is
expected to appear in the zero disorder limit c 0. Indeed, in this limit gn (x 1/c
) n sgn(x) and the ns on both sides of Eq. (46) cancel, leading again to the
free fermion result. The solution of Eq. (46) in the limit n 0 gives the disorder averaged
free energy of lines in a random environment following Eqs. (10) and (41). Expanding
gn (x) to first order in n, we get from Eq. (46)
kj W =
N

l=1
l=j

(kj kl )
c
+ coth
,
kj kl
c
(48)

2
2 1 of
where we have used the representation coth(x) = 1/x + 2x
m=1 (m + x )
the hyperbolic cotangent. It is easily seen that this equation corresponds to the discrete
494
momenta version of the integral equation derived by a different technique for treating
the n 0 limit in Ref. [8]. In this reference the corresponding integral equation is
solved in the dilute limit /c 0 yielding a free energy with steric interaction between
lines proportional to 2 (instead of the behavior 3 due to thermal fluctuations). From
Eq. (48), the free energy can be obtained at all densities [30].
Now consider the thermodynamic limit N , W of Eq. (46), while keeping
the density = W/N fixed. The employed technique is analogous to that of Lieb
and Liniger [20], using the fact that the function 7(kj ) = [W (kj +1 kj )]1 becomes
continuous in this limit. The limiting behavior gives the density of states in the sense that
W 7(k) dk = number of allowed k-values in the interval [k, k + dk]. For a given density ,
the boundary (Fermi) momentum K kN = k1 is determined by the condition
K
7(k) dk = ,
(49)
while the ground state energy is given by

h 2 nW
h 2 c2
n 1 n2 W +
E0 =
24m
2m
K
k 2 7(k) dk.
(50)
Calculating the differences of Eq. (46) between all adjacent momenta kj +1 and kj and
expanding with respect to the difference kj +1 kj , yields an integral equation for 7(k),

K
nk =
K
gn
k k
7 k dk ,
c
(51)
for all k [K, K]. This integral equation is not amenable to a closed form solution
because of the complicated form of its kernel gn (x). However, the kernel is non-singular,
and thus a numerical treatment seems feasible, which we leave for a future publication [30].
Instead, in the next section we will show that taking the limit of small /c leads to an
interesting simplification which paves the way for a description of the replicated lines in
terms of a pure, weakly interacting Bose gas.
4. Mapping to a weakly repulsive Bose gas

In this section, we develop a novel mapping of the dilute disordered line lattice onto a
weakly interacting pure Bose gas. This mapping then provides a valuable tool to obtain
exact results for the probability distribution of the free energy in the important limits of
low density or high disorder strength c. This is the relevant limit for the critical
behavior at the transition to a line-free system, e.g., for flux lines expelled at Hc1 from
superconductors or for domain walls at the commensurateincommensurate transitions in
absorbed layers or charge density waves.
495
From Eq. (49) one observes that the cutoff momentum K goes to zero in the limit of
0. Therefore the behavior of gn (x) for x 1 determines the solution of the integral
Eq. (51) in the limit /c 1. This limit has to be analyzed very carefully since we have to
keep the n dependence of gn (x) to all orders to obtain a complete expansion of the ground
state energy E0 in n, which is exact to first order in /c 1. Expressing the arctan of the
m = 0 part of Eq. (47) in terms of its inverse argument, and expanding the remaining terms
in x, we obtain

2n
x
+x
+ O x3 .
gn (x) = sgn(x) 2 arctan
(52)
n
m(m + n)
m=1
Note that we are interested in an expansion of the ground state energy E0 around n = 0.
Independent on how small the ratio /c and therefore the argument of gn (x) is, the final
expansion in n for fixed /c tests arbitrarily small values of n and thus can render the
arctan(x/n) in Eq. (52) of order one. Therefore, an expansion of this term is not justified.
In contrast, the remaining terms with m 1 in Eq. (52) tend always to zero for both x 1
and n 1 and can be safely neglected. Thus, retaining the first two terms of Eq. (52) only
(see Fig. 1) and taking the derivative of Eq. (51) with respect to k, we obtain the integral
equation
K
K
(nc)2

2nc
7 k dk = 27(k) n,

2
+ (k k )
(53)
which yields the complete n dependence of 7(k) to first order in /c exactly. The crucial
observation is that this equation is identical to that obtained by Lieb and Liniger for their
exact Bethe ansatz solution of the one-dimensional Bose gas with repulsive delta-function
interactions [20].
To obtain a physical understanding of the relation between the replicated line lattice
and the Bose gas, we change variables as follows: Define the function 7b (k) = 7(k/n)/n,
the cutoff momentum Kb = nK, the mass mb = nm and the rescaled interaction strength
cb = n2 c. With this new variables Eqs. (53), (49) read
Kb
Kb

2cb
7b k dk = 27b (k) 1,
2

2
cb + (k k )
Kb
7b (k) dk = .
(54)
Kb
The bosonic ground state energy, i.e., the second term of Eq. (50) becomes
h 2 W
E0,b =
2mb
Kb
k 2 7b (k) dk.
(55)
Kb
These equations represent the exact ground state (with energy E0,b ) of a Bose gas with
repulsive interactions, described by the Hamiltonian
N
2

2
b = h
H
+
2c
(xi xj ),
0,b
2mb
xi2
i=1
i<j
(56)
496
Table 1
The usual correspondence of the parameters of a (d 1)-dimensional Bose gas and a d-dimensional
lattice of lines without a random potential. The interaction potential between vortex lines maps to the
Boson pair potential. In d = 2 dimensions, non-crossing lines are modeled by hard core bosons or,
equivalently, free fermions
Model
Bose gas
Non-random lines
Mass
Plancks constant
Inverse temperature
mb
g
h
T
h qm
L
Table 2
The correspondence of parameters in the novel analogy between the one-dimensional Bose gas
and the two-dimensional replica theory of lines. The density of bosons and (non-replicated) lines,
respectively, is in both descriptions given by
Model
Bose gas
Replicated lines
Mass
Plancks constant
Inverse temperature
Contact interaction
Effective interaction
mb
nm ng
h
T
h
qm
L
c0,b (repulsive)
nc0 n/2T (attractive)
cb 2mb c0,b /h 2
n2 c n2 g/T 3
for N with fixed density of bosons [20]. Due to the definition of c = 2mc0 /h 2 the
amplitude of the interaction is given by c0,b = nc0 . The correspondence can be summarized
as follows: The exact ground state solution of the quantum Hamiltonian in Eq. (11) for
attracting particles, each appearing in n colors, is in the dilute limit /c 1, and to all
orders in n around n = 0, the same as the ground state solution of the Bose gas Hamiltonian
of Eq. (56) with repulsive interactions. For comparison with the usual mapping between
elastic lines and world lines of bosons for pure systems, both the correspondence of
parameters for the pure case and the novel analogy of quantities in the random system
are summarized in Tables 1 and 2.
The correspondence between the Bose gas and the replica theory of lines can be
interpreted physically as follows: As we have seen in Section 3.1, the Bethe ansatz solution
can be interpreted as n lines of different colors forming a cluster due to the pair attraction
of amplitude c0 . This suggests considering these clusters as new particles of mass nm =
ng, then appearing as the components of the Bose gas described by Eq. (56). To obtain
insight in the effective strength of the contact interaction between these composite bosons,
consider two clusters, each composed of n lines of different colors. Since particles of the
same color avoid each other like free fermions or, equivalently, bosons with hard core
repulsion, some inter-cluster interactions are screened, see Fig. 2. There are n! orderings
of particles within a cluster, each ordering providing the same number of pair interactions
given by 1 + + (n 1) = 12 n(n 1) due to screening. Finally, we have to take into
account that in the quantum Hamiltonian (11), which corresponds to the replica theory of
lines, the interaction sum runs implicitly for two given clusters over all different colors
= instead of running over color pairs < only. Therefore, the total amplitude of the
contact interaction between two clusters is given by n(n 1)c0 , which yields a repulsion
497
Fig. 2. The origin of the effective clustercluster interaction: For n = 2 there is for both configurations
only one inter-cluster interaction possible since the black particle of the first cluster is blocked by
the black particle of the second cluster from interacting with the dark gray particle. For n = 3 only
two typical configurations are shown. It is easy to convince oneself that each of the 3! configurations
allows for three interactions suggesting 1 + + (n 1) = 12 n(n 1) pair interactions for n-particle
clusters.
of amplitude nc0 in the limit n 1, in agreement with our findings (see Table 2). That
this naive estimate produces the correct result to first order in /c is presumably related to
the fact that the internal structure of the clusters does not matter in the dilute limit since
the distance between clusters is much larger than cluster size 1/c.
Having established the analogy between the replica theory of lines and the Bose gas,
one can take advantage of the many results for the one-dimensional Bose gas accumulated
during the past decades [20,3133]. In the following, we focus on the ground state energy
E0,b as given by Eq. (55) to obtain the free energy Fn of the replicated line system using
Eq. (9). The only dimensionless intensive variable in the Bose gas problem is = cb / =
n2 g/T 3 . Thus the ground state energy can be written as
h 2
(57)
W 3 e( ),
2mb
where e( ) is a dimensionless monotonically increasing function of [20]. Since we are
interested in the limit n 1, we have to consider the weak coupling limit of the Bose gas.
In the limit cb 0, the integral equation (54) has the diverging 4 solution [20]

1
Kb2 k 2 .
7b (k) =
(58)
2cb
This zero-order result corresponds to the n = 0 limit of 7(k), which is a well-defined
expression according to the mapping between the two models, and given by
1 T3 2
lim 7(k) =
(59)
K k2,
n0
2 g

with K = 2 g/T 3 . This result coincides with that obtained in Ref. [8] by solving an
integral equation which itself is valid for n = 0 only. Using Eqs. (54), (55) one gets Kb =
2 1/2 , e( ) = and, therefore, the non-trivial (disorder dependent) part of the disorder
averaged free energy of the lines is given by

1 g
E0 (n)
2
L=
[F ]d = lim
(60)
W L,
n0
n
2T 12 T 3
E0,b =
4 The divergence is related to the fact that as c 0 the kernel of the integral equation is a representation for
b
2 (k) leading to 2 7b (k) 1 2 7b (k). Therefore, 7b (k) has to grow unbounded in this limit.
498
using Eqs. (10), (50), (57), and the correspondences of parameters in Table 2. Note the
difference from the case without random potential, in which the non-linear term in [F ] is
proportional to 3 .
5. Distribution of the free energy in the dilute limit

The preceding section demonstrates how the lowest order term in the energy of the Bose
gas yields the first moment or cumulant of the free energy of the random line system in the
limit 0. As discussed in connection with a scaling hypothesis for PDFs of disordered
system [17,34], cumulants of the free energy are also singular at critical points. To calculate
these cumulants the complete n dependence of the replica free energy Fn is needed. Due
to the mapping between replicated lines and bosons, Fn can be obtained order by order
around n = 0 by perturbation theory for the Bose gas in the weak coupling limit cb =
n2 g/T 3 1. As realized by Lieb and Liniger [20], and as the zero-order solution (see
Eq. (58)) suggests, the complete solution of the integral equation (54) is highly singular at
cb = 0. The physical meaning of this singularity in the context of the line system is related
to the relevance of the random potential. Any randomness, however weak, leads to a glassy
state that is fundamentally different from the pure case.
The singular behavior of the integral equation seems to preclude any direct analytical
way of extracting higher order corrections to the zero-order result. Fortunately, for small ,
Bogoliubovs perturbation theory [35] can be used to calculate the ground state energy
of the Bose gas. One reason for trusting Bogoliubovs theory for small is that in the
case of a delta-function potential the ground state energy is given correctly by this theory
in the limit of high boson density [20]. Note that the important small n limit maps
onto the high density limit of the Bose gas as far as the coupling strength n2 /
is concerned, although we are studying the low density limit of the line system. Indeed,
agreement between the exact ground state energy obtained numerically from the integral
equation (54) and Bogoliubovs first-order perturbation theory has been confirmed for
small by Lieb and Liniger [20]. Later Takahashi [31] used the correlated basis function
method to calculate the second-order correction to the ground state energy, which again
coincides with the second-order energy in Bogoliubovs perturbation theory [36], and is
in perfect agreement with a numerical high-accuracy solution of the integral equation (54)
down to 105 [37]. Using the analytical results for first and second order terms, the
next two orders can be obtained by this numerical solution, and the ground state energy
summarized, in terms of the rescaled function introduced in Eq. (57), as

1
1
4 3/2
2 2 + b5 5/2 + b6 3 + O 7/2 ,
+
e( ) =
(61)
3
6
with numerical coefficients [37]
b5 = 0.001588,
b6 = 0.000171.
(62)
It is interesting to note that the ground state energy is non-analytic in the coupling strength
and thus has to be considered as an asymptotic expansion for small .
499
The replica free energy Fn of the line system can now be evaluated by applying the
mapping between bosons and lines. First, we note that Eq. (61) provides an explicit proof
that Fn is indeed an analytic function of n since n2 . While this property of Fn is
usually assumed in replica theories, it cannot be proved in most cases, and had not been
checked before for random line lattices. In terms of the cluster size
ld =
T3
1
=
,
c g
(63)
which sets the crossover length scale from pure to disorder dominated behavior for a single
line [8], the disorder dependent part of the replica free energy can be written as

LW n
4 2
1 n2 ld + n(ld )2
n (ld )3/2
Fn,d = 2
3
ld 2T 12

6
1
1
3
4
1/2
5
n ld + b5 n (ld ) + b6 n + O n ,
+
(64)
6 2
in the dilute limit ld1 . This exact result for the replica free energy has the scaling form

2
G n(ld )1/2 ,
(65)
T
where G(x) is a polynomial. This result is in agreement with dimensional arguments given
in Ref. [8], which motivated a scaling theory for cumulants of thermodynamic quantities in
random systems [17]. Thus this scaling theory is strikingly confirmed by our exact results.
The cumulants of the free energy of the line lattice at fixed density can be read off
from Fn . They can be written in the general form
Fn = LW n
Fp

c
= p
WL
T p2 (ld )(5p)/2,
ld2
(66)
with coefficients
3
1
4
1
1 = ,
2 =
,
3 = 2 ,
2
3
4
5 = 0.01026.
4 = 0.01906,
(67)
Consistent with the central limit theorem all cumulants of the free energy density vanish in
the thermodynamic limit as [f p ]c (LW )1p . However, for a large but finite system the
cumulants are finite and show a non-trivial dependence on the density, or on the chemical
potential within a grand canonical description. Close to the critical point we have
c and [f p ]c ( c )(5p)/2 . Therefore, as c + the first four cumulants
vanish continuously at the critical point, the fifth cumulant approaches a constant and
higher order cumulants show increasing divergence. 5
Of course, since we are discussing finite systems, the singular behavior is cutoff when
the correlation length approaches the system size. Due to the anisotropy of the line system,
5 A numerical solution of the integral equation (54) indicates that the coefficient b in Eq. (61) is negative
7
and about one order of magnitude smaller than b6 [37]. This suggests that the sixth and presumably all higher
cumulants have indeed non-vanishing contributions W L.
500
there are two correlation lengths. The first one is given by the mean distance between
lines, = 1/, whereas the second one is set by the mean longitudinal distance between
3/2
collisions of lines, = (gT /)1/2 . The critical regime corresponds to W ,
L. We consider a fixed ratio of longitudinal and transversal system sizes so that
they are in agreement we the anisotropic scaling of the system, i.e., L = (W/ )3/2 ,
allowing for an anisotropy parameter . In the critical regime, the moments of the free
energy density are then given by [f p ]c = p (/T W 2 )p yielding universal relative
cumulants
[f p ]c
(68)
= 2p p 1p ,
[f ]p
which depend only on the geometry of the system via the anisotropy parameter . This
result is a consequence of the relevance of disorder, and demonstrates the complete
destruction of self-averaging at the critical point, since otherwise the relative cumulants
vanish with increasing system size [34]. Numerical simulations for systems of appropriate
size ratios should be suitable to test this universality. Another interesting property of the
PDF of the free energy is its asymmetry. Since the third and fifth cumulant are both
negative, there is a higher probability for the system to be in a state with a free energy
which is smaller than the average value. This asymmetry gets more pronounced as the
system gets closer to the critical point.
Some comments on the above results for the PDF are in order. These comments are
of more general nature, and are given here in order to clarify the status of the replica
method itself. The replica approach provides a tool to calculate integer moments [Z n ] of
the partition function. However, Z itself is unphysical, and the desired PDF of ln Z has to be
deduced using analytical continuation of the moments [Z n ] to non-integer n, and applying
the cumulant series representation in Eq. (10). There are two potential problems related to
this kind of procedure. First, an unique PDF can be deduced from the integer moments of
the PDF only if the [Z n ] grow slower than n! for large n [38]. One might conclude that this
condition is not fulfilled for the random line system because of [Z n ] exp(E0 (n)L/T )
and E0 (n) has a contribution proportional to n(1 n2 ), see Eq. (50). The origin of this
contribution can be traced back to the energy of individual clusters in the Bethe ansatz.
Therefore, the same situation appears for the single line system. In this context, it has
been noted that the result for E0 (n) given above is not expected to be valid for arbitrarily
large n [39]. One is lead to this conclusion by the observation that the cluster size goes
to zero if n tends to infinity, reflecting the tendency of attractive bosons to collapse. The
issue of an attracting 1D Bose gas has been discussed briefly by Lieb and Liniger [20]
in connection with their Bethe ansatz solution. They conclude that for the attractive Bose
gas it is not clear that the Bethe ansatz wavefunction which includes the n-string solution
exhausts all solutions of the Schrdinger equation. When the cluster size ld goes to zero,
it seems to be no longer justified to construct the wavefunction as a product of only twoparticle functions exp(|x1 x2 |/ ld ), as done for the n-string, since they are extremely
localized. The second potential difficulty of the replica method has its origin in the order
in which the thermodynamic limit and the n 0 limit are taken. To apply the cumulant
series in Eq. (10), we must expand the analytically continued function ln[Z n ] around n = 0
501
for a fixed system size. However, in calculating [Z n ] in terms of the quantum system, first
we consider at fixed n the thermodynamic limit to eliminate all exited states. For a single
line in random media the two limits do not commute due to the non-trivial sample-tosample variation of the free energy [39], and the free energy cumulants have to be obtained
indirectly by a scaling ansatz for the replica free energy [17]. For the line lattice considered
here, the situation is different. Now, the extensive replica free energy Fn , obtained by first
taking the thermodynamic limit and then the limit of small n, has contributions at any order
in n, and thus the two limits are more likely to commute.
Finally, we would like to mention that our mapping between bosons and elastic lines
provides also new information about the weak coupling limit of the ground state energy
of the one-dimensional Bose gas. Since cumulants [F p ]c of even order p have to be
positive, all coefficients of non-integer powers of in the expansion of the ground state
energy in Eq. (61) have to be negative, see Eq. (10). This is an important constraint on the
various approximative methods in use for the Bose gas. Recently, different self-consistent
approximations have been used to calculate the ground state energy based on a so-called
local-field correction approach [40,41]. In view of the fact that the two approaches give
different signs for the 7/2 term in Eq. (61), our sign constraint rules out the solution of
Ref. [40] from being correct.
6. Correlation functions
The preceding study of the free energy of the line lattice shows that disorder induces
critical behavior which is distinct from the pure case. Moreover, disorder is expected to lead
to correlations in the system which are different from pure 2D line lattices. In the context
of flux lines in high-Tc superconductors, Fisher predicted a vortex glass phase in which
disorder locks the flux lines into one of many possible metastable states [42]. He showed
that the random line lattice can be mapped onto the two-dimensional random-field XY
(RFXY) model with vortices excluded. The correlation functions of the low temperature
phase of the RFXY model, and the vortex glass phase, and their collective excitations
have been studied extensively during the last decade. For the RFXY model, analytical
approaches include replica-symmetric renormalization group (RG) techniques [4345],
and one-step replica-symmetry breaking variational anstze [46] and RG methods [47,
48]. The analytical studies of the random line lattice are mainly based on RFXY model
results. They consist in scaling arguments [49], linear continuum elasticity models [50], RG
techniques [5153], and variational methods including replica-symmetry breaking [53] and
without use of replicas [54]. However, the results obtained within the different approaches
are not consistent. Whereas all variational methods predict for the continuum displacement
field u(x, y) of the lines, correlations of the form [(u(x, y) u(0, 0))2 ] ln(x 2 + y 2 ),
RG approaches yield additional corrections growing as ln2 (x 2 + y 2 ), which vanish as the
non-glassy high-temperature phase is approached. A critical comparison of the various
results is given in Ref. [55]. One should also keep in mind the range of validity of the
502
different approaches. The variational methods are expected to hold throughout the whole
glassy phase, while the RG results are applicable only close to the glass transition point.
Numerical simulations of glassy line lattices in two dimensions are also inconclusive.
For both the variational method predictions [56,57], and the RG results [5861], qualitative
agreement was found in simulations. The numerical calculations are either hampered by
slow glassy dynamics, and/or by a large length scale for the crossover to a regime where
differences between the two analytical methods become significant. More recently, Zeng
et al. [62] found good quantitative agreement, albeit in the vicinity of the glass transition,
between the RG results and their numerical studies of the correlation functions via a
mapping to a discrete dimer model with quenched disorder.
The analogy between the line lattice and the quantum model of U (n) fermions has
been used as an alternative to RG for obtaining information about correlations in the
glassy phase [15,16]. Refs. [15] and [16] both use the replica-symmetric Bethe ansatz, but
employ different techniques in handling the n 0 limit, and make different assumptions.
As a consequence, the two approaches yield different results for the asymptotic decay
of the line lattice densitydensity correlation function, i.e., [(x, y)(0, 0)] 2
(x 2 + y 2 )1 [15] compared to [(x, y)(0, 0)] 2 (x 2 + y 2 )1/2 [16]. Although the
decay exponent of the first result is in agreement with the variational anstze, the amplitude
calculated in Ref. [15] does not agree with them. Neither result shows the logarithmic
corrections predicted by the RG approaches.
In this section we reconsider the displacement and density correlations in the glassy
phase by exploring the mapping of the random line lattice to the pure Bose gas. While the
free energy of the former is related to the ground state energy of the latter, correlations
in the statistical system are related to the low energy excitations of the quantum problem.
Therefore, we calculate the low-energy excitations of the U (n) fermion model in the limit
n 0. In the following, we consider only replica-symmetric or color-less excitations,
which do not break apart the n-particle clusters corresponding to the bosons. This
restriction is motivated by the fact that for n > 1 only singlet states are expected to
be gap-less [63], due to the finite energy associated with breaking up a cluster. With
this assumption, the excitations above the ground state consist of cluster-particles and
cluster-holes. Under the mapping 6 of the replicated lines or U (n) fermions to the Bose
gas, these excitations become the usual particle and hole states studied by Lieb [20]. He
calculated the double spectrum associated with particle and hole elementary excitations
exactly by Bethe ansatz. Particle states can be excited with any momentum p, but the hole
state spectrum exists only for momenta p satisfying the condition |p| . Both spectra
have in common a linear dispersion relation at small momenta with the same velocity of
sound,

>p,h (p) = vs |p| + O p2 ,
(69)
where vs in the weak coupling limit is given by
6 That the mapping holds also for the excitation spectrum follows from the fact that the response of the Fermi
sea (shifts of the discrete wave vectors kj ) to excitations is completely determined by the phase shifts in an infinite
system associated with transposing two wave vectors of the Bethe wave function, see, e.g., Ref. [20].
h
1/2
vs =
=
mb
503

2c0,b
.
mb
(70)
Since under the mapping both c0,b n and mb n, the excitation spectra of the replicated
lines or U (n) fermions remain linear in the n 0 limit, and the velocity of sound is given
by

.
lim vs
(71)
n0
gT
A linear dispersion relation in the n 0 limit was also obtained in Refs. [15,16] by solving
directly a linear integral equation which determines the shifts of the Bethe ansatz momenta
kj under particle/hole excitations. However, in these references different assumptions
and approximations were made to go from the excitation spectra to the densitydensity
correlation function of the line lattice, leading to contradictory results.
Having the mapping to the pure Bose gas at hand, it is tempting to reanalyze the density
density correlations of the line lattice from a new perspective. Below we will show that the
mapping gives direct results only for certain correlations in the replicated system, from
which information about correlations in the original random system has to be deduced.
Therefore, the remaining part of the section is guided more by the question of how much
we can learn about the correlations in the line system from the knowledge of the density
correlation function of the Bose gas, rather then giving final exact results for the line
lattice correlations. The densitydensity correlation function of the 1D Bose gas at zero
temperature has been studied extensively by several methods in the past, see, e.g., Ref. [33].
Since the dispersion relation is linear for small p, conformal field theory (CFT) can be
employed to obtain the asymptotics of the correlation function. The decay exponent b
is given by the conformal dimensions, which are determined by the spectra of low-lying
excitations. For the class of Luttinger liquids, including the Bose gas, the CFT approach
is equivalent to Haldanes effective harmonic fluid description of 1D quantum fluids [32].
This description is based on the fact that the elementary excitations may be regarded as
bosons which represent long-wavelength density fluctuations. The Hamiltonian for these
bosons can be expressed as a sum of harmonic oscillators. Therefore, correlation functions
can be calculated by simple Gaussian averages. In particular, from the Luttinger liquid
approach and CFT, the asymptotic time-dependent correlation function of the local density
b(x, t) of the 1D Bose gas is known at any coupling strength cb as [32,64]

b x 2 vs2 t 2
cos(2x)
+A 2
+ h.h.,
b(x, t)b(0, 0) = 2
2
2
2
2
2
4 (x + vs t )
(x + vs2 t 2 )b /2
(72)
where the coefficient A is not known exactly and higher harmonics (h.h.) are not shown
explicitly. The coupling strength enters the exponent b only via the velocity of sound,
which is known perturbatively for small cb , and numerically for all cb , from the Bethe
ansatz [20]. This exponent also determines the amplitude of the homogeneous part of the
correlations, and is given by
b =
2 h
.
mb vs
(73)
504
We now examine to what extent the knowledge of the densitydensity correlation

functions of the Bose gas provides information about the random line lattice correlations.
First consider the correlation function of the replicated U (n) fermion system of mean
density n. Its local particle density r(x, y) is given by the sum over all densities (x, y)
of particles of a specific color , i.e.,
r(x, y) =
n
(x, y).
(74)
=1
Since all replicas or colors are equivalent, we obtain in terms of the density fluctuations
(x, y) = (x, y) , the relation

r(x, y)r(0, 0) = n2 2 + n (x, y) (0, 0) + n(n 1) (x, y) (0, 0) .
(75)
On length scales larger then the cluster size ld , see Eq. (63), the simple relation r(x, y) =
nb(x, y) between boson and U (n) fermion densities holds, since bosons correspond to
n-clusters of fermions. Thus, we obtain

1
b(x, y)b(0, 0) = 2 +
(x, y) (0, 0) (x, y) (0, 0)

n
+ (x, y) (0, 0) ,
(76)
where we have assumed that all off-diagonal correlations ( = ) are identical. The
disorder and thermally averaged densitydensity correlation function of the original line
system can be calculated by taking the n 0 limit of one of the n equivalent correlation
functions for particles of the same color,

(x, y)(0, 0) = lim (x, y) (0, 0) .
(77)
n0
Can we extract information about this correlation function from the Bose gas correlations?
Assuming that correlations of the density fluctuations (x, y) have a finite limit for n
0, it follows from Eq. (76) that we can expect two types of terms in the n 0 limit of
Eq. (72): Contributions diverging as 1/n, and those which saturate. Counting powers of
n shows that we cannot identify the diagonal correlations needed in Eq. (77), but only the
difference between diagonal and off-diagonal correlations. However, this difference can be
calculated both in the RG approach and the variational ansatz (VA), and is therefore of
interest, too. To check for the diverging contributions proportional to 1/n in Eq. (72), we
have to map the Bose gas exponent b to the corresponding exponent of the replicated
line lattice. Using the result of Eq. (71) for the mapped velocity of sound, h T and
mb ng, we get in the limit n 0 the diverging exponent

2 T 3 2
=
=
(78)
ld .
n
g
n
For the last expression, we have used the definition of the length scale ld given in
Eq. (63). Focusing on the homogeneous part of the line density correlations, we obtain,
by comparing the diverging part proportional to 1/n, the correlation function

505

(x/a)2 (y/ )2
2
(x, y) (0, 0) (0, 0) =
ld
+ o.t.,
2
((x/a)2 + (y/ )2 )2
(79)
where o.t. stands for omitted oscillating terms. The

length scales in the last expression are
a = 1/, the mean separation of lines, and = gT / 3 corresponding to 1/vs , the
mean longitudinal distance between collisions of lines.
Before we discuss the oscillating contributions of the correlation function, we would
like to compare our result for the difference correlations in Eq. (79) with the predictions
made by RG techniques and VA. Both, RG and VA are constructed on the basis of linear
elasticity theory, which describes the line lattice in terms of a scalar displacement field
u(x, y). To represent the coupling of the random potential to the line density in terms
of the displacement field, the line density in each replica has to be decomposed into
harmonics as [32,50,65]

(x, y) = 1 x u (x, y)

cos 2m x u (x, y) .
(80)
m=0
In this representation, higher order terms in x u (x, y) are usually ignored since u (x, y)
varies slowly on coarse-grained scales. By retaining only the coupling of the first harmonic
(m = 1) to the random potential, one obtains for the displacement field u(x, y) the
Hamiltonian of the RFXY model [42]. For this model, the RG approach [44] gives the
result

r
2 a2
T
2 2 r
x u (x, y)x u (0, 0) = 2 2
(81)
ln
+ ln
,
x 4 2J
a
a

2 a2
r
x u (x, y)x u (0, 0) = 2 2 2 ln2
(82)
,
x 4
a
where the ln2 contribution is obtained by an expansion to lowest order in = 1 T /Tg
which measures the distance from the glass transition at Tg = 4J . Here J is the elastic
constant of the (isotropic) RFXY model and r 2 = x 2 + (c11 /c44 )y 2 the rescaled squared
distance. J can be expressed in terms of the elastic constant c11 and c44 , which are related
to compressions and tilts of the line lattice, by [50,55]
a2
(83)
c11 c44 .
4 2
In contrast, VA [53] yields a simple logarithmic growth of displacements with = 0. The
amplitude of the logarithm in Eq. (81) then depends on whether one allows for replica
symmetry breaking or not. The replica symmetric result for the amplitude agrees with the
RG result in Eq. (81), as compared to the universal amplitude a 2 /2 2 , which is obtained
in case of replica symmetry breaking.
The relation between densitydensity correlations and displacement correlations follows
from Eq. (80) as

(x, y) (0, 0) = 2 1 + x u (x, y)x u (0, 0) + o.t.,
(84)
J=
where we have not written explicitly the oscillatory contributions which are generated by
harmonics of order m > 1 in the density decomposition. Using this relation, we can extract
506
displacement correlations from our Bethe ansatz result in Eq. (79). Comparing Eq. (79)
and Eq. (84), we deduce
2

2 a2
x
y 2 1/2
x u (x, y) x u (0, 0) x u (0, 0) = 2
(85)
ld ln 2 + 2
.
x 2
a

To complete the comparison with RG and VA, the elastic constants have to be determined
in the low density limit 1/ ld , where the above prediction holds. The tilt modulus is
given by c44 = g/a, whereas the compression modulus c11 is determined by the mutual
interaction potential W (x) of the lines. Due to entropic and steric repulsions, the original
contact potential becomes long-ranged. The interaction potential can be read off from
the free energy density f (a = 1/) since in a grand-canonical description with chemical
potential one has
g(/c 1) + W (a)
.
(86)
a
The elastic constant c11 is now determined by the curvature of the free energy density,
evaluated at the mean line separation a, as
f (a) =

(87)
.
T a2
To calculate the last expression, which is valid in the low density limit a ld , we have
used the result for the free energy in Eq. (60), which yields the non-trivial part W (a)/a
of the free energy density. From this result for the elastic constants we obtain J =
(ag/T )1/2 /4 2 , and hence the amplitude of the ln term of the RG result in Eq. (81)
becomes
a2
a2T
=
ld ,
(88)
8 3 J
2
which is in complete agreement with the Bethe ansatz result of Eq. (85). Moreover, there
is also agreement between the two results regarding the anisotropy parameter, since =
(gT / 3 )1/2 = a(c44/c11 )1/2 in the dilute limit. This means that the velocity of sound
is related to the elastic constants by the mapping between bosons and replicated lines
according to

1/2
c11
vs .
(89)
c44
c11 = x 2 f (x)|x=a = aW (a) =
It is interesting to note that exactly the same relation holds in the case of a pure (1 + 1)dimensional line lattice, which can be mapped to world-lines of free fermions [14] with
T /g.
vs = h/m
Compared to the VA, our Bethe ansatz result for the difference correlations [x u (x, y)
(x u (0, 0) x u (0, 0))] is only in agreement with the replica symmetric case.
A universal amplitude of the logarithm as predicted by the VA with replica symmetry
breaking is not reproduced by our approach. However, the agreement of the Bethe ansatz
result with the RG predictions is not complete. If the RG is assumed to give the correct
result, then there should appear also a slower decaying contribution proportional to
507
ln(r)/r 2 in the Bose gas correlations in Eq. (72) which remains finite in the n 0 limit.
This contribution corresponds to the last off-diagonal term in Eq. (76), for which this
slower decay follows from the RG result in Eq. (82).
There are two potential reasons for the absence of such kind of contributions in our Bethe
ansatz approach. The first one is not related to our mapping, but can be traced back to the
fact that there are indeed logarithmic corrections to Haldanes harmonic fluid approach
for the 1D Bose gas. These corrections vanish in the strong coupling limit cb and
have been calculated explicitly so far only in a 1/cb expansion by Korepin [66] using the
quantum inverse scattering method and by Berkovich and Murthy [64,67] by means of
conformal field theory. However, these results predict, in agreement with each other to first
order in 1/cb , logarithmic corrections only to the oscillating term of Haldanes result in
Eq. (72). Therefore these results do not account for ln contributions to the homogeneous
part of the densitydensity correlation function. But we are interested in the weak coupling
limit cb n2 0, where the actual form of the corrections is not yet known. The second
potential, and we think more likely reason for the absence of corrections is related to
the fact that we apply our mapping to obtain the excitation spectra of the U (n) fermions
from the Bose gas spectra by assuming that only non cluster breaking replica symmetric
excitations are gap-less. This assumption is certainly justified for integer n due to the finite
energy necessary to break up a cluster. But if n goes to zero, cluster breaking excitations
might become gap-less, providing a possible source of additional contributions to the
asymptotic densitydensity correlations. Note that these cluster breaking excitations can be
absolutely replica symmetric since n particles of mutual different colors can be transferred
from different clusters across the Fermi surface to form together a color neutral excitation.
Taking into account this type of excitations is, of course, very interesting, 7 but not possible
within the mapping to bosons, which itself cannot be broken apart.
Finally, we come back to the oscillating term in the Bose gas correlation function of
Eq. (72), and its implications for the correlations in the line lattice. This term and higher
harmonics are quite important in view of experimental determinations of the translational
order of line lattices. Scattering experiments measure the Fourier transform of the density
density correlation function around a reciprocal lattice vector 2m with integer m.
Therefore, these experiments can determine the value of the line lattice analogue , of the
exponent b of the Bose gas correlations. Using the mapping between lines and bosons,
we have seen that b 1/n (see Eq. (78)) and therefore
,
(90)
in the n 0 limit. This result is in strong contrast to the case of a pure line lattice with
= 2. The infinite exponent means that the oscillating terms in the density correlation
function decay faster than any power law due to destruction of quasi-long-range order by
the random potential. Exactly the same result was obtained by Cardy and Ostlund in their
RG approach [43], but it is in contradiction to all VA approaches. Although a mean squared
7 The splitting of clusters has been considered by Parisi for a single line (N = 1) to study the possibility of
replica symmetry breaking [68].
508
relative displacement which grows slightly faster than a logarithmic increase cannot be
obtained directly from the homogeneous part of the Bose gas density correlations, the fact
that the Bose gas exponent b is mapped onto an infinite for the line lattice might be
interpreted as a signature of the presence of these contributions. However, this argument
has to be viewed with care since 1/n means that the oscillating terms of the density
density correlation function have an essential singularity in the n 0 limit.
7. Conclusion and discussion

In this paper, we studied the probability distribution function of the free energy of
a lattice of lines, distorted due to thermal and disorder fluctuations. Using the replica
approach in combination with an analytically continued nested Bethe ansatz, we calculate
exactly the cumulants of the free energy in the dilute limit. Our results confirm a
previously proposed scaling form for the replica free energy in terms of the the replica
number n, as a scaling field [17]. These exact results were obtained by establishing a
novel analogy between the random line lattice and the weakly interacting 1D Bose gas
with delta-function interactions. Based on this analogy, we related the results for the
densitydensity correlation function of the Bose gas, known from Haldanes effective
theory of 1D quantum liquids, and from conformal field theory, to correlation functions
of the line lattice. Our result for the difference between diagonal and off-diagonal replicacorrelations is in agreement with RG predictions, and a replica symmetric variational
ansatz for an elastic model of the line lattice, if one takes into account the correct
disorder dependence of elastic constants obtained by the Bethe ansatz approach for the
free energy. The Bethe ansatz results do not agree with the variational ansatz with replica
symmetry breaking. The comparison with the RG results of Cardy and Ostlund [43] and
Goldschmidt and Houghton [44], which predict a faster than logarithmic growth of relative
line displacements, shows that our mapping to a Bose gas cannot capture the corresponding
contributions. Potential sources for these additional logarithmic terms within the Bethe
ansatz provide avenues for future explorations.
Finally, it might be interesting to note that all results obtained in this paper are valid if
the length scale ld , setting the crossover to random behavior for a single line, is much
larger than the correlation length 0 of the random potential. When this condition is
fulfilled, 0 does not show up in the final results, and can be safely set to zero from
the beginning as we did throughout the paper, see Eq. (3). However, if temperature is
decreased, the length scale ld = T 3 /g also decreases, and gets equal to 0 at the crossover
temperature T = (0 g)1/3 , which is of the order of the smallest random energy barrier
[69]. Therefore, our results apply directly to the high temperature regime T T . They
can also be extended to the low temperature limit T T , using a method developed
by Korshunov and Dotsenko [70] to calculate the replica free energy of a single line in a
random potential with finite 0 . By splitting the n-cluster for a single line into n/k separate
blocks with k T /T , they were able to show that the ground state energy of their model
with finite 0 can be obtained from the model with -function interactions by the simple
509
substitutions g kg, k 2 and n n/k. Since the internal structure of individual

clusters remains unchanged by considering a finite density of clusters (see the anstze in
Eqs. (30)(32)), our results can be translated easily to the low-temperature regime by the
above substitutions.
Acknowledgements
We would like to thank E. Frey, A. Hanke, V.E. Korepin, T. Nattermann, V.L. Prokovsky
and W. Zwerger for discussions and conversations. This work was supported by the
Deutsche Forschungsgemeinschaft under grant No. EM70/1-3 (T.E.), and by the National
Science Foundation grants No. DMR-98-05833 and No. PHY99-07949 (M.K.).
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]
T. Halpin-Healy, Y.-C. Zhang, Phys. Rep. 254 (1995) 215.

M. Kardar, G. Parisi, Y.C. Zhang, Phys. Rev. Lett. 56 (1986) 889.
D.A. Huse, C.L. Henley, D.S. Fisher, Phys. Rev. Lett. 55 (1985) 2924.
T. Hwa, M. Lssig, Phys. Rev. Lett. 76 (1996) 2591.
T. Hwa, Nature 399 (1999) 17.
C.A. Bolle et al., Nature 399 (1999) 43.
T. Davis, J. Cardy, Nucl. Phys. B 570 (2000) 713.
M. Kardar, Nucl. Phys. B 290 [FS20] (1987) 582.
B. Derrida, H. Spohn, J. Stat. Phys. 51 (1988) 817.
D.A. Gorokhov, G. Blatter, Phys. Rev. Lett. 82 (1999) 2705.
E. Brunet, B. Derrida, Phys. Rev. E 61 (2000) 6789.
W.W. Mullins, R.F. Sekerka, J. Appl. Phys. 35 (1964) 444.
P.G. de Gennes, J. Chem. Phys. 48 (1968) 2257.
V.L. Pokrovsky, A.L. Talapov, Phys. Rev. Lett. 42 (1979) 65.
A.M. Tsvelik, Phys. Rev. Lett. 68 (1992) 3889.
L. Balents, M. Kardar, Nucl. Phys. B 393 (1993) 480.
T. Emig, M. Kardar, Phys. Rev. Lett. 85 (2000) 2176.
C.N. Yang, Phys. Rev. Lett. 19 (1967) 1312.
M. Takahashi, Prog. Theor. Phys. 44 (1970) 899.
E.H. Lieb, W. Liniger, Phys. Rev. 130 (1963) 1605;
E.H. Lieb, Phys. Rev. 130 (1963) 1616.
D.R. Nelson, H.S. Seung, Phys. Rev. B 39 (1989) 9153.
R.P. Feynman, A.R. Hibbs, Quantum Mechanics and Path Integrals, McGraw-Hill, New York,
1965;
R.P. Feynman, Statistical Mechanics, Benjamin, Reading, 1972.
H.A. Bethe, Z. Phys. 71 (1931) 205.
H.B. Thacker, Rev. Mod. Phys. 53 (1981) 253.
M. Gaudin, Phys. Lett. 24A (1967) 55.
D.S. Koltun, Phys. Rev. C 36 (1987) 2047.
B. Sutherland, Phys. Rev. Lett. 20 (1968) 98.
F.G. Tricomi, Integral Equations, Interscience, New York, 1957.
I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series and Products, Academic Press, Orlando,
1980.
510
[30] T. Emig, M. Kardar, in preparation.

[31] M. Takahashi, Prog. Theor. Phys. 53 (1975) 386;
M. Takahashi, Thermodynamics of One-dimensional Solvable Models, Cambridge Univ. Press,
Cambridge, 1999.
[33] V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method and
Correlation Functions, Cambridge Univ. Press, Cambridge, 1997.
[34] A. Aharony, A.B. Harris, Phys. Rev. Lett. 77 (1996) 3700.
[35] N.N. Bogoliubov, J. Phys. USSR 11 (1947) 23.
[36] D.K. Lee, Phys. Rev. A 4 (1971) 1670;
D.K. Lee, Phys. Rev. A 9 (1974) 1760.
[37] T. Emig, M. Kardar, unpublished.
[38] N.I. Akhiezer, The Classical Moment Problem, Oliver and Boyd, London, 1965.
[39] M. Kardar, in: F. David, P. Ginzparg, J. Zinn-Justin (Eds.), Fluctuating Geometries in Statistical
Mechanics and Field Theory, Proceedings, Elsevier, Amsterdam, 1996.
[40] A. Gold, Z. Phys. B 91 (1993) 397.
[41] E. Demirel, B. Tanatar, Phys. Rev. B 59 (1999) 9271.
[42] M.P.A. Fisher, Phys. Rev. Lett. 62 (1989) 1415.
[43] J.L. Cardy, S. Ostlund, Phys. Rev. B 25 (1982) 6899.
[44] Y.Y. Goldschmidt, A. Houghton, Nucl. Phys. B 210 (1982) 155.
[45] M. Paczuski, M. Kardar, Phys. Rev. B 43 (1991) 8331.
[46] S.E. Korshunov, Phys. Rev. B 48 (1993) 3969.
[47] P. Le Doussal, T. Giamarchi, Phys. Rev. Lett. 72 (1994) 1530.
[48] J. Kierfeld, J. Phys. I (Paris) 5 (1995) 379.
[49] D.S. Fisher, M.P.A. Fisher, D.A. Huse, Phys. Rev. B 43 (1991) 130.
[50] T. Nattermann et al., Europhys. Lett. 16 (1991) 295.
[51] J. Toner, Phys. Rev. Lett. 67 (1991) 2537;
J. Toner, Phys. Rev. Lett. 68 (1992) 3367.
[52] T. Nattermann, I. Lyuksyutov, Phys. Rev. Lett. 68 (1992) 3366.
[53] T. Giamarchi, P. Le Doussal, Phys. Rev. B 52 (1995) 1242.
[54] H. Orland, Y. Shapir, Europhys. Lett. 30 (1995) 203.
[55] T. Nattermann, S. Scheidl, Adv. Phys. 49 (2000) 607.
[56] G.G. Batrouni, T. Hwa, Phys. Rev. Lett. 72 (1994) 4133.
[57] D. Cule, Y. Shapir, Phys. Rev. Lett. 74 (1995) 114.
[58] D.J. Lancaster, J.J. Ruiz-Lorenzo, J. Phys. A 28 (1995) L577.
[59] E. Marinari, R. Monasson, J.J. Ruiz-Lorenzo, J. Phys. A 28 (1995) 3975.
[60] C. Zeng, A.A. Middleton, Y. Shapir, Phys. Rev. Lett. 77 (1996) 3204.
[61] H. Rieger, U. Blasum, Phys. Rev. B 55 (1997) R7394.
[62] C. Zeng, P.L. Leath, T. Hwa, Phys. Rev. Lett. 83 (1999) 4860.
[63] I. Affleck, F.D.M. Haldane, Phys. Rev. B 36 (1987) 5291.
[64] A. Berkovich, G. Murthy, Phys. Lett. A 142 (1989) 121.
[65] T. Nattermann, Phys. Rev. Lett. 64 (1990) 2454.
[66] V.E. Korepin, Commun. Math. Phys. 94 (1984) 93.
[67] A. Berkovich, Nucl. Phys. B 356 (1991) 655.
[68] G. Parisi, J. Phys. (Paris) 51 (1990) 1595.
[69] T. Nattermann, W. Renz, Phys. Rev. B 38 (1988) 5184.
[70] S.E. Korshunov, V.S. Dotsenko, J. Phys. A 31 (1998) 2591.

Boundary bootstrap principle in two-dimensional

integrable quantum field theories
Valentina Riva a,b
a Dipartimento di Scienze Chimiche, Fisiche e Matematiche, Universit dellInsubria, Como, Italy
b International School for Advanced Studies,via Beirut 4, 34014 Trieste, Italy
Abstract
We study the reflection amplitudes of affine Toda field theories with boundary, following the ideas
developed by Fring and Koberle [A. Fring, R. Koberle, Nucl. Phys. B 421 (1994) 159; A. Fring,
R. Koberle, Nucl. Phys. B 419 (1994) 647; A. Fring, R. Koberle, Int. J. Mod. Phys. A 10 (1995) 739]
and focusing our attention on the En series elements, because of their interesting structure of higher
order poles. We also investigate the corresponding minimal reflection matrices, finding, with respect
to the bulk case, a more complicated relation between the spectra of bound states associated to the
minimal and to the dressed amplitudes. 2001 Published by Elsevier Science B.V.
PACS: 11.55.Ds
Keywords: Affine Toda field theory; Conformal field theory; Reflection matrices; Boundary bound states
1. Introduction
A two-dimensional quantum field theory with boundary can be basically defined in two
ways [1]. Let us consider the semi-infinite euclidean plane, x (, 0], y (, ),
where the y-axis represents the boundary. In the Lagrangian approach one writes the action
in the form

A=
0
dy

dx a(, ) +

d
dy b B , B ,
dy
(1.1)
where a and b are local functions respectively of the bulk and boundary fundamental
fields , B (B (y) = (x, y)|x=0). In the perturbed conformal field theory approach
one writes the symbolic action
E-mail address: riva@fm.sissa.it (V. Riva).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 6 7 - 5
512
V. Riva / Nuclear Physics B 604 [FS] (2001) 511536

A = ACFT+CBC +
0
dy

dx (x, y) +
dy B (y),
(1.2)
where ACFT+CBC is the action of a conformal field theory (CFT) on the semi-infinite
plane with certain conformal boundary conditions (CBC), and , B are specific bulk
and boundary relevant fields.
If a bulk theory is integrable, i.e., if it possesses an infinite set of mutually commutative
integrals of motion, the scattering processes are purely elastic, and the S-matrix factorizes
into two-particle amplitudes. This can be generalized to the case with boundary defining, in
addition to the bulk S-matrix, an opportune boundary reflection matrix which will factorize
into single-particle amplitudes. In every integrable boundary field theory we expect to find
a certain number of different reflection matrices, corresponding to the various boundary
conditions compatible with integrability.
In the perturbed conformal field theory approach the integrable boundary conditions
must be associated in the ultraviolet limit to some of the possible conformal ones. It
was demonstrated by Cardy [2] that, in the case of Virasoro minimal models, conformal
boundary conditions are in one-to-one correspondence with the primary fields of the
examined CFT. In fact, imposing the absence of energy or momentum flux across the
boundary, one can write the physical boundary states as
|l =
Slj

|j ,
S0j
j
(1.3)
where the indices l and j label the primary operators of the theory (0 refers to the identity),
Slj is the modular matrix and the states |j , called Ishibashi states, are constructed from
the states in the irreducible quotient of the Verma module j . A fundamental quantity related
to a boundary state |A is the so-called ground state degeneracy g, defined as [3]
gA =
0|A ,
(1.4)
where |0 is the ground state of the bulk Hamiltonian. If the state |A is of the form (1.3),
g has the generally noninteger value
SA0
gA = .
S00
(1.5)
In the bulk case, the two-particle scattering amplitudes can be exactly calculated (up
to the so-called CDD ambiguity) as solutions of a number of constraints, which are
expressed by unitarity, crossing, YangBaxter and bootstrap equations. In the presence of
a boundary, the reflection amplitudes are constrained by analogous equations, which relate
them to the bulk S-matrix. In this way, starting from a theory whose bulk scattering matrix
is known, one can investigate which are the possible reflection amplitudes.
If the examined theory is a perturbed minimal model, we know explicitly from the
primary fields content of the corresponding CFT which are the conformal boundary
conditions that we have to recover in the UV limit.
513
On the contrary, the affine Toda field theory related to an algebra G of rank r has in
general a description in terms of the Lagrangian
L=
r
r
2 m2
1
j 2
ni ei ,
2
j =1
(1.6)
i=0
where the j s are r bosonic fields, m and are real parameters, i are the simple roots

of G, and 0 = ri=1 ni i is the opposite of the highest root. We can now define a
boundary field theory with Lagrangian
LB = (x)L (x)B()
(1.7)
and investigate which reflection amplitudes are compatible with the bulk S-matrix.
Restrictions on the boundary interaction B() due to the integrability requirement have
been discussed in [8,9]. Also this theory can be seen as a perturbation of a certain CFT,
which has however a spectrum of primary operators generally much richer than in minimal
models. Furthermore, this CFT possesses a larger symmetry than just conformal invariance,
and in this case the structure of the boundary states for a boundary condition that is only
conformally invariant is not yet known.
2. Boundary affine Toda field theories

In a two-dimensional integrable field theory it is possible to parameterize the momentum
of a particle in terms of a single variable , called rapidity:
pa = ma (cosh a , sinh a ).
(2.1)
Assuming that the effect of the boundary is to reverse the momentum and preserve the
energy of a particle which scatters off it, we define the reflection amplitude K( ) as
the proportionality factor which relates a single-particle out state to the in state with
reversed momentum:
|a, out = Ka ( )|a, in.
(2.2)
With this definition we restrict our attention to theories with distinguishable particles,
otherwise the reflection factor Ka should be a matrix to allow for a mixing of states.
The main equations for the reflection amplitudes have been derived in [1] and discussed
in [47,10]. The unitarity condition
Ka ( )Ka ( ) = 1
(2.3)
is a straightforward generalization of the bulk one, while the boundary crossing equation
Ka ( )Ka ( + i) = Saa (2 )
(2.4)
required more attention, and will be of great importance in our considerations (Fig. 1).
514
Fig. 1. The boundary crossing-unitarity relation.
In the case of distinguishable particles the YangBaxter equations

Ka (a )Sab (b + a )Kb (b )Sab (b a )
= Sab (b a )Kb (b )Sab (b + a )Ka (a )
(2.5)
are trivially satisfied. On the contrary, a strong constraint is given by the bootstrap
equations

Kc ( ) = Ka i u bac Sab 2 i u bac + i u abc Kb + i u abc ,
(2.6)
with u cab ucab , where the bulk fusing angles ucab correspond to poles of Sab at
= iucab creating the bound state c, and are related to the masses of the particles by
m2c = m2a + m2b + 2ma mb cos ucab .
In affine Toda field theories the S-matrix elements are always of the form

s x+1 ( )s x1 ( )
sinh 12 ( + i)
h
h
,
s ( ) =
{x} =
.
s x+1B ( )s x1+B ( )
sinh 12 ( i)
h
(2.7)
x {x} ,
with
(2.8)
(B() [0, 2] is a coupling constant dependent parameter, and h is the Coxeter number
of the algebra G.) Fring and Koberle [5] have demonstrated that in this case the boundary
crossing equation (2.4) has factorized solutions of the form
Ka ( ) =
Kx ( ),
Kx ( ) =
s 1xh ( ) s 1xh ( )
2h
2h
s 1xBh ( ) s 1x+Bh ( )
2h
(2.9)
2h
where the blocks Kx ( ) are in one-to-one correspondence to the blocks {x}2 in Sa a up to

a shift of 2h in x. In order to determine which of the blocks are shifted by 2h and which
are not, one has to write a bootstrap equation (2.6) involving only one particular Ka ( ),
find its most general solution, and then compute the other Kb ( ) consistently.
Given a solution Ka ( ) of Eqs. (2.3)(2.6), a function of the form
Ka ( )a ( )
is again a solution if a satisfies the homogeneous equations

a + i
a i
= 1,
2
2

c ( ) = a i u bac b + i u abc .
(2.10)
(2.11)
(2.12)
515
Some possible choices of these functions (called CDD-factors and analyzed in [7]) are:
Ka ( + i)
.
1. a ( ) =
Ka ( )
x+2h ()
Every block x ( ) = KxK(+i)
= KK
satisfies Eqs. (2.11) and (2.12), so that the
x ()
x ()
new solution
obtained from the first one by shifting all the xs by 2h.
canbe
1
2. a ( ) = Sab ( ) , b.

3. a ( ) = b Sab ( ).
What we expect is that these CDD-factors relate different sets of reflection amplitudes
corresponding to the various boundary conditions compatible with integrability.
For the theories in exam we will analyze the two solutions of the form (2.9) Ka ( ) and
Ka ( + i), related by the first kind of CDD-ambiguity mentioned, calling minimal the
one with the minimum number of poles in the physical strip.
As noticed in [1], the boundary can exist in several stable states, and the presence of
poles in Ka ( ) indicates the possibility for particle a to excite it. For reflection amplitudes,
the physical strip is the region 0 Im 2 of the complex -plane (Fig. 2).
We will call a the fusing angle related to an odd-order pole with positive residue at
a = ia in the reflection amplitude Ka of particle a on the boundary state , creating the

state . The corresponding bootstrap equation for the boundary bound states is

Kb ( ) = Sab + ia
(2.13)
Kb ( )Sab ia
,
and the energies of the two states | and | are related by

E = E + ma cos a
.
(2.14)
The blocks Kx ( ) have poles at

1 x h
B 1 + x + h
i (mod 2i) and B =
i (mod 2i) (2.15)
2h
2h
and zeroes at
=
1 + x + h
i (mod 2i)
2h
and
0 B
1 B x h
i (mod 2i).
2h
(2.16)
Fig. 2. Boundary excitation.
516
Only shifted blocks can give poles and 0 B zeroes in the physical strip. Contrary to
the case of the bulk S-matrix, the coupling constant dependent B poles of the unshifted
blocks can move inside the strip 0 Im , but they are confined inside the interval
0
2 Im (they can reach the value i 2 only in the trivial cases B = 0, 2); the
zeroes of unshifted blocks are also located in 2 Im .

Fring and Koberle have demonstrated in [6] that every excited state reflection amplitude
obtained by Eq. (2.13) can be expressed in terms of the ground state one as

Sab ( ) Ka0 ( ),
Ka ( ) =
(2.17)
b
where all the bs are of the same colour with respect to the bicolouration of the Dynkin
diagram of G, and the corresponding energy levels are related by
1
E = E0 +
(2.18)
mb .
2
b
2.1. E6 -affine Toda field theory

A complete analysis of the minimal solution was performed in [6]. The bootstrap
closes on eight boundary bound states, and there are six different energy levels, two of
which are degenerate.
The bulk S-matrix of this model has been found in [11]. We label the six particles
according to the following Dynkin diagram
whose extension by 0 possesses a Z3 symmetry, being invariant under the transformations

1 6 0 ,
2 3 5 ,
4 4 .
(2.19)
Omitting the -dependence, the minimal set of reflection amplitudes is:

K1 = K5 K35 ,
(2.20)
K2 = K1 K7 K11 K29 ,
(2.21)
K3 = K3 K5 K7 K11 K29 K33 ,
(2.22)
2
K35 ,
K4 = K1 K3 K52 K7 K92 K27 K29 K31
(2.23)
with K6 = K1 and K5 = K3 .
Fixing E0 = 0, the energy levels are:
m2
,
E = 0.796 m =
2
517
m3 + m5
,
2
m2 + m3 + m5
E = 2.971m =
,
2
where the mass parameter m is defined by m21 = (3 3 )m2 . Levels and are
degenerate, due to the equality m3 = m5 ; they correspond, respectively, to the states 1 , 2
and 1 , 2 , with
m3
,
2
m2 + m3
E = 1.884 m =
,
2
E = 1.087 m =
E = 2.175 m =
Kb 1 ( ) = Sb2 ( )Sb5 ( )Kb0 ( ),
Kb 2 ( ) = Sb2 ( )Sb3 ( )Kb0 ( ).
Kb 1 ( ) = Sb3 ( )Kb0 ( ),
Kb 2 ( ) = Sb5 ( )Kb0 ( ),
We list the fusing angles in Table 1 following the conventions of [6]. Each entry in the
table indicates a fusing angle as a multiple of i
12 ; the left column refers to the particle type
which scatters off the boundary in the state indicated in the first row. The superscript refers
to the state the boundary is changing into, and the subscript refers to the order of the pole,
if multiple. As in all the other cases we will examine, the residues have always the same
sign as B varies in [0, 2], they vanish at the extremes of the interval and sometimes also in
B = 1.
The second solution gives the same number of boundary bound states, with energies
m1
,
E = 0.563 m =
2
m1 + m6
m1 + m4
E = 1.126 m =
,
E = 2.101 m =
,
2
2
m4
m1 + m4 + m6
,
E = 2.664 m =
.
E = 1.538 m =
2
2
Levels and are degenerate, with
Kb1 ( ) = Sb6 ( )Kb0 ( ),
Kb1 ( ) = Sb4 ( )Sb1 ( )Kb0 ( ),
Kb2 ( ) = Sb1 ( )Kb0 ( ),
Kb2 ( ) = Sb4 ( )Sb6 ( )Kb0 ( ).
The fusing angles are listed in Table 2.

The two solutions have different behaviours with respect to the Z3 symmetry of the
extended Dynkin diagram. In fact, the boundary bound states obtained from the minimal
Table 1
a\
11
12 51
2 6
21 42
431
1 33 5
35
22 41
432
12
11 52
31
432 61
23 43
552
63 2
1
431 62
63 1
551
23 43
32
1 53
45 63
65
631
632
45
652
59
45
651
1 53
518
Table 2
a\
41
62
2 4
22
432
43
631
1 3 60
12 61
11 62
1 53 6
33 6
61
62
11 52
1 53
23 43 60
12 51
42
332
452 63 1
451 63 2
331
1 53
2 4
61
21
67
691
552
57
632
431
57
551
47 65
692
611
43
solution also enjoy this symmetry, while the ones obtained from the second solution dont.
This is an indication that the minimal solution should correspond to the free boundary
condition or to another boundary condition which preserves the Z3 symmetry, while in the
second case an operator which breaks this symmetry lives on the boundary.
The bulk S-matrix has been found in [12]. Ordering the particles with increasing mass
and omitting the -dependence, the minimal set of reflection amplitudes is:
K1 = K1 K9 K17 ,
K2 = K1 K7 K11 K53 ,
K3 = K1 K5 K7 K9 K47 K13 K17 ,
K4 = K1 K3 K7 K9 K45 K11 K15 K53 ,
K5 = K1 K3 K5 K7 K43 K92 K11 K47 K13 K51 K17 ,
2
K51 K17 ,
K6 = K1 K3 K52 K7 K43 K92 K45 K11 K47 K13
2
3
2
2
K11
K47 K13 K49
K15
K53 .
K7 = K1 K32 K52 K41 K73 K43 K92 K45
Fixing E0 = 0 and defining M the mass of the lightest particle, the corresponding energy
levels are:
m5
,
E = 1.266 M =
2
m1 + m3
m1 + m3 + m5
E = 1.440 M =
,
E = 2.706 M =
,
2
2
m1 + m6
m1 + m5 + m6
,
E = 3.206 M =
,
E = 1.940 M =
2
2
m3 + m6
m3 + m5 + m6
,
E = 3.645 M =
.
E = 2.380 M =
2
2
In Table 3 we list the fusing angles as multiples of i
18 . The signs refer to the Z2
symmetry of the extended Dynkin diagram, choosing the convention in which the boundary
ground state is even. All the poles in the reflection amplitudes are consistent with the
519
Table 3
0+
a\
2 6
4
1 73 9
1
2+
3 9
1 9
43
2 63
4+
1 5
5+
2 4 6
2 6
7+
1 33 53 7
1 73 9
23 43
65
57 95
53 93
65
45 85
77 95
75 93
95
97
97
2 63
33 53 95
67 85
1 53 93
93
85
65
55 9
95
67
89
89
79 97
711 99
911
913
915
change of parity induced in the boundary by the particles which create the bound states, so
that this solution corresponds to a boundary condition which preserves parity.
Starting from the second solution, we obtain the same number of boundary bound states,
and energy levels:
m2
,
2
m4
,
E = 0.985 M =
2
m2 + m4
E = 1.627 M =
,
2
m7
,
E = 1.851 M =
2
E = 0.643 M =
m2 + m7
,
2
m4 + m7
E = 2.836 M =
,
2
m2 + m4 + m7
E = 3.478 M =
.
2
E = 2.494 M =

In this case the boundary bound states dont have a definite parity, and this again signals
the presence on the boundary of an operator which breaks this symmetry.
Table 4
a\
1 5 90
1 7 9
3 5 93
2+
4 6 90
2 4 9
1 3 7 90
1 5 93
4+
2 6 90
2 63 83 9
5+
1 53 90
33 93
6 1 33 53 73 90
7+ 23 45 65 83 90
1 55 75 93
47 67 9
63 9

1 73 93
23 43 9
1 75 93
35 55 95
69 87 9
3 93
1 53 93
1 73 93
53 95
95
2 83 9
43 63 9
43 9
65 9
1 55 95
33 75 95
97
77 97
99
23 9
65 9
67 87 9
97
99
99
911
57 97
79 97
711 99
911
913
611 9
811 9
815 9
77 95
520
The bulk S-matrix has been found in [13]. Ordering the particles with increasing mass
and omitting the -dependence, the minimal set of reflection amplitudes is
K1 = K1 K11 K19 K89 ,
K2 = K1 K7 K11 K13 K77 K19 K23 K89 ,
K3 = K1 K3 K9 K11 K71 K13 K17 K19 K79 K21 K87 K29 ,
K4 = K1 K5 K7 K9 K11 K71 K13 K15 K75 K17 K19 K79 K21 K83 K25 K29 ,
2
2
2
K71 K13 K73 K15
K17 K77 K19
K79 K21 K81
K5 = K1 K3 K5 K7 K9 K69 K11
K23 K85 K27 K89 ,

2
2
2
2
2
K6 = K1 K3 K5 K7 K67 K92 K11
K71 K13
K73 K15 K75 K17
K77 K19 K79
K21
K23 K83 K25 K87 K29 ,

2
2
3
2
2
3
2
2
K7 = K1 K3 K52 K72 K67 K92 K69 K11
K71
K13
K73 K15
K75
K17
K77 K19
K79
2
2
2
K21
K81 K23 K83
K25
K87 K29 ,
2
4
2
3
3
4
2
3
3
4
K8 = K1 K32 K52 K65 K73 K67 K93 K69
K11
K71
K13
K73
K15
K75
K17
K77
K19
2
2
3
3
2
2
K79
K21
K81
K23
K83 K25 K85
K27
K89 .
Fixing E0 = 0 and defining M the mass of the lightest particle, the corresponding energy
levels are:
m3
,
2
m4
,
E = 1.2024 M =
2
m6
,
E = 1.6092 M =
2
m7
E = 1.9456 M =
,
2
m3 + m4
,
E = 2.1969 M =
2
m3 + m6
,
E = 2.6037 M =
2
m4 + m6
,
E = 2.8116M =
2
m3 + m7
E = 2.9401 M =
,
2
E = 0.9945 M =
m4 + m7
,
2
m6 + m7
E = 3.5547 M =
,
2
m3 + m4 + m6
E = 3.8061 M =
,
2
m3 + m4 + m7
E = 4.1425 M =
,
2
m3 + m6 + m7
E = 4.5493 M =
,
2
m4 + m6 + m7
E = 4.7572 M =
,
2
m3 + m4 + m6 + m7
E = 5.7517 M =
.
2
E = 3.1480 M =
We list now the fusing angles as multiples of
i
30
(see Table 5).
521
Table 5
a\
3 13
1 7 11 15
1 9 15
1 7
1 5 15
3 73 133 15
2 6 10
2 63 103
2 83 103
4 6 8 10
2 63 83 123
43 63 105
1 3 7 11
2 4 63 10 12
2 43 63 83 103 12
1 33 55 75 95 113 13
1 73 9 15

23 4 6 123 14
63 83 103
33 53 115 133 15
2 85 103
45 65 107 125
1 79 99 117 155
1 95 153
1 75 115 155
43 65 105 125

2
3 43 127 145

4
5 87 107 145
6
7 109 127

3 7
9 119 137 155
59 1111 139 157
1 5 15
3 7 113 15
3 133 15
1 73 153
5 133 15
2 1 73 113 153
1 93 153
3 75 93 153
135 153
1 75 155
23 63 125
23 145
2 85 105
4
3 63 127 145
63 105
4 43 85 105 145
2 65 85 125
65 107
6
5 127
87 107

5 3
3 75 117 155

5
5 137 155
119 157
1 99 159
35 1111 139 157
107
23 8
9 149
45 1211 149
109
47 1011 149
69 1013 1211
10
13
1213
11
17 1315 1513
917 1515
67 107 127
23 89 129
9
13 1311 159
1 93 153
2 1 53 115 155
7
5 155
1115 1513
3 73 153
115 153
13
5 153

9
5 137 155
1 157
1 15
5
155
77 157
15
7
159
14
9
109
23
8
7
45 10
9 149
67 129
89
109
159
79 11
11 1511
1511
13
13 1511
1115 1513
1213
1413
1013
1017
1417
1219
1521
1523
6 69 10
11 1211
813 1413
12
15
1317 1515
1319 1517
1121 1519
107
6
5 107
65 12
9
713 1113 1511
1 113 153
15
15
1517
1217
1325 15
23
1529
522
Table 6
a\
6 10 150
2 6 12 15
8 10 15
4 6 10 12 150
4 10 15
2 4 123 15
1 7 11 150
3 7 113 15
1 73 93 153
1 3 13 150
1 3 73 113 153
1 53 133 153
2 63 83 103 150
43 63 105 143 15
2 65 85 105 15
1 53 73 113 150
1 75 95 153
33 75 135 153
1 33 75 95 133 150
1 55 97 135 153
1 77 117 155
0
2
3 45 67 87 107 125 143 15

23 69 89 109 127 15
47 6
9 1011 129 147 15
2 8 15
4 6 103 15
4 6 123 15
4 103 143 15
2 4 83 143 15
63 103 123 15
103 15
2 125 15
3 73 153
1 115 155
53 115 153
33 53 117 135 153
155
1 95 135 155
1 157
1 55 97 137 157
65 107 15
23 65 127 15
33 77 97 137 155
1 77 117 157
1 5
7 119 157
611 1013 1211 15
7
9 159
35 79 139 157
1311 159
711 1113 1511
8
13 1013 1411 15
815 1015 15
10
17 1415 15
63 103 15
4 143 15
2 63 123 15
43 63 105 125 15
2 85 15
43 105 15
1 75 115 157
3
3 137 155
75 117 157

3
3 137 157
1 159
89 109 15
6
7
8
1 79 99 159
23 6
7 129 15

4
5 65 129 147 15
67 1211 15
57 1111 159
711 9
11 1511
9
13 1313 1511
1115 1513
1217 1415 15
1019 15
83 103 15
2 83 15

4
3 127 15
43 147 15
1 7
7 97 159

33 7
7 119 159
139 159
77 159
11
11 1511
1011 1411 15
811 10
11 15
1511
1313 1511
915 1315 1513
1515
1517
12
21 15
1223 15
1219 15
1013 15
1315 1513
523
Table 6 continued
a\
6
3 105 15
65 10
7 127 15
107 15
11
9 159
1111 159
1311 1511
1513
1313 15
13
1515
15
14
15 15
15
15
1115 1515
1517
15
17
1519
13
19 1517
1321 1519
15
21
1523
1423 15
15
1427 15
15
63 12
5 15
105 15
15
129 15
15
15
11
1511
Starting from the second solution, we obtain the same number of boundary bound states,
and energy levels:
m1
,
2
m2
,
E = 0.8090 M =
2
m1 + m2
E = 1.3090 M =
,
2
m5
,
E = 1.4781 M =
2
m1 + m5
,
E = 1.9781 M =
2
m2 + m5
,
E = 2.2872 M =
2
m8
E = 2.3917 M =
,
2
m1 + m2 + m5
,
E = 2.7872 M =
2
E = 0.5000 M =
m1 + m8
,
2
m2 + m8
E = 3.2007 M =
,
2
m1 + m2 + m8
E = 3.7007 M =
,
2
m5 + m8
E = 3.8698 M =
,
2
m1 + m5 + m8
E = 4.3698 M =
,
2
m2 + m5 + m8
E = 4.6789 M =
,
2
m1 + m2 + m5 + m8
E = 5.1789 M =
.
2
E = 2.8917 M =
3. Pole interpretation
In the procedure of applying the bootstrap Eq. (2.13) we have considered all the odd
order poles with positive residue. This property, however, is necessary but not sufficient for
the creation of a boundary bound state. In fact, Dorey, Tateo and Watts [14] have proposed
two kinds of mechanisms which can describe some of the poles without involving new
boundary bound states.
524
Fig. 3. First mechanism.
The first one is a u-channel mechanism (see Fig. 3), and it is invoked when an excited
state reflection factor Ka has a pole at the same place a = ia as the pole in Ka which
generated .
The pole in exam doesnt excite the boundary to a new bound state, but simply
corresponds to going back from to . This rule, reasonable but non-properly founded,
has a clear explanation in the case of a theory with defect [15], where not just reflection but
also transmission is allowed. In that case it is shown with an explicit example how a pole
at = i with this property can be neglected, because, although it has positive residue in
the reflection amplitude, its residue in the transmission one is negative. At the same time,
both amplitudes have a positive residue pole at = i( ), which exactly corresponds to
going back to the original boundary state. Unfortunately, integrable defect theories seem
to apply only to quasi-free systems.
The other possibility is a boundary generalization of the ColemanThun mechanism: in
some cases a pole can be described by on-shell diagrams, different from the one in Fig. 1,
which correspond to multiple rescattering processes. These methods have also been applied
in [16] and [17], with the hypothesis that, if an alternative diagram can be drawn, then the
pole in exam doesnt correspond to the creation of a boundary bound state.
In general, the order of a certain diagram is given by P 2L, where P and L
are, respectively, the number of propagators and loops. When dealing with a boundary,
however, vertex factors can also be given by reflection amplitudes. If these amplitudes
have poles (or zeroes) at the rapidities dictated by the on-shell condition, then their effect
will be to raise (or lower) the order of the diagram.
As we have seen, the reflection amplitudes of the form (2.9) never have couplingindependent zeroes in the physical strip, hence the order of a given diagram could only
be raised by their insertions. In this case, the only two diagrams which can describe a first
order pole are represented in Fig. 4. If we call a the pole we are interested in, we can see
that it is possible to draw Type 1 diagram (already introduced in [1]) if there is a particle b
such that iuaab = a + i 2 and such that the corresponding reflection amplitude Kb ( ) has
an odd order pole with positive residue at = i 2 . Type 2 diagram can be drawn if there are
525
Fig. 4. First order diagrams.
Fig. 5. Type 3 (second order).
Fig. 6. Type 4 (third order).
two particles b and c such that uabc < 2 and a = i(uabc + ucab ); the amplitude Kb ( )
has to be evaluated at b = iuabc .
A second order diagram, which will describe third or higher order poles of Ka , is
represented in Fig. 5. This diagram can be drawn if there are b and c such that Kc has
b
a pole at c = iuac a creating the boundary bound state or, if Im(c ) > 2 , such
that Kc has a pole at i c creating . The amplitude Kb has to be evaluated at b =

c
iuab + a i .
A third order diagram, with 13 propagators and 5 loops, is represented in Fig. 6. This
diagram can be drawn if there are all the opportune bulk fusing angles, and one has to
evaluate the two amplitudes Kb and Kd at the rapidities dictated by the on-shell condition.
We will now investigate if some of our poles can be described by these mechanisms.
In none of the examined reflection amplitudes u-channel diagrams can explain any
pole. However, many generalized Coleman-Thun diagrams can be drawn, with interesting
consequences.
526
3.1. Analysis of the E7 pole structure

3.1.1. Minimal solution
We start considering the reflection matrix in the ground state. Type 1 diagram can never
be drawn (none of the seven particles couples to the boundary at = i 2 ). If a = 1, 2,
neither can Type 2, lacking b and c such that uabc < 2 .
However, many poles of the remaining amplitudes can be explained by this diagram; we
list all the possible corresponding choices of b and c in Table 7.
The two triple poles of K70 are described by this diagram because, in the case (b, c) =
4
, and in the other two cases the fusing
(6, 1), K60 has a double pole at 6 = iu716 = i 18
7
angle u36 corresponds to a triple pole of the S-matrix.
In this way, excluding the excitation diagrams, the only boundary bound state that we
can get from the ground state is . If we repeat the above procedure for the Kb amplitudes,
we can exclude the creation of other boundary bound states. In Table 8 we list which poles
are explained by the three kinds of diagram seen. All the Type 3 diagrams mentioned have
as external boundary state, and the ground state as intermediate one. Poles of high order
can be described by many different diagrams, but we have listed just one of the possible
2
admits also a description in terms of
choices. As an example, the triple pole at 5 = i 18
5
is due to
Type 3 diagram with (b, c) = (7, 2). The order seven for the pole at 7 = i 18
1
0
5
the facts that K7 has a simple pole at 7 = i 18 , and u57 corresponds to a fifth-order pole
of the S-matrix.
3.1.2. Second solution
As we will see, in this case the above mechanisms cannot explain a number of poles
sufficient to exclude the existence of some boundary bound states.
Lets start from the ground state. Type 1 diagram cannot be drawn, because there are not
the appropriate bulk fusing angles. If a = 1, 2, we already know that neither can Type 2,
Table 7
a
33
53
(b, c)
(2, 1)
(1, 1)
(1, 3)
(3, 1)
(1, 5)
(5, 1)
(1, 6)
(6, 1)
(6, 3)
(2, 5)
(3, 6)
(4, 1)
(2, 3)
Table 8
a
43
73
23
Type
(b, c)
(4)
(2)
(2, 1)
(1, 1)
(4, 4)
43
65
57
(1, 3)
(3, 1)
(5, 1)
(7, 5)
527
Table 9
a
a
(b, c)
3
(1, 2)
5
1
(2, 2)
33
(3, 2)
1
(1, 4)
53
(4, 2)
7
23
(4, 4)
45
(5, 4)
but for the remaining particles it explains the poles listed in Table 9.
In this way, from the ground state we get the excited stated , , . It is now easy to
see that in the corresponding amplitudes there are simple poles which cannot be explained
with alternative diagrams, and which excite the boundary to all the other bound states ,
, , . These are
Ka
K1
K3
K1
K1
3.2. The same analysis for E6 and E8

In these cases, none of the four solutions admits a reduction of the boundary bound states
number by means of generalized ColemanThun diagrams.
In the E6 case, this can be easily seen if we notice that particles 1, 2 and 6, for which
Type 2 diagram is not allowed, have simple poles at rapidities which forbid also Type 1
diagram, and are able to generate the whole set of boundary bound states.
An analogous mechanism works in the E8 case, because Type 2 diagram, when applied
to light particles, describes a narrow range of possible poles.
4. Perturbed minimal models

It is now interesting to see if the results obtained for affine Toda field theories can be
extended to minimal models perturbations.
It is known that the minimal parts of the Toda S-matrices, defined as

min
( ) =
s x+1 ( )s x1 ( ),
Sab
(4.1)
x
correspond to the scattering amplitudes of certain perturbed conformal field theories. The
E6 , E7 and E8 Toda theories are related, respectively, to the thermal perturbation of the
tricritical 3-state Potts model, the thermal perturbation of the tricritical Ising model and the
magnetic perturbation of the Ising model.
In the bulk theory, dressing a minimal S-matrix with coupling constant-dependent
CDD-factors doesnt induce any change in the bound states spectrum. These factors, in
fact, dont introduce new poles in the physical strip, and dont alter the sign of the existing
ones residues.
528
Starting from a Toda S-matrix of the form (2.8), where B is real and varies in [0, 2], we
can recover its minimal part performing the so-called roaming limit, which consists in
taking B = 1 + iC and sending the real quantity C to infinity.
In general, the residue of a given pole is a real function of C which preserves the same
sign it had for real B; its limit as C tends to infinity is the value dictated by the minimal
S-matrix.
Reflection matrices of the form (2.9) are manifestly factorized in a minimal part, which
satisfies equations (2.3)(2.6) with the minimal S-matrix, and a set of coupling constantdependent factors, which admit a roaming limit of the same form as for the {x} blocks.
We already know from (2.15) that these factors dont introduce new poles in the physical
strip 0 Im 2 .
As we have seen in Eq. (2.17), a general reflection amplitude is a product of the two
kinds of blocks {x} and Ky ( ). Let us assume that Ky ( ) has a pole at 0 = i h ; the
CDD-factors of the block {x} , evaluated at that rapidity, give:

sin 2h
( x 1 + B) sin 2h
( x + 1 B)
1

.
=
(4.2)
s x+1B (0 ) s x1+B (0 ) sin 2h
( + x + 1 B) sin 2h
( + x 1 + B)
h
If = x, this quantity is always positive, while if = x we have

(1 B)
sin2 2h
1

,
=
s x+1B (0 ) s x1+B (0 )
sin 2h (2x + 1 B) sin 2h
(2x 1 + B)
h
(4.3)
which is always negative as B varies in [0, 2], and vanishes in B = 1. If we now

parameterize B = 1 + iC, as a function of C we have

cos h ( x) cosh h C
1

,
=
(4.4)
s x+1B (0 ) s x1+B (0 ) cos h ( + x) cosh h C
h
which is a positive quantity for every C, and if = x vanishes in C = 0. This means that,
in the presence of a block {}, the corresponding pole can have different signs whether we
are considering the minimal or the Toda reflection amplitude.
This phenomenon is not present in the bulk theory, because S-matrix poles are always
located at positions shifted by 1 with respect to the blocks parameters x.
An analogous behaviour is produced by the CDD-factors of a block Ky ( ), evaluated at
0 = i h with = 3hy
2 . This, however, never happens for the En series elements, because
all the parameters y are odd, but the poles are always located at entire multiples of i h .
From this analysis we can conclude that a reflection amplitude pole located at 0 = i h
will have a residue with different sign in the minimal and in the Toda theory if this reflection
amplitude has an odd number of {} blocks.
If we think to the bulk situation this new possibility seems problematic. We have to
remember, however, that the two integrable field theories defined by a Toda Lagrangian
and a minimal model perturbation correspond in the UV limit to completely different
conformal field theories. Hence, although they share the minimal part of the S-matrix,
they could be governed by very different integrable boundary conditions, and this might
529
become manifest in distinct bound states structures of the corresponding reflection

amplitudes.
We will now study this phenomenon in the three examined theories.
4.1. The tricritical Ising model
Analyzing the minimal solution, we find many situations of the type described above,
always corresponding to poles with negative residue in the Toda theory, and positive in the
minimal one. In Table 10 we list the additional fusing angles with the usual conventions.
These poles create twenty new different boundary bound states, with reflection amplitudes
of the form

1
Ka ( ) =
(4.5)
Sab ( ) Ka0 ( ),
b
where the bs can now have different colours with respect of the bicolouration of the
Dynkin diagram, and the corresponding energy levels are related by
1
(mb ).
E = E0 +
(4.6)
2
b
If we analyze the reflection amplitudes on this new boundary states, we can see that their
poles generate a cascade of other bound states, with the possibility to have

1
n
Sab ( ) Ka0 ( ),
E = E0 +
(n mb ),
Ka ( ) =
(4.7)
2
b
with n = 1, 2, . . . .
This seems an indication that in this case the bootstrap doesnt close on a finite number
of boundary states.
For the states examined we have checked that the mentioned u-channel mechanism
cannot be applied to any new pole. However, if we consider the generalized Coleman
Thun diagrams we can explain all the new poles introduced on the eight Toda states by
the roaming limit. This seems an amazing coincidence, because we can always use
Type 2 diagram, and exactly particles 1 and 2, for which this diagram cant be drawn,
Table 10
a\
1
2
3
4
5
6
7
3
4
1, 3, 5
3
4
3
3
4
3, 5
1
43
5
1
43
1, 5
3
1
1, 3
43
4
1, 3
1, 3
43
530
dont produce any new pole. In this way we are also left with the two possibilities of a
bootstrap closing on two or eight boundary bound states.
Essentially the same situation arises with the second solution: the roaming limit
introduces many new positive residue poles in a similar way, and again ColemanThun
diagrams can describe all of them. This time also particles 1 and 2 generate new states, but
we can use both Type 1 and Type 2 diagrams.
4.2. The same analysis for E6 and E8
Also in the E8 case the roaming limit produces many additional positive residue poles,
which seem to indicate a non-closing bootstrap, but again we have an almost incredible
coincidence between the new poles and the ones we can explain with ColemanThun
diagrams, so that for both solutions these mechanisms allow the bootstrap to close on
sixteen states.
In the E6 case the situation is more delicate. As before, the roaming limit introduces
many new poles, and again the bootstrap presumably doesnt close. The problem is that
now the ColemanThun diagrams can describe almost all these poles, but in both solutions
four of them remain unexplained. For the minimal solution these are simple poles located
3
in K2 with = 1 , 2 , 1 , 2 , and they generate a cascade of states. In the
at = i 12
3
in K4 with = 1 , 2 , 1 , 2 , are
second solution case these poles, located at = i 12
triple, so that it is more difficult to conclude that they dont admit alternative diagrams.
However, we have tried to explain them with all kinds of diagram mentioned (including
Type 4), but we havent succeeded.
This seems to indicate that both the solutions analyzed dont have a physical meaning
for the minimal model. At this point, in order to confirm the whole construction, we need
to find at least one physical reflection matrix, using the CDD-ambiguity mentioned.
5. CDD-ambiguity: the E6 case

We start considering the minimal reflection matrix (2.20)(2.23). The problem with
3
the poles = i 12
in K2 ( = 1 , 2 , 1 , 2 ) is that they generate states characterized by
a mixing of the Dynkin diagram colourations, because

3
3
Sb2 + i Sb2 i = Sb1 ( )Sb6 ( ).
(5.1)
12
12
Their appearance is due to the fact that every minimal block {x} evaluated at = i xh is
a negative quantity, hence it changes the residue sign as it enters the reflection amplitude
expression at some excited state. This doesnt happen in the Toda theory because another
compensating negative sign arises from the coupling-dependent factors. It is then clear that,
adding a S-matrix factor to the initial reflection amplitude, we will hardly overcome this
problem, because we will get a more complicated pole structure and presumably we will
recover most of the previous boundary bound states.
531
The best strategy seems then to divide the reflection amplitude by an opportune S-matrix
element. The Dynkin colour which characterize the boundary bootstrap for the minimal
solution is the one of particles 2, 3 and 5, and we will start with the simplest choice, i.e.,
the self-conjugate particle 2. We then define:
a0 ( ) = S 1 ( )Ka0 ( )
K
a2
(5.2)
The fusing angles are listed in Table 11, where the brackets mean that the
corresponding pole can be explained by ColemanThun diagrams, of the type indicated
by the external subscript. We have kept the same letters for the boundary states, because
the relation between the excited state reflection amplitudes and the ground state ones is the
same as in the case of the initial solution (2.20)(2.23).
0 deserve a separate discussion. They would create again states
The poles at 11,2 in K1,6
on which the reflection amplitude of particle 2 has a change of sign from negative to
3
, due to the appearance of the block {3}. We
positive in the residue of the pole at i 12
already know that Type 1 diagram cannot be used, and there are not b and c such that
u2bc < 2 . However, we know that if a scattering amplitude Sbc ( ) has a pole at =
iuabc , then the amplitude Sbc
( ) will have a corresponding crossed-channel pole at
= i( uabc ). In this way, we could try to draw a crossed version of Type 2 diagram,
represented in Fig. 7.
We will indicate the modified fusing angles of this diagram by a tilde; these are related
to the direct-channel ones in the following way:
u bac = ubac ,
u abc = uabc ,

u cab = ucab + uabc u abc = ucab + 2uabc ,

a = i u abc + u cab = i uabc + ucab .
The boundary crossing equation (2.4) implies that the antiparticle reflection amplitude
to be evaluated at b = u abc is
Sbb (2 )
(5.3)
.
Kb ( )
This diagram can effectively explain the two poles in exam, with (a, b, c) = (1, 6, 3) and
(a, b, c) = (6, 1, 5).
Kb ( + i) =
Table 11
a\
1
2
3
4
5
6
11
(3)1
(2)1 6
(1)2 (23 )2 (3)2
(3)2
3
(3)2
12
(1)2 (23 )2 (3)2

(3)1
532
Fig. 7. Crossed version of Type 2 diagram.
Fig. 8. Crossed version of Type 3 diagram.
The same idea can be applied to Type 3 diagram, when the version to be used is the one
with Im (c ) > 2 , which again requires b and c such that uabc < 2 . The modified fusing
angles are defined as before, and the process is represented in Fig. 8.
Also this crossed diagram, although of second order, has an immediate application to
the two simple poles mentioned, because now the reflection amplitudes can have zeroes in
1
term. If we choose (a, b, c) = (1, 5, 4) (and (a, b, c) =
the physical strip, due to the Sa2
1
(6, 3, 4)), = 0 and = , we can describe the poles at a = i 12
, because particle 4
S55 (2)
S33 (2)
3
couples to the ground state at i c = i 12 and K () = K () has a simple zero in
5
2
b = i 12
.
In this way, we have a bootstrap closing on the two states 0 and : the fact that we have
skipped states 1,2 and 1,2 let us conclude that the same situation arises in the Toda theory
and in the minimal model.
6. Consequences of crossed diagrams

It is now interesting to see what happens if we extend the use of crossed diagrams,
fundamental in the last discussion, also to the other sets of reflection amplitudes examined.
First of all we summarize the various possibilities given by the application of the directchannel diagrams, listing in Table 12 the number of states on which the boundary bootstrap
closes (or presumably doesnt, if we indicate ) in the various systems analyzed. The
first column refers to the kind of theory: the voices Toda field theory and Minimal
Model discriminate between dressed and minimal scattering matrices, with a bootstrap
carried on all odd-order poles with positive residue, while with + ColemanThun we
mean the exclusion of boundary bound state creations if alternative diagrams can be drawn.
The first row indicates the six solutions examined (two for each algebra), with (1) and (2)
referring, respectively, to the minimal solution and to the one shifted by i .
Obviously the new diagrams increase the number of explicable poles, most of all for
light particles, but the principal novelty is that their order can also be lowered, because
in general the crossed reflection amplitudes (5.3) can have zeroes in the physical strip,
533
Table 12
(1)
E6
Toda field theory
Toda + ColemanThun
Minimal Model (MM)
MM + ColemanThun
8
8
(2)
E6
8
8
(1)
E7
8
2
2/8
(2)
E7
8
8
(1)
E8
16
16
16
(2)
E8
16
16
16
even if this is impossible for the direct-channel ones. However, this is true only if we are
on excited states, because on the ground state (5.3) exactly corresponds to going from the
minimal solution to the shifted one or vice versa.
This implies that if a simple pole of Kb0 cannot be explained by ColemanThun (normal
or crossed) Type 1 or Type 2 diagrams, then we can conclude that it creates a boundary
bound state, but this argument is not valid on excited states. On these states, in fact, even
if we dont find an opportune diagram to describe a certain pole (simple or multiple), we
cannot conclude that this pole corresponds to a boundary excitation, because we cannot
check all possible order diagrams with the opportune zeroes insertions.
We will now describe the combined effect of normal and crossed diagrams on the
various amplitudes considered.
Lets start with the E6 minimal solution. In the Toda case we are able to reduce the
number of boundary states from 8 to 4, getting a bootstrap which closes on 0, , 1 , 2 .
This doesnt solve the minimal model problem, because it seems again that the poles at =
3
in K2 1,2 generate an infinite cascade of boundary states; we couldnt find opportune
i 12
crossed diagrams to describe this simple poles, but as we have explained we could not
investigate all the possibilities.
As it regards the second solution, from the ground state we are sure to obtain states
and , skipping 1,2 , but we cannot exclude the subsequent creation of 1,2 and . Hence
for the Toda theory we can conclude that the bootstrap closes on a number of states between
3 and 6, while we have not a definite answer for the minimal model.
For the other two algebras we already know that the use of standard ColemanThun
diagrams gives the same boundary states spectrum in the Toda theory and in the perturbed
minimal model.
The bootstrap generated by the E7 minimal solution remains unchanged on the two
states 0 and . With the second solution we certainly have the creation from the ground
state (without , and ), but we cannot decide what happens with the following states ,
and ; the bootstrap will then close on a number of states between 2 and 5.
The E8 case, finally, is very similar. The minimal solution generates states , , , ,
from the ground state, hence the bootstrap will close on a number of states between 6 and
all the 16. With the second solution, instead, we are sure to get states and avoiding ,
and , so that the uncertainty is between 3 and 13 states.
We summarize these results in Table 13, with the same conventions as described before.
Every entry of the form n m means that we cannot decide on how many states the
534
bootstrap closes, but this number should lie between n and m. The additional solution
(3)
indicated by E6 is the one with ground state amplitudes (5.2).
7. Discussion
We have performed a detailed analysis of the boundary states structures arising from
the reflection amplitudes found by Fring and Koberle, showing how generalized Coleman
Thun mechanisms can have interesting consequences compared with a blind iteration of the
bootstrap on all odd-order poles with positive residue. However, these on-shell methods are
not sufficient to outline a clear and definitive picture of the phenomenon. We are in fact left
with various kinds of problems.
The first one is to understand if the possibility of drawing generalized ColemanThun
diagrams really excludes the creation of a boundary bound state. This seems reasonable
in cases where the bootstrap closes only under this assumption, as for perturbed minimal
models, but we have seen that, for the minimal solution in the E7 Toda theory, the
alternative is between two closing bootstraps, one on eight states, and the other on two.
Another eventuality is that the same ground state reflection amplitudes correspond to
distinct boundary conditions, whose different physical properties become manifest in the
interpretation of certain poles. This phenomenon was recognized in [14] in the case of the
scaling LeeYang model, knowing independently the different spectra from a boundary
generalization of the truncated conformal space approach.
The direct strategy to face this problem would be to calculate the residues of the various
diagrams and compare them with the actual residue of the corresponding pole in the
reflection amplitude, but it is not known how to treat this perturbative calculations in the
presence of a boundary.
Another delicate point is the use of crossed diagrams, which are so important in the E6
case; again it would be necessary to calculate their contribute to the residues. Furthermore,
the possibility of inserting in these diagrams crossed reflection amplitudes with zeroes in
the physical strip makes it very difficult to conclude something about many poles, and
alternative methods are essential to check the existence of the related boundary states.
Finally, in this context we have no way to understand which is the boundary condition
related to a certain reflection matrix, and the best we can do is just to notice whether
an eventual symmetry of the systems is preserved or not by the corresponding excited
boundary states structure. This is a particularly delicate problem, especially in the light of
the difference between the bound states spectra displayed by the minimal and the Toda
Table 13
(1)
Toda
MM
E6
E6
(2)
E6
(3)
E7
(1)
E7
(2)
E8
(1)
E8
(2)
4
4
36
3
2
2
2
2
25
25
616
616
313
313
535
reflection amplitudes. It could be, in fact, that the two basic solutions analyzed in the
E6 case are related to boundary conditions that in the UV limit correspond to conformal
boundary conditions present in the CFT associated to the Toda Lagrangian but not in the
tricritical 3-state Potts model.
To solve these problems, we intend to proceed in the future work performing indirect
checks on certain properties of the theory, using TBA equations [18,19] and analyzing
one-point function behaviours [20]. The basic idea is to perform the UV limit in order to
calculate the g-function (1.4) related to a certain set of reflection amplitudes. The various
possible choices of excited boundary states should produce different values of g, which in
the case of minimal models can be compared to the ones given by the primary operators
present in the CFT using Eq. (1.5). In this way it should be possible to identify the physical
meaning of the reflection matrices, associating them to specific boundary conditions, and
to deduce which of the initial boundary bound states are really involved in the bootstrap.
Acknowledgements
Im grateful to G. Delfino, D. Fioravanti and R. Tateo for very useful discussions and
suggestions, and to G. Mussardo for having introduced me into the field and for his
continuous interest and contribution to the development of this work. I also thank S.I.S.S.A.
for a pre-graduate grant and for hospitality.
References
[1] S. Ghoshal, A.B. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 3841;
S. Ghoshal, A.B. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 4353, Erratum.
[2] J.L. Cardy, Adv. Studies Pure Math. 19 (1989) 127;
J.L. Cardy, Nucl. Phys. B 324 (1989) 581.
[3] I. Affleck, A. Ludwig, Phys. Rev. Lett. 67 (1991) 161.
[4] A. Fring, R. Koberle, Nucl. Phys. B 421 (1994) 159.
[5] A. Fring, R. Koberle, Nucl. Phys. B 419 (1994) 647.
[6] A. Fring, R. Koberle, Int. J. Mod. Phys. A 10 (1995) 739.
[7] R. Sasaki, Reflection bootstrap equations for Toda field theory, Plenary talk at the International
Conference on Interface between Physics and Mathematics IPM93, Hangzhou, China, YITP/U93-33, hep-th/9311027.
[8] E. Corrigan, P.E. Dorey, R.H. Rietdijk, R. Sasaki, Phys. Lett. B 333 (1994) 83.
[9] P. Bowcock, E. Corrigan, P.E. Dorey, R.H. Rietdijk, Nucl. Phys. B 445 (1995) 469.
[10] E. Corrigan, Integrable field theory with boundary conditions, Talk given at the International
Workshop on Frontiers in Quantum Field Theory, Urumqi, P.R. China, 1996, hep-th/9612138.
[11] G. Sotkov, C.J. Zhu, Phys. Lett. B 229 (1989) 391.
[12] P. Christe, G. Mussardo, Nucl. Phys. B 330 (1990) 465.
[13] V.A. Fateev, A.B. Zamolodchikov, Int. J. Mod. Phys. A 5 (1990) 1025.
[14] P. Dorey, R. Tateo, G. Watts, Phys. Lett. B 448 (1999) 249.
[15] G. Delfino, G. Mussardo, P. Simonetti, Nucl. Phys. B 432 (1994) 518.
[16] G.W. Delius, G.M. Gandenberger, Nucl. Phys. B 554 (1999) 325.
[17] P. Mattsson, P. Dorey, J. Phys. A 33 (2000) 9065.
536
[18] A. LeClair, G. Mussardo, H. Saleur, S. Skorik, Nucl. Phys. B 453 (1995) 581.
[19] P. Dorey, A. Pocklington, R. Tateo, G.M.T. Watts, Nucl. Phys. B 525 (1998) 641.
[20] R. Konik, A. LeClair, G. Mussardo, Int. J. Mod. Phys. A 11 (1996) 2765.

Legendres relation and the quantum equivalence

osp(4|4)1 = osp(2|2)2 su(2)0
Miraculous J. Bhaseen
Physics Department University of Oxford, Theoretical Physics, 1 Keble Road, Oxford, OX1 3NP, UK
Received 22 December 2000; accepted 4 April 2001
Abstract
Using explicit results for the four-point correlation functions of the WessZuminoNovikov
Witten (WZNW) model we discuss the conformal embedding osp(4|4)1 = osp(2|2)2 su(2)0 .
This embedding has emerged in Bernard and LeClairs recent paper cond-mat/003075. Given that
the osp(4|4)1 WZNW model is a free theory with power law correlation functions, whereas the
su(2)0 and osp(2|2)2 models are CFTs with logarithmic correlation functions, one immediately
wonders whether or not it is possible to combine these logarithms and obtain simple power laws.
Indeed, this very concern has been raised in a revised version of cond-mat/003075. In this paper
we demonstrate how one may recover the free field behaviour from a braiding of the solutions
of the su(2)0 and osp(2|2)2 KnizhnikZamolodchikov equations. We do this by implementing a
procedure analogous to the conformal bootstrap programme Nucl. Phys. B 241 (1984) 333. Our
ability to recover such simple behaviour relies on a remarkable identity in the theory of elliptic
integrals known as Legendres relation. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
Since the pioneering work of Belavin, Polyakov and Zamolodchikov [2], two-dimensional conformally invariant field theories have attracted a great deal of attention (see, for
example, [3]). A prominent rle in this arena is played by the WZNW models [46]
possessing additional Lie algebraic symmetry. In condensed matter physics, these models
have been used to describe the critical points of many low-dimensional physical systems.
In particular, in the theory of disordered systems, averaging over disorder may lead to field
theories defined over supermanifolds [7] and it is natural to expect that WZNW models
based on Lie superalgebras 1 may describe their critical points [1,1015]. In addition,
E-mail address: bhaseen@thphys.ox.ac.uk (M.J. Bhaseen).
1 For more information on Lie superalgebras we refer the reader to the work of Kac [8] and the dictionary on
superalgebras [9].
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 7 8 - X
538
M.J. Bhaseen / Nuclear Physics B 604 [FS] (2001) 537550
averaging over disorder, may lead to the special class of logarithmic conformal field
theories (LCFTs) [11,1523] originally introduced by Saleur [24] and Gurarie [25]. The
LCFTs have been extensively studied in both statistical physics and string theory [26
59]. Aside from their relevance in critical disordered systems, they have emerged in
problems of gravitational dressing [26,27], D-branes [6065], SeibergWitten models [66,
67], magnetohydrodynamics [6872] and the HaldaneRezayi quantum hall state [73].
A recent example of a supersymmetric LCFT has been considered in a paper by Bernard
and LeClair [1]. An essential ingredient in their analysis is the observation that the
Sugawara energymomentum tensor for the osp(4|4)1 Lie superalgebra, separates into two
commuting pieces:
Tosp(4|4)1 = Tosp(2|2)2 + Tsu(2)0 .
(1.1)
The conformal field theories appearing on the left and right hand sides of this expression
are quantum equivalent, if and only if, they possess the same operator algebra. 2 As the
authors of [1] are acutely aware, the decomposition (1.1) is a necessary, but not sufficient,
condition to ensure this quantum equivalence. In addition to (1.1) one also requires the
coincidence of the four-point functions of primary fields [74]. However, the osp(2|2)2
and su(2)0 WZNW models are known to contain logarithmic operators [11,33,38] whereas
the osp(4|4)1 WZNW model is known have power law correlation functions [1]. One
immediately enquires whether or not it is possible to combine these logarithms and obtain
simple power laws. In this paper we demonstrate that it is in fact possible to combine these
logarithms and achieve a reversion to power law behaviour.
The structure of this paper is as follows: in Section 2 we study the four-point functions
of the su(2)0 WZNW model. We note that these conformal blocks may be expressed
in terms of the complete elliptic integrals a representation which turns out to be
extremely fruitful. In Section 3 we study the four-point correlation functions of the socalled [0, 1/2] representation of osp(2|2) [75,76]. The KnizhnikZamolodchikov equation
for this representation has been provided by Maassarani and Serban [11]. As we discuss
more fully in [23], the conformal blocks undergo a dramatic simplification at k = 2,
and it is possible to implement the conformal bootstrap [2] on a reduced set of solutions
to the KnizhnikZamolodchikov equation. Remarkably, this reduced subset assumes a
simple form in terms of the complete elliptic integrals. In Section 4 we demonstrate how
Legendres relation [77,78] enables one to braid the conformal blocks of Sections 2 and 3
so as to reproduce the free field behaviour of osp(4|4)1 . Finally we present concluding
remarks and technical appendices.
2. The su(2)0 WZNW model

Following Knizhnik and Zamolodchikov [79] we consider the four-point functions of
the WZNW model
2 For a discussion of quantum equivalence see Subsection 3.7 of the book by Fuchs [74].

F , zi , z i = g 1 , 1 z1 , z 1 g 2 , 2 z2 , z 2 g 3 , 3 z3 , z 3 g 4 , 4 z4 , z 4 ,
539
(2.1)
where g transforms in the fundamental representation of su(N) and g = g 1 transforms

in the conjugate representation. We use the symbol to denote the ordered sequence of
su(N) tensor indices 1 , 2 , 3 , 4 . Conformal invariance allows us to write

2h ,
F , zi , z i = z14 z23 z 14 z 23
(2.2)
z, z ,
F
where h is the conformal dimension the field g(z, z ) (which in the case of su(2)0 is 1/8),
zij = zi zj , and the anharmonic ratios z and z are defined as
z12 z34
z 12 z 34
,
z =
.
z14 z32
z 14 z 32
The correlation function (2.2) admits the invariant decomposition
z=
m

Ii Ij Fij z, z ,
F , z, z =
(2.3)
(2.4)
ij =1
with the m = 2 su(N) invariant tensors I1 = 1 ,2 3 ,4 and I2 = 1 ,3 2 ,4 . The four

scalar functions Fij satisfy the coupled first order equations

1
1
dF
= P+
Q F,
(2.5)
dz
z
z1
where F denotes the 2 2 matrix Fij , and the matrices P and Q are given by

N 1 N
1
0
1
1
,
,
Q=
P=
2N
2N N N 2 1
0
1
(2.6)
where = 12 (N + k). In order to study the case su(2)0 we set N = 2 and k = 0 in

Eq. (2.6). Suppressing the antiholomorphic index j from the entries of F , one may reduce
the first-order differential equation (2.5) in this case to the following pair of equations

2
4z(1 z) F1
+ (3 4z) 8z(1 z)F1

F1 = 0,
(2.7a)
(3 2z)
F1 .
F2 = 2zF1

(2.7b)
2(1 z)
It is easily seen that (2.7a) admits the two solutions 3

F1(1) = z3/4 (1 z)1/4 E(z) K(z) ,
F1(2)
=z
3/4
(1 z)
1/4
E(1 z),
(2.8a)
(2.8b)
where K(z) is the complete elliptic integral of the first kind and E(z) is the complete
elliptic integral of the second kind as defined in Eqs. (A.3) and (A.4). The representation
(2.8) is particularly useful 4 owing to the very simple manner in which the elliptic integrals
3 Upon the change of variables F (z) = [z(1 z)]1/4 H (z) Eq. (2.7a) reduces to the canonical form of the
1
hypergeometric equation z(1 z)H
+ [c (a + b + 1)z]H abH = 0 with a = 1/2, b = 3/2 and c = 2. This

1
3
(1)
has solutions H =2 F1 [ 2 , 2 ; 2; z] and H (2) =2 F1 [ 12 , 32 ; 1; 1 z] which may be expressed in terms of the
elliptic integrals using the identities (A.7) and (A.10).

4 This becomes even more apparent when we study the case of osp(2|2) .
2
540
behave under differentiation with respect to the parameter z see Eq. (A.5). Applying
(2.7b) to these solutions yields:
F2 = z1/4 (1 z)3/4 E(z),

F2(2) = z1/4 (1 z)3/4 E(1 z) K(1 z) .
(1)
(2.9a)
(2.9b)
The results (2.8) and (2.9) agree with those of [33,38], but are presented in a form which,
for our purposes, is more convenient. In particular, solutions (2.8a) and (2.9a) develop
logarithms upon analytic continuation into the vicinity of z = 1; likewise solutions (2.8b)
and (2.9b) develop logarithms in the vicinity of z = 0. In Section 3 we perform a similar
analysis for the osp(2|2)2 WZNW model.
3. The osp(2|2)2 WZNW model

Following Maassarani and Serban [11] we consider the four-point functions of the
WZNW model

F , zi , z i = g 1 , 1 z1 , z 1 g 2 , 2 z2 , z 2 g 3 , 3 z3 , z 3 g 4 , 4 z4 , z 4 ,
(3.1)
where g transforms in the so-called [0, 1/2] representation of the Lie superalgebra
osp(2|2) and once again we use the symbol to denote the ordered sequence of indices
1 , 2 , 3 , 4 . The representation theory of osp(2|2) has been considered in [75,76]. In
particular, the [0, 1/2] representation is four-dimensional and the index i takes on the
values 1, 2, 3, 4. In the notation of [11], the indices 1, 4 are even (bosonic) while 2, 3 are
odd (fermionic). 5 As in Section 2, conformal invariance allows us to write (3.1) in the
form (2.2) where this time h = 1/8 for the field g(z, z ) transforming under [0, 1/2] [11].
The correlation function (3.1) also admits the invariant decomposition where the m = 3
osp(2|2) invariant tensors derived in [11] are reproduced in Appendix B. The nine scalar
functions Fij satisfy the coupled first-order equations (2.5) where this time F denotes the
3 3 matrix Fij and the matrices P and Q are given by
1
0
2
1
0
0
1
1
1
1 ,
P = 2 3 2
(3.2)
,
Q= 2
1
2
x
x
4 8
1
4 4
1
where x = 2 k. The two free parameters and correspond to the arbitrary relative
normalizations of the su(2) doublet (|1, |4) and the two singlets |2 and |3 which together
provide a four-dimensional basis for the [0, 1/2] representation of osp(2|2) [75,76].
Suppressing the antiholomorphic index j from the entries of F , one may reduce the firstorder differential equation (2.5) in this case to the following set of equations [11]
5 For reader unfamiliar with the Lie superalgebras we note that the Hilbert space basis states (upon which
matrix representations may be formed) carry a so-called grading the state vectors themselves may be either
bosonic or fermionic.
x 3 z3 (1 z)3 F3
+ x 2 (1 + 2x)z2(1 z)2 (1 2z)F3

+ xz(1 z) 1 x + 2xz 2x(2 + x)z(1 z) F3

+ 1 x + 2z + 2xz(1 z) F3 (z) = 0,
2 2

1
x D F3 (z) + 2x(1 z)DF3 (z) + (1 2z)F3 (z) ,
F2 (z) =
4 xz(1 z)

1
F1 (z) =
xDF3 (z) F3 (z) + (z 2)F2 (z),
4
541
(3.3)
(3.4)
(3.5)
where D = z(1 z) d/dz. The authors of [11] tackle (3.3) by introducing ancillary
functions F from which one may obtain the F3 via the differential relation F3 (1
xD)F . The F satisfy a third-order equation, which the authors of [11] solve in terms of
generalized hypergeometric functions see Eqs. (68)(70) of [11]. One of these solutions
is of the form 6
1

1/x

1
1
2
F0 (z) = z(1 z)
(3.6)
3 F2 2 , x , 1 x ; 1, 1 x ; 4z(1 z) .
In particular, for the study of osp(2|2)2 we must set x = 4. The generalized hypergeometric function appearing in (3.6) reduces under these circumstances to an ordinary hypergeometric function. Using the well known identities satisfied by the ordinary hypergeometric
functions, one is able to obtain from this result, two independent solutions to Eq. (3.3) in
terms of the elliptic integrals (see Appendix A.2):
F3(1) =
E(z)
,
[z(1 z)]1/4
F3(2) =
E(1 z) K(1 z)
.
[z(1 z)]1/4
(3.7)
These solutions may be verified straightforwardly using (A.5). Applying (3.4) and (3.5) to
these solutions yields:
K(z)
,
4 [z(1 z)]1/4
K(1 z)
(2)
F2 =
,
4 [z(1 z)]1/4
(1)
F2 =
zK(z)
,
4 [z(1 z)]1/4
zK(1 z)
(2)
F1 =
.
4 [z(1 z)]1/4
(1)
F1 =
(3.8)
(3.9)
As we demonstrate in [23], one may satisfy the demands of single-valuedness and crossing
symmetry the so-called conformal bootstrap [2] on the subspace of functions (3.7)
(3.9). In Section 4 we shall demonstrate how one may braid the su(2)0 conformal blocks
given by (2.8) and (2.9) with those of the osp(2|2)2 model appearing in (3.7)(3.9), so as
to produce simple power laws.
4. Legendres relation and osp(4|4)1

4.1. Embedding blocks
We begin by defining embedding blocks
6 We note the small but significant typing error in the last indicial argument of this function as it appears in
Eq. (68) of [11]. The other two solutions (69) and (70) are correct, however.
542
Eij (z) =
2

(a)
(b)
cijab Fi, su (z)Fj, osp (z),
(4.1)
a,b=1
which are a simple braiding of the (holomorphic) su(2)0 conformal blocks, Fi,(a)
su (z),
(a)
and the (holomorphic) osp(2|2)2 conformal blocks, Fi, osp (z). The cijab are, as yet,
undetermined coefficients which we shall determine by imposing single-valuedness and
crossing symmetry. In order to impose the condition of single-valuedness on the embedding
blocks, we consider the monodromy transformations of its constituent conformal blocks.
A monodromy transformation of a function of z consists in letting z circulate around some
other point (typically a singular point). We define

C0 F z, z = lim F ze2it , z e2it ,
(4.2)
t 1

C1 F z, z = lim F 1 + (z 1)e2it , 1 + z 1 e2it .
(4.3)
t 1
Using the analytic continuation formulae for the hypergeometric functions it is straightforward to see that
A
(b)
C0 Fi,(a)
(4.4)
A (z) = g0 ab Fi, A (z), i = 1, 2, 3,

(a)
(b)
A
C1 Fi, A (z) = g1 ab Fi, A (z), i = 1, 2, 3,
(4.5)
where the algebra index A is either su or osp, and the monodromy matrices g0A and g1A are
given by

i
0
i 2
g0su =
,
g1su =
,
2 i
0 i

i 0
i 2
osp
osp
g0 =
(4.6)
,
g1 =
.
2 i
0 i
osp
We notice that these matrices are related by gisu = gi . Demanding that the embedding
functions be invariant under both C0 and C1 , requires that cij11 = cij22 = 0 and cij12 = cij21
(= cij say). In other words,
(1)

(2)
(2)
(1)
Eij (z) = cij Fi, su (z)Fj, osp (z) Fi, su (z)Fj, osp (z) .
(4.7)
Since there are two invariant tensors in the su(2) sector (i = 1, 2), and three in the osp(2|2)
sector (j = 1, 2, 3), this defines a total of six functions Eij up to individual normalization.
In our explicit evaluation of these embedding blocks, we are able to utilize the elegant
identity in the theory of elliptic integrals which is now termed Legendres relation [77,78]:
E(z)K(1 z) + E(1 z)K(z) K(z)K(1 z) =
.
2
(4.8)
Straightforward substitution of the su(2)0 and osp(2|2)2 conformal blocks into (4.7)
reveals that the elliptic integrals always appear in the manner prescribed in Legendres
relation (4.8) not only are the embedding blocks single-valued, but they do not contain
543
any elliptic integrals. Absorbing numerical factors into the coefficients cij , and defining
Eij = cij eij , the renormalized embedding blocks may be summarized:
1
e12 = e13 = ,
z
e11 = 1,
e21 =
z
,
1z
e22 =
1
,
1z
e23 = 0.
(4.9)
In Section 4.2 we shall present the free field correlation functions of the osp(4|4)1 WZNW
model, and demonstrate how one may recover these results by combining the embedding
blocks with the tensorial structure of osp(2|2) and su(2).
4.2. Free fields and osp(4|4)1
The current algebra osp(4|4)1 admits a simple representation in terms of free fields [1].
We introduce the complex fields i , and their Hermitian conjugates i , where i = 1, 2,

is a species index, and the index = 1, 2, denotes boson and fermion, respectively. These
eight fields have the following non-trivial operator product expansions
i1 ,i2 1 ,2
i1 ,i2 1 ,2
,
i11 (z)i22 (w) = (1)1
,
(4.10)
zw
zw
and furnish a representation of osp(4|4)1. Let us consider the holomorphic correlation
function

Fi (z1 , . . . , z4 ) = i11 (z1 )i22 (z2 )i33 (z3 )i44 (z4 ) ,

(4.11)
i11 (z)i22 (w) =
where i denotes the ordered sequence i1 , i2 , i3 , i4 and similarly denotes 1 , 2 , 3 , 4 .

Conformal invariance restricts this correlation function to be of the form
Fi (z1 , . . . , z4 ) = [z14 z23 ]2h Fi (z),
(4.12)
where in this case h = 1/2. Setting z2 = 0, z3 = 1, z4 = , and correspondingly z1 = z,

one may extract the conformal block Fi (z):

Fi (z) = lim w i11 (z)i22 (0)i33 (1)i44 (w) .

(4.13)
w
This correlation function may be evaluated with the aid of Wicks theorem

Fi (z) = lim w i11 (z)i22 (0) i33 (1)i44 (w)

w

+ (1)(2 1)(3 1) i11 (z)i33 (1) i22 (0)i44 (w) ,
(4.14)
where the parity factor (1)(2 1)(3 1) contributes a minus sign if both fields 2 and 3 are
fermionic. Using the operator product expansions given in (4.10) one obtains
(1)3 1 i
(1)3 (2 1) i
I1, su 1 ,2 3 ,4 +
I2, su 1 ,3 2 ,4 .
(4.15)
z
z1
We shall now demonstrate how one may recover this result from the decomposition
Fi (z) =
Fi (z) =
2
3

j =1 k=2
Ij,i su Ik,
osp cj k ej k (z),
(4.16)
544
where denotes the ordered sequence of indices 1 , 2 , 3 , 4 , and we have introduced

the conjugate indices 1 = 4 and 2 = 3. We note in particular that the sum over the
osp(2|2) invariant tensors begins at k = 2 since the explicit form (B.1a) reveals that I1 = 0
for i = 1, 2. As we demonstrate in Appendix C, crossing symmetry demands that c22 =
c12 and c13 = 4 c12 . Substituting these explicit relationships into (4.16) along with
the eij given in Eq. (4.9) one finds

1
1
i
i
F (z) = c12 I1, su I2, osp 4 I3, osp + I2, su I2, osp
(4.17)
.
z
z1
Now, using the explicit form of the osp tensors (B.1b) and (B.1c) one may easily verify
that

3 1
(4 )g 1 ,2 3 ,4 ,
I2, osp 4 I3,
(4.18)
osp = (1)
3 (2 1)
I2,
(4.19)
(4 )g 1 ,3 2 ,4 ,
osp = (1)

where g is given by g = 4i=1 i /2 2, one sees that up to an irrelevant normalization,
Eq. (4.17) reduces to (4.15). That is to say, we have succeeded in braiding the su(2)0 and
the osp(2|2)2 conformal blocks, in a manner prescribed by single-valuedness and crossing
symmetry, so as to recover the free field correlation function of osp(4|4)1 .
5. Conclusions
In this paper we have demonstrated how one may braid the logarithmic conformal
blocks of the su(2)0 and osp(2|2)2 WZNW models so as to recover the free field
correlation functions of the osp(4|4)1 model. It is an interesting open problem to relate
these observations to the Hilbert space non-factorization arguments presented in [1].
Acknowledgements
The author would like to thank J.-S. Caux, V. Gurarie, I.I. Kogan, C. Pepin,
A.M. Tsvelik, for stimulating and valuable discussions. The author would also like to thank
EPSRC for financial support.
Appendix A. Hypergeometric series and elliptic integrals

A.1. Generalities and su(2)0
We gather here some useful properties of the hypergeometric series and their connections
to the complete elliptic integrals. A good general reference is [78]. More details may be
found in the monographs [80,81] or in the various books on special functions [8285].
A generalized hypergeometric series is of the form (see [83], 9.14(1))
p Fq [a1 , . . . , ap ; b1 , . . . , bq ; z] =

(a1 ) (ap )n zn
,
(b1 )n (bq )n n!
545
(A.1)
n=0
where the so-called Pochammer symbol is defined as

(a)n = a(a + 1) (a + n 1) =
)(a + n)
.
)(a)
(A.2)
When p = 2 and q = 1 we obtain the ordinary hypergeometric series. It is well known in

the theory of elliptic integrals that (see [83], 8.113(1) and 8.114(1) and note our different
conventions)

K(z) = 2 F1 12 , 12 ; 1; z
(A.3)
E(z) = 2 F1 12 , 12 ; 1; z ,
2
2
where K(z) is the complete elliptic integral of the first kind and E(z) is the complete
elliptic integral of the second kind 7 (see [83], 8.110(1) and the subsequent discussion)
1

K(z) =
0
1
(1 x 2 )(1 zx 2 )
dx,
1
1 zx 2
dx.
E(z) =
1 x2
(A.4)
The elliptic integrals have rather simple properties under differentiation with respect to the
parameter z (see [83], 8.123(2) and 8.132(4) and note our different conventions):
dE(z) E(z) K(z)
dK(z) E(z) (1 z)K(z)
=
,
=
.
(A.5)
dz
2z(1 z)
dz
2z
Given the relations (A.3) it is possible to express many other hypergeometric functions in
terms of the elliptic integrals. In particular, the definition (A.1) implies that
a1 a2
(A.6)
2 F1 [a1 + 1, a2 + 1; b1 + 1; z],
2 F1 [a1 , a2 ; b1 ; z] =
b1
and thus from (A.3) and (A.5) one obtains

1 3

4
(A.7)
E(z) K(z) .
2 F1 2 , 2 ; 2; z =
z
Further, using the Gauss recursion formula (see [83], 9.137(3))
(2 z + z)2 F1 [, ; ; z] + ( )2 F1 [, 1; ; z]
+ (z 1)2 F1 [, + 1; ; z] = 0,
with = = 1/2, = 1, one finds that
1 3
1 1

1
2 F1 2 , 2 ; 1; z = (1 z) 2 F1 2 , 2 ; 1; z .
(A.8)
(A.9)
Recalling (A.3) we may recast this in terms of elliptic integrals

1 3
2 F1 2 , 2 ; 1; z
2E(z)
.
(1 z)
(A.10)
7 Note that many texts on the theory of elliptic integrals (including Gradshteyn and Ryzhik [83]) denote the
parameter z by k 2 the so-called modulus. Although
this is purely a mater of convention, we find it preferable
to deal with functions of z, rather than k 2 , or indeed k.
546
A.2. Quadratic transformations and osp(2|2)2

Amongst the many relations satisfied by the ordinary hypergeometric functions are the
quadratic transformations

1
2,
2;
+
F
;
z
,
(a)

2
1
2
1
(A.11)

2 F1 , ; + + 2 ; 4z(1 z) =
1
2 F1 2, 2; + + 2 ; 1 z , (b)
where form (a) is valid inside the loop of the lemniscate |4z(1 z)| = 1 surrounding z = 0,
and form (b) is valid inside the loop surrounding z = 1 see, for example, 4(iii) and 4(v)
on page 97 of [80]. In particular, when applied to the ancillary function appearing in (3.6)
with x = 4 (and = 1/4, = 3/4, = 1) one obtains

z(1 z) 1/4 2 F1 1 , 3 ; 1; z ,
(a)
2
2
F0 =
(A.12)

z(1 z) 1/4 F 1 , 3 ; 1; 1 z. (b)
2 1
2 2
Using the Gauss recursion formula (see [83], 9.137(4))
2 F1 [, 1; ; z] 2 F1 [ 1, ; ; z] + ( )z2 F1 [, ; + 1; z] = 0,
(A.13)
with = 1/2, = 3/2, = 1, together with the results (A.3) and (A.7) one finds that

2
(A.14)
2E(z) K(z) .
12 , 32 ; 1; z =
Substituting this result into (A.12) and recalling that the solutions to the osp(2|2)2
KnizhnikZamolodchikov equation (3.3) are obtained through the differential relation
F3 (1 4D)F0 , one may utilize (A.5) and arrive at the solutions (3.7) stated in the
text.
2 F1
Appendix B. Invariant tensors for osp(2|2)

The non-vanishing components of the three invariant tensors for the [0, 1/2] representation of osp(2|2) are equal to the components of the three vectors [11]
I1 = (1144) + (1234)4 + (1324)4 (1414) + (2143)4
+ (2233)16 2 2 + (2323)16 2 2 (2413)4 + (3142)4
+ (3232)16 2 2 + (3322)16 2 2 (3412)4 (4141)
(4231)4 (4321)4 + (4411),
(B.1a)
I2 = (1234)4 (1243)4 + (1324)4 (1342)4

(1414) + (1441) (2134)4 + (2143)4 + (2233)32 2 2
+ (2323)16 2 2 + (2332)16 2 2 (2413)4 + (2431)4 (3124)4
+ (3142)4 + (3223)16 2 2 + (3232)16 2 2 + (3322)32 2 2
(3412)4 + (3421)4 + (4114) (4141) + (4213)4
(4231)4 + (4312)4 (4321)4 ,
547
(B.1b)
I3 = (1234) (1243) + (1324) (1342) + (1423) + (1432) (2134) + (2143)

+ (2233)8 + (2314) + (2323)8 + (2332)8 (2341) (2413)
+ (2431) (3124) + (3142) + (3214) + (3223)8 + (3232)8 (3241)
+ (3322)8 (3412) + (3421) (4123) (4132) + (4213)
(4231) + (4312) (4321).
(B.1c)
For example, one finds that I13412 = 4 whilst I11441 = 0.

Appendix C. Crossing symmetry
The correlation function (4.11) has the following transformation under the permutation
of the fields 2 and 3 (coordinates and indices) crossing symmetry:
i (z1 , z3 , z2 , z4 ),
Fi (z1 , z2 , z3 , z4 ) = PF
(C.1)
where denotes the sequence of indices 1 , 2 , 3 , 4 , denotes the permuted sequence

= (1)(2 1)(31) denotes the parity
of indices 1 , 3 , 2 , 4 (and similarly for i), and P
of the interchange. Interchanging z2 and z3 induces the transformation z 1 z (2.3) and
thus by Eq. (4.12), the crossing symmetry constraint may be also be written
i (1 z).
Fi (z) = PF
(C.2)
Returning to our embedding decomposition (4.16) and introducing the following tensors
(see Eq. (A.5) of [11])

Jh,
osp = PIk, osp ,
Jj,i su = Ij,i su ,
(C.3)
the crossing symmetry constraint takes the form

2
3

Ij,i su Ik,
osp cj k ej k (z) =
j =1 k=2
2
3

Jj,i su Jk, osp cj k ej k (1 z).
(C.4)
j =1 k=2
These new tensors admit the decompositions 8

J1, su = I2, su ,
J2, su = I1, su ,
J1, osp = I2, osp 4 I3, osp,
J3, osp = I3, osp.
(C.5)
Substituting these decompositions into Eq. (C.4) and equating the coefficients of Ij, su Ik, osp
on both sides, one finds the following identities which must be satisfied by the
combinations cij eij (z) if crossing symmetry is to be satisfied:
c12 e12 (z) = c22 e22 (1 z),
(C.6a)

8 These relations for osp(2|2) follow from Eq. (A.6) of [11] with the additional fact that I 1 ,2 ,3 ,4 = 0 for
1, osp
i = 1, 2 with the identifications 1 = 4, 2 = 3.
548
c13 e13 (z) = 4 c22 e22 (1 z) + c23 e23 (1 z),
(C.6b)
c22 e22 (z) = c12 e12 (1 z),
(C.6c)
c23 e23 (z) = 4 c12 e12 (1 z) + c13 e13 (1 z).
(C.6d)
Inserting the explicit expressions for eij (4.9) one finds that (C.6a) implies c22 =
c12 , and (C.6b) implies c13 = 4 c22 . Eqs. (C.6c) and (C.6d) simply reproduce these
identifications.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
D. Bernard, A. LeClair, cond-mat/003075.

A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Nucl. Phys. B 241 (1984) 333.
P.D. Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Springer, 1997.
J. Wess, B. Zumino, Phys. Lett. 37B (1971) 95.
S.P. Novikov, Sov. Math. Dock. 24 (1981) 222.
E. Witten, Commun. Math. Phys. 92 (1984) 455.
K. Efetov, Supersymmetry in Disorder and Chaos, Cambridge Univ. Press, Cambridge, 1997.
V.G. Kac, Functional Anal. Appl. 9 (1975) 263.
L. Frappat, P. Sorba, A. Sciarrio, hep-th/9607161.
D. Bernard, (Perturbed) conformal field theory applied to 2d disordered systems: an introduction, in: Low-dimensional Applications of Quantum Field Theory, NATO Adv. Sci. Inst. Ser. B
Phys., Vol. 362, Plenum, New York, 1997, hep-th/9509137.
Z. Maassarani, D. Serban, Nucl. Phys. B 489 (1997) 603, hep-th/9605062.
C. Mudry, C. Chamon, X.-G. Wen, Nucl. Phys. B 466 (1996) 383, cond-mat/9509054.
S. Guruswamy, A.L. Clair, A.W.W. Ludwig, Nucl. Phys. B 583 (2000) 475, cond-mat/9909143.
M.R. Zirnbauer, hep-th/9905054.
M.J. Bhaseen, I.I. Kogan, O.A. Soloviev, N. Taniguchi, A.M. Tsvelik, Nucl. Phys. B 580 (2000)
688, cond-mat/9912060.
J.-S. Caux, I.I. Kogan, A.M. Tsvelik, Nucl. Phys. B 466 (1996) 444, hep-th/9511134.
J.-S. Caux, N. Taniguchi, A.M. Tsvelik, Phys. Rev. Lett 80 (1998) 1276, cond-mat/9711109.
J.-S. Caux, N. Taniguchi, A.M. Tsvelik, Nucl. Phys. B 525 (1998) 671, cond-mat/9801055.
J.-S. Caux, Phys. Rev. Lett. 81 (1998) 4196, cond-mat/9804133.
I.I. Kogan, A.M. Tsvelik, Mod. Phys. Lett. A 15 (2000) 931, hep-th/9912143.
J. Cardy, cond-mat/9911024.
V. Gurarie, A.W.W. Ludwig, cond-mat/9911392.
M.J.Bhaseen, J.-S. Caux, I.I. Kogan, A.M. Tsvelik, in preparation.
H. Saleur, Nucl. Phys. B 382 (1992) 486, hep-th/9111007.
V. Gurarie, Nucl. Phys. B 410 (1993) 535, hep-th/9303160.
A. Bilal, I.I. Kogan, hep-th/9407151.
A. Bilal, I.I. Kogan, Nucl. Phys. B 449 (1995) 569, hep-th/9503209.
M. Flohr, Int. J. Mod. Phys. A 11 (1996) 4147, hep-th/9509166.
H.G. Kausch, hep-th/9510149.
A. Shafiekhani, M.R.R. Tabar, Int. J. Mod. Phys. A 12 (1997) 3723, hep-th/9604007.
M. Flohr, Int. J. Mod. Phys. A 12 (1997) 1943, hep-th/9605151.
M.R. Gaberdiel, H.G. Kausch, Phys. Lett. B 386 (1996) 131, hep-th/9606050.
J.-S. Caux, I. Kogan, A. Lewis, A.M. Tsvelik, Nucl. Phys. B 489 (1997) 469, hep-th/9606138.
M.R.R. Tabar, A. Aghamohammadi, M. Khorrami, Nucl. Phys. B 497 (1997) 555, hepth/9610168.
549
[35] A. Aghamohammadi, A. Alimohammadi, M. Khorrami, Mod. Phys. Lett. A 12 (1997) 1349,

hep-th/9611009.
[36] F. Rohsiepe, hep-th/9611160.
[37] A.M. Ghezelbash, V. Karimipour, Phys. Lett. B 402 (1997) 282, hep-th/9704082.
[38] I. Kogan, A. Lewis, Nucl. Phys. B 509 (1998) 687, hep-th/9705240.
[39] M.R.R. Tabar, S. Rouhani, Phys. Lett. B 431 (1998) 85, hep-th/9707060.
[40] M. Flohr, Nucl. Phys. B 514 (1998) 523, hep-th/9707090.
[41] M. Khorrami, A. Aghamohammadi, M.R.R. Tabar, Phys. Lett. B 419 (1998) 179, hepth/9711155.
[42] E.V. Ivashkevich, J. Phys. A 32 (1999) 1691, cond-mat/9801183.
[43] I.I. Kogan, A. Lewis, Phys. Lett. B 431 (1998) 77, hep-th/9802102.
[44] N.E. Mavromatos, R.J. Szabo, Phys. Lett. B 430 (1998) 94, hep-th/9803092.
[45] A.M. Ghezelbash, M. Khorrami, A. Aghamohammadi, Int. J. Mod. Phys. A 14 (1999) 2581,
hep-th/9807034.
[46] M.R. Gaberdiel, H.G. Kausch, Nucl. Phys. B 538 (1999) 631, hep-th/9807091.
[47] V. Gurarie, Nucl. Phys. B 546 (1999) 765, cond-mat/9808063.
[48] K. Kaviani, A.M. Ghezelbash, Phys. Lett. B 469 (1999) 81, hep-th/9902104.
[49] I.I. Kogan, Phys. Lett. B 458 (1999) 66, hep-th/9903162.
[50] Y.S. Myung, H.W. Lee, JHEP 9910 (1999) 009, hep-th/9904056.
[51] Sanjay, Mod. Phys. Lett. A 14 (1999) 1413, hep-th/9906099.
[52] S. Moghimi-Araghi, S. Rouhani, Lett. Math. Phys. 53 (2000) 49, hep-th/0002142.
[53] H.G. Kausch, Nucl. Phys. B 570 (2000) 699, hep-th/0003029.
[54] I.I. Kogan, J.F. Wheater, Phys. Lett. B 486 (2000) 353, hep-th/0003184.
[55] H. Hata, S. Yamaguchi, Phys. Lett. B 482 (2000) 283, hep-th/0004189.
[56] A. Nichols, Sanjay, Nucl. Phys. B 597 (2001) 633, hep-th/0007007.
[57] F. Kheirandish, M. Khorrami, hep-th/0007073.
[58] S. Moghimi-Araghi, S. Rouhani, M. Saadat, hep-th/0008165.
[59] M.A.I. Flohr, hep-th/0009137.
[60] N.E. Mavromatos, R.J. Szabo, Phys. Rev. D 59 (1999) 104018, hep-th/9808124.
[61] N.E. Mavromatos, R.J. Szabo, hep-th/9811116.
[62] J. Ellis, N.E. Mavromatos, E. Winstanley, Phys. Lett. B 476 (2000) 165, hep-th/9909068.
[63] A. Campbell-Smith, N.E. Mavromatos, Phys.Lett. B 488 (2000) 199, hep-th/0003262.
[64] N.E. Mavromatos, E. Winstanley, Int. J. Mod. Phys. A 16 (2001) 251, hep-th/0006022.
[65] G.K. Leontaris, N.E. Mavromatos, hep-th/0011102.
[66] A. Cappelli, P. Valtancoli, L. Vergnano, Nucl. Phys. B 524 (1998) 469, hep-th/9710248.
[67] M. Flohr, Phys. Lett. B 444 (1998) 179, hep-th/9808169.
[68] M.R.R. Tabar, S. Rouhani, Nuovo Cimento B 112 (1997) 1079, hep-th/9507166.
[69] M. Flohr, Nucl. Phys. B 482 (1996) 567, hep-th/9606130.
[70] M.R.R. Tabar, S. Rouhani, Europhys. Lett. 37 (1997) 447, hep-th/9606143.
[71] M.R.R. Tabar, S. Rouhani, hep-th/9606154.
[72] S. Skoulakis, S. Thomas, Phys. Lett. B 438 (1998) 301, cond-mat/9802040.
[73] V. Gurarie, M. Flohr, C. Nayak, Nucl. Phys. B 498 (1997) 513, cond-mat/9701212.
[74] J. Fuchs, Affine Lie Algebras and Quantum Groups, Cambridge Univ. Press, Cambridge, 1992.
[75] M. Scheunert, W. Nahm, V. Rittenberg, J. Math. Phys. 18 (1977) 155.
[76] M. Scheunert, W. Nahm, V. Rittenberg, J. Math. Phys. 18 (1977) 146.
[77] A.M. Legendre, Exercices de Calcul Intgral sur divers ordres de transcendantes et sur les
quadratures, 1811, Paris.
[78] E.T. Whittaker, G.N. Watson, A Course of Modern Analysis, Cambridge Univ. Press,
Cambridge, 1952.
[79] V.G. Knizhnik, A.B. Zamolodchikov, Nucl. Phys. B 247 (1984) 83.
550
[80] W.N. Bailey, Generalized Hypergeometric Series, Cambridge Tracts in Mathematics and
Mathematical Physics, Vol. 32, Cambridge Univ. Press, Cambridge, 1935.
[81] L.J. Slater, Generalized Hypergeometric Functions, Cambridge Univ. Press, Cambridge, 1966.
[82] A.P. Prudnikov, Y.A. Brychkov, O.I. Marichev, Integrals and Series Vol. 3, Gordon and Breach
Science Publishers, 1990.
[83] I.S. Gradshteyn, I.M. Ryzhik, Tables of Integrals, Series, and Products, Academic Press, 1965.
[84] A. Erdeyli, Higher Transcendental Functions: The Bateman Manuscript Project Vol. 1,
McGraw-Hill Book Company Inc., 1953.
[85] M. Abramowitz, I. Stegun, Handbook of Mathematical Functions, Dover, 1965.

Characterization of one-dimensional Luttinger

liquids in terms of fractional exclusion statistics
Yong-Shi Wu a , Yue Yu b , Huan-Xiong Yang b,c
a Department of Physics, University of Utah, Salt Lake City, UT 84112, USA
b Institute of Theoretical Physics, Chinese Academy of Sciences, Beijing 100080, China
c Department of Physics, Zhejiang University, Hangzhou 310027, China
Received 9 February 2000; accepted 9 March 2001
Abstract
We develop a bosonization approach to study the low temperature properties of one-dimensional
gas of particles obeying fractional exclusion statistics (FES). It is shown that such ideal gas
reproduces the low-energy excitations and asymptotic exponents of a one-component Luttinger liquid
(with no internal degrees of freedom). At low energy (or temperature), the bosonized effective theory
is identified to the c = 1 conformal field theory (CFT) with compactified radius determined by the
statistics parameter . Moreover, this CFT can be put into a form of the harmonic fluid description
for Luttinger liquids, with the Haldane controlling parameter identified with the statistics parameter
(of quasi-particle excitations). Thus we propose to use the latter to characterize the fixed points
of 1-d Luttinger liquids. Such a characterization is further shown to be valid for generalized ideal
gas of particles with mutual statistics in momentum space and for non-ideal gas with Luttingertype interactions: In either case, the low temperature behavior is controlled by an effective statistics
varying in a fixed-point line. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
It is well-known that the Landau theory of Fermi liquids fails to describe most of
one-dimensional (1-d) interacting many-body systems. To provide a substitute, Haldane
proposed, years ago, the concept of the Luttinger liquid [1], defined by a set of lowlying excitations and critical exponents of the asymptotic correlation functions. Like Fermi
liquids, there is a (pseudo-)Fermi surface for the quasiparticle-like excitations in Luttinger
liquids, so that the classification of low-lying excitations is similar to that in Fermi liquids.
However, the exponents of the asymptotic correlation functions (at low temperature) are
distinct from those for Fermi liquid theory. For one-component systems (without internal
E-mail address: yyu@itp.ac.cn (Y. Yu).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 2 0 - 1
552
Y.-S. Wu et al. / Nuclear Physics B 604 [FS] (2001) 551579
degrees of freedom), the low-energy or low-temperature behavior of a Luttinger liquid

is controlled by a single parameter, the Haldane controlling parameter. It controls not
only all exponents, but also the velocity ratios between different types of elementary
excitations. The Fermi liquid theory is a special case of the Luttinger liquids with the
Haldane parameter = 1.
In recent years, the failure of Landaus theory of Fermi liquids to describe several newly
discovered strongly correlated electron systems have revived the interests in the theory of
Luttinger liquids. Among other questions, compared with Fermi liquids, one would like
very much to know the answer to the following questions:
What is the physical meaning of Haldanes controlling parameter? Or more precisely,
how to use physical properties of low-lying excitations to characterize the concept of
Luttinger liquids?
Does Haldanes theory of Luttinger liquids possess universality in one dimension, just
like Landaus theory of Fermi liquids in three dimensions? Or equivalently, in terms
of modern language of renormalization group, does the Luttinger liquids describe the
infrared (or low-energy) fixed points in 1-d systems?
In what directions could one expect to go for generalizing the concept of Luttinger
liquids to higher than one dimensions?
In short, a characterization of 1-d Luttinger liquids, other than using a bunch of
excitations and exponents, is in demand for gaining more insights and looking for possible
generalization.
To achieve this, let us recall what motivated Landaus concept of Fermi liquids, which
is known to describe an infrared fixed point (or a universality class) of interacting electron
systems. The basic idea behind it is based on the following organizing principle for
interacting many-body systems: At low temperature, the low-lying excited states of an
interacting many-body system above a stable ground state can be viewed as consisting of
weakly coupled elementary excitations. Here weakly coupled only means that the total
energy can be written as a sum of single-particle (dressed) energies, while the dispersion
of the dressed energy may well depend on the total particle number, a signal of remnant
interactions between the quasiparticles. According to Landau, the ground state and the lowlying excited states of a Fermi liquid are approximately, to a good accuracy at sufficiently
low temperature, described by those of an ideal Fermi gas with dressed energy for the
quasiparticles.
We note the significant role played by the ideal Fermi gas distribution (with dressed
energy) in this description of Fermi liquids. Actually it is the ideal Fermi gas that gives
a characterization to the Fermi liquid fixed point, and a meaning to the universality of
the concept of Fermi liquids. This inspires us to try to give a characterization of the 1-d
Luttinger liquids along a similar line of thoughts, namely, using a properly generalized
concept of exclusion statistics, of which a special case is the usual Fermi statistics.
Because the concept of quantum statistics in statistical mechanics is independent of the
dimensionality of a system, a characterization of 1-d infrared fixed points using statistics, if
successful, would shed light on how to generalize to higher-dimensional non-Fermi liquids.
553
Fortunately, a generalization as such has been available recently, under the name of
fractional exclusion statistics (FES). It is based on a new combinatoric rule for the manybody state counting [2,3], which is essentially an abstraction and generalization of Yang
Yangs state counting [4,5] in 1-d soluble many-body models. FES has been shown to be
applicable to elementary excitations in a number of exactly solvable models for strongly
correlated systems [2,3,59], anyons in the lowest Landau level [3,10], and quasiparticle
excitations in the fractional quantum Hall effect [2,3,7,11]. The thermodynamics of the socalled generalized ideal gas (GIG) associated with FES have been studied [3] in a general
framework.
Inspired by these results, the thoughts along the lines indicated in the above paragraphs
have led two of present authors [12] to propose that at least for some strongly correlated
systems or non-Fermi liquids, their low-energy or low-temperature fixed point may be
described by a GIG associated with FES, similar to the way that of the Fermi liquid fixed
point by the ideal Fermi gas [13]. As a testimony to this proposition, a sketchy proof
was given in that short letter [12] that the low-T critical properties of the 1-d Luttinger
liquids are exactly reproduced by those of 1-d ideal excluson gas (IEG), if one identifies
the Haldane parameter of the former with the statistics parameter of the latter. (We call
the particles obeying the FES without mutual statistics exclusions.) Therefore, IEG can be
used to describe the fixed points of the Luttinger liquids. In this paper, we will present our
results obtained in [12] in details, much of which was not published before.
A main tool we use in this study is bosonization of the 1-d excluson systems at low T ,
la Tomonaga [14] and Mattis and Lieb [15]. To bosonize an IEG system is a little
bit tricky, because at low temperature the linearized dispersion of dressed energy versus
pseudo-momentum has different slope outside and inside the pseudo-Fermi sea: There is
refraction at both pseudo-Fermi points. In spite of this, we still manage to construct welldefined density fluctuation operators that obey the U (1) current algebras and physically
describe free phonons. Then, the TomonagaMattisLieb bosonization applies, resulting a
bosonized effective field theory, in agreement with Haldanes harmonic fluid description
of the Luttinger liquid [16]. Then the asymptotic correlation functions and their exponents
can be systematically calculated. In this way, the critical properties of IEG reproduce those
of the Luttinger liquids.
A remarkable consequence of our bosonization is that the low energy behavior of
IEG is controlled by a c = 1 conformal field theory (CFT) on a cycle S 1 with the
compactified radius [1722] R = 2/. The duality relation R 2/R has an explicit
physical interpretation, i.e., the particle-hole duality in IEG, 1/ [5,6].
The fact that the low-T behavior of IEG is controlled by a conformally invariant
theory is significant, implying that indeed IEG provides a characterization of infrared fixed
points, having the conformal invariance as required by renormalization group. We have
also studied the effects of mutual statistics between different pseudomomenta and of the
Luttinger-type (density-density) interactions among exclusons. In either case, the low-T
behavior is controlled by an effective statistics eff for excitations near the Fermi points,
the same way as in the case of IEG. In one dimension both the momentum-independent
part of interactions and change in chemical potential are relevant perturbations [13,23],
554
leading to a continuous shift in the fixed-point line parameterized by . All these will be
explained in details in the present paper.
To make this paper self-contained, we devote the next two sections, Section 2 and
Section 3, to reviewing the Luttinger liquid theory and the GIG associated with FES,
respectively. In Section 4, we discuss the low-energy behavior of the IEG system and
achieve its bosonization. In Section 5, the generalization to the GIG with mutual statistics
as well as the non-ideal gas with FES are provided. The last section is dedicated to
conclusions and discussions.
2. Luttinger liquid
The Luttinger liquid, which describes a very large class of one-dimensional interacting
many-body systems, is introduced because of the infrared divergence of certain vertices in
the Fermi liquid description of the 1-d systems. Some pioneering works have been done
in the Luttinger model before the Luttinger liquid concept [15,24]. The model has been
exactly solved by using the bosonization technique [15]. Haldane [1] re-solved the model
with the following important observations:
(i) Besides a linearized spectrum of non-zero mode excitations, i.e., the density
fluctuations (sound waves), there are two kinds of zero mode excitations, singleparticle excitations by adding extra particles to the system and persistent currents
by making Galileo boosts.
(ii) There is a fundamental relation among the velocities of these three types of
excitations
vs = vN vJ ,
(2.1)
where vs is the sound velocity, vJ the current velocity and vN a velocity related to
the change in particle number. The velocity ratios define a controlling parameter,
e2 , by
vN = vs e2 ,
vJ = vs e2 .
(2.2)
(iii) The above defined controlling parameter measures the essential renormalized
coupling constant, and is the unique parameter that determines the exponents of
power-law decay in the zero-temperature correlation functions.
Based on these observations, Haldane defined the Luttinger liquids as 1-d systems that
have similar behavior (i)(iii) at low temperature just like the Luttinger model. In this way
the Luttinger liquids are characterized through their excitations and the exponents of the
asymptotic correlation functions.
To be more precise, recall that the Luttinger model describes a one-dimensional
interacting fermion system with the Hamiltonian

1
dx dy V (x y)(x)(y).
H = dx ||2 +
(2.3)
2
In the low energy limit, the Hamiltonian (2.3) can be bosonized as

1
H = vs
|q|bq bq + (/L) vN M 2 + vJ J 2 ,
2
q
555
(2.4)
where bq are the standard boson annihilation operators, and M and J the operators
corresponding to adding extra particle and boosting persistent currents, whose eigenvalues
obey the following selection rule,
(1)J = (1)M ,
(2.5)
which rules out the odd charge boson excitations. The total momentum of the system also
has a bosonized form

qbq bq ,
P = kF + (/L)M J +
(2.6)
q
with kF being the Fermi momentum.

Eqs. (2.1), (2.2), (2.4)(2.6) turned to be universally valid for the description of the lowenergy properties of gapless interacting one-dimensional spinless fermion systems even
for those not exactly soluble with a conserved current J . This universality class is named
as the Luttinger liquid by Haldane [1]. The Luttinger liquid has a model-independent
representation, namely, the harmonic fluid description [16], which is convenient for
calculating the correlation functions. The results of the harmonic fluid representation are
listed in Appendix A, for later use to be compared with our bosonization theory of the IEG.
The Haldane theory of Luttinger liquids is based on the significant observation that the
low-T behavior of the Luttinger model is universal. Naturally arises the question: why is it
so? In this paper we intend to answer this question by pointing out a profound coincidence
of the low-T behavior of the Luttinger model and that of ideal excluson gas (IEG), i.e.,
ideal gas of particles obeying fractional exclusion statistics: the universality of the former
is due to that of the latter.
3. Generalized ideal gas

In quantum mechanics, there are two ways to define the statistics of particles. One is in
terms of the symmetry of the many-body wave function under particle exchange. The other
is based on the state counting. Here we are interested in the latter definition. As is wellknown, bosons and fermions have different counting for many-body states, or different
statistical weights W : the number of quantum states of N particles occupying a group of
G states is, for bosons and fermions, respectively, given by
Wb =
(G + N 1)!
N! (G 1)!
or Wf =
G!
.
N! (G N)!
(3.1)
A simple interpolation between bosons and fermions is given by [2,3]

W=
[G + (N 1)(1 )]!
,
N! [G N (1 )]!
(3.2)
556
with = 0 corresponding to bosons and = 1 to fermions. The physical meaning of this

equation is the following: by assumption, the statistical weight remains to be a single
combinatoric number, so one can count the states by thinking of the particles effectively
either as bosons or as fermions, with the effective number of available single-particle states
being linearly dependent on the particle number:
G(b)
eff = G (N 1)
(f )
or Geff = G (1 )(N 1).

(b)
(3.3)
(f )
Obviously, for genuine bosons (or fermions), Geff (or Geff ) is independent of the particle
number. In all other cases, either of the two Geff is linearly dependent on the particle
number. This is the defining feature of the FES. The statistics parameter tells us, on
the average, how many single-particle states that a particle can exclude others to occupy.
A proper understanding of this has been discussed in [25]. Thus, the expression (3.2) for the
statistical weight, W , formulates a generalized Pauli exclusion principle, as first recognized
by Haldane [2].
It is easy to generalize this state counting to more than one species, labeled by the
index i:

Gi + Ni 1 j ij (Nj ij ) !

.
W=
(3.4)

i (Ni )! Gi 1
j ij (Nj ij ) !
Here Gi is the number of states when the system consists of only a single particle of
species i. By definition, the diagonal ii is the self-exclusion statistics of species i, while
the non-diagonal ij (for i = j ) is the mutual-exclusion statistics. Note that ij , which
Haldane [2] called statistical interactions, may be asymmetric in i and j . The interpretation
is similar to that of the one-species case: The number of available single-particle states for
species i, in the presence of other particles, is again linearly dependent on particle numbers
of all species:

(f )
ij (Nj j i ) or Geff,i = G(b)
G(b)
(3.5)
eff,i = Gi
eff,i + Ni 1.
j
The definition (3.2) or (3.4) starts with a postulated form for the statistical weight, and
thus is more direct and convenient for the purpose of formulating quantum statistical
mechanics. One of us [3] has first formulated the quantum statistical mechanics by
proposing the notion of generalized ideal gas (GIG): A GIG satisfies the following two
conditions:
(i) The total energy (eigenvalue) is always of the form of a simple sum, in which the
ith term is linear in the particle number Ni :

E=
(3.6)
Ni i0 ,
i
with i0 identified as the energy of a particle of species i.

(ii) The state-counting (3.4) for statistical weight W is applicable. When there are no
statistical interactions (i.e., ij = 0 for i = j ), we have the usual ideal gas, which
we call as IEG.
557
With the assumptions (3.6) and (3.4), the thermodynamics of a GIG can be worked out
by the usual techniques in statistical mechanics. Consider a grand canonical ensemble at
temperature T and with chemical potential i for species i, whose partition function is
given by

0
Z=
(3.7)
W ({Ni }) exp
Ni i i /T .
{Ni }
As usual, we expect that for very large Ni , the summation has a very sharp peak around
i }. Using the Stirling formula,
the set of most-probable (or mean) particle numbers {N
i /Gi , and
introducing the average occupation number per state defined by ni N
maximizing

(3.8)
ln W +
Gi ni i i0 /T = 0,
ni
i
one obtains the equations that determine the most-probable distribution of ni

(ij wj + gij )nj = 1,
(3.9)
with gij ij Gj /Gi , and wi being determined by the functional equations

wj ji
0
(1 + wi )
= e(i i )/T .
1 + wj
(3.10)
The thermodynamic potential = T ln Z and the entropy S are then given by

1 + ni j gij nj

P V = T
Gi log
= T
Gi ln 1 + wi1 , (3.11)
1 j gij nj
i

0
1 + ni j gij nj
i i

+ ln
Gi ni
S=
T
1 j gij nj
i
0 i
+ ln(1 + wi1 ) .
Gi ni i
=
T
(3.12)
Other thermodynamic functions follow straightforwardly. As usual, one can easily verify
i 2 )/N
i 2 , of the occupation numbers are negligible, which
that the fluctuations, (Ni 2 N
justifies the validity of the above approach.
4. Bosonization of 1-d ideal excluson gas

Let us first consider the simplest case, the 1-d IEG without internal degrees of freedom.
We expect to obtain a continuous interpolation between the usual ideal Bose and ideal
Fermi gas. Moreover, we want to show that the low-energy behavior of the IEG reproduces
that of the Luttinger liquid and, therefore, provides a better characterization of the infrared
fixed points associated with the Luttinger liquid.
558
4.1. Ideal excluson gas

Consider a GIG of N particles on a ring with size L. Single-particle states are labeled
by pseudo-momenta ki . The total energy and momentum are given by

E=
ki2 ,
(4.1)
P=
ki .
According to (3.5), in the thermodynamic limit the hole density, a (k, T ) (or the density
of available single-particle states) is linearly dependent on the particle density, (k, T ). By
definition, the statistics interaction matrix is given by
(ki , kj ) =
1a (ki )
.
1(kj )
(4.2)
Or in the thermodynamic limit, one has

(k, k ) = a (k)/(k ).
(4.3)
The system is called an IEG of statistics (with no mutual statistics between different
momenta), if
(ki , kj ) = (ki kj ),
(4.4)
or (3.5) reads
(kj ) =

1
1
+ (1 )
(kj ki )(ki )1k,
2
L
(4.5)
i=j
which, in the thermodynamic limit, can be simply written as [5]

a (k, T ) + (k, T ) = 0 (k, T ),
(4.6)
where 0 (k) 1/2 is the bare density of single-particle states. Thus, = 1 corresponds to
fermions, and = 0 to bosons. The thermodynamic potential, now reads, in terms of (3.11)
T
=
2

dk ln 1 + w(k, T )1 ,
(4.7)
with the function w(k, T ) a (k)/(k) satisfying an algebraic equation,

1
2
w(k, T ) 1 + w(k, T )
= e(k )/T .
(4.8)
Firstly, we consider the ground state, in which the particles are distributed in a
finite and origin-symmetric interval in the pseudo-momentum space. The (pseudo-)Fermi
momentum is defined by
kF2 = ,
(4.9)
and its value is fixed by the average particle density d0 = N0 /L in the ground state,
kF
kF
dk (k) = d0 .
(4.10)
559
Because holes are absent in the ground state, the particle density in the ground state is
easily obtained from (4.6),
1
, for |k| < kF ,
(k) = 2
(4.11)
0,
for |k| > kF .
Hence, one has
2

= d0 .
kF = d0 ,
(4.12)
Then the ground state energy and momentum are given by

E0
=
L
kF
kF
1
dk (k)k = 2 2 d03 ,
3
P0 =
kF
dk (k)k = 0.
(4.13)
kF
Now let us examine possible excitations in an IEG. First there are density fluctuations
due to particle-hole excitations, i.e., sound waves with velocity (see Subsection 4.2)
vs = vF 2kF .
(4.14)
Besides, by adding extra M particles to the ground state, one can create particle excitations,
and by Galileo boosts a persistent current. We observe that the velocities of these three
classes of elementary excitations in IEG also satisfy the fundamental relation (2.1). Indeed,
shifting N0 to N = N0 + M, the change in the ground state energy is
1
1
M E0 = 2 2 (N/L)3 2 2 (N0 /L)3
3
3

= 2 d02 M + (kF )M 2 + O M 3 /L3 ,
(4.15)
while a persistent current, created by the boost of the Fermi sea k k + J /L, leads to
the energy shift
kF +J
/L
J E0 =
kF
dk (k)k
2
kF +J /L
dk (k)k 2 = (kF /)J 2 .
(4.16)
kF
Therefore, the total change in energy, due to charge and current excitations, is
E0 M =

vF M 2 + 1 J 2 .
2L
The total momentum change due to the current excitations is

J
M
= d0 +
J.
P0 =
L
L
(4.17)
(4.18)
If we denote the variation in free energy as F0 = E0 M, and identify as the

controlling parameter e2 in the Luttinger liquid theory, (4.17) just recuperates the zero-
560
mode contributions 1 in (2.4). Comparing (4.17) with (2.4) we identify the velocities vN
and vJ to be
vN = vF ,
vJ = vF /.
(4.19)
Then we see the velocity relation (2.1), i.e., vs = vN vJ , that Haldane used to characterize
the Luttinger liquids, is satisfied in IEG. The selection rule (2.5) also holds for the IEG,
since the system should correspond to the ideal Fermi gas if = 1.
Encouraged by this relationship between the IEG and Luttinger liquids, we want to
calculate the critical exponents of IEG to see whether they reproduce those of the Luttinger
liquids. This motivates to develop a bosonization for the density fluctuations in IEG.
4.2. Low energy limit and bosonization
Following Yang and Yang [4], we introduce the dressed energy 4(k, T ) by writing
w(k, T ) = e4(k,T )/T .
(4.20)
The point is that the grand partition function ZG , corresponding to the thermodynamic
potential (4.7), is of the form of that for an ideal system of fermions with a complicated,
T -dependent energy dispersion given by the dressed energy:

ZG =
(4.21)
1 + e4(k,T )/T .
k
However, this fermion representation is not very useful, because of the implicit
T -dependence of the dressed energy. To simplify, we consider the low-T limit. By using
the dressed energy, (4.8) reads

4(k, T ) = k 2 T (1 ) ln 1 + e4(k,T )/T .
(4.22)
Because there is no singularity in 4(k, T ) at T = 0, the zero temperature dressed energy is
given by

k 2 kF2 /, |k| < kF ,
4(k) =
(4.23)
|k| > kF .
k 2 kF2 ,
Denote
4(k, T ) = 4(k) + 4(k,
T ),
(4.24)
where
4 (k, 0) = 0.
In the low-T limit, one has
2

k
1 1 T ln 1 + e|4(k)|/T ,
4(k, T ) =

k 2 (1 )T ln 1 + e|4(k)|/T ,
(4.25)
|k| < kF ,
|k| > kF ,
(4.26)
1 It is easy to understand why it is F rather than E that exactly contains the zero-mode contributions
0
0
in (2.4). In writing down the Hamiltonian (2.4), we have shifted the zero point of the energy to the Fermi surface.
561
Hence,

1 1 T ln 1 + e|4(k)|/T , |k| < kF ,

4 (k) =
( 1)T ln 1 + e|4(k)|/T ,
|k| > kF .
(4.27)
For low energies, one can consider only the excitations around the Fermi surface,
(T )
T
L
2
1
2

1
dk ln 1 + e4(k)/T +
2
kF
dk 4(k)
kF
kF

dk
4 (k, T )
1 + e4(k)/T

dk ln 1 + e|4(k)|/T
kF
kF +

dk ln 1 + e|4(k)|/T ,
(4.28)
kF
where the first term on the right-hand side of the last equality is recognized as (0)/L.
The cut-off is of order O(T /vs ) (actually, a few times of T /vs ). Mathematically, we take
the limit of T 0 followed by 0. Using the integral formula
2

,
dx ln 1 + ex =
12
we have the low-T thermodynamic potential

(T ) (0)
T 2
=
,
L
L
6vs
(4.29)
which implies that the theory is cut-off independent at low temperature. Notice that
F = N . Because we only consider the particlehole excitations near the Fermi
surface contribute to thermal excitations, N(T ) N(0) = 0, which can be checked by
an explicit calculation in terms of the definition of (k, T ). Thus, we have
T 2
F (T ) F (0) (T ) (0)
=
.
L
L
L
L
6vs
(4.30)
This means that the low energy behavior of the IEG is controlled by a c = 1 CFT [26]. This
result can be verified by a finite-size scaling in the spatial direction,
FL (0) F (0)
vs
(4.31)
= 2,
L
L
6L
where FL (0) is the zero temperature free energy for a system with size L. (For details,
see [9].)
The above relation agrees with the finite-size scaling of a conformally invariant system
with central charge c = 1. So we want to see whether the low-energy effective theory
of the IEG is really a CFT. Let us start with the grand partition function (4.21). At low
562
temperature, the solution (4.26) leads to 4(k,

T ) = O(T e|4|/T ), so one can simply replace
4(k, T ) with 4(k) in the grand partition function:

ZG
(4.32)
1 + e4(k) .
k
Note that the dressed energy with k outside the Fermi points kF has a slope different
from that with k inside kF . The former is 2/L and the latter 2/L. It is necessary to
keep this in mind for writing down the correct ground state wave functions of the excluson
system. Now that the dispersion 4(k) is T -independent, the grand partition function in the
low-T limit can be expressed in a fermionic representation as
ZG = Tr eHeff ,
(4.33)
where the effective Hamiltonian is given by

Heff =
4(k)ckck ,
(4.34)
where ck are fermionic creation operators. We also see that 4(kF ) = 0, which can be used
to define the Fermi momentum.
Physically, it is the phonon excitations that dominate the low-energy behavior of the
system. In the low-T limit, it is enough to consider the density fluctuations only near
the Fermi points, k kF , where the left- and right-moving sectors are separable and
decoupled:
Heff = H+ + H .
(4.35)
Besides this, another important simplification for excitations near Fermi points in the
low-T limit is that their energy, H , has a linearized dispersion:

vF (k kF ),
|k| > kF ,
4 (k) =
(4.36)
vF (k kF )/, |k| < kF .
We note the refraction at k = kF , which implies to create a particle with pseudomomentum k and to create a hole with k cost different energies, even if |k kF | =
|k kF |. The reason for this is that k is not the actual momentum carried by ck , as we
will see soon.
The key thing for bosonization is to construct a density fluctuation operator. Taking into
account the different slopes for dressed energy inside and outside the Fermi points, the
density fluctuation operator at k kF is constructed as follows:

q(+) = e
: ck+q ck : +
: ck+q
ck : +
: ckkF
ck : ,
k>kF
k<kF q
kF q<k<kF
+kF +q
(4.37)
q()
for q > 0. A similar density operator

q() =
: ckq
ck : +
k<kF
k>kF +q
can also be defined at k kF ,

: ckq
ck : +
: ck+kF
kF +q>k>kF
kF q
ck : .
(4.38)
563
To define the normal ordering we write, e.g.,

ck , k > k F ,
ck =
dk , k < kF ,
(4.39)
where dk is understood as a creation operator of a hole. Then normal ordering is done as

usual: putting the annihilation operators to the right of the creation ones. Hence we have,
e.g.,

: ck+q ck : +
: dk+q dk : +
: ckkF
dk : .
q(+) =
k>kF
k<kF q
kF q<k<kF
+kF +q
(4.40)
Within the Tomonaga approximation, 2 in which commutators are taken to be their
ground-state expectation value, we obtain

() ()
0| q() , q() |0
q , q

L
qq,q .
=
(4.41)
0|ck+q ck+q
1=
|0 = q,q
2
kF q<k<kF
kF q<k<kF
Also, the commutators between Heff and q() are

H , q() 0| H , q() |0 = vF qq() .
(4.42)
(4.41) and (4.42) describe 1-d free phonons with the sound velocity vs = vF (so we have
proved (4.14)). Introducing normalized boson annihilation operators

bq = 2/qL q(+) ,
(4.43)
bq = 2/qL q() ,
and adding back the zero mode contributions, the bosonized Hamiltonian satisfying (4.41)
is given by

1
1 2
HB = vs
(4.44)
q bq bq + bq bq +
,
M + J
2L
q>0
which agrees with the bosonized Hamiltonian (2.4) in the Luttinger liquid theory.
In passing, we make a comment on linearization of the dressed energy dispersion. When
we did this, we changed the ground state energy, because we assumed that for all k the
spectrum is linear in k. However, we changed neither the ground state wave function,
nor the low-T physics. On the other hand, the linearized spectrum was valid only for
phonon excitations, it has nothing to do with the zero-mode excitations. So, after the
linearized phonon part of the Hamiltonian is bosonized, we had to add back the zero-mode
excitations.
The construction of the bosonized momentum operator is a bit more tricky, because ck
does not carry a momentum k. Each term in (4.37) should carry the same momentum q,
2 Alternatively, we may extend the Fermi sea to that for two separate right- and left-moving fermions with
linear dispersion for k covering the whole real axis in each sector. Then our commutation relations are exact
(see [15]).
564
therefore, the fermion created by ck carries a dressed momentum p, which is related to k

by
k kF + (kF /), k > kF ,

|k| < kF ,
p(k) = k/,
(4.45)
k + kF (kF /), k < kF .

In terms of this variable, the linearized dressed energy 4(p) is of a simple form: 4 (p) =
vs (p pF ), with pF = kF /. The bosonized total momentum operator, corresponding

to the fermionized P = k p(k) ck ck , is

P=
(4.46)
q bq bq bq bq + d0 + M/L J.
q>0
We see that the fundamental velocity relation, the bosonized Hamiltonian and momentum, and the selection rule of the quantum numbers in the Luttinger liquid theory can all
be reproduced in IEG if we identify
e2 .
(4.47)
To say that IEG can be used to characterize the renormalization group fixed points of
Luttinger liquids, we still need to check the conformal invariance of the bosonized theory
of IEG, and to verify the critical properties of IEG reproduce those of the Luttinger liquids.
4.3. Effective field theory and conformal invariance
To check conformal invariance, we need to rewrite the above bosonized effective
Halmitonian (4.44) into a form of field theory in coordinate space. Employing the Fourier
transformation, the density operator can be written as

q iqx
MR
+
e bq + eiqx bq ,
R (x) =
(x) = R (x) + L (x),
L
2L
q>0

q iqx
ML
+
e
L (x) =
(4.48)
bq + eiqx bq ,
L
2L
q>0
where MR,L are given by M = MR + ML and bq = bq for q > 0.

The boson field (x), which is conjugated to (x) and satisfies

(x), (x ) = i(x x ),
(4.49)
is given by
(x) = R (x) + L (x),

0 JR x
+
+i
R (x) =
2
L
q>0

0 JL x
L (x) =
+
+i
2
L
q>0

iqx
e bq eiqx bq ,
2qL

iqx
e
bq eiqx b q ,
2qL
(4.50)
565
with J = JR + JL . We have to assign the quantum numbers such that there are only two
independent each other in MR,L and JR,L . A consistent choice is
MR = JR ,
ML = JL .
(4.51)
Then,
J = JR + JL ,
M = JR JL .
(4.52)
Here 0 is an angular variable conjugated to M: [0 , M] = i. The Hamiltonian (2.4)

becomes
1
H=
2
L

2
dx vN (x)2 + vJ / x (x) ,
(4.53)
or by a field rescaling,
H=
vs
L

2
1
dx (x)2 + x X(x) ,
4
(4.54)
where
(x) = 1/2 / 2 (x),
X(x) =
1/2
2
(x).
(4.55)
With X(x, t) = eiH t X(x)eiH t , the Lagrangian density reads

L=
vs
X(x, t) X(x, t).
4
(4.56)
This is the Lagrangian density of a free scalar field theory in 1 + 1 dimensions. Writing
the corresponding operators as the functionals of the scalar field X(x, t), all correlation
functions can be obtained by using the propagators of XR (x, t) and XL (x, t),

1
XR (x, t)XR (0, 0) = ln(x vs t),
2

1
XL (x, t)XL (0, 0) = ln(x + vs t).
2
(4.57)
The statistics of an operator in the theory can also be inferred by the commutators of the
scalar fields,

i
XR,L (x), XR,L (x ) = (x x ).
2
(4.58)
We recognize that L (4.56) is the Lagrangian of a c = 1 CFT [17], consistent with the
finite-size scaling (4.30). Alternatively, it is easy to check that the theory is invariant under
the conformal transformations generated by a set of the Virasoro generators
Lm =
1
nm n ,
2 n=
1
L m =
nm n ,
2 n=
(4.59)
566
where the oscillators m = m1/2 bq and m = m1/2 bq for m = qL/2 > 0 being integers.
0 = (/2L)1/2 [J 1/2 M1/2 ] and 0 = (/2L)1/2 [J 1/2 + M1/2 ]. The generators
obey the Virasoro algebra with the central charge c = 1,
tot tot

1 3
Lm , Ln = (m n)Ltot
(4.60)
m m m+n,0 ,
m+n +
12
with Ltot
m = Lm + Lm .
Since 0 is an angular variable, there is a hidden invariance in the theory under
+ 2 . The field X is thus said to be compactified on a circle, with a radius that is
determined by the exclusion statistics [9,19]:
X X + 2R,
R 2 = 2/.
(4.61)
Noting the selection rule (2.5), the Hamiltonian has a duality

1/,
M J,
(4.62)
which has referred to the particlehole duality [5,6]. Using the CFT terminology, this
duality is represented as the duality of the compactified radii,
R 2/R.
(4.63)
This is nothing but the duality relation in the usual c = 1 CFT [17] compactifed on a
circle [17].
4.4. Correlation functions
The CFT description of the IEG offers a better understanding for the space of quantum
states in the theory. States V [X]|0 or operators V [X] are allowed only if they respect the
invariance (4.61),
V [X + 2R] V [X],
(4.64)
with a given boundary condition restriction. Here, a Fermi or a Bose operator obeys the
periodic boundary condition (PBC). So quantum numbers of quasiparticles are strongly
constrained, in particular by the selection rule for zero-mode quantum numbers. For
example, the primary fields obeying the PBC in the CFT are given by

1
1
M,J (x) f J, X0 : ei(M/R+ 2 J R)XR (x) ei(M/RJ 2 )XL (x) : ,

0
0
f J, X0 = eiJ (1/RR/2)XR eiJ (1/RR/2)XL ,
(4.65)
where the prefactor f (J, X0 ) makes the fields satisfy the PBC, M and J eigenvalues of
the number and current operators, and X0 = Mx/L. The field carries the charge M and
current J . The conformal dimensions of the fields are

2
2
1 M 1
M 1
h=
(4.66)
+ JR +
JR
= M 2 + J 2 1 .
2
R
2
R
2
The statistics of the field can be calculated by using (4.58) and the statistics factors are

2
2

exp i
(4.67)
M1/2 + J /1/2 M1/2 J /1/2
= (1)MJ .
4
567
Consider the charge-1 primary fields, with M = 1. Therefore, they can only be fermions
since J = odd due to the selection rule. The general charge-1 fermion operator is a linear
combination of the charge-1 primary fields. A careful construction of the allowed fermion
field with unit charge leads to
F (x, t) = (x)1/2
eiOm : ei(1/R+ 2 (2m+1)R)XR (x ) : : ei(1/R 2 (2m+1)R)XL(x+ ) : ,
m=
(4.68)
where the prefactor f has been suppressed and the hermitian, constant-valued operators
Om satisfy 3
[Om , Om ] = i(m m ).
(4.69)
The multi-sector density operator is the linear combination of those primary fields with
M = 0 and J = even,

1
: exp i R2m XR (x) XL (x) : .
(x)
= F (x)F (x) = (x)

(4.70)
2
m
All the secondary fields in the CFT follow by considering the sound wave contribution
to the conformal weight of the fields.
The correlation functions can easily be calculated by using the CFT techniques. For
examples, the densitydensity and single particle correlation functions are as follows,

1
1
1
2
(x,
t)(0,
0) d0 1 +
+ 2
(2 d0 )2 xR2
xL

1
Am
cos 2 d0 mx ,
+
(4.71)
m2 /
[x
x
]
R
L
m=1
and

G(x, t) F (x, t)F (0, 0)
d0
Bm
m=
1
(1/2 +(2m+1)1/2 )2 /4
xR
1
(1/2 (2m+1)1/2 )2 /4
xL
ei(2(m+/2)d0x+t ) ,
(4.72)
where xR,L = x vs t and Am and Bm regularization-dependent constants.

Usually a physical quantity, e.g., a boson field, satisfies the periodic boundary conditions
(PBC). Hence, a charge-1 bosonic excitations are not allowed in the theory, because it
is anti-periodic. However, as we know, an anyon field neednt to obey the PBC. So in
the theory, there may be allowed anyonic excitations. A charge-1 anyonic (or exclusonic)
operator is a primary field that does not obey the PBC,
(x) = : F (x)e(1/RR/2)(XR(x)XL (x)) : .
(4.73)
3 Here operators O give rise to the Klein factors necessary for correct commutation relation for and .
m
F
F
568
The anyon commutation relation is easy to check:

(x) (x ) eisgn(xx ) (x ) (x) = 0,
for x = x .
(4.74)
In other words, the anyon field carries a fractional current. Or by the M J -duality, the
anyon with integer J carries a fractional charge. The correlation function of the singleanyon reads

G(x, t; ) (x, t) (0, 0)
(4.75)
d0
a
Bm
m=
(m+)2 /
xR
m2 /
xL
ei(2(m+/2)x+t ).
(4.76)
This correlation function coincides with the asymptotic one [7] in the CalogeroSutherland
model. We see that
(i) if = 1, (4.76) consists with (4.72);
(ii) there are no boson excitations ( = 0) because G(x, t; 0) = 0;
(iii) and moreover, > 0 is implied since (4.76) will diverge at the long distance if < 0;
(iv) look at m = 0. The critical exponents can be reads out,
f = + 1 ,
= 2.
Thus,
f > ,
if < 1,
f < ,
if > 1;
(v) the multi-sector density operator for exclusons is the same as that of the fermion.
|0 ( 1 )|0, with charge 1/ alone is not

The single-hole state, i.e., 1/
allowed. The minimum allowed multi-hole state is given by
(x1 ) 1/
(xp )|0,
1/
if = p/q is rational. One may obtain, e.g.,

p
p
p
1/ (x, t) 1/ (0, 0) G(x, t; 1/) .
A more interesting allowed operator is what creates q particle excitations accompanied
by p hole excitations:
q
p

(x, t) .
n(x,
t) = (x, t) 1/
We note the similarity of this operator to Reads order parameter [27] for fractional
quantum Hall fluids (in bulk). Its correlation function can be calculated by using Wicks
theorem:
p

q
n(x,
t)n(0,
0) G(x, t; ) G(x, t; 1/) .
(4.77)
If the contribution from the m = 0 sector dominates, then one gets

n(x,
t)n(0,
0) (x vs t)(p+q) .
569
5. Two extensions
Now we proceed to go beyond IEG. Two extensions will be discussed in this section: the
one-component GIG with the mutual statistics, and the non-ideal gas with the Luttingertype interactions. In either case, we will show that the low-temperature behavior is that of
an IEG, controlled by a single effective statistics parameter eff , whose value depends
on the mutual statistics and the coupling constants in the interactions.
5.1. Generalized ideal gas with mutual statistics
We turn to discussing the effects of mutual statistics. Consider a GIG with the statistics
matrix (4.3) in momentum space given by
g(k k ) = (k k ) + (k k ).
(5.1)
Here (k) = (k) is a smooth function. (k k ) stands for mutual statistics between
particles with different momenta; for IEG (k) = ( 1)(k). The thermodynamic
properties of GIG is given by Eq. (4.7), but now w(k, T ) satisfies integral equation [3,5]
which, in terms of the dressed energy (4.20), is of the form

4(k, T ) = 40 (k) + T

dk

(k k ) ln 1 + e4(k ,T )/T ,
2
(5.2)
where 40 (k) k 2 . In the low-T limit, it can be proven by the iteration [9] that 4(k, T ) =
4(k) + O(T 2 /vs ), where 4(k) is zero-temperature dressed energy given below. At T = 0,
the Fermi momentum kF is determined by
4(kF ) = 0.
(5.3)
Introduce
kF
( )[kF , kF ]
kF
dk
(k)(k),
2
kF
( )(k; kF , kF ]
kF
dk
(k k )(k ).
2
(5.4)
(5.5)
Then both (k) and 4(k) in the ground state satisfy an integral equations like
(k) = 0 (k) ( )(k; kF , kF ].
(5.6)
The dressed momentum p(k) is related to (k) by

dp(k) = 2(k)dk,
p(k) = p(k).
(5.7)
The ground state energy is given by

E0 /L = (40 )[kF , kF ].
(5.8)
570
Using the equation satisfied by (k), it can be expressed by the dressed energy
E0 /L = (4 0 )[kF , kF ].
(5.9)
The above equations are of the same form as those in the thermodynamic Bethe
ansatz [4], hence the Luttinger-liquid relation [28], vs = vN vJ , remains true. A simple
proof is sketched as follows. The sound velocity is well-known:
vs = 4(pF )/pF .
(5.10)
The charge velocity is given by

vN = vs z(kF )2 ,
(5.11)
where the dressed charge z(k) [28] is given by the solution to the integral equation
z(k) = 1 ( z)(k; kF , kF ].
(5.12)
This relation can be easily derived from the definitions

vN = L/N0 ,
z(k) = 4(k)/.
(5.13)
To create a persistent current, let us boost the Fermi sea by

kF kF + ,
(5.14)
where = z(kF )/L(kF ). Then the total energy of the state with the persistent current is
E /L = (40 )[kF + , kF + ] = (4 0 )[kF + , kF + ],
(5.15)
where
(k) = 0 (k) ( )(k; kF + , kF + ]
(5.16)
4 (k) = 40 (k) ( 4 )(k; kF + , kF + ].
(5.17)
and
Now, using the last expression for E and substituting 4 in (5.15), we have

2
4 (kF ) 0 (kF ) + (0 2F )(kF ; kF , kF ]
2

2
(5.18)
4 (kF ) 0 (kF ) + (0 2F )(kF ; kF , kF ] .
2
E /L = (4 0 )[kF , kF ] +
Here F (k, k ) is determined by

1
1
F (k, k ) =
(k, k )
2
2

kF
dk (k, k )F (k , k ).
(5.19)
kF
On the other hand, we note that the equation for 0 (k) can be rewritten as
(k) = 0 (k) (0 2F )(k; kF , kF ].
(5.20)
571
Thus, we have
E E0 = L2 4 (kF )(kF ) = (2/L)vs z(kF )2 .
(5.21)
This verifies vJ = vs z(kF )2 . In view of Eq. (4.19), at low energies, the GIG looks like an
IEG with
eff = z(kF )2 .
(5.22)
It can be shown that it is the effective statistics (5.22) that controls the low-T critical
properties of GIG, as does for IEG. Linearization near the Fermi points and bosonization
of the low-energy effective Hamiltonian go the same way as before for IEG. The only
difference now is that the slope of the linearized dispersion for the dressed energy 4 (k) =
4 (kF )(k kF ) + = vs (p(k) pF ) + , is smooth at k kF . So bosonization
is standard and the bosonized Hamiltonian is the same as Eq. (4.44) for IEG, only with
replaced by eff . However, before going to the bosonization we need an effective
Hamiltonian of the fermions with the dressed energy. Unlike the IEG, in the GIG case,
4(k, T ) = 4(k) + O(T 2 /vs ). Now, we work out the T -expansion of 4(k, T ) explicitly in
the low-T limit:

4(k, T ) = 4(k) + 4(k,
T ) + O T 3 /vs2 .
(5.23)
One finds that
4 (k, T ) =
T 2
f (k),
64 (kF )
(5.24)
with the function f determined by

f (k) = (kF k) ( f )(k; kF , kF ]

= (kF k) ( )(k; kF , kF ] + ( ) (k; kF , kF ] + . (5.25)
Note that the equation that (k) obeys can be rewritten as

(k)
=1
0
kF
kF

dk (k k ) + ( )(k ; kF , kF ]

8
+ ( ) (k ; kF , kF ] + .
(5.26)
Integrating (5.25) over k and comparing with (5.26), one has

kF
kF
(kF )
dk
f (k) = 1
,
2
0
(5.27)

T 2
dk
4(k,
T)=
1 2(kF ) .
2
64 (kF )
(5.28)
and then
kF
kF
572
Substituting (5.23) into the thermodynamic potential (4.7), we have

(T )
T
=
L
2

1
dk ln 1 + e4(k)/T +
2
dk
4(k,
T ).
1 + e4(k)/T
(5.29)
In the low-T limit, the first term in the last equation gives
(0)
T 2

,
L
64 (kF )
with
kF
(0)
1
=
L
2
dk4(k)
kF
and the second term is approximately given by (5.28). Thus,

(T ) (0)
T 2
(5.30)
=
,
L
L
6vs
which proves the central charge c = 1 CFT behavior of the theory at the low energy.
We may also confirm this from the finite size scaling in the spatial direction. To see this,
we consider the discrete version of the equation in which the density L (ki ) obeys

1
L (ki ) =
(5.31)
(ki kj ).
2
j =i
Using the relation between discrete sum and integration

N2
In
1
=
f
L
L
n=N1
one has
L (k)
(N2 1/2)/L
dx f (x) +
(N1 +1/2)/L

1
f (N1 1/2)/L f (N2 + 1/2)/L
24L2

+ O 1/L3 ,
(5.32)

1
1
1
(k k )
( L )(k; kF , kF ]
2
24L2 (kF )
dk
kF

1
1
(k k )
+
.
24L2 (kF )
dk
kF
(5.33)
Denote
L = + 1 ,
(5.34)
order O(1/L0 )
where (k) is of the

and 1 (k) is determined by
and 1 (k) the order
O(1/L2 ).
Then, (k) is as defined

(k k )
1
1
dk
24L2 (kF )
k
F

1
1
(k k )
+
( 1 )(k; kF , kF ].
24L2 (kF )
dk
kF
1 (k) =
(5.35)
573
The corresponding thermodynamic potential reads

1
L (0)
=
40 k(Ii /L)
L
L
i
kF
dk L (k)40 (k) +
=
kF
kF
=
1
24L2 (k
F)

40 (k)k 40 (k)k
F
kF
dk (k)40 (k) +
kF
dk 1 (k)40 (k)
kF

1
4 (k)k 40 (k)k .
+
(5.36)
F
F
24L2 (kF ) 0
The first term of the last equation is (0)/L and the rest, using (5.35), can be written as
1

40 (k) + (1)( 40 )
24L2 (kF ) k

+ (1)2 ( ) 40 + (k; kF , kF ]k=kF
1

+
40 (k) + (1)( 40 )
2
24L (kF ) k

+ (1)2 ( ) 40 + (k; kF , kF ]k=kF .
(5.37)
Recall the equation that 4(k) obeys, one has immediately,
4 (k)kF 4 (k)kF
vs
L (0) (0)
=
= 2
(5.38)
L
L
2(kF )
12L2
6L
as desired.
Similar to the case of IEG, we also could have a fermion representation of the grand
partition function with the temperature-dependent spectrum. To derive the low-energy
effective theory, however, one rewrites the thermodynamic potential (5.29) in the low-T
limit as
(T ) (0)
2T (kF )I (kF , T ),
(5.39)
L
L
where
kF +

dk ln 1 + e|4(k)|/T =
I (kF , T ) =
pF +
pF
kF

dp
(p) ln 1 + e|4(k((p))|/T . (5.40)
2
That is,
(T )
=
L
kF
kF
T
dk
4(k)
2
2
pF

dp ln 1 + e|4(k(p))|/T ,
where p is the physical (dressed) momentum. The grand partition function reads

ZG
1 + e4(k(k )) ,
k
(5.41)
pF +
(5.42)
574
where k = k for |k| < kF and k = p for |k| > kF . Now, we can have an effective
Hamiltonian because 4(k) is T -independent,

Heff =
(5.43)
4 k(k ) ck ck .
k
Similar to the IEG case, the low-T excitations can be considered by taking the
linear approximation near the Fermi points, and after bosonization the zero-temperature
excitations should be added back. The way to bosonize the linear Hamiltonian is also
similar to the case of IEG. Because the dressed energy is smooth at the Fermi points now,
the bosonization is even simpler. The density fluctuation operators are simply given by

: cp+q cp : ,
q() =
: cpq
cp : .
q(+) =
(5.44)
kkF
kkF
()
The commutators among q and H are

() ()

q , q
0| q() , q() |0 =

pF q<p<pF
0|cp+q cp+q
|0
= q,q
pF q<p<pF
1=
L
qq,q
2
(5.45)
and

H , q() 0| H , q() |0 = vF qq() .
(5.46)
Introducing the normalized bosonic annihilation operators

bq = 2/qL q() ,
bq = 2/qL q(+) ,
(5.47)
and adding back the zero-mode contributions, the bosonized Hamiltonian satisfying (5.46)
is given by

1
1
2
2
eff M +
,
q bq bq + bq bq +
J
HB = vs
(5.48)
2L
eff
q>0
which agrees with the bosonized Hamiltonian (2.4) in the Luttinger liquid theory. We see
that with replaced by eff , the bosonized Hamiltonian for the GIG is the same as that for
the IEG. So, all consequences we have obtained from the bosonized Hamiltonian in the IEG
case can be applied to the GIG case. Especially, there is an (allowed) eff describing the
particle excitation near the Fermi surface with both anyon and exclusion statistics being
eff . In this sense, one may say that the effect of mutual statistics is to renormalize the
statistics matrix.
Here we remark that in IEG, (k, k ) = ( 1)(k k ) is not smooth, so the dressed
charge has a jump at kF : z(kF+ ) = 1 and z(kF ) = 1 for kF = kF 0+ . The general
Luttinger-liquid relation is of the form
1

,
vJ = vs z kF+ z kF .
vN = vs z kF+ z kF
(5.49)
575
5.2. Non-ideal gas

Finally, we examine non-ideal gases, e.g., with general Luttinger-type densitydensity
interactions,

H = Heff + HI ,
Uq q q + q q + Vq q q + q q , (5.50)
HI =
L
q0
where Heff is given by (5.43) describing a GIG, and q and q are the excluson density
fluctuations near kF , respectively. After bosonization, the total Hamiltonian remains
bilinear in densities:
H = HB + HI

1
=
q (vs + Uq ) bq bq + bq bq + bq bq + bq b q
2

q>0
+ Vq bq bq + bq bq + bq bq + bq bq
+
2

1
vN M 2 + vJ J 2 +
U0 MR + ML2 + 2V0 MR ML
vs q. (5.51)
2L
L
q>0
Using the Bogoliubov transformation, the Hamiltonian can be easily diagonalized

1

q aq aq + a q a q + (/L) vN M 2 + vJ J 2 + E0 ,
2
q>0

E0 =
(q vs q),
H=
(5.52)
q>0
where
aq = cosh 0 bq sinh 0 bq ,
a q = cosh 0 bq sinh 0 bq ,
(5.53)
and the renormalized velocities are

1/2
vs vs = (vs + U0 )2 V02 ,
vN vN = vs e20 ,
vJ vJ = vs e20 .
(5.54)
with the controlling parameter 0 determined by

tanh(2 0 ) =
vJ vN 2V0
.
vJ + vN + 2U0
(5.55)
Thus, the Luttinger-liquid relation (4.19) survives with eff of GIG renormalized to
eff = e20 .
(5.56)
Note that the new fixed point depends both on the position of the Fermi points and on the
interaction parameters U0 and V0 , leading to non-universal exponents.
576
6. Discussions and conclusions
In conclusion, we have shown that 1-d IEG (without mutual statistics) exactly
reproduces the low-energy and low-T properties of (one-component) Luttinger liquids.
This gives rise to the following physical picture: At low temperature, the Luttinger
liquids can be approximately thought of as an IEG consisting of quasiparticle excitations.
Introducing mutual statistics or/and Luttinger-type interactions among these excitations
only shifts the value of eff . Thus the essence of Luttinger liquids is to have an IEG obeying
FES as their fixed point. This is our characterization of Luttinger liquids in terms of FES.
In this way, we have explicitly answered the three questions raised in the introduction
about Luttinger liquids:
The physical meaning of the Haldanes controlling parameter is the quasiparticles
effective statistics, eff .
The Luttinger liquids, more precisely, the IEG, indeed describe the infrared (or lowenergy) fixed points in 1-d systems, since their effective field theory at low energy
is conformally invariant. However, these fixed points are not isolated; they form
a fixed-point line. Both the chemical potential and coupling constants are relevant
perturbations that can drive the fixed point to move along the line, corresponding to
the renormalization of the effective statistics eff and leading to non-universal
exponents.
It is conceivable that some strongly correlated systems, exhibiting non-Fermi liquid
behavior, in two or higher dimensions may also be characterized as having a GIG with
appropriate statistics matrix as their low-energy or low-temperature fixed point. This
is because the concept of exclusion statistics is independent of spatial dimeniosnality
of the system.
Moreover, we also showed that the effective field theory of 1-d IEG is a CFT with central
charge c = 1 and compactified radius R = 2/ on S 1 . The particle-hole duality of the

exclusons is corresponding to the usual duality R 2/R. The CFT explanation makes a
better understanding of the single-particle operators, especially, the anyonic (or exclusonic)
ones. Also, the CFT techniques provide a systematic way to calculate the correlation
functions.
Finally, we observe several additional implications of this work: (1) Our bosonization
and operator derivation of CFT at low energies or in low-T limit can be applied to Bethe
ansatz solvable models, including the long-range (e.g., CalogeroSutherland) one [9].
(2) Here we have only consider one-species cases, i.e., with excitations having no internal
quantum numbers such as spin. Our bosonization and characterization of Luttinger liquids
are generalizable to GIG with multi-species, with the effective statistics matrix related to
the dressed charge matrix [9]. (3) The chiral current algebra in Eqs. (4.41) and (4.42) with
= 1/m coincides with that derived by Wen [29] for edge states in = 1/m fractional
quantum Hall fluids. So these edge states and their chiral Luttinger-liquid fixed points can
be described in terms of chiral IEG.
577
Acknowledgements
The authors thank referee to point out a misunderstanding for our selection rule of the
quantum numbers. This work was supported in part by the US NSF grant PHY-9309458,
PHY-9970701 and NSF of China.
Appendix A. The harmonic fluid description

In coordinate space, there is a harmonic fluid description [16] of the Luttinger liquid.
Instead of the representation that Haldane originally used, we prefer the rightleftmoving representation. The density operator can be written as the Fourier transformations

iqx

q
MR
+
e bq + eiqx bq ,
R (x) =
(x) = R (x) + L (x),
2
L
2Le
q>0

iqx

q
ML
+
e
L (x) =
(A.1)
bq + eiqx bq ,
2
L
2Le
q>0
where MR,L are given by M = MR + ML and bq = bq for q > 0.

The boson field (x), which is conjugated to (x) and satisfies

(x), (x ) = i(x x ),
(A.2)
is given by
(x) = R (x) + L (x),

0 JR x
R (x) =
+
+i
2
L
q>0

0 JL x
+
+i
L (x) =
2
L
q>0

e2 iqx
e bq eiqx bq ,
2qL

e2 iqx
e
bq eiqx bq ,
2qL
(A.3)
with J = JR + JL . We have to assign the quantum numbers such that there are only two
independent variables in MR,L and JR,L . A consistent choice is
MR = JR ,
ML = JL .
(A.4)
Then,
J = JR + JL ,
M = JR JL .
(A.5)
Here 0 is an angular variable conjugated to M: [0 , M] = i. The Hamiltonian (2.4)

becomes
1
H=
2
L
0

2
dx vN (x)2 + vJ / x (x) ,
(A.6)
578
or by a field rescaling,
vs
H=
L

2
1
dx (x)2 + x X(x) ,
4
(A.7)
where
(x) = e (x)/ 2,
X(x) =

2 e (x).
With X(x, t) = eiH t X(x)eiH t , the Lagrangian density reads

vs
L=
X(x, t) X(x, t),
4
which describes a free scalar field theory in 1 + 1 dimensions.
(A.8)
(A.9)
References
[1] F.D.M. Haldane, J. Phys. C 14 (1981) 2585.
[2] F.D.M. Haldane, Phys. Rev. Lett. 67 (1991) 937;
F.D.M. Haldane, in: N. Kawakami, A. Okiji (Eds.), Proc. 16th Taniguchi Symposium, SpringerVerlag, 1994.
[3] Y.S. Wu, Phys. Rev. Lett. 73 (1994) 922.
[4] C.N. Yang, C.P. Yang, J. Math. Phys. 10 (1969) 1115.
[5] D. Bernard, Y.S. Wu, in: M.L. Ge, Y.S. Wu (Eds.), Proc. 6th Nankai Workshop, World Scientific,
1995.
[6] C. Nayak, F. Wilczek, Phys. Rev. Lett. 73 (1994) 2740.
[7] Z.N.C. Ha, Nucl. Phys. B 435 (1995) 604.
[8] Y. Hatsugai, M. Kohmoto, T. Koma, Y.S. Wu, Phys. Rev. B 54 (1996) 5358.
[9] Y. Yu, H.X. Yang, Y.S. Wu, cond-mat/9911141.
[10] A. Dasnierea de Veigy, S. Ouvry, Phys. Rev. Lett. 72 (1994) 600.
[11] Y.S. Wu, Y. Yu, Y. Hatsugai, M. Kohmoto, Phys. Rev. B 57 (1998) 9907.
[12] Y.S. Wu, Y. Yu, Phys. Rev. Lett. 75 (1995) 890.
[13] See, e.g., R. Shankar, Rev. Mod. Phys. 66 (1994) 129.
[14] S. Tomonaga, Prog. Theor. Phys. 5 (1950) 544.
[15] D.C. Mattis, E.H. Lieb, J. Math. Phys. 6 (1965) 304.
[17] See, e.g., R. Dijkgraaf, E. Verlinde, H. Verlinde, Commun. Math. Phys. 115 (1988) 649.
[18] N. Kawakami, S.K. Yang, Phys. Rev. Lett. 67 (1990) 2493.
[19] R. Shankar, M.V.N. Murthy, Phys. Rev. Lett. 72 (1994) 3629.
[20] S. Iso, Nucl. Phys. B 443 (1995) 581.
[21] S. Iso, S.J. Rey, Phys. Lett. B 352 (1995) 111.
[22] R. Caracciolo, A. Lerda, G.R. Zemba, Phys. Lett. B 352 (1995) 304.
[23] See, e.g., H.J. Schulz, in: E. Akkermans, G. Montambaux, J. Pichard, J. Zinn-Justin (Eds.),
Proceedings of Les Houches Summer School LXI, Elsevier, Amsterdam, 1995, p. 533.
[24] J.M. Luttinger, J. Math. Phys. 4 (1963) 1154.
[25] Y.S. Wu, in: Y.M. Cho (Ed.), Proceedings of the Second Pacific Winter School for Theoretical
Physics, Sorak Mountains, Korea, January 1924, 1995, World Scientific, 1996, pp. 2759.
[26] I. Affleck, Phys. Rev. Lett. 56 (1986) 746.
[27] N. Read, Phys. Rev. Lett. 62 (1988) 86.
[28] F.D.M. Haldane, Phys. Lett. A 81 (1981) 153;

N.M. Bogoliubov, A.G. Izergin, V.E. Korepin, J. Phys. A 20 (1987) 5361.
[29] X.G. Wen, Phys. Rev. Lett. 64 (1990) 2206.
579

The Baxter equation for quantum discrete

Boussinesq equation
Kazuhiro Hikami
Department of Physics, Graduate School of Science, University of Tokyo, Hongo 7-3-1, Bunkyo,
Tokyo 113-0033, Japan
Abstract
Studied is the Baxter equation for the quantum discrete Boussinesq equation. We explicitly
construct the Baxter Q operator from a generating function of the local integrals of motion of the
affine Toda lattice field theory, and show that it solves the third order operator-valued difference
equation. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
In analysis of the integrable models, realized is the importance of the Baxter equation
since it was applied to the eight-vertex model (XYZ model) to compute the spectrum [1].
The Baxter equation is the functional difference relation among the transfer matrix t () and
the auxiliary function Q() which is called the Baxter Q operator, and is powerful enough
to determine the spectrum of the transfer matrix only from certain analytic conditions of
these functions.
Later Gaudin and Pasquier explicitly constructed the Baxter Q operator in an integral
form for the quantum periodic Toda lattice [2]. It was also shown [2,3] that in the
classical limit the logarithm of the Q operator is the generating function of the Bcklund
transformation, i.e., the Bcklund transformation is given by the classical limit of the
evolution equation, O QOQ1 (see for review Ref. [4]). This correspondence suggests
that the integrable dynamical system on spacetime lattice can be constructed by use of the
Q operator [5,6]. Recent studies indicate that the Q operator is useful enough to construct
the eigenfunction of the Toda lattice [7].
Since then the explicit form of the Baxter Q operator has been derived for other
integrable models such as the XXX model [8,9], the dimer self-trapping model [10], the
E-mail address: hikami@phys.s.u-tokyo.ac.jp (K. Hikami).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 0 4 - 8
K. Hikami / Nuclear Physics B 604 [FS] (2001) 580602
581
quantum KdV model [11], and the Volterra model (discrete KdV equation) [12,13]. All
these models are governed by the quantum algebra associated to s2 . In these cases as was
pointed out in studies of the integrability of the conformal field theory [11], the Baxter
2 ) Uq (s
2 ), with some
equation may be related with the universal R matrix, R Uq (s
2 ) algebra and an auxiliary
quantum finite-dimensional representation for the first Uq (s
infinite-dimensional q-oscillator representation for the second part. In fact proposed was
the universal procedure [14] to derive the Baxter equation from the universal R matrix
by use of the q-oscillator representation. Following this strategy it was shown that the Q
operator [12] for the Volterra model can be constructed from the universal R matrix for the
2 ) with the q-oscillator representation [15].
quantum algebra Uq (s
In this paper we construct the Baxter Q operator for the quantum discrete Boussinesq
3 ) symmetry. In the classical limit the discrete Boussinesq
equation which has Uq (s
equation is given as
dLn
= Wn + Wn1 + Ln (Ln+1 Ln1 ),
dt
dWn
= Wn (Ln+2 Ln1 ).
(1.1)
dt
We stress that Ln and Wn do not mean the Fourier modes of the generators L(x) and
W (x) but the discretization thereof. We note that the higher flow of Eq. (1.1) reduces in a
continuum limit [16] to
du
= uxx 2wx ,
dt
2
2
dw
= wxx + uxxx + uux .
(1.2)
dt
3
3
This set of equations gives the Boussinesq equation which is generated from the
pseudo-differential operator, L = 3 + u + w. In this sense we call Eq. (1.1) as the
discrete Boussinesq equation. The discrete Boussinesq equation (1.1) was first introduced
in Ref. [17] from studies of a discretization of the W3 algebra, motivated from the
close connection between the soliton equations and the conformal field theory [18
20]. The integrability of the discrete Boussinesq equation follows from the fundamental
commutation relation,

T(x) , T(y) = r(x, y), T(x) T(y) ,
(1.3)
where we omit the explicit form of the Z3 symmetric classical r
matrix T(x) is given from the Lax matrix as [16]
2
x xLn+1

1
W
W
Ln (x),
Ln (x) =
T(x) =
1
0
3
Wn
n
0
1
matrix. The monodromy
Wn
0 .
0
(1.4)
As a result we can construct the integrals of motion from the transfer matrix Tr T(x); some
of them are

1
(Ln )2 + Ln Ln+1 Wn .
H0 =
log Wn ,
H1 =
Ln ,
H2 =
2
n
n
n
582
As is well known, the discrete Boussinesq equation has also an intimate relationship
with a lattice analogue of the affine Toda field theory. We have constructed a generating
function Q of the local integrals of motion of the affine Toda lattice field theory [21], and
have also shown that this also commute with the transfer matrix of the discrete Boussinesq
equation. The function Q is constructed in terms of the quantum dilogarithm function as the
intertwining operator of the vertex operators. As was conjectured there, we shall show here
that this generating function is indeed the Baxter Q operator for the discrete Boussinesq
equation.
This paper is organized as follows. In Section 2 we briefly review the integrable structure
of the quantum discrete Boussinesq equation. Recalling the relationship with the affine
Toda field theory, we construct the transfer matrix and the generating function Q() of
the local integrals of motion for the affine Toda lattice field theory. In Section 3 we
explicitly show that the generating function Q plays a role of the Baxter Q operator for
the transfer matrix of the discrete Boussinesq equation. Based on the duality of the affine
Toda lattice field theory, we show in Section 4 that we have the dual Baxter equation for
the discrete Boussinesq equation. In Section 5 we briefly comment on the continuum limit
of the discrete Boussinesq equation, and see a relationship between the previously known
results. In Section 6 we consider the generalization for the N -reduced discrete KP equation
N ) structure. Last section is devoted for the concluding
which has the quantum group Uq (s
remarks and discussions.
2. Quantum Boussinesq equation

2.1. Lattice vertex operator and transfer matrix
To define the quantization of the discrete Boussinesq hierarchy (1.4), we use two sets of
the discretized free fields n(1) and n(2) , which respectively correspond to the Toda fields
associated to the simple roots 1 and 2 . We use the free field associated to the maximal
root as
n(0) = n(1) n(2) .
(2.1)
The commutation relations are defined as a discrete analogue of the usual relations of the
free chiral fields by 1
(a) (a)
m , n = 2i , for m > n, a = 1, 2,

(1) (2)
i ,
for n m,
n , m =
(2.2)
i , for n > m.
Throughout this paper we set the deformation parameter q as
q = ei .
(2.3)
(1)
1 We note that the dynamical variables and in Ref. [22] are defined by = en i , =
n
n
n
n
(1)
(2)

e n +n i , with q = ei .
583
Hereafter for our convention we often use a difference of the lattice free field
(a)
n(a) = n(a) n+1
,
(2.4)
whose commutation relations become local.

With these settings, we define the discrete screening charges as in the case of the
conformal field theory (see, e.g., Ref. [23]);
(a)
en ,
Q(a) =
(2.5)
n
3 ) is fulfilled;
for a = 0, 1, 2. We can see that the Serre relation of the algebra Uq (s

2
(a) 2 (b)
Q q + q 1 Q(a) Q(b) Q(a) + Q(b) Q(a) = 0,
Q
(2.6)
for a = b. The generators of the quantum W3 algebra are then defined so as to commute
with these screening charges. Note that the dual screening charges as a summation of the
dual vertex operators,
(a)
(a) =
Q
(2.7)
e n ,
n
commute with Q(a) , and that they satisfy the same Serre relation (2.6) replacing q with
q = e
(2.8)
As was shown in Ref. [16], the Lax matrix (1.4) for the discrete Boussinesq equation is
gauge-equivalent with
1 (1) 1 (2)
e3 n 3 n
(1) 2
(2)
1
Ln () = g()
e 3 n + 3 n
2
e e
1
e
e
1
g()Cn X(),
(1)
e 3 n
(2)
13 n
1
e
e
(2.9)
where we have modified the spectral parameter by x 3 = 1 e3 from Eq. (1.4). The
function g() is a normalization function which will be defined below. We introduce
diagonal elements of the matrix Cn as

Cn = diag cn(1) , cn(2), cn(3) ,
(2.10)
which corresponds to the weights in the vector representation 3 of s3 . Following the
standard way of studies in the integrable systems, we define the monodromy matrix and
the transfer matrix as
T() =

n
Ln (),
t1 () = Tr T().
(2.11)
584
The monodromy matrix satisfies the YangBaxter equation [22],

1
R( )T()T() = T()T()R( ).
(2.12)
Here the R-matrix is Z3 Z3 symmetric, and reads as
a
b c
c
b
c b
R() =
a
,
b c
c
b
c b
a
(2.13)
with
a = sh
3
1
2 2 i ,

1
b = e 6 i sh 32 ,

1
c = e 2 sh 12 i ,

1
b = e 6 i sh 32 ,

1
c = e 2 sh 12 i .
It follows that the transfer matrix commute with each other,

t1 (), t1 () = 0,
(2.14)
which supports the quantum integrability of our system.

We note that under the quasi-periodic boundary condition,
(a)
n+L
= n(a) + L iP (a) ,
for a = 1, 2,
(2.15)
the transfer matrix (2.11) can be rewritten as

t1 () = Tr e
Li(P (1) h3 +P (2) h2 )
L

g()e
1
3 i
Dn X()D1
n
(2.16)
n=1
where 3 3 diagonal matrix Dn is given by

Dn = exp n(1) h3 + n(2)h2 ,
1
ha = Eaa 11.
3
Matrix Eij is (Eij )mn = i,m j,n .

For our purpose to construct the Baxter equation for the discrete Boussinesq equation,
we use another transfer matrix which is associated with a representation 3 . Generally
the monodromy matrix for this type of representation can be constructed from one for the
585
vector representation by the fusion method [24], and in the s3 case the Lax matrix is
simply defined from Ln () ((Ln ())1 )t , where t denotes a transpose [25];
1 (1) + 1 (2)
e 3 n 3 n
(1)
(2)
1
2
Ln () = g()
e 3 n 3 n
2
(1)
e 3 n
(2)
+ 13 n
e 1 0
0
e 1
1 0
e
1
g()C
n X(),
(2.17)
where as before g()
is a normalization function. One sees that each element of C1

n
corresponds to weights of 3 . Correspondingly the monodromy matrix and the transfer
matrix for this representation is respectively defined by
T() =

Ln (),
t2 () = Tr T().
(2.18)
We note that under the boundary condition (2.15) the transfer matrix t2 () has a form,

t2 () = Tr e
Li(P (1) h3 +P (2) h2 )
L

g()e
1
3 i
D1
n X()Dn .
(2.19)
n=1
The commutativity of this transfer matrix follows from Eq. (2.12) as follows. We have
the fundamental commutation relation for T() from Eq. (2.12) as

(R( ))1
t1 t2
1
2
2
1
t t

T()T() = T()T() (R( ))1 1 2 ,
(2.20)
where ta denotes a transpose over the ath space. We see the commutativity of the transfer
matrix;

t2 (), t2 () = 0.
(2.21)
Furthermore, Eq. (2.12) also gives

(R( ))t2
1
2
2
1
1
1

T()T() = T()T() (R( ))t2
,
which proves the commutativity among two transfer matrices,

t1 (), t2 () = 0.
(2.22)
(2.23)
2.2. Fundamental L operator

We shall construct the generating function of the local integrals of motion for the affine
Toda lattice field theory.
For the Heisenberg operators p and q satisfying
[p,
q]
= 2i ,
586
operator as [22] 2
L
q,
L(;
p)
=
(q)
(p)
1
1
.
( + p)
(2 + q)
( q)
( p)
(2.24)
Here is a spectral parameter, and commute with every quantum operators, and the
functions () and () are, respectively, defined by

dx
eix
() = exp
(2.25)
,
4 sh( x) sh(x) x
R+i0
() = () ().
(2.26)
The integral () was first introduced by Faddeev [27] as a non-compact version of the
quantum dilogarithm function. See Appendix A for properties of these functions, and we
especially recall that the function () is the Gaussian (A.3). By use of the difference
equation (A.4), we have [22]

1
q 2i , p)
q,
L(;
= L(;
p)
e + ei +q + ei +p
i
+
q
1+e
+ ei +p

1
q,
1 + ei +q + e+i +p L(;
p),
=
e + ei +q + ei +p
(2.27)

1
+i
+
q
i
+
p
q,
q,
+e
L(;
p 2i ) = L(;
p)
1+e
e + ei +q + ei +p

1
q,
e + ei +q + ei +p L(;
p).
=
i
+
q
i
+
p
1+e
+e
(2.28)
As two operators n(1) and n(2) are the Heisenberg operators, [n(1), n(2) ] = 2i ,
we denote for brevity

; n(1), n(1) + n(2) .
Ln (; ) = L
(2.29)
We collect useful properties of this operator in Appendix B. This operator acts as the
intertwining operator for the vertex operators satisfying [21]
(a)
(a)

(a)
(a)
e n + e en+1 Ln (; ) = Ln (; ) e en + en+1 ,
(2.30)
for a = 0, 1, 2. The duality (A.2) of the integral (2.25) indicates that the operator Ln (; )
also intertwines the dual vertex operator;
(a)
(a)
n(a)
(a)

e
(2.31)
+ e e n+1 Ln (; ) = Ln (; ) e e n + e n+1 .
What is important here is that the fundamental L operator (2.29) itself satisfies the Yang
Baxter equation [28],
Ln (; )Ln+1 ( + ; )Ln (; ) = Ln+1 (; )Ln ( + ; )Ln+1 (; ).
(2.32)
2 Strictly speaking, the operator (2.24) is a non compact analogue of that was introduced in Ref. [22], i.e., we
replace the q-exponential function (A.1) with the Faddeev integral integral (2.25) [26].
587
Thus, when we set the Q operator as

Q() =

f ()Ln (; ),
(2.33)
with a normalization f (), we see that they commute between themselves with arbitrary
spectral parameters:

Q(), Q() = 0.
(2.34)
Furthermore, the fundamental L operator also satisfies [22]
Ln ( ; )Ln+1 ()Ln () = Ln+1 ()Ln ()Ln ( ; ),
(2.35)
which proves that the transfer matrices t1 () and t2 () commute with the Q operator (2.33),

Q(), t1 () = Q(), t2 () = 0.
(2.36)
In conclusion we have a commutative family; the transfer matrices t1 () and t2 (),
and the operator Q(). We note that intertwining relations (2.30) and (2.31) proves the
commutativity between the Q operator and the screening charges,

(a) = 0.
Q(), Q(a) = Q(), Q
(2.37)
As a consequence, the Q operator is a generating function of the local integrals of motion
3 Toda field theory [21], whose Hamiltonian and its dual are
of the affine s
H=
Q(a) ,
a=0
=
H
(a) .
Q
a=0
See Refs. [29,30] for the geometrical interpretation of the screening charges in the classical
limit of the sine-Gordon system.
3. Baxter equation
We shall show that the Q operator defined in Eq. (2.33) plays a role of the Baxter Q
operator for the transfer matrices (2.11) and (2.18) of the quantum discrete Boussinesq
equation.
3.1. Step 0
We prepare some formulae of the fundamental L operator (2.29); using elements of
(1) (2) (3)
matrix Cn = diag(cn , cn , cn ) in Eq. (2.9), we have

1
e 3 i Ln + 23 i ;
(1)

(2)
= L(1)
+ cn(3)
n (; ) cn e + cn e
(1)

(2)
(3)
= L(2)
n (; ) cn + cn e + cn e
(1)

+ cn(2) + cn(3)e ,
= L(3)
(3.1)
n (; ) cn e
588

1
e 3 i e3 1 Ln 23 i ;
1 (1)

= cn(2)
Ln (; )e L(2)
n (; )
1 (2)

= cn(3)
Ln (; )e L(3)
n (; )
1 (3)

Ln (; )e L(1)
= cn(1)
n (; ) .
(3.2)
Here we have defined L(a)

n (; ) by
(a) 1
(a)
L(a)
Ln (; )cn+1 ,
n (; ) = cn+1
which, using the commutation relations (2.2), are explicitly written as

(1)
2
(1)
(2)
2

L(1)
n (; ) = L ; n 3 i , n + n + 3 i ,

(1)
(1)
(2)
2
4

L(2)
n (; ) = L ; n 3 i , n + n 3 i ,

(1)
(1)
(2)
4
2

L(3)
n (; ) = L ; n + 3 i , n + n + 3 i .
See Appendix C for proof of Eqs. (3.1) and (3.2).
3.2. Step 1
We first consider the product of the Q operator (2.33) and the transfer matrix t1 (). As
we know from the commutation relations (2.2) of the lattice free field that a difference field
(a)
(a)
n does not commute only with nearest neighbor n1 , we get
Q()t1 () = Tr
= Tr

n

f ()g()Ln (; )Cn+1 X()

f ()g()M1 C1
n+1 Ln (; )Cn+1 X()Cn M.
(3.3)
Here a matrix M denotes a gauge transformation. We substitute
1
,
M = 1 1
1
1
into Eq. (3.3), and compute each element directly. We see that both (2, 1) and (3, 1)
components vanishes due to Eq. (3.1). Thus we find that a trace of the product of 3 3
matrix reduces to a sum of the product of (1, 1) component and a trace of the product of
remaining 2 2 matrices;
Q()t1 () =

1
f ()g()e 3 i Ln + 23 i ;
+ Tr

n
f ()g()An (),
(3.4)
where
An () =
Ln (; )e + Ln (; )e
(1)
(3)
Ln (; )e + Ln (; )
(2)

c
n
.
cn(3)
(1)
(2)
589
Ln (; ) + Ln (; )e
(1)
(3)
Ln (; ) + Ln (; )e
(1)
(2)
We now rewrite above 2 2 matrix An (). Using Eq. (3.2), we have

cn(1)e + cn(3) cn(2)e
13 i 3
An () = e
e 1
cn(1)e
cn(1)

(2)

cn
Ln 23 i ;
(3)
cn
1
1 (1)1 2 i
cn(3)

+ cn
e 3
cn(1)
3
(3) 1
(2)1 2 i
=e e 1
cn
cn
e 3

(3)

2
Ln 3 i ;
.
(1)
Ln 23 i ;
In the last equality, we have used the commutation relations (2.2) and a condition
1
cn(1)cn(2)cn(3) = e 3 i .
As a result we get
Q()t1 ()

1
f ()g()e 3 i Ln + 23 i ;
= Tr

n
1
2
+ cn(1) e 3 i
(3) 1
f ()g()e
cn

(3)

2
Ln 3 i ;

.
(1)
Ln 23 i ;
3

e 1
cn(3)
1
1
cn(1)
(2)1 2 i
e 3
cn
(3.5)
We keep aside this identity for a while.

3.3. Step 2
In the same manner with Step 1, we study a product of the Q operator and the transfer
matrix t2 (). We have
Q()t2 ()

1
f ()g()L
= Tr
n (; )Cn+1 X()
n
= Tr

1
1
f ()g()C
n+1 Ln (; )Cn+1 X()Cn
= Tr

n
f ()g()M
590
(1) 1
cn
e L(3)
n (; )
0
(1)1 (1)
Ln (; )
cn
1
cn(2) L(2)
n (; )
(2)1 (1)
cn
e Ln (; )
0
0

(3) 1 (3)
cn
Ln (; ) M1 .
(3) 1 (2)
cn
e Ln (; )

In the last expression, we have introduced matrix M for gauge-transformation. When we

substitute
1
,
M=
1
1 1 1
we see that both (3, 1) and (3, 2) elements vanish due to Eq. (3.2), and that a trace of the
product of 3 3 matrices can be rewritten as a sum of the product of (3, 3)-element and a
trace of the product of 2 2 matrices;
Q()t2 () =

13 i 3
f ()g()e
e 1 Ln 23 i ;
+ Tr

f ()g()B
n (),
(3.6)
where 2 2 matrix Bn () is given by

(1)1 (3)
cn
e Ln (; )
Bn () = (3)1 (3)
Ln (; )
cn
1

cn(2) L(2)
n (; )
(2)1 (1)
1
.
cn
e Ln (; ) + cn(3) L(3)
n (; )
We see that by applying Eq. (3.2) to (2, 2)-element of above matrix Bn () we have
Tr

f ()g()B
n ()
= Tr

n
f ()g()M
(1) 1
(3)
e Ln (; )
cn
(3)1 (3)
Ln (; )
cn
(2)1 (2)

cn
Ln (; )
(3)1 (2) 1 (2)
M1 ,
Ln (; )
cn
e + cn
where the invertible matrix M is included again for gauge-transformation. Substituting

1 1
M=
,
1
and erasing L(2)
n (; ) using Eq. (3.2), we get
Tr

f ()g()B
n ()
(1) 1 (3)1
cn
e + cn
(3) 1
= Tr
f ()g()
cn
n
(3)

Ln (; )
.
(1)
Ln (; )

1
cn(1)
(2) 1
e
cn
591
As a result we obtain an identity,

Q()t2 ()

13 i 3
e 1 Ln 23 i ;
f ()g()e
(1) 1 (3)1
cn
e + cn
(3) 1
= Tr
f ()g()
c
n
n
(3)

Ln (; )
.
L(1)
n (; )

1
cn(1)
(2) 1
e
cn
(3.7)
3.4. Step 3
By comparing Eq. (3.5) and Eq. (3.7), we find that 2 2 matrices in the right hand sides
have same form with each other. By erasing these 2 2 matrices, we obtain an identity for
operators t1 (), t2 (), and Q();

f ()
13 i
Q()t1 ()
Q + 23 i

g()e
2
n f + 3 i

f ()g()
3
=

e e 1 Q 23 i t2 23 i
2
2
n f 3 i g 3 i

32i

f ()
13 i 3
4
e
1
e
1
Q
i
.
g()e

3
4
n f 3 i
After setting normalization functions as
f () = 1,
g() = e+ 3 i ,

1
g()
= e 3 i e3+2i 1 ,
we see that the operator Q() solves the difference equation,

Q( + 2i ) t1 + 43 i Q + 43 i

+ t2 + 23 i Q + 23 i + 23 i Q() = 0,
(3.8)
where the function () is defined by

L

() = e3 1 e3+2i 1 .
As we know from Eq. (2.36) that the transfer matrices t1,2 () and the Q operator commute
each other, the operator valued difference equation (3.8) can be regarded as the third
order difference equation in usual sense, once we apply both hand sides to simultaneous
eigenfunction.
We expect in general the third order difference equation (3.8) has three linearly
independent solutions, which we write Qa () for a = +, 0, . We then introduce the
function P() as the quantum Wronskian of solutions;

Qa () Qa 2 i
3

,

Pab () =
(3.9)
Qb () Qb 23 i
592
for a, b = 0, (we thus have three independent functions Pab ()). It is straightforward to
see that the function Pab () satisfies the third order difference equation,

P( + 2i ) t2 + 23 i P + 43 i

+ + 23 i t1 + 23 i P + 23 i () + 23 i P() = 0.
(3.10)
From two difference equations (3.8) and (3.10), we get the Baxter equation for the
discrete Boussinesq equation:

P 23 i
P + 23 i Q 23 i
2
+
t1 () = 3 i
P()
P()
Q()

2
Q + 3 i
,
+
(3.11a)
Q()

P + 43 i
Q + 23 i
P()
t2 () =
+
()

Q()
P + 23 i
P + 23 i

2
Q 3 i
.
+ ()
(3.11b)
Q()
Once we have a set of functional equations (3.11a), (3.11b) we can follow a strategy in
Ref. [25] to obtain the spectrum of these transfer matrices; we suppose

Q() =
(3.12)
sh (1)
sh (2)
P() =
n ,
n ,
n
and substitute them into Eqs. (3.11a), (3.11b). From the condition of analyticity of the
transfer matrices ta () we get the nested Bethe ansatz equations. The precise analysis of
these equations is for future studies. As a consequence, we have diagonalized the transfer
matrices t1,2 () by use of the auxiliary functions Q() and P().
4. Duality
We have seen that the Q operator (2.33) satisfies the third order difference equation (3.8)
whose coefficients are given by two transfer matrices t1 () and t2 () for the quantum
discrete Boussinesq equation. As was noticed in Appendix A the Q operator has a property
of duality which follows from Eq. (2.31) and Eq. (A.2);
2
,
n(a) n(a) ,
(4.1)
.
Above duality implies the existence of the dual Baxter equation; we have the dual third
order difference equation,

Q( + 2i) u1 + 43 i Q + 43 i

+ 2 i Q() = 0,
+ u2 + 23 i Q + 23 i
(4.2)
3
where
=
()

1 e
3+2i

1
L
593
The operator Q() is as same with Eq. (2.33), and the transfer matrix ua () is defined by
u1 () = Tr U(),
U() =
u2 () = Tr U(),
U() =

n

Mn (),
(4.3)
Mn (),
(4.4)
with the dual Lax matrix M() and M(),

( 13 n(1) 13 n(2) )
e
(1)
(2)
1
( + 2 )
Mn () = h()
e 3 n 3 n
e
e
1

1
e ,
e

e
1
e
( 13 n(1) + 13 n(2) )
e
(1)
(2)
( 1 2 )
Mn () = h()
e 3 n 3 n
(1)
(2)
( 32 n 13 n )
(1) 1
(2)
2
( 3 n + 3 n )
1
0
e
.
1
0
e
1
0
e
To get the dual Baxter equation, another operator P should be slightly different from
Eq. (3.9),

2

ab () = Qa () Qa 32 i ,
P
(4.5)
Qb () Qb i
where Qa () denotes three linearly independent solutions of the difference equation (4.2).
This solves the third order difference equation,

+ 2i) u2 + 2 i P
+ 4 i
P(
3
3

+ 2 i ()
= 0.
+ 2 i u1 + 2 i P

+ 2 i P()
+
(4.6)
3
As a result, we obtain the dual Baxter equations,

+ 2 i Q 2 i
2 i

P
P
3
3
3
2

+
u1 () = 3 i

Q()
P()
P()

2
Q + 3 i
,
+
(4.7a)
Q()

+ 4 i

P
Q + 23 i
P()
3

u2 () =
+ ()

+ 2 i
+ 2 i
Q()
P
P
3
3

Q 23 i

,
+ ()
(4.7b)
Q()
from which we get the nested Bethe ansatz equations only from the analytic property of
the transfer matrices ua ().
594
We remark that this duality of the Baxter equation also appeared in studies of the discrete
KdV equation [31], where given was the interpretation from the viewpoint of the algebraic
geometry.
5. Continuum limit
We briefly study the continuum limit of the quantum discrete Boussinesq equation. In
(a)
a continuum limit, the lattice free field n is replaced by (a) (x), and the commutation
relations (2.2) then reduce into

(a)
(x), (b) (y) = Cab i sgn(x y),
(5.1)
where C is the Cartan matrix of s3 ,

2 1
C=
.
1 2
The quasi-periodic boundary condition (2.15) denotes
(a) (x + L) = (a) (x) + L iP (a) ,
which indicates the mode expansion of the free fields as [32]
(b) (x) = iQ(b) + iP (b) x +
(b)
an
n=0
ei L nx .
We can take the continuum limit of the transfer matrices t1,2 () from Eq. (2.16) and
Eq. (2.19). After a gauge-transformation, we get

1 L
g()e+ 3 i
t1 ()
Tr eiL(P
(1) h P (2) h )
1
3
P exp
L
dx
g()e

+ 13 i L
1
(1)
e+ 6 i e (x)
1
(0)
e2 6 i e (x)
e2 6 i e (x)
0
1
(2)
e+ 6 i e (x)
(1)
e+ 6 i e (x)
2 16 i (2) (x) ,
e
e
0
(0)
(5.2)
t2 ()
Tr eiL(P
(1) h +P (2) h )
1
3
P exp e
61 i
L
0
0
0
dx
e
(0) (x)
(1)
e (x)
0
0
0
e
(2) (x)
(5.3)
595
where P denotes the path operator ordering. One finds that the transfer matrix t2 () in
Eq. (5.3) essentially coincides with that was proposed in Ref. [32] as a quantized Drinfeld
Sokolov reduction.
Correspondingly a difference of the free fields (2.4) becomes the differential of the free
field,
(a)
(x),
x
and the Baxter Q operator for the quantum Boussinesq equation can be given by the
path ordering of the operator Ln (; ). As a result we recover the third order difference
equation (3.8). At this stage we are not sure how to relate the Q operator with the universal
R matrix.
n(a)
6. Generalization: N -reduced discrete KP equation

As the Boussinesq equation (1.2) is the 3-reduced KP equation, we can introduce the
N Toda field
quantum N -reduced discrete KP equation which is related with the affine s
theory. In the classical case, the Lax matrix for the discrete N -reduced KP equation is given
by [33,34]
LW
n (x) =
N
1
Wn(N)
x N1 x N2 W (2)
N3 W (3)
n+N2 x
n+N3
0
0
1
0
1
0
..
0
1
.
..
..
..
.
0
.
0
..
.
..
(N1)
(1)N2 xWn+1
0
0
..
.
..
.
1
(N)
(1)N1 Wn
0
0
..
.
..
.
0
(6.1)
where the generators of the lattice WN algebra are defined as
1
Wn(s) =
(i)

s1
N1
k=0 1 +
i=1 exp n+k

(i1 )
(is1 )
(i2 )
exp n+s1
+ n+s2
+ + n+1
1i1 <i2 <<is1 N1

(i1 )
(i2 )
(is )
,
exp n+s1 + n+s2 + + n
1i1 <i2 <<is N1
for s = 2, 3, . . . , N 1,

N1 (i)
exp
i=1 n+Ni
(N)

Wn =
(i) .

N1
N1
1
+
exp
n+k
k=0
i=1
(6.2)
596
Here n(a) is defined in terms of the free fields n(a) (for a = 1, 2, . . . , N 1) as

n(a) =
n(b) ,
b=1
and the quantum algebra for the lattice free fields reads as
(a) (a)
m , n = 2i , for m > n,

(a) (a+1)
i ,
for n m,
n , m
=
i , for n > m.
(6.3)
We note that the transformation (6.2) is the so-called Miura transformation. We set
n(0) =
N1

n(a),
a=1
and the screening charges are defined by (for a = 0, 1, . . . , N 1)

(a)
(a)
(a) =
en ,
Q
e n .
Q(a) =
n
(6.4)

(a)
The sum of the screening charges, N1
a=0 Q , corresponds to the Hamiltonian of the
affine Toda field theory.
In terms of the free fields, the quantum Lax matrix is then defined by [34]

N1

(a)
Ln () = exp
(6.5)
ha n X(),
a=1
where N N matrices ha and X() are defined by

ha = Ea,a
1
11,
N
X() =
N1

ek Ck ,
C = E1,N +
k=0
N1

Ek+1,k .
k=1
The spectral parameter is related with x in Eq. (6.1) by

eN = 1 x N .
As usual the monodromy and the transfer matrix are, respectively, given by
T() =

Ln (),
t1 () = Tr T().
(6.6)
This transfer matrix generates the integrals of motion of the discrete N -reduced KP
equation.
The fundamental L operator for sN case is defined by [22]
Ln (; ) =
1
1
1

(N1)
(N2)
(1)
+ n
2 + n
(N 1) + n

n(2)
n(N1)
n(1)

(1)
(2)
(N1)
n n
n
(6.7)
597
This operator not only satisfies the YangBaxter equation (2.32), but fulfills the intertwining relations for the (dual) vertex operators [35];
(a)
(a)
(a)

(a)
e n + e en+1 Ln (; ) = Ln (; ) e en + en+1 ,
(6.8)
(a)
(a)
(a)

n(a)
+ e e n+1 Ln (; ) = Ln (; ) e e n + e n+1 ,
e
(6.9)
for a = 0, 1, . . . , N 1. We can easily see that the Q operator
Q() =

Ln (; ),
(6.10)
commute with the screening charges

(a) = 0,
Q(), Q(a) = Q(), Q
(6.11)
which proves that the Q operator generates the local integrals of motion for the discrete
N Toda field theory.
analogue of the affine s
As we have clarified for s3 case in preceding sections, the generating function of the
local integrals of motion for the affine Toda field theory becomes the Q operator for the
quantum Boussinesq equation. Thus it is natural to conjecture that the Q operator (6.10)
satisfies the N th order difference equation,
Q( + 2i )
N1

Nk
tk + 2 Nk
N i Q + 2 N i + +
2
N i
Q() = 0,
k=1
(6.12)
with some function (). Here the transfer matrix t1 () is as Eq. (6.6) and corresponds
to the vector representation. Other transfer matrices tk>1 () are for [1k ], and can
be constructed by the fusion method from the transfer matrix t1 () with a suitable
normalization. Especially, we have

N1

(a)
exp
ha n X() ,
tN1 () Tr
n
a=1
with X() = 1 e C1 . The duality of the Q operator also suggests the dual difference
equation,
Q( + 2i)
N1

Nk

uk + 2 Nk
N i Q + 2 N i + +
2
N i

Q() = 0,
k=1
(6.13)
where the transfer matrices uk () follows from tk () under a duality transform (4.1).
We have explicitly constructed the Baxter Q operator for the quantum discrete
3 ). As far as the author knows, the explicit form of
Boussinesq equation which has Uq (s
the Q operator has been known only for the quantum integrable systems associated to s2 ,
598
and our Q operator is the first example which solves the third order difference equation.
In our construction, the Q operator originates from the generating function of the local
3 Toda field theory, and we hope
integrals of motion for a discrete analogue of the affine s
this can be applied for other Lie algebras, to which some attempts have been made [36].
Important structure is that it has a kind of duality, and that the Q operator satisfies the dual
Baxter equation. It was shown [31] in the case of the discrete KdV equation that the dual
Baxter equations can be interpreted as the duality between homologies and cohomologies
of quantized affine hyper-elliptic Jacobian from the view point of the algebraic geometry.
It will be interesting to give such interpretation in our case. In such studies the separation
of variables (see Ref. [37] for review, and Ref. [38] for s3 case) will be useful.
Also in s2 case, the universal procedure to derive the Baxter equation was proposed [14], where the Q operator was constructed from the universal R matrix with the
infinite-dimensional q-oscillator representation. This may be true for sN case, but we are
not sure how to relate our Q operator (6.10) with the universal R matrix.
The Baxter equation for s2 appeared in studies of the second order ordinary differential
equations; the spectral determinants for the Schrdinger equation is identified with the
Q operator for the CFT [3941]. Such correspondence is also studied for the higher
order differential equation [42,43]. This ODE/IM correspondence seems to come from
a similarity between the Stokes multiplier and the Lax matrix, and our results will help
further investigations on these topics.
Acknowledgement
The author would like to thank Rei Inoue for stimulating discussions.
Appendix A. Quantum dilogarithm function

We review a q-deformation of the dilogarithm function (2.25);

dx
eix
.
() = exp
4 sh( x) sh(x) x
R+i0
This integral was introduced by Faddeev [27], and it corresponds to a non-compact

analogue of the q-exponential function,

(1)k wk

2n+1
Sq (w) =
(A.1)
1+q
w = exp
.
k(q k q k )
n=0
k=1
We list properties of the integral (2.25) in order [44].

Duality:

2 / () = .
(A.2)
599
Zero points:

1

= i (2m + 1) + (2n + 1) m, n Z0 .
zeros of ()
Inversion relation:
() () ()

1 2 2 + 2
+
.
= exp
2i 2
6
(A.3)
A reason why we use notation is that it is proportional to the Jacobi theta function
when we replace the integral () with Sq (w), Sq (w)Sq (w1 ).
Difference equations:
( + i )
1
=
.
( i ) 1 + e
(A.4)
The duality (A.2) gives another type of difference equation:

( + i)
1
=
.
( i) 1 + e
Pentagon relation [44,45]:
(A.5)
(p)
(q)
= (q)
(p + q)
(p),
(A.6)
where p and q are the Heisenberg operators satisfying a commutation relation,

[p,
q]
= 2i .
(A.7)
This identity is a key to apply the integral (2.25) to the quantization of the Teichmller
space [45] and to the construction of the invariant of 3-manifold [46,47].
Appendix B. Properties of the fundamental L operator
We study the property of the fundamental L operator (2.29).
8. Z3 symmetry
(1)
(2)
(0)
(1)
(2)
By construction the free fields n , n , and n = n n correspond to roots 1 ,

2 , and 0 . As the fundamental L operator Ln (; ) is defined as the intertwining operators
for these Z3 symmetric vertex operators (2.30), we can suppose the L operator itself has
the Z3 symmetry. This can be checked by recursive uses of the pentagon identity (A.6) as

Ln (; ) = n(2) n(1) 2 + n(1)

1 (2) (1)
n n
+ n(1) + n(2)

= + n(1) + n(2) n(2) 2 + n(2)
600

1 (1)

n(1)
n + n(2) n(2)

= n(1) + n(1) + n(2) 2 n(1) n(2)

n(2)

n(1) n(1) + n(2) .
As a result, we see that

; n(2), n(1)
; n(1), n(0) = L
Ln (; ) L

; n(0), n(2) .
=L
(B.1)
9. Unitarity
Recalling the fact that the function (A.3) is proportional to the Gaussian, we can
check that the fundamental L operator satisfies the unitarity condition,
Ln (; )Ln (; ) = e
1 2
2i
(B.1)
Appendix C. Proof of Eqs. (3.1), (3.2)

We sketch the outline of a proof of Eqs. (3.1), (3.2).
To prove Eq. (3.1), we first compute as follows;
1
L(1)

n (; )
Ln + 23 i ;

(1)

(1)
(2)
+ 23 i n n n 23 i
=
(1)
(1)
(2)
n + n
n

(1)

(1)
(2)
2 + n + 43 i
n + n + 23 i

(1)
(1)
(2)
2 + n 23 i 23 i n n

+ 23 i n(1) n(2)
1
=
(1)
(1)
1
(2)
n + n
1 + e2+ 3 i +n

(1) 1
n(1) + n(2) + 43 i
1
e 3 n 9 i

(1)
(2)
4
3 i n n
1
=
(1)
(1)
(2)
+ 23 i
2+ 31 i +n
e
+e
+ en +n +i
(1)

(2)
n + n + 43 i 1 n(1) 1 i
9
(1)
e3
(2)
n + n + 2i
(1) 1
(2)
1
1
= e 3 n + 3 n
.
(1)
(1)
(2)
+ 13 i
2+ 32 i +n
e
+e
+ en +n
(C.1)
601
This proves the first equality in Eq. (3.1). Remaining equalities follow from Eq. (2.27).
To prove Eq. (3.2), we shift parameters in Eq. (3.1);

1
; n(1) + 2 i , n(1) + n(2) 2 i
L

3
3
2
Ln 3 i ;

1
= e 3 i cn(1)e2 + cn(2) + cn(3)e .
(C.2)
Using Eq. (3.1) and above identity, we can prove the first equality in Eq. (3.2) as follows;
(2)
L(1)
n (; )e Ln (; )

(1)

(2)
(3)
= L(2)
n (; ) cn + cn e + cn e

2
(2)
= L(2)
cn
n (; ) e e
(1)
e
cn(1)e + cn(2)e + cn(3)
1
(2)
(3)
cn e + cn e + cn

; n(1) + 2 i , n(1) + n(2) 2 i
= e3 1 cn(2)L
3
3
1
(1)
cn e2 + cn(2) + cn(3) e

1

= e3 1 e 3 i cn(2)Ln 23 i ; .

Other equalities directly follow from the Z3 symmetry (B.1).
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, London, 1982.
V. Pasquier, M. Gaudin, J. Phys. A: Math. Gen. 25 (1992) 5243.
V.B. Kuznetsov, E.K. Sklyanin, J. Phys. A: Math. Gen. 31 (1998) 2241.
E.K. Sklyanin, in: J. Harnad, G. Sabidussi, P. Winternitz (Eds.), Integrable Systems: From
Classical to Quantum, CRM Proceedings and Lecture Notes, Vol. 26, AMS, Providence, 2000,
pp. 227250.
L.D. Faddeev, A.Yu. Volkov, in: A.I. Bobenko, R. Seiler (Eds.), Discrete Integrable Geometry
and Physics, Oxford Univ. Press, Oxford, 1998, pp. 301320.
R. Inoue, K. Hikami, J. Phys. Soc. Jpn. 67 (1998) 1163.
S. Kharchev, D. Lebedev, Lett. Math. Phys. 50 (2000) 53.
S. Derkachov, J. Phys. A: Math. Gen. 32 (1999) 5299.
G.P. Pronko, Commun. Math. Phys. 212 (2000) 687.
V.B. Kuznetsov, M. Salerno, E.K. Sklyanin, J. Phys. A: Math. Gen. 33 (2000) 171.
V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, Commun. Math. Phys. 190 (1997) 247.
A.Yu. Volkov, Lett. Math. Phys. 39 (1997) 313.
K. Hikami, R. Inoue, J. Phys. Soc. Jpn. 68 (1999) 376.
A. Antonov, B. Feigin, Phys. Lett. B 392 (1997) 115.
A. Antonov, Theor. Math. Phys. 113 (1997) 1520.
K. Hikami, R. Inoue, J. Phys. A: Math. Gen. 30 (1997) 6911.
A.A. Belov, K.D. Chaltikian, Phys. Lett. B 309 (1993) 268.
T. Eguchi, S.-K. Yang, Phys. Lett. B 224 (1989) 373.
B.A. Kupershmidt, P. Mathieu, Phys. Lett. B 227 (1989) 245.
R. Sasaki, I. Yamanaka, Commun. Math. Phys. 108 (1987) 691.
602
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
K. Hikami, Phys. Lett. B 443 (1998) 202.

K. Hikami, Nucl. Phys. B 505 (1997) 749.
P. Di Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Springer, New York, 1997.
P.P. Kulish, N.Yu. Reshetikhin, E.K. Sklyanin, Lett. Math. Phys. 5 (1981) 393.
P.P. Kulish, N.Yu. Reshetikhin, J. Sov. Math. 34 (1986) 1948.
K. Hikami, in: Physics and Combinatorics 1999, World Scientific, 2001.
L.D. Faddeev, Lett. Math. Phys. 34 (1995) 249.
K. Hikami, Chaos Solit. Fract. 9 (1998) 853.
B. Enriquez, B.L. Feigin, Theor. Math. Phys. 103 (1995) 738.
S.V. Kryukov, Theor. Math. Phys. 105 (1995) 1359.
F.A. Smirnov, J. Phys. A: Math. Gen. 33 (2000) 3385.
V.A. Fateev, S.L. Lukyanov, Int. J. Mod. Phys. A 7 (1992) 1325.
K. Hikami, R. Inoue, J. Phys. Soc. Jpn. 68 (1999) 776.
K. Hikami, Chaos Solit. Fract. 9 (1998) 1773.
K. Hikami, J. Phys. Soc. Jpn. 68 (1999) 55.
R. Inoue, K. Hikami, Nucl. Phys. B 581 (2000) 761.
E.K. Sklyanin, Prog. Theor. Phys. Suppl. 118 (1995) 35.
E.K. Sklyanin, Zap. Nauch. Semin. POMI 205 (1993) 166.
P. Dorey, R. Tateo, J. Phys. A: Math. Gen. 32 (1999) L419.
J. Suzuki, J. Phys. A: Math. Gen. 32 (1999) L183.
V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, J. Stat. Phys. 102 (2001) 567.
P. Dorey, C. Dunning, R. Tateo, J. Phys. A: Math. Gen. 33 (2000) 8427.
J. Suzuki, J. Phys. A: Math. Gen. 33 (2000) 3507.
L.D. Faddeev, R.M. Kashaev, A.Yu. Volkov, hep-th/0006156.
L. Chekhov, V.V. Fock, Theor. Math. Phys. 120 (1999) 1245.
R.M. Kashaev, Mod. Phys. Lett. A 10 (1995) 1409.
K. Hikami, RIMS Kkyroku 1172 (2000) 44.

Baxter operator for HofstadterHarper

Hamiltonians
O. Lipan
Division of Physics, Mathematics, and Astronomy, California Institute of Technology, Pasadena, CA 91125, USA
Received 11 July 2000; accepted 9 March 2001
Abstract
We find the Baxter Q-operator for Hamiltonians governing the dynamics of a charged particle on
a lattice in a magnetic field. 2001 Published by Elsevier Science B.V.
PACS: 71.23.Ft; 71.30.+h; 72.40.Hm
Keywords: Baxter Q-operator; HarperHofstadter Hamiltonian
1. Introduction
The problem of Bloch electrons on a lattice in a magnetic field can be studied by the
use of Quantum Inverse Scattering Method as was shown in [1,2]. The main result of
these papers is the relation between the Hofstadter Hamiltonian and the inhomogeneous
XXZ chain. The Bethe ansatz equation is formulated through the Baxter vector. However,
the Baxter operator was not taken into consideration. We aim in this paper to construct the
Baxter Q-operator for Hamiltonians which generalize the Hofstadter Hamiltonian. Besides
its intrisec importance as being a fundamental operator of the Quantum Inverse Scattering
Method, the Baxter Q-operator appeared as an important object in the Sklyanin method of
separation of variables [5].
2. Quantum variables
In this section we introduce the quantum variables, i.e., the magnetic translations, [1,7
10,13]. Consider an open chain of N sites. Two operators, Xn and Yn , are attached to each
site, n = 1, . . . , N , and satisfy the Weyl permutation relation
Xn Yn = q 2 Yn Xn ,
E-mail address: olipan@its.caltech.edu (O. Lipan).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 1 7 - 1
(2.1)
604
O. Lipan / Nuclear Physics B 604 [FS] (2001) 603615
with q being a complex number. Later in the paper q will be a primitive root of unity. The
variables from different sites commute between themselves
Xi Yj = Yj Xi ,
i = j.
(2.2)
The magnetic translations are defined as

wn = Xn Yn1 ,
n = 1, . . . , N.
(2.3)
They satisfy a Weyl commutation relation between adjacent neighbors

wn wn+1 = q 2 wn+1 wn ,
(2.4)
where n runs from 1 to N and it is considered (mod N + 1), so wN w1 = q 2 w1 wN . The w

variables commute for non-nearest neighbor sites
wm wn = wn wm ,
2 |m n| N 2.
(2.5)
We will have in mind that the variables w are represented as unitary matrices so wn =
wn1 , n = 1, . . . , N , where represents the complex conjugation. For an odd number of
sites, N = odd, the product
C1 = wN wN1 w1 ,
(2.6)
commutes with all ws. For an even number of sites, N = even, there are two such products
that commutes with all ws
C1 = wN1 wN3 w1 ,
(2.7)
C2 = wN wN2 w2 .
(2.8)
These operators are Casimir operators. For example, if N = 4,

C1 = w3 w1 ,
(2.9)
C2 = w4 w2 .
(2.10)
The Casimir operators can be treated like numbers with respect to the algebra generated by
wn , so for the case N = 4 here are only two independent operators, w1 and w2
w3 = C1 w11 ,
(2.11)
w4 = C2 w21 .
(2.12)
If q is a root of unity, q N = 1, then in addition to these two Casimir operators, the N th

powers of the ws are also Casimir operators
N
N
wn = wn wm
,
wm
(2.13)
for every m, n from the set {1, . . . , N}. This means that wnN can be treated like numbers
also. These numbers specify a point in the Brillouin zone.
605
3. The Hamiltonians
Our purpose here is to introduce the L-operator and to show how a whole class of
Hamiltonians can be generated from them. Important physical examples will close this
section.
The 2-by-2 matrix

qan Yn Xn bn Yn
L(An |Xn , Yn ) =
(3.1)
,
cn Xn
dn
is called L-operator [1]. This is the fundamental object in the quantum inverse scattering
method; it is the quantum version of the Lax operator. The operator L(An |Xn , Yn ) is a
2-by-2 matrix in the auxiliary space C2 (here C is the set of complex numbers). The matrix
elements of the L-operator are operators in the Hilbert space Hn associated with the lattice
point n, so L(An |Xn , Yn ) is an operator in Hn C2 . All the spaces Hn , n = 1, . . . , N , are
isomorphic.
The matrix

qan bn
An =
(3.2)
cn dn
depends on the complex parameters (an , bn , cn , dn ) and on another complex number
which is called the spectral parameter. Sometimes we will emphasize the -dependence by
writing An ().
The operators Ln are used to construct the monodromy matrix
(AN (), . . . , A1 ()) = LN (AN ()|XN , YN ) . . . L1 (A1 ()|X1 , Y1 ),
(3.3)
which is a product of N 2-by-2 matrices. The monodromy matrix is a 2-by-2 matrix with
elements that are functions of the whole chain quantum variables XN , . . . , X1 , YN , . . . , Y1 .
The trace of the monodromy matrix on C2 depends only on the variables w1 , . . . , wN and
generates the needed set of Hamiltonians
T () = TrC2 =
[N/2]
2k Hk .
(3.4)
k=0
Those Hamiltonians Hk , k = 0, . . . , [N/2], that are not constants, describe the motion
of charged particles on a lattice in a magnetic field. These Hamiltonians depend on the
parameters (an , bn , cn , dn ), n = 1, . . . , N. Different physical Hamiltonians for hopping
electrons on a lattice can be obtained by an appropriate choice of these parameters. Let
us consider some examples.
(i) The first interesting example appears for N = 3, studied in [1].
In this case there will be 12 parameters (an , bn , cn , dn ), n = 1, 2, 3, plus the spectral
parameter . The trace of the monodromy matrix is
T () = d1 d2 d3 + a1 a2 a3 qw3 w2 w1
+ 2 (e1 w1 + qf1 w3 w2 + e2 w2 + qf3 w1 w3 + e3 w3 + qf3 w2 w1 ),
(3.5)
606
where we make the notation

ei = ci di+1 bi+2 ,
(3.6)
fi = ai+1 ci+2 bi+3 ,
(3.7)
with i = 1, 2, 3 taken modulo 3. The coefficient of 0 is a constant, so it does not represent

a Hamiltonian. The Hamiltonian is the coefficient of 2 and it can be further simplified if
we make use of the Casimir operator w3 w2 w1 = C, (2.6), to eliminate the w3 variable
H = e1 w1 + qCf1 w11 + e2 w2 + qCf2 w21 + Ce3 w11 w21 + qf3 w2 w1 .
(3.8)
To simplify even more let us consider C = q 1 . The hermiticity condition requires ei =

fi , where means complex conjugation. Finally
H = e1 w1 + e1 w11 + e2 w2 + e2 w21 + q 1 e3 w11 w21 + qe3 w2 w1 .
(3.9)
This Hamiltonian describes an hopping electron in a triangular lattice. It has first been
studied by Claro and Wannier [3] and is a model for crystals with a coupling to 6 neighbors.
See [4] for details.
When e3 = f3 = 0 we obtain the HarperHofstadter Hamiltonian. For this case the
parameter matrices can be chosen as follows

0
q k
q
A1 =
(3.10)
,
A2 =
,
A3 =
,
0
k 1
1
and the resulting Hamiltonian is

H = w1 + w11 + k w2 + w21 .
(3.11)
The real number k represents the anisotropy of the system.

(ii) The next number of sites is N = 4. This time there are two Casimir operators, (2.9),
so only w1 and w2 will survive. Again, there is only one Hamiltonian, generated by the
second power of :
H = (e4 + f4 C2 )w1 + (e2 C1 + f2 C1 C2 )w11
+ (e1 + f1 C1 )w2 + (e3 C2 + f3 C1 C2 )w21 + g1 qw2 w1
+ g3 q 1 C1 C2 w11 w21 + g2 qC1 w11 w2 + g4 q 1 C2 w21 w1 ,
(3.12)
where
ei = bi ci+1 di+2 di+3 ,
(3.13)
fi = ai ai+1 ci+2 bi+3 ,
(3.14)
gi = ai ci+1 di+2 bi+3 ,
(3.15)
with i = 1, 2, 3, 4 taken modulo 4. This variables are not independent. For example,
a relation between them is
gi gi+1 = fi ei .
(3.16)
607
The Hamiltonian looks simple if we change the variables

1
w1 = C1 2 w1 ,
(3.17)
12
w2 = C2 w2 .
(3.18)
The constants C1 and C2 can be hidden inside the coefficients

1
e1 = C1 2 e1 ,
1
2
f1 = C1 f1 ,
gi
= gi ,
e2 = C2 2 ,
1
2
f2 = C2 ,
e3 = C1 2 e3 ,
1
2
f3 = C1 f3 ,
e4 = C2 2 e4 ,
1
2
f4 = C2 f4 ,
i = 1, 2, 3, 4.
(3.19)
(3.20)
(3.21)
12
Multiply the Hamiltonian (3.12) by (C1 C2 ) to obtain

H = e4 + f4 w1 + e2 + f2 w1 1 + e1 + f1 w2 + e3 + f3 w2 1
+ g1 q 1 w1 w2 + g3 qw2 1 w1 1 + g2 qw1 1 w2 + g4 q 1 w2 1 w1 .
(3.22)
The Hamiltonian describes the dynamics of an electron on a square lattice which has the
freedom to hop on the sides of the unit lattice as well as on its diagonals.
The Hofstadter Hamiltonian can be obtained as a reduction of (3.12) if we choose C1 =
1/2
1/2
C2 = 1 so (C1 = C2 = i) and ei = fi , i = 1, . . . , 4. For this it is sufficient to take the
following matrix of parameters

q
A1 = A3 =
(3.23)
,
1

q
.
A2 = A4 =
(3.24)
1
The result is

H = 2 q 1 w1 w2 + qw21 w11 2 qw11 w2 + q 1 w21 w1 .
(3.25)
The procedure employed in these two examples can be continued for an arbitrary number
of sites. Besides playing the role of generating hopping Hamiltonians, the L-operators
produce quantum conserved quantities for those Hamiltonians. The method of quantum
inverse scattering is used in the next section to obtain the conserved quantities.
4. Baxter operator
In this section we introduce the R-matrix subjected to the YangBaxter equation [11,
12]. Let us remember the two spaces which had already appeared in the previous section:
the auxiliary space C2 and the local quantum space Hn . The L-operator acts on Hn C2 .
In the following formulas it is better to use the symbol Lna for the L-operator, where n
stands for Hn and a for the auxiliary space C2 . There are two R-matrices in connection
with the L-operator.
608
(i) The first one, Ra1 a2 , acts on C2 C2 , i.e., on two copies of the auxiliary space. This
is a 4-by-4 matrix
0
0
0
q 1 q 1
q 1 ( 1 )
0
0
q q 1
,
Ra1 a2 =
(4.1)
1
1
q q
0
0
q( )
0
0
0
q 1 q 1
and does not depend on the parameters (an , bn , cn , dn ) from the L-operator. Between the
L-operator and the R-matrix R the following relation holds

Lna1 ()Lna2 () = Lna2 ()Lna1 ()Ra1 a2

.
Ra1 a2
(4.2)
The product Lna1 Lna2 is usually written in a tensor notation: Ln () Ln (). An important
consequence of the relation (4.2) is the fact that the Hamiltonians generated by the trace of
the monodromy matrix form a commutative set
[Hk , Hl ] = 0,
(4.3)
for each allowed value of k and l, see (3.4).

(ii) Now we introduce the second R-matrix which will generate an operator, Q(), which
depends nontrivially on and commutes with the Hamiltonians Hk . This second R-matrix
acts on Hn Hf . The index f corresponds to another quantum Hilbert space, Hf , called
the fundamental space. The fundamental space is isomorphic to the local quantum space,
Hf Hn . This R-matrix is called the fundamental R-matrix and it carries the indices n
and f. The commutation relation now reads like:

An
An
t
t
L(An |Xn , Yn )L(|Xf , Yf ) = L( |Xf , Yf )L(An |Xn , Yn )Rnf
.
Rnf
(4.4)
The L-operators in (4.4) are multiplied together by usual matrix multiplication. We will
write the tensor product Xn Yf simply Xn Yf , where indices make it clear on which space
each operator acts. This R-matrix depends on the parameters entering the two L-operators.
The An matrix is given by (3.2) and the matrix is

q
=
.
(4.5)
The matrices At , t are the transpose matrices of A and, respectively, . For the
fundamental R-operator we choose the following form

An
An
Rnf
(4.6)
= Pnf rn
|Xn Yn ,
where Pnf is the permutation operator in Hn Hf . For example,

Pnf Xf Yl = Xn Yl Pnf ,
(4.7)
for an arbitrary Hilbert space Hl on which Yl acts. Due to Eq. (4.4), the function
rn ( An |Xn Yn ) must be a solution to the following equation

dn + cn Xn Yf
rn An |qXn Yn
.
An
=
rn |q 1 Xn Yn
bn + an Xn Yf
609
(4.8)
The functional relation shows explicitly that the r-operator depends on the spectral
parameter and only through the ratio /. For the case when

q 1
=
,
(4.9)
1 1
we will simply write Rnf (An , ). This notation means that the spectral parameter in An
is . If it is not otherwise specified, from now on we will work with given by (4.9). We
introduce now the trace over the fundamental space Hf of the product along the chain of
the fundamental R-operators

Q(AN , . . . , A1 , ) = TrHf RNf (AN , ) . . . R1f (A1 , ) .
(4.10)
This is the Baxter Q-operator. Like T (), Q() depends only on variables wn , n =
1, . . . , N. The central property of the trace (4.10) is

Q(AN , . . . , A1 , )T (AN , . . . , A1 , ) = T AtN , . . . , At1 , Q(AN , . . . , A1 , ). (4.11)
From this relation we deduce the following commutation relation
t

Q AN , . . . , At1 , Q(AN , . . . , A1 , ), T (AN , . . . , A1 , ) = 0.
(4.12)
This commutation relation is valid for every , so we obtained the proposed operator.
If we choose symmetric matrices, An = Atn , n = 1, . . . , N , then the commutation relation
becomes simpler

Q(AN , . . . , A1 , ), T (AN , . . . , A1 , ) = 0.
(4.13)
Baxter operators commute between themselves for different spectral parameters

t
Q AN , . . . , At1 , Q(AN , . . . , A1 , ), Q AtN , . . . , At1 , Q(AN , . . . , A1 , ) = 0.
(4.14)
If An are symmetric matrices, the commutation gets simplified

Q(AN , . . . , A1 , ), Q(AN , . . . , A1 , ) = 0.
(4.15)
The commutation relations between the Baxter operators follow from the YangBaxter
relation for the fundamental R-matrix. Consider three quantum spaces Hi , i = 1, 2, 3. The
YangBaxter relation is

t
t

t
A
A
B
B
A
A
R12
(4.16)
R13
= R23
R13
R12
.
R23
B
C
Ct
C
Ct
Bt
As usual, for R13 ( CAt ), the indices 13 show that the operator acts nontrivially on the space
H1 H3 and as an identity on H2 . Choosing the three matrices of parameters A, B, C,
in different ways it is possible to generate nontrivial algebra, like TemperelyLieb for
example. It remains to solve the functional relation (4.8) and then to find the trace (4.10).
610
5. R-matrix
In order to find the fundamental R-matrix, we have to solve a functional equation of the
type
r(, qw)
d + cw
.
=
1
r(, q w) b + aw
(5.1)
The interesting physical applications are those for which the modulus of q is 1. For an odd
root of unity
q 2l+1 = 1
(5.2)
the solution to the functional equation is, [13]:

r(, w) = 1 +
l

2
q k Ck wk + Ck wk ,
(5.3)
k=1
where
Ck =
k

a cq 2(j 1)
,
d bq 2j
Ck =
j =1
k

d bq 2(j 1)
.
a cq 2j
(5.4)
j =1
The functional relation defines the r-operator up to a function of , so g()r(, w) is also

a solution. We will call the solution (5.3) the normalized solution because the constant term
is normalized to 1. There are two useful special cases when a = b = c = d = 1. If = 0
the r-operator becomes the theta-function
r(0, w) = (w) = 1 +
l
q k wk ,
(5.5)
k=l
and for
l
r(, w) = 1 (w) = 1 +
q k wk .
2
(5.6)
k=l
With the following notation

(a, c) =
l

a cq 2j ,
(5.7)
j =1
we can write the coefficients Cl in a compact form

Cl =
(a, qc)
,
(d, b)
Cl =
(d, qb)
.
(a, c)
(5.8)
The trace over the fundamental space Hf can be explicitly computed so that the Baxter
Q-operator can be expressed as
Q() = SrN (, wN ) r1 (, w1 ),
(5.9)
611
where S is the shift operator

Swn = wn+1 S
(5.10)
and the *-product rN (, wN ) r1 (, w1 ) is defined by

l
kN
q 2kN k1 q kN +k1 CkN wN
rN1 (, wN1 ) r2 (, w2 )Ck1 w1k1 .
2
(5.11)
kN ,k1 =l
The star product, see [13], is a usual product for all terms except the first and the last which
are connected via q 2kn k1 . This is a consequence of the trace. Another interesting fact is
that the shift operator S can be expressed in terms of ws variables. For an odd number of
sites, equivalently for an odd number of ws, we have
S = (wn ) (wN+n ) (wN+n+1 ),
(5.12)
so the number of terms in S is one more than the number of ws. The starting wn is arbitrary.
For an even number of sites
S = (wn ) (wN1 ).
(5.13)
This time the number of terms is one less than the number of ws. The periodicity condition,
q 2l+1 = 1, constrains the parameters a, b, c, d to lie on an algebraic curve of Fermat type.
Indeed, from
N

r(, q j +1 w)
=1
r(, q j 1 w)
(5.14)
j =1
we get
N cN wN + d N = a N wN + N bN .
(5.15)
Recall that wN are considered numbers, (2.13), which represent points in the Brillouin
zone. If (5.15) is to be valid for every , we are forced to consider only a few points in
the Brillouin zone, which is a drawback. We also need a family of commuting operators
Q(), so must be kept as a continuous parameter. How to accomodate these two opposed
requirements will be seen in the next section.
6. Gauge freedom
We use in this section the gauge invariance of the transfer matrix to solve the problem
raised at the end of the last section [1]. Consider the following gauge transformation
1

n+1
0
0
n

An
,
An =
(6.1)
0
n+1
0 n1
for n = 1, . . . , N and N+1 = 1 . The trace of the monodromy matrix is invariant with
respect to this gauge transformation, and so are the generated Hamiltonians (3.4). Instead,
612
the r-matrix is not gauge invariant. The coefficients (a, b, c, d) however are transformed,
and the functional equations, n = 1, . . . , N , become
r(, qwn )
dn n+1 n1 + cn n+1 n wn
=
.
1 1
1
r(, q 1 wn ) bn n+1
n + an n+1
n wn
(6.2)
The Fermat curves are also gauge dependent

N N
N N
n
nN + dnN N+1 nN = wnN anN n+1
n + N bnN n+1
n ,
wnN N cnN n+1
(6.3)
and it is this gauge dependence that solves the problem, because we have N equations for N + 1 variables , 1 , . . . , N . The goal would be to find a parametrisation
(p), 1 (p), . . . , N (p), with p a complex parameter, for the Fermat curves. Then p will
play the role of the spectral parameter. This task is difficult to accomplish. We will be
confined to consider only those points in the Brillouin zone for which is the spectral
parameter. This will happen if

bn dn cn1 dn1 N
.
wn2N =
(6.4)
an cn an1 bn1
These points are called rational points. For rational points the gauge parameters are given
by

an bn N
4N
n+1 =
(6.5)
.
cn d n
It is easy to check that the rational points (6.4) are gauge independent. For N = 4 the gauge
invariance becomes evident if we write (6.4) as

gn+2 en N
.
wn2N =
(6.6)
gn fn+2
For this example and for rational points we can find the connection between the two
Casimir constants and the parameters an , . . . , dn

e1 e3 2N
e2 e4 2N
C12N =
,
C22N =
.
(6.7)
f1 f3
f2 f4
For the rational points the coefficients of the solution to the functional equation (6.2)
become

k 1 bi ci 1/2 q 2(j 1)

ai di
(i)
k
Ck = i
(6.8)
,
1/2

1 abii dcii
q 2j
j =1
for k = 1, . . . , l. For negative values of k the formula is the same just wik goes into wik .
Here i is defined as that constant for which wiN = iN . We turn our attention now to the
Baxter operator and ask the following question: for what values of (, 1 , . . . , N ) does the
star product (5.11) become a normal product. This question is a consequence of the fact
that the star product is less manageable than the normal product. It is quite possible that in
some specific problems we can tune the parameters so the star product becomes a normal
613
product. In order for this to happen, it is sufficient for one of the r-operators of the star
product to become the constant 1. From (5.4) and (6.3) we get
ai bi
(6.9)
,
ci d i
so the ratio (ai bi )/(ci di ) must be independent of i. This is not a constrain because the
Hamiltonians (3.4) are not affected if bn and cn , n = 1, . . . , N , are simultaneously changed
with the same amount; the bn , cn , n = 1, . . . , N , can than be chosen to make the ratio
(ai bi )/(ci di ) independent of i.
4
2 = i+1
=
7. Baxter T Q relation
In this section we apply Baxters method to diagonalize the set of commuting
Hamiltonians describing hopping electrons in a magnetic field [6]. The Baxter T Q
relation which plays the key part is

Q(A N , . . . , A 1 , )T AN , . . . , A1 , q 1/2

= A()(wN w1 )l+1 Q A N , . . . , A 1 , q 1

+ B()(wN w1 )l Q A N , . . . , A 1 , q ,
(7.1)
where the two complex functions are
A() =
B() =
N

l
2 q 2(j 1)
1

an cn n+1
2
q l n an n+1
qcn n+1
n2
,
2
dn n+1
bn q 2j
n=1
j =1
N

q l n1
2
n=1
l
an dn 2 q 2 bn cn
1
an n+1
qcn n+1
j =1
n2
2
dn n+1
bn q 2(j 1)
2 q 2j
an cn n+1
(7.2)
(7.3)
The proof of (7.1) is similar to the proof of the Baxter equation in [13]. The Baxter T Q
relation drastically simplifies if we consider the following special case, which is sufficiently
general to incorporate many Hamiltonians
an = dn ,
b n = cn .
(7.4)
Usually it is required that the Baxter Q-operator to be polynomial in the spectral

parameter . This can be done by working not with the normalized solution (5.3), but
with the polynomial solution
rpoly = (a, c)r.
(7.5)
For the polynomial solution, the Baxter relation looks as follows

N

an cn (wN w1 )l+1 Q q 1
q Nl+1 Q()T q 1/2 =
n=1
N

an + qcn (wN w1 )l Q(q).
n+1
(7.6)
614
Here the parameters are the gauged ones

an = an n+1 n1 ,
cn
1 1
= cn n+1
n ,
bn = bn n+1 n ,
(7.7)
dn
(7.8)
For an even number of sites,
= dn n+1 n1 .
N = 2n , the product of
all ws can be reduced to
(wN w1 )l+1 = q 2(N1) C2 C1 ,
(7.9)
where C2 , C1 are the Casimir numbers (2.7). The Baxter relation is then
N

q Nl+1 Q()T q 1/2 =
an cn (C2 C1 )l+1 Q q 1
n=1
N

an + qcn (C2 C1 )l Q(q).
(7.10)
n+1
For an odd number of sites N = 2N + 1, the Baxter relation is

N

an cn C1l+1 Q q 1
q Nl+1 Q()T q 1/2 =
n=1
N

an + qcn C1l Q(q),
(7.11)
n+1
where C1 is the Casimir number (2.6). Once we have the Baxter T Q relation, we can
obtain the Bethe ansatz equations. We already explained that, for a generic point in the
Brillouin zone, we first have to find a parametrization of the curve (6.3) and then proceed
to Bethe ansatz equations. For the rational points, however, the parameter is and the
Bethe ansatz equations are

Q() =
(7.12)
1
,
k
k
C 2l+1
N

n=1
j qk
an k cn
=
,
an + qk cn
j q 1 k
(7.13)
where C is C2 C1 for the N = even and C1 for N = odd. We choose n = 1 as a solution of

(6.5). The proof of the Baxter T Q relation (7.1) goes along traditional lines [13]. Here we
only mention the properties of the r-operator (5.3) which are essential to prove the T Q
relation
1

w2
r(, w) = N()r q 1 , qw ,
qaw + b
(7.14)
w 2 (qcw + d)r(, w) = P ()r(q, qw),
(7.15)
with
N() = q l
q 1 a c
Cl ,
ad 2 bc
P () = q l
q 1 ad 2 qbc
Cl .
q 1 c
(7.16)
615
8. Conclusions
We showed how to construct the Baxter Q-operator and the T Q relation for
Hamiltonians that are generalizations of the Hofstadter Hamiltonian. It would be
interesting to use this operator from the perspective of the Sklyanin method of separation
of variables [5].
Acknowledgements
I express my gratitude to P.B. Wiegmann for stimulating and helpful discussions.
References
[1] L.D. Faddev, R.M Kashaev, Generalized Bethe-Ansatz equations for Hofstadter problem,
CMP 169 (1) (1995) 181191.
[2] P.B. Wiegmann, A.V. Zabrodin, Quantum group and magnetic translations. Bethe-Ansatz for
the AsbelHofstadter problem, Nucl. Phys. B 422 (3) (1994) 495514.
[3] F.H. Claro, G.H. Wannier, Phys. Rev. B 19 (1979) 6068.
[4] C. Kreft, Spectral analysis of Hofstadter-like models, Thesis, Berlin, 1995, D83.
[5] E.K. Sklyanin, Separation variables, Prog. Theor. Phys. Suppl. No. 118 (1995) 3560.
[6] R.J. Baxter, Exacly Solved Models in Statistical Mechanics, Academic Press, New York, 1982.
[7] V.V. Bazhanov, Yu.G. Stroganov, Chiral Potts model as a descendant of the six-vertex model,
J. Stat. Phys. 59 (1990) 799817.
[8] L.D. Faddeev, Discrete HeisenbergWeyl group and modular group, Lett. Math. Phys. 34
(1995) 249254, hep-th/9504111.
[9] L.D. Faddeev, A.Yu. Volkov, Abelian current algebra and the Virasoro algebra on the lattice,
Phys. Lett. B 315 (1993) 311318.
[10] L.D. Faddeev, A.Yu. Volkov, Hirota equation as an example of integrable symplectic map, Lett.
Math. Phys 32 (1994) 125136.
[11] A.Yu. Volkov, L.D. Faddeev, The quantum method of the inverse problem on a discrete space
time, Teor. Mat. Fiz. 92 (2) (1992) 837842.
[12] V.O. Tarasov, L.A Takhtadzhyan, L.D. Faddeev, Local Hamiltonians for integrable quantum
models on a lattice, Teor. Mat. Fiz. 57 (2) (1983) 163181.
[13] A.Yu. Volkov, Quantum lattice KdV equation, Lett. Math. Phys. 39 (1997) 313329.

CUMULATIVE AUTHOR INDEX B601B604
Ahn, C.
Akemann, G.
Alford, M.
Alonso-Alberca, N.
ALPHA Collaboration
lvarez, E.
Amoros, G.
Anastasiou, C.
Anastasiou, C.
Arutyunov, G.
Astier, P.
Autiero, D.
B601 (2001) 539

B601 (2001) 77
B602 (2001) 61
B602 (2001) 329
B603 (2001) 180
B603 (2001) 286
B602 (2001) 87
B601 (2001) 318
B601 (2001) 341
B602 (2001) 238
B601 (2001) 3
B601 (2001) 3
Bajnok, Z.
Baldisseri, A.
Baldo-Ceolin, M.
Banner, M.
Bassompierre, G.
Behrndt, K.
Bellini, M.
Benatti, F.
Benslama, K.
Besson, N.
Bhaseen, M.J.
Bialas, A.
Bialas, P.
Bijnens, J.
Bintruy, P.
Bird, I.
Blumenfeld, B.
Bobisut, F.
Boer, D.
Bouchez, J.
Boyd, S.
Braun, V.M.
Bueno, A.
Bunyatov, S.
Burda, Z.
Bytsko, A.G.
B601 (2001) 503

B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B601 (2001) 49
B604 (2001) 441
B602 (2001) 541
B601 (2001) 3
B601 (2001) 3
B604 (2001) 537
B603 (2001) 218
B603 (2001) 369
B602 (2001) 87
B604 (2001) 32
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B603 (2001) 195
B601 (2001) 3
B601 (2001) 3
B603 (2001) 69
B601 (2001) 3
B601 (2001) 3
B602 (2001) 399
B604 (2001) 455
Cachazo, F.
Camilleri, L.
B603 (2001)
B601 (2001)
3
3
0550-3213/2001 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 4 6 - 2
Cardini, A.
Castro-Alvaredo, O.A.
Cattaneo, P.W.
Cavasinni, V.
Cerdeo, D.G.
Cervera-Villanueva, A.
Chandrasekharan, S.
Chim, L.
Cognola, G.
Colangelo, G.
Collazuol, G.
Conforto, G.
Conta, C.
Contalbrigo, M.
Cousins, R.
Cox, J.
Cski, C.
Curio, G.
B601 (2001) 3
B604 (2001) 367
B601 (2001) 3
B601 (2001) 3
B603 (2001) 231
B601 (2001) 3
B602 (2001) 61
B601 (2001) 539
B602 (2001) 383
B603 (2001) 125
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B602 (2001) 61
B604 (2001) 312
B602 (2001) 172
Dalmazi, D.
Damgaard, P.H.
Daniels, D.
de Azcrraga, J.A.
De Foss, L.
Degaudenzi, H.
Deger, N.S.
Del Prete, T.
De Santo, A.
Dignan, T.
Di Lella, L.
Do Couto e Silva, E.
Domenech-Garret, J.L.
Dorey, P.
Dorey, P.
Dumarchez, J.
B601 (2001) 77
B601 (2001) 77
B601 (2001) 3
B604 (2001) 75
B603 (2001) 413
B601 (2001) 3
B604 (2001) 343
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B601 (2001) 395
B603 (2001) 581
B603 (2001) 582
B601 (2001) 3
Ellis, M.
Emig, T.
Erlich, J.
Evslin, J.
B601 (2001) 3
B604 (2001) 479
B604 (2001) 312
B602 (2001) 486
618
Fabris, J.C.
Fazio, T.
Feldman, G.J.
Feng, J.L.
Ferrari, R.
Ferrre, D.
Flaminio, V.
Floreanini, R.
Forte, S.
Fradkin, E.
Fraternali, M.
Frau, M.
Fring, A.
Frizzo, A.
Frolov, S.
B602 (2001) 644

B601 (2001) 3
B601 (2001) 3
B602 (2001) 307
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B602 (2001) 541
B602 (2001) 585
B601 (2001) 591
B601 (2001) 3
B602 (2001) 39
B604 (2001) 367
B604 (2001) 92
B602 (2001) 238
Gabrielli, E.
Gaillard, J.-M.
Gaillard, M.K.
Gangler, E.
Garousi, M.R.
Gasser, J.
Gehrmann, T.
Gehrmann, T.
Geiser, A.
Geppert, D.
Gherghetta, T.
Gibin, D.
Gilkey, P.B.
Glover, E.W.N.
Glover, E.W.N.
Gninenko, S.
Godley, A.
Gmez, C.
Gomez-Cadenas, J.-J.
Gorbunov, D.S.
Gosset, J.
Gling, C.
Gouanre, M.
Grant, A.
Graziani, G.
Greene, B.R.
Grojean, C.
Guglielmi, A.
Gukov, S.
B603 (2001) 231

B601 (2001) 3
B604 (2001) 32
B601 (2001) 3
B602 (2001) 527
B603 (2001) 125
B601 (2001) 248
B601 (2001) 287
B601 (2001) 3
B601 (2001) 3
B602 (2001) 3
B601 (2001) 3
B601 (2001) 125
B601 (2001) 318
B601 (2001) 341
B601 (2001) 3
B601 (2001) 3
B603 (2001) 286
B601 (2001) 3
B602 (2001) 213
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B604 (2001) 181
B604 (2001) 312
B601 (2001) 3
B601 (2001) 49
Hagner, C.
Hjcek, P.
Hjcek, P.
Hambye, T.
Hara, T.
Hautmann, F.
Hernndez, L.
Hernando, J.
Hikami, K.
Hubbard, D.
B601 (2001) 3
B603 (2001) 531
B603 (2001) 555
B602 (2001) 23
B602 (2001) 499
B604 (2001) 391
B603 (2001) 286
B601 (2001) 3
B604 (2001) 580
B601 (2001) 3
Huerta, M.
Hurst, P.
Hyett, N.
B601 (2001) 591

B601 (2001) 3
B601 (2001) 3
Iacopini, E.
Ilha, A.
Intriligator, K.
Iso, S.
Iucci, A.
Izquierdo, J.M.
B601 (2001) 3
B604 (2001) 426
B603 (2001) 3
B604 (2001) 121
B601 (2001) 607
B604 (2001) 75
Joseph, C.
Juget, F.
Jurco, B.
B601 (2001) 3
B601 (2001) 3
B604 (2001) 148
Kamani, D.
Kardar, M.
Kaya, A.
Ketov, S.V.
Khalil, S.
Kiefer, C.
Kiem, Y.
Kim, J.E.
Kim, Y.
Kimura, Y.
Kirsanov, M.
Kirsten, K.
Klemm, D.
Klimov, O.
Koerber, P.
Kokkonen, J.
Korchemsky, G.P.
Kovzelev, A.
Krasnoperov, A.
Krause, A.
Kuznetsov, V.
B601 (2001) 149

B604 (2001) 479
B604 (2001) 343
B604 (2001) 256
B603 (2001) 231
B603 (2001) 531
B601 (2001) 27
B602 (2001) 346
B602 (2001) 467
B604 (2001) 121
B601 (2001) 3
B601 (2001) 125
B601 (2001) 380
B601 (2001) 3
B603 (2001) 413
B601 (2001) 3
B603 (2001) 69
B601 (2001) 3
B601 (2001) 3
B602 (2001) 172
B601 (2001) 3
Lacaprara, S.
Lachaud, C.
Lakic, B.
Lanza, A.
LaRotonda, L.
Laveder, M.
Lazaroiu, C.I.
Lazaroiu, C.I.
Leo, C.R.
Lee, H.M.
Letessier-Selvon, A.
Leutwyler, H.
Levy, J.-M.
Li, K.
Li, M.
Liccardo, A.
Ling, Y.
Linssen, L.
Lipan, O.
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B603 (2001) 497
B604 (2001) 181
B602 (2001) 514
B602 (2001) 346
B601 (2001) 3
B603 (2001) 125
B601 (2001) 3
B601 (2001) 607
B602 (2001) 201
B602 (2001) 39
B601 (2001) 191
B601 (2001) 3
B604 (2001) 603

Ljubicic, A.
Long, H.N.
Long, J.
Lupi, A.
B601 (2001) 3
B601 (2001) 361
B601 (2001) 3
B601 (2001) 3
Ma, E.
Ma, J.P.
Macfarlane, A.J.
Magnea, L.
Manashov, A.N.
Mangano, M.L.
March-Russell, J.
Marchionni, A.
Martelli, F.
Martn-Delgado, M.A.
McInnes, B.
Mchain, X.
Meessen, P.
Meggiolaro, E.
Mendiburu, J.-P.
Merlatti, P.
Meyer, J.-P.
Mezzetto, M.
Mishra, S.R.
Molke, H.
Moon, S.-H.
Moorhead, G.F.
Moreau, G.
Morel, A.
Mueller, A.H.
Munier, S.
Muoz, C.
Musto, R.
B602 (2001) 23
B602 (2001) 572
B604 (2001) 75
B604 (2001) 92
B603 (2001) 69
B602 (2001) 585
B602 (2001) 307
B601 (2001) 3
B601 (2001) 3
B601 (2001) 569
B602 (2001) 132
B601 (2001) 3
B602 (2001) 329
B602 (2001) 261
B601 (2001) 3
B602 (2001) 453
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B603 (2001) 180
B602 (2001) 467
B601 (2001) 3
B604 (2001) 3
B603 (2001) 369
B603 (2001) 427
B603 (2001) 427
B603 (2001) 231
B602 (2001) 39
Nagao, T.
Nan, C.M.
Naumov, D.
Navelet, H.
Ndlec, P.
Nefedov, Yu.
Nelson, B.D.
Nguyen-Mau, C.
Nielsen, H.B.
NOMAD Collaboration
Norrbin, E.
B602 (2001) 622

B601 (2001) 607
B601 (2001) 3
B603 (2001) 218
B601 (2001) 3
B601 (2001) 3
B604 (2001) 32
B601 (2001) 3
B604 (2001) 405
B601 (2001) 3
B603 (2001) 297
Oleari, C.
Oleari, C.
Oller, J.A.
Orestano, D.
Ortn, T.
B601 (2001) 318

B601 (2001) 341
B602 (2001) 641
B601 (2001) 3
B602 (2001) 329
Palla, L.
Park, D.H.
Pastore, F.
Peak, L.S.
B601 (2001) 503

B601 (2001) 27
B601 (2001) 3
B601 (2001) 3
619
Pearce, P.A.
Pelinson, A.M.
Pennacchio, E.
Perez, E.
Peschanski, R.
Pessard, H.
Petersson, B.
Petersson, B.
Petkou, A.C.
Petkou, A.C.
Petkova, V.B.
Petriello, F.J.
Petrov, K.
Petti, R.
Placci, A.
Polesello, G.
Polesello, G.
Pollmann, D.
Polyarush, A.
Pomarol, A.
Popov, B.
Poulsen, C.
B601 (2001) 539

B602 (2001) 644
B601 (2001) 3
B604 (2001) 3
B603 (2001) 218
B601 (2001) 3
B602 (2001) 399
B603 (2001) 369
B601 (2001) 380
B602 (2001) 238
B603 (2001) 449
B601 (2001) 169
B603 (2001) 369
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B604 (2001) 3
B601 (2001) 3
B601 (2001) 3
B602 (2001) 3
B601 (2001) 3
B601 (2001) 3
Rathouit, P.
Reisz, T.
Remiddi, E.
Remiddi, E.
Resco, P.
Rtey, A.
Rey, S.-J.
Rico, J.
Ridolfi, G.
Riva, V.
Rivelles, V.O.
Roda, C.
Rodriguez-Laguna, J.
Romano, R.
Rubbia, A.
Russo, J.G.
Russo, R.
B601 (2001) 3
B603 (2001) 369
B601 (2001) 248
B601 (2001) 287
B603 (2001) 286
B604 (2001) 281
B602 (2001) 467
B601 (2001) 3
B602 (2001) 585
B604 (2001) 511
B602 (2001) 514
B601 (2001) 3
B601 (2001) 569
B602 (2001) 541
B601 (2001) 3
B602 (2001) 109
B604 (2001) 92
Sabella, G.
Sakai, N.
Salvatore, F.
Sanchis-Lozano, M.A.
Sarkar, U.
Sato, H.-T.
Schahmaneche, K.
Schmidt, B.
Schupp, P.
Sekino, Y.
Sethi, S.
Sevior, M.
Sevrin, A.
Sevrin, A.
Sezgin, E.
B602 (2001) 453

B602 (2001) 413
B601 (2001) 3
B601 (2001) 395
B602 (2001) 23
B601 (2001) 27
B601 (2001) 3
B601 (2001) 3
B604 (2001) 148
B602 (2001) 147
B602 (2001) 307
B601 (2001) 3
B603 (2001) 389
B603 (2001) 413
B604 (2001) 343
620
Shapiro, I.L.
Shmakova, M.
Sierra, G.
Sillou, D.
Siopsis, G.
Sjstrand, T.
Smolin, L.
Smolin, L.
Soa, D.V.
Sokal, A.D.
Soler, F.J.P.
Sozzi, G.
Starinets, A.O.
Stasto, A.M.
Steele, D.
Stiegler, U.
Stipcevic, M.
Stolarczyk, Th.
Sundell, P.
B602 (2001) 644

B601 (2001) 49
B601 (2001) 569
B601 (2001) 3
B601 (2001) 380
B603 (2001) 297
B601 (2001) 191
B601 (2001) 209
B601 (2001) 361
B601 (2001) 425
B601 (2001) 3
B601 (2001) 3
B601 (2001) 425
B603 (2001) 427
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B604 (2001) 343
Tabaczek, J.
Takcs, G.
Takanishi, Y.
Takayanagi, T.
Talavera, P.
Tanaka, K.
Tani, T.
Tanii, Y.
Tareb-Reyes, M.
Tateo, R.
Tateo, R.
Taylor, G.N.
Tejeda-Yeomans, M.E.
Tejeda-Yeomans, M.E.
Tereshchenko, V.
Theis, U.
Tomizawa, S.
Toropin, A.
Torrente-Lujan, E.
Toublan, D.
Touchard, A.-M.
Tovey, S.N.
Tran, M.-T.
Troost, J.
Troost, W.
Tsesmelis, E.
B602 (2001) 399

B601 (2001) 503
B604 (2001) 405
B603 (2001) 259
B602 (2001) 87
B604 (2001) 121
B602 (2001) 434
B604 (2001) 343
B601 (2001) 3
B603 (2001) 581
B603 (2001) 582
B601 (2001) 3
B601 (2001) 318
B601 (2001) 341
B601 (2001) 3
B602 (2001) 367
B602 (2001) 413
B601 (2001) 3
B603 (2001) 231
B603 (2001) 343
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B603 (2001) 389
B603 (2001) 389
B601 (2001) 3
Ulrichs, J.
B601 (2001)
Vacavant, L.
Vafa, C.
Valdata-Nappi, M.
Valuev, V.
Van Neerven, W.L.
Vannucci, F.
Varadarajan, U.
Varvell, K.E.
Vassilevich, D.V.
Veltri, M.
Verbaarschot, J.J.M.
Verbaarschot, J.J.M.
Vercesi, V.
Vermaseren, J.A.M.
Vidal-Sitjes, G.
Vieira, J.-M.
Vinogradova, T.
Vogt, A.
B601 (2001) 3
B603 (2001) 3
B601 (2001) 3
B601 (2001) 3
B603 (2001) 42
B601 (2001) 3
B602 (2001) 486
B601 (2001) 3
B601 (2001) 125
B601 (2001) 3
B601 (2001) 77
B603 (2001) 343
B601 (2001) 3
B604 (2001) 281
B601 (2001) 3
B601 (2001) 3
B601 (2001) 3
B603 (2001) 42
Wgner, F.
Wakatsuki, K.
Wang, J.E.
Weber, F.V.
Weisse, T.
Wess, J.
Wiese, U.-J.
Wilczek, F.
Wilson, F.F.
Winton, L.J.
Wolff, U.
Wotzasek, C.
Wu, Y.-S.
B601 (2001) 503

B604 (2001) 121
B602 (2001) 486
B601 (2001) 3
B601 (2001) 3
B604 (2001) 148
B602 (2001) 61
B602 (2001) 307
B601 (2001) 3
B601 (2001) 3
B603 (2001) 180
B604 (2001) 426
B604 (2001) 551
Xiong, Z.
B602 (2001) 289
Yabsley, B.D.
Yang, H.-X.
Yang, J.M.
Yoneya, T.
Yu, Y.
B601 (2001) 3
B604 (2001) 551
B602 (2001) 289
B602 (2001) 499
B604 (2001) 551
Zaccone, H.
Zemba, G.R.
Zerbini, S.
Zuber, J.-B.
Zuber, K.
Zuccon, P.
B601 (2001) 3
B601 (2001) 591
B602 (2001) 383
B603 (2001) 449
B601 (2001) 3
B601 (2001) 3

Nucl - Phys.B v.604

Uploaded by

Copyright:

Available Formats

You might also like

Nucl - Phys.B v.604

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nucl - Phys.B v.604

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 604 (2001) 331

Resonant sneutrino production in supersymmetry

Received 7 March 2001; accepted 12 April 2001

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

Fig. 2. Cross-sections for the

1 production as a function of the parameter through various

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

LEP results at s = 196 GeV [32].

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

main kinematically allowed decays of the lightest chargino are 1 10 lp p and 1

1 10 dp up (with branching fractions of 3 11% and 2 33%, respectively)

mW + m 0 , mH + m 0 , ml, m > m > mqL,R , the dominant chargino decay is 1

10 dp up since it takes place via the two-body decay 1 qL,R qp .

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

coupling value is exactly the opposite of the W3 L (W3 u L u) coupling value

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

3.2. Standard Model background

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

4. Analysis reach in various models

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

4.2. Analysis reach in the MSSM

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

excluded by the preliminary LEP results at s = 196 GeV [32].

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

4.3. Analysis reach in mSUGRA

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

1 or l by LEP running at s = 200 GeV.

was performed for this model, corresponding to the parameter values:

m1/2 = 220 GeV,

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

G. Moreau et al. / Nuclear Physics B 604 (2001) 331

[24] N. Ghodbane, S. Katsanevas, P. Morawitz, E. Perez, SUSYGEN 3.0/06, lyoinfo.in2p3.fr

Nuclear Physics B 604 (2001) 3274

One loop soft supersymmetry breaking terms

50A-5101, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA

P. Bintruy et al. / Nuclear Physics B 604 (2001) 3274

2. General form of one loop supersymmetry breaking terms

1 We keep only the terms of leading order in m

P. Bintruy et al. / Nuclear Physics B 604 (2001) 3274

As a sign of spontaneous breaking of supersymmetry, the gravitino

P. Bintruy et al. / Nuclear Physics B 604 (2001) 3274

if supersymmetry breaking is moduli mediated (F S = 0) with the moduli stabilized at